2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
40 #include <linux/sort.h>
41 #include <linux/pfn.h>
42 #include <linux/backing-dev.h>
44 #include <asm/tlbflush.h>
45 #include <asm/div64.h>
49 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
52 nodemask_t node_online_map __read_mostly
= { { [0] = 1UL } };
53 EXPORT_SYMBOL(node_online_map
);
54 nodemask_t node_possible_map __read_mostly
= NODE_MASK_ALL
;
55 EXPORT_SYMBOL(node_possible_map
);
56 unsigned long totalram_pages __read_mostly
;
57 unsigned long totalreserve_pages __read_mostly
;
59 int percpu_pagelist_fraction
;
61 static void __free_pages_ok(struct page
*page
, unsigned int order
);
64 * results with 256, 32 in the lowmem_reserve sysctl:
65 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
66 * 1G machine -> (16M dma, 784M normal, 224M high)
67 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
68 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
69 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
71 * TBD: should special case ZONE_DMA32 machines here - in those we normally
72 * don't need any ZONE_NORMAL reservation
74 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
76 #ifdef CONFIG_ZONE_DMA32
84 EXPORT_SYMBOL(totalram_pages
);
87 * Used by page_zone() to look up the address of the struct zone whose
88 * id is encoded in the upper bits of page->flags
90 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
91 EXPORT_SYMBOL(zone_table
);
93 static char *zone_names
[MAX_NR_ZONES
] = {
95 #ifdef CONFIG_ZONE_DMA32
104 int min_free_kbytes
= 1024;
106 unsigned long __meminitdata nr_kernel_pages
;
107 unsigned long __meminitdata nr_all_pages
;
108 static unsigned long __initdata dma_reserve
;
110 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
112 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
113 * ranges of memory (RAM) that may be registered with add_active_range().
114 * Ranges passed to add_active_range() will be merged if possible
115 * so the number of times add_active_range() can be called is
116 * related to the number of nodes and the number of holes
118 #ifdef CONFIG_MAX_ACTIVE_REGIONS
119 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
120 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
122 #if MAX_NUMNODES >= 32
123 /* If there can be many nodes, allow up to 50 holes per node */
124 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
126 /* By default, allow up to 256 distinct regions */
127 #define MAX_ACTIVE_REGIONS 256
131 struct node_active_region __initdata early_node_map
[MAX_ACTIVE_REGIONS
];
132 int __initdata nr_nodemap_entries
;
133 unsigned long __initdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
134 unsigned long __initdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
135 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
136 unsigned long __initdata node_boundary_start_pfn
[MAX_NUMNODES
];
137 unsigned long __initdata node_boundary_end_pfn
[MAX_NUMNODES
];
138 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
139 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
141 #ifdef CONFIG_DEBUG_VM
142 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
146 unsigned long pfn
= page_to_pfn(page
);
149 seq
= zone_span_seqbegin(zone
);
150 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
152 else if (pfn
< zone
->zone_start_pfn
)
154 } while (zone_span_seqretry(zone
, seq
));
159 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
161 #ifdef CONFIG_HOLES_IN_ZONE
162 if (!pfn_valid(page_to_pfn(page
)))
165 if (zone
!= page_zone(page
))
171 * Temporary debugging check for pages not lying within a given zone.
173 static int bad_range(struct zone
*zone
, struct page
*page
)
175 if (page_outside_zone_boundaries(zone
, page
))
177 if (!page_is_consistent(zone
, page
))
183 static inline int bad_range(struct zone
*zone
, struct page
*page
)
189 static void bad_page(struct page
*page
)
191 printk(KERN_EMERG
"Bad page state in process '%s'\n"
192 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
193 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
194 KERN_EMERG
"Backtrace:\n",
195 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
196 (unsigned long)page
->flags
, page
->mapping
,
197 page_mapcount(page
), page_count(page
));
199 page
->flags
&= ~(1 << PG_lru
|
209 set_page_count(page
, 0);
210 reset_page_mapcount(page
);
211 page
->mapping
= NULL
;
212 add_taint(TAINT_BAD_PAGE
);
216 * Higher-order pages are called "compound pages". They are structured thusly:
218 * The first PAGE_SIZE page is called the "head page".
220 * The remaining PAGE_SIZE pages are called "tail pages".
222 * All pages have PG_compound set. All pages have their ->private pointing at
223 * the head page (even the head page has this).
225 * The first tail page's ->lru.next holds the address of the compound page's
226 * put_page() function. Its ->lru.prev holds the order of allocation.
227 * This usage means that zero-order pages may not be compound.
230 static void free_compound_page(struct page
*page
)
232 __free_pages_ok(page
, (unsigned long)page
[1].lru
.prev
);
235 static void prep_compound_page(struct page
*page
, unsigned long order
)
238 int nr_pages
= 1 << order
;
240 page
[1].lru
.next
= (void *)free_compound_page
; /* set dtor */
241 page
[1].lru
.prev
= (void *)order
;
242 for (i
= 0; i
< nr_pages
; i
++) {
243 struct page
*p
= page
+ i
;
245 __SetPageCompound(p
);
246 set_page_private(p
, (unsigned long)page
);
250 static void destroy_compound_page(struct page
*page
, unsigned long order
)
253 int nr_pages
= 1 << order
;
255 if (unlikely((unsigned long)page
[1].lru
.prev
!= order
))
258 for (i
= 0; i
< nr_pages
; i
++) {
259 struct page
*p
= page
+ i
;
261 if (unlikely(!PageCompound(p
) |
262 (page_private(p
) != (unsigned long)page
)))
264 __ClearPageCompound(p
);
268 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
272 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
274 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
275 * and __GFP_HIGHMEM from hard or soft interrupt context.
277 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
278 for (i
= 0; i
< (1 << order
); i
++)
279 clear_highpage(page
+ i
);
283 * function for dealing with page's order in buddy system.
284 * zone->lock is already acquired when we use these.
285 * So, we don't need atomic page->flags operations here.
287 static inline unsigned long page_order(struct page
*page
)
289 return page_private(page
);
292 static inline void set_page_order(struct page
*page
, int order
)
294 set_page_private(page
, order
);
295 __SetPageBuddy(page
);
298 static inline void rmv_page_order(struct page
*page
)
300 __ClearPageBuddy(page
);
301 set_page_private(page
, 0);
305 * Locate the struct page for both the matching buddy in our
306 * pair (buddy1) and the combined O(n+1) page they form (page).
308 * 1) Any buddy B1 will have an order O twin B2 which satisfies
309 * the following equation:
311 * For example, if the starting buddy (buddy2) is #8 its order
313 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
315 * 2) Any buddy B will have an order O+1 parent P which
316 * satisfies the following equation:
319 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
321 static inline struct page
*
322 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
324 unsigned long buddy_idx
= page_idx
^ (1 << order
);
326 return page
+ (buddy_idx
- page_idx
);
329 static inline unsigned long
330 __find_combined_index(unsigned long page_idx
, unsigned int order
)
332 return (page_idx
& ~(1 << order
));
336 * This function checks whether a page is free && is the buddy
337 * we can do coalesce a page and its buddy if
338 * (a) the buddy is not in a hole &&
339 * (b) the buddy is in the buddy system &&
340 * (c) a page and its buddy have the same order &&
341 * (d) a page and its buddy are in the same zone.
343 * For recording whether a page is in the buddy system, we use PG_buddy.
344 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
346 * For recording page's order, we use page_private(page).
348 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
351 #ifdef CONFIG_HOLES_IN_ZONE
352 if (!pfn_valid(page_to_pfn(buddy
)))
356 if (page_zone_id(page
) != page_zone_id(buddy
))
359 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
360 BUG_ON(page_count(buddy
) != 0);
367 * Freeing function for a buddy system allocator.
369 * The concept of a buddy system is to maintain direct-mapped table
370 * (containing bit values) for memory blocks of various "orders".
371 * The bottom level table contains the map for the smallest allocatable
372 * units of memory (here, pages), and each level above it describes
373 * pairs of units from the levels below, hence, "buddies".
374 * At a high level, all that happens here is marking the table entry
375 * at the bottom level available, and propagating the changes upward
376 * as necessary, plus some accounting needed to play nicely with other
377 * parts of the VM system.
378 * At each level, we keep a list of pages, which are heads of continuous
379 * free pages of length of (1 << order) and marked with PG_buddy. Page's
380 * order is recorded in page_private(page) field.
381 * So when we are allocating or freeing one, we can derive the state of the
382 * other. That is, if we allocate a small block, and both were
383 * free, the remainder of the region must be split into blocks.
384 * If a block is freed, and its buddy is also free, then this
385 * triggers coalescing into a block of larger size.
390 static inline void __free_one_page(struct page
*page
,
391 struct zone
*zone
, unsigned int order
)
393 unsigned long page_idx
;
394 int order_size
= 1 << order
;
396 if (unlikely(PageCompound(page
)))
397 destroy_compound_page(page
, order
);
399 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
401 VM_BUG_ON(page_idx
& (order_size
- 1));
402 VM_BUG_ON(bad_range(zone
, page
));
404 zone
->free_pages
+= order_size
;
405 while (order
< MAX_ORDER
-1) {
406 unsigned long combined_idx
;
407 struct free_area
*area
;
410 buddy
= __page_find_buddy(page
, page_idx
, order
);
411 if (!page_is_buddy(page
, buddy
, order
))
412 break; /* Move the buddy up one level. */
414 list_del(&buddy
->lru
);
415 area
= zone
->free_area
+ order
;
417 rmv_page_order(buddy
);
418 combined_idx
= __find_combined_index(page_idx
, order
);
419 page
= page
+ (combined_idx
- page_idx
);
420 page_idx
= combined_idx
;
423 set_page_order(page
, order
);
424 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
425 zone
->free_area
[order
].nr_free
++;
428 static inline int free_pages_check(struct page
*page
)
430 if (unlikely(page_mapcount(page
) |
431 (page
->mapping
!= NULL
) |
432 (page_count(page
) != 0) |
446 __ClearPageDirty(page
);
448 * For now, we report if PG_reserved was found set, but do not
449 * clear it, and do not free the page. But we shall soon need
450 * to do more, for when the ZERO_PAGE count wraps negative.
452 return PageReserved(page
);
456 * Frees a list of pages.
457 * Assumes all pages on list are in same zone, and of same order.
458 * count is the number of pages to free.
460 * If the zone was previously in an "all pages pinned" state then look to
461 * see if this freeing clears that state.
463 * And clear the zone's pages_scanned counter, to hold off the "all pages are
464 * pinned" detection logic.
466 static void free_pages_bulk(struct zone
*zone
, int count
,
467 struct list_head
*list
, int order
)
469 spin_lock(&zone
->lock
);
470 zone
->all_unreclaimable
= 0;
471 zone
->pages_scanned
= 0;
475 VM_BUG_ON(list_empty(list
));
476 page
= list_entry(list
->prev
, struct page
, lru
);
477 /* have to delete it as __free_one_page list manipulates */
478 list_del(&page
->lru
);
479 __free_one_page(page
, zone
, order
);
481 spin_unlock(&zone
->lock
);
484 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
486 spin_lock(&zone
->lock
);
487 zone
->all_unreclaimable
= 0;
488 zone
->pages_scanned
= 0;
489 __free_one_page(page
, zone
,order
);
490 spin_unlock(&zone
->lock
);
493 static void __free_pages_ok(struct page
*page
, unsigned int order
)
499 for (i
= 0 ; i
< (1 << order
) ; ++i
)
500 reserved
+= free_pages_check(page
+ i
);
504 if (!PageHighMem(page
))
505 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
506 arch_free_page(page
, order
);
507 kernel_map_pages(page
, 1 << order
, 0);
509 local_irq_save(flags
);
510 __count_vm_events(PGFREE
, 1 << order
);
511 free_one_page(page_zone(page
), page
, order
);
512 local_irq_restore(flags
);
516 * permit the bootmem allocator to evade page validation on high-order frees
518 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
521 __ClearPageReserved(page
);
522 set_page_count(page
, 0);
523 set_page_refcounted(page
);
529 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
530 struct page
*p
= &page
[loop
];
532 if (loop
+ 1 < BITS_PER_LONG
)
534 __ClearPageReserved(p
);
535 set_page_count(p
, 0);
538 set_page_refcounted(page
);
539 __free_pages(page
, order
);
545 * The order of subdivision here is critical for the IO subsystem.
546 * Please do not alter this order without good reasons and regression
547 * testing. Specifically, as large blocks of memory are subdivided,
548 * the order in which smaller blocks are delivered depends on the order
549 * they're subdivided in this function. This is the primary factor
550 * influencing the order in which pages are delivered to the IO
551 * subsystem according to empirical testing, and this is also justified
552 * by considering the behavior of a buddy system containing a single
553 * large block of memory acted on by a series of small allocations.
554 * This behavior is a critical factor in sglist merging's success.
558 static inline void expand(struct zone
*zone
, struct page
*page
,
559 int low
, int high
, struct free_area
*area
)
561 unsigned long size
= 1 << high
;
567 VM_BUG_ON(bad_range(zone
, &page
[size
]));
568 list_add(&page
[size
].lru
, &area
->free_list
);
570 set_page_order(&page
[size
], high
);
575 * This page is about to be returned from the page allocator
577 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
579 if (unlikely(page_mapcount(page
) |
580 (page
->mapping
!= NULL
) |
581 (page_count(page
) != 0) |
597 * For now, we report if PG_reserved was found set, but do not
598 * clear it, and do not allocate the page: as a safety net.
600 if (PageReserved(page
))
603 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
604 1 << PG_referenced
| 1 << PG_arch_1
|
605 1 << PG_checked
| 1 << PG_mappedtodisk
);
606 set_page_private(page
, 0);
607 set_page_refcounted(page
);
608 kernel_map_pages(page
, 1 << order
, 1);
610 if (gfp_flags
& __GFP_ZERO
)
611 prep_zero_page(page
, order
, gfp_flags
);
613 if (order
&& (gfp_flags
& __GFP_COMP
))
614 prep_compound_page(page
, order
);
620 * Do the hard work of removing an element from the buddy allocator.
621 * Call me with the zone->lock already held.
623 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
625 struct free_area
* area
;
626 unsigned int current_order
;
629 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
630 area
= zone
->free_area
+ current_order
;
631 if (list_empty(&area
->free_list
))
634 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
635 list_del(&page
->lru
);
636 rmv_page_order(page
);
638 zone
->free_pages
-= 1UL << order
;
639 expand(zone
, page
, order
, current_order
, area
);
647 * Obtain a specified number of elements from the buddy allocator, all under
648 * a single hold of the lock, for efficiency. Add them to the supplied list.
649 * Returns the number of new pages which were placed at *list.
651 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
652 unsigned long count
, struct list_head
*list
)
656 spin_lock(&zone
->lock
);
657 for (i
= 0; i
< count
; ++i
) {
658 struct page
*page
= __rmqueue(zone
, order
);
659 if (unlikely(page
== NULL
))
661 list_add_tail(&page
->lru
, list
);
663 spin_unlock(&zone
->lock
);
669 * Called from the slab reaper to drain pagesets on a particular node that
670 * belongs to the currently executing processor.
671 * Note that this function must be called with the thread pinned to
672 * a single processor.
674 void drain_node_pages(int nodeid
)
680 for (z
= 0; z
< MAX_NR_ZONES
; z
++) {
681 struct zone
*zone
= NODE_DATA(nodeid
)->node_zones
+ z
;
682 struct per_cpu_pageset
*pset
;
684 if (!populated_zone(zone
))
687 pset
= zone_pcp(zone
, smp_processor_id());
688 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
689 struct per_cpu_pages
*pcp
;
693 local_irq_save(flags
);
694 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
696 local_irq_restore(flags
);
703 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
704 static void __drain_pages(unsigned int cpu
)
710 for_each_zone(zone
) {
711 struct per_cpu_pageset
*pset
;
713 if (!populated_zone(zone
))
716 pset
= zone_pcp(zone
, cpu
);
717 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
718 struct per_cpu_pages
*pcp
;
721 local_irq_save(flags
);
722 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
724 local_irq_restore(flags
);
728 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
732 void mark_free_pages(struct zone
*zone
)
734 unsigned long pfn
, max_zone_pfn
;
737 struct list_head
*curr
;
739 if (!zone
->spanned_pages
)
742 spin_lock_irqsave(&zone
->lock
, flags
);
744 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
745 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
746 if (pfn_valid(pfn
)) {
747 struct page
*page
= pfn_to_page(pfn
);
749 if (!PageNosave(page
))
750 ClearPageNosaveFree(page
);
753 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
754 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
757 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
758 for (i
= 0; i
< (1UL << order
); i
++)
759 SetPageNosaveFree(pfn_to_page(pfn
+ i
));
762 spin_unlock_irqrestore(&zone
->lock
, flags
);
766 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
768 void drain_local_pages(void)
772 local_irq_save(flags
);
773 __drain_pages(smp_processor_id());
774 local_irq_restore(flags
);
776 #endif /* CONFIG_PM */
779 * Free a 0-order page
781 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
783 struct zone
*zone
= page_zone(page
);
784 struct per_cpu_pages
*pcp
;
788 page
->mapping
= NULL
;
789 if (free_pages_check(page
))
792 if (!PageHighMem(page
))
793 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
794 arch_free_page(page
, 0);
795 kernel_map_pages(page
, 1, 0);
797 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
798 local_irq_save(flags
);
799 __count_vm_event(PGFREE
);
800 list_add(&page
->lru
, &pcp
->list
);
802 if (pcp
->count
>= pcp
->high
) {
803 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
804 pcp
->count
-= pcp
->batch
;
806 local_irq_restore(flags
);
810 void fastcall
free_hot_page(struct page
*page
)
812 free_hot_cold_page(page
, 0);
815 void fastcall
free_cold_page(struct page
*page
)
817 free_hot_cold_page(page
, 1);
821 * split_page takes a non-compound higher-order page, and splits it into
822 * n (1<<order) sub-pages: page[0..n]
823 * Each sub-page must be freed individually.
825 * Note: this is probably too low level an operation for use in drivers.
826 * Please consult with lkml before using this in your driver.
828 void split_page(struct page
*page
, unsigned int order
)
832 VM_BUG_ON(PageCompound(page
));
833 VM_BUG_ON(!page_count(page
));
834 for (i
= 1; i
< (1 << order
); i
++)
835 set_page_refcounted(page
+ i
);
839 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
840 * we cheat by calling it from here, in the order > 0 path. Saves a branch
843 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
844 struct zone
*zone
, int order
, gfp_t gfp_flags
)
848 int cold
= !!(gfp_flags
& __GFP_COLD
);
853 if (likely(order
== 0)) {
854 struct per_cpu_pages
*pcp
;
856 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
857 local_irq_save(flags
);
859 pcp
->count
= rmqueue_bulk(zone
, 0,
860 pcp
->batch
, &pcp
->list
);
861 if (unlikely(!pcp
->count
))
864 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
865 list_del(&page
->lru
);
868 spin_lock_irqsave(&zone
->lock
, flags
);
869 page
= __rmqueue(zone
, order
);
870 spin_unlock(&zone
->lock
);
875 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
876 zone_statistics(zonelist
, zone
);
877 local_irq_restore(flags
);
880 VM_BUG_ON(bad_range(zone
, page
));
881 if (prep_new_page(page
, order
, gfp_flags
))
886 local_irq_restore(flags
);
891 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
892 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
893 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
894 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
895 #define ALLOC_HARDER 0x10 /* try to alloc harder */
896 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
897 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
900 * Return 1 if free pages are above 'mark'. This takes into account the order
903 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
904 int classzone_idx
, int alloc_flags
)
906 /* free_pages my go negative - that's OK */
907 unsigned long min
= mark
;
908 long free_pages
= z
->free_pages
- (1 << order
) + 1;
911 if (alloc_flags
& ALLOC_HIGH
)
913 if (alloc_flags
& ALLOC_HARDER
)
916 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
918 for (o
= 0; o
< order
; o
++) {
919 /* At the next order, this order's pages become unavailable */
920 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
922 /* Require fewer higher order pages to be free */
925 if (free_pages
<= min
)
932 * get_page_from_freeliest goes through the zonelist trying to allocate
936 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
937 struct zonelist
*zonelist
, int alloc_flags
)
939 struct zone
**z
= zonelist
->zones
;
940 struct page
*page
= NULL
;
941 int classzone_idx
= zone_idx(*z
);
945 * Go through the zonelist once, looking for a zone with enough free.
946 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
950 if (unlikely(NUMA_BUILD
&& (gfp_mask
& __GFP_THISNODE
) &&
951 zone
->zone_pgdat
!= zonelist
->zones
[0]->zone_pgdat
))
953 if ((alloc_flags
& ALLOC_CPUSET
) &&
954 !cpuset_zone_allowed(zone
, gfp_mask
))
957 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
959 if (alloc_flags
& ALLOC_WMARK_MIN
)
960 mark
= zone
->pages_min
;
961 else if (alloc_flags
& ALLOC_WMARK_LOW
)
962 mark
= zone
->pages_low
;
964 mark
= zone
->pages_high
;
965 if (!zone_watermark_ok(zone
, order
, mark
,
966 classzone_idx
, alloc_flags
))
967 if (!zone_reclaim_mode
||
968 !zone_reclaim(zone
, gfp_mask
, order
))
972 page
= buffered_rmqueue(zonelist
, zone
, order
, gfp_mask
);
976 } while (*(++z
) != NULL
);
981 * This is the 'heart' of the zoned buddy allocator.
983 struct page
* fastcall
984 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
985 struct zonelist
*zonelist
)
987 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
990 struct reclaim_state reclaim_state
;
991 struct task_struct
*p
= current
;
994 int did_some_progress
;
996 might_sleep_if(wait
);
999 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
1001 if (unlikely(*z
== NULL
)) {
1002 /* Should this ever happen?? */
1006 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1007 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1012 wakeup_kswapd(*z
, order
);
1016 * OK, we're below the kswapd watermark and have kicked background
1017 * reclaim. Now things get more complex, so set up alloc_flags according
1018 * to how we want to proceed.
1020 * The caller may dip into page reserves a bit more if the caller
1021 * cannot run direct reclaim, or if the caller has realtime scheduling
1022 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1023 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1025 alloc_flags
= ALLOC_WMARK_MIN
;
1026 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1027 alloc_flags
|= ALLOC_HARDER
;
1028 if (gfp_mask
& __GFP_HIGH
)
1029 alloc_flags
|= ALLOC_HIGH
;
1031 alloc_flags
|= ALLOC_CPUSET
;
1034 * Go through the zonelist again. Let __GFP_HIGH and allocations
1035 * coming from realtime tasks go deeper into reserves.
1037 * This is the last chance, in general, before the goto nopage.
1038 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1039 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1041 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1045 /* This allocation should allow future memory freeing. */
1047 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1048 && !in_interrupt()) {
1049 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1051 /* go through the zonelist yet again, ignoring mins */
1052 page
= get_page_from_freelist(gfp_mask
, order
,
1053 zonelist
, ALLOC_NO_WATERMARKS
);
1056 if (gfp_mask
& __GFP_NOFAIL
) {
1057 congestion_wait(WRITE
, HZ
/50);
1064 /* Atomic allocations - we can't balance anything */
1071 /* We now go into synchronous reclaim */
1072 cpuset_memory_pressure_bump();
1073 p
->flags
|= PF_MEMALLOC
;
1074 reclaim_state
.reclaimed_slab
= 0;
1075 p
->reclaim_state
= &reclaim_state
;
1077 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
1079 p
->reclaim_state
= NULL
;
1080 p
->flags
&= ~PF_MEMALLOC
;
1084 if (likely(did_some_progress
)) {
1085 page
= get_page_from_freelist(gfp_mask
, order
,
1086 zonelist
, alloc_flags
);
1089 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1091 * Go through the zonelist yet one more time, keep
1092 * very high watermark here, this is only to catch
1093 * a parallel oom killing, we must fail if we're still
1094 * under heavy pressure.
1096 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1097 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1101 out_of_memory(zonelist
, gfp_mask
, order
);
1106 * Don't let big-order allocations loop unless the caller explicitly
1107 * requests that. Wait for some write requests to complete then retry.
1109 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1110 * <= 3, but that may not be true in other implementations.
1113 if (!(gfp_mask
& __GFP_NORETRY
)) {
1114 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
1116 if (gfp_mask
& __GFP_NOFAIL
)
1120 congestion_wait(WRITE
, HZ
/50);
1125 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1126 printk(KERN_WARNING
"%s: page allocation failure."
1127 " order:%d, mode:0x%x\n",
1128 p
->comm
, order
, gfp_mask
);
1136 EXPORT_SYMBOL(__alloc_pages
);
1139 * Common helper functions.
1141 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1144 page
= alloc_pages(gfp_mask
, order
);
1147 return (unsigned long) page_address(page
);
1150 EXPORT_SYMBOL(__get_free_pages
);
1152 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1157 * get_zeroed_page() returns a 32-bit address, which cannot represent
1160 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1162 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1164 return (unsigned long) page_address(page
);
1168 EXPORT_SYMBOL(get_zeroed_page
);
1170 void __pagevec_free(struct pagevec
*pvec
)
1172 int i
= pagevec_count(pvec
);
1175 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1178 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1180 if (put_page_testzero(page
)) {
1182 free_hot_page(page
);
1184 __free_pages_ok(page
, order
);
1188 EXPORT_SYMBOL(__free_pages
);
1190 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1193 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1194 __free_pages(virt_to_page((void *)addr
), order
);
1198 EXPORT_SYMBOL(free_pages
);
1201 * Total amount of free (allocatable) RAM:
1203 unsigned int nr_free_pages(void)
1205 unsigned int sum
= 0;
1209 sum
+= zone
->free_pages
;
1214 EXPORT_SYMBOL(nr_free_pages
);
1217 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1219 unsigned int sum
= 0;
1222 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1223 sum
+= pgdat
->node_zones
[i
].free_pages
;
1229 static unsigned int nr_free_zone_pages(int offset
)
1231 /* Just pick one node, since fallback list is circular */
1232 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1233 unsigned int sum
= 0;
1235 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1236 struct zone
**zonep
= zonelist
->zones
;
1239 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1240 unsigned long size
= zone
->present_pages
;
1241 unsigned long high
= zone
->pages_high
;
1250 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1252 unsigned int nr_free_buffer_pages(void)
1254 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1258 * Amount of free RAM allocatable within all zones
1260 unsigned int nr_free_pagecache_pages(void)
1262 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1265 static inline void show_node(struct zone
*zone
)
1268 printk("Node %ld ", zone_to_nid(zone
));
1271 void si_meminfo(struct sysinfo
*val
)
1273 val
->totalram
= totalram_pages
;
1275 val
->freeram
= nr_free_pages();
1276 val
->bufferram
= nr_blockdev_pages();
1277 val
->totalhigh
= totalhigh_pages
;
1278 val
->freehigh
= nr_free_highpages();
1279 val
->mem_unit
= PAGE_SIZE
;
1282 EXPORT_SYMBOL(si_meminfo
);
1285 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1287 pg_data_t
*pgdat
= NODE_DATA(nid
);
1289 val
->totalram
= pgdat
->node_present_pages
;
1290 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1291 #ifdef CONFIG_HIGHMEM
1292 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1293 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1298 val
->mem_unit
= PAGE_SIZE
;
1302 #define K(x) ((x) << (PAGE_SHIFT-10))
1305 * Show free area list (used inside shift_scroll-lock stuff)
1306 * We also calculate the percentage fragmentation. We do this by counting the
1307 * memory on each free list with the exception of the first item on the list.
1309 void show_free_areas(void)
1312 unsigned long active
;
1313 unsigned long inactive
;
1317 for_each_zone(zone
) {
1318 if (!populated_zone(zone
))
1322 printk("%s per-cpu:\n", zone
->name
);
1324 for_each_online_cpu(cpu
) {
1325 struct per_cpu_pageset
*pageset
;
1327 pageset
= zone_pcp(zone
, cpu
);
1329 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1330 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1331 cpu
, pageset
->pcp
[0].high
,
1332 pageset
->pcp
[0].batch
, pageset
->pcp
[0].count
,
1333 pageset
->pcp
[1].high
, pageset
->pcp
[1].batch
,
1334 pageset
->pcp
[1].count
);
1338 get_zone_counts(&active
, &inactive
, &free
);
1340 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1341 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1344 global_page_state(NR_FILE_DIRTY
),
1345 global_page_state(NR_WRITEBACK
),
1346 global_page_state(NR_UNSTABLE_NFS
),
1348 global_page_state(NR_SLAB_RECLAIMABLE
) +
1349 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1350 global_page_state(NR_FILE_MAPPED
),
1351 global_page_state(NR_PAGETABLE
));
1353 for_each_zone(zone
) {
1356 if (!populated_zone(zone
))
1368 " pages_scanned:%lu"
1369 " all_unreclaimable? %s"
1372 K(zone
->free_pages
),
1375 K(zone
->pages_high
),
1377 K(zone
->nr_inactive
),
1378 K(zone
->present_pages
),
1379 zone
->pages_scanned
,
1380 (zone
->all_unreclaimable
? "yes" : "no")
1382 printk("lowmem_reserve[]:");
1383 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1384 printk(" %lu", zone
->lowmem_reserve
[i
]);
1388 for_each_zone(zone
) {
1389 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1391 if (!populated_zone(zone
))
1395 printk("%s: ", zone
->name
);
1397 spin_lock_irqsave(&zone
->lock
, flags
);
1398 for (order
= 0; order
< MAX_ORDER
; order
++) {
1399 nr
[order
] = zone
->free_area
[order
].nr_free
;
1400 total
+= nr
[order
] << order
;
1402 spin_unlock_irqrestore(&zone
->lock
, flags
);
1403 for (order
= 0; order
< MAX_ORDER
; order
++)
1404 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1405 printk("= %lukB\n", K(total
));
1408 show_swap_cache_info();
1412 * Builds allocation fallback zone lists.
1414 * Add all populated zones of a node to the zonelist.
1416 static int __meminit
build_zonelists_node(pg_data_t
*pgdat
,
1417 struct zonelist
*zonelist
, int nr_zones
, enum zone_type zone_type
)
1421 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1426 zone
= pgdat
->node_zones
+ zone_type
;
1427 if (populated_zone(zone
)) {
1428 zonelist
->zones
[nr_zones
++] = zone
;
1429 check_highest_zone(zone_type
);
1432 } while (zone_type
);
1437 #define MAX_NODE_LOAD (num_online_nodes())
1438 static int __meminitdata node_load
[MAX_NUMNODES
];
1440 * find_next_best_node - find the next node that should appear in a given node's fallback list
1441 * @node: node whose fallback list we're appending
1442 * @used_node_mask: nodemask_t of already used nodes
1444 * We use a number of factors to determine which is the next node that should
1445 * appear on a given node's fallback list. The node should not have appeared
1446 * already in @node's fallback list, and it should be the next closest node
1447 * according to the distance array (which contains arbitrary distance values
1448 * from each node to each node in the system), and should also prefer nodes
1449 * with no CPUs, since presumably they'll have very little allocation pressure
1450 * on them otherwise.
1451 * It returns -1 if no node is found.
1453 static int __meminit
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1456 int min_val
= INT_MAX
;
1459 /* Use the local node if we haven't already */
1460 if (!node_isset(node
, *used_node_mask
)) {
1461 node_set(node
, *used_node_mask
);
1465 for_each_online_node(n
) {
1468 /* Don't want a node to appear more than once */
1469 if (node_isset(n
, *used_node_mask
))
1472 /* Use the distance array to find the distance */
1473 val
= node_distance(node
, n
);
1475 /* Penalize nodes under us ("prefer the next node") */
1478 /* Give preference to headless and unused nodes */
1479 tmp
= node_to_cpumask(n
);
1480 if (!cpus_empty(tmp
))
1481 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1483 /* Slight preference for less loaded node */
1484 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1485 val
+= node_load
[n
];
1487 if (val
< min_val
) {
1494 node_set(best_node
, *used_node_mask
);
1499 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1501 int j
, node
, local_node
;
1503 int prev_node
, load
;
1504 struct zonelist
*zonelist
;
1505 nodemask_t used_mask
;
1507 /* initialize zonelists */
1508 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1509 zonelist
= pgdat
->node_zonelists
+ i
;
1510 zonelist
->zones
[0] = NULL
;
1513 /* NUMA-aware ordering of nodes */
1514 local_node
= pgdat
->node_id
;
1515 load
= num_online_nodes();
1516 prev_node
= local_node
;
1517 nodes_clear(used_mask
);
1518 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1519 int distance
= node_distance(local_node
, node
);
1522 * If another node is sufficiently far away then it is better
1523 * to reclaim pages in a zone before going off node.
1525 if (distance
> RECLAIM_DISTANCE
)
1526 zone_reclaim_mode
= 1;
1529 * We don't want to pressure a particular node.
1530 * So adding penalty to the first node in same
1531 * distance group to make it round-robin.
1534 if (distance
!= node_distance(local_node
, prev_node
))
1535 node_load
[node
] += load
;
1538 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1539 zonelist
= pgdat
->node_zonelists
+ i
;
1540 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1542 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1543 zonelist
->zones
[j
] = NULL
;
1548 #else /* CONFIG_NUMA */
1550 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1552 int node
, local_node
;
1555 local_node
= pgdat
->node_id
;
1556 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1557 struct zonelist
*zonelist
;
1559 zonelist
= pgdat
->node_zonelists
+ i
;
1561 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
1563 * Now we build the zonelist so that it contains the zones
1564 * of all the other nodes.
1565 * We don't want to pressure a particular node, so when
1566 * building the zones for node N, we make sure that the
1567 * zones coming right after the local ones are those from
1568 * node N+1 (modulo N)
1570 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1571 if (!node_online(node
))
1573 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1575 for (node
= 0; node
< local_node
; node
++) {
1576 if (!node_online(node
))
1578 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1581 zonelist
->zones
[j
] = NULL
;
1585 #endif /* CONFIG_NUMA */
1587 /* return values int ....just for stop_machine_run() */
1588 static int __meminit
__build_all_zonelists(void *dummy
)
1591 for_each_online_node(nid
)
1592 build_zonelists(NODE_DATA(nid
));
1596 void __meminit
build_all_zonelists(void)
1598 if (system_state
== SYSTEM_BOOTING
) {
1599 __build_all_zonelists(NULL
);
1600 cpuset_init_current_mems_allowed();
1602 /* we have to stop all cpus to guaranntee there is no user
1604 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
1605 /* cpuset refresh routine should be here */
1607 vm_total_pages
= nr_free_pagecache_pages();
1608 printk("Built %i zonelists. Total pages: %ld\n",
1609 num_online_nodes(), vm_total_pages
);
1613 * Helper functions to size the waitqueue hash table.
1614 * Essentially these want to choose hash table sizes sufficiently
1615 * large so that collisions trying to wait on pages are rare.
1616 * But in fact, the number of active page waitqueues on typical
1617 * systems is ridiculously low, less than 200. So this is even
1618 * conservative, even though it seems large.
1620 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1621 * waitqueues, i.e. the size of the waitq table given the number of pages.
1623 #define PAGES_PER_WAITQUEUE 256
1625 #ifndef CONFIG_MEMORY_HOTPLUG
1626 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1628 unsigned long size
= 1;
1630 pages
/= PAGES_PER_WAITQUEUE
;
1632 while (size
< pages
)
1636 * Once we have dozens or even hundreds of threads sleeping
1637 * on IO we've got bigger problems than wait queue collision.
1638 * Limit the size of the wait table to a reasonable size.
1640 size
= min(size
, 4096UL);
1642 return max(size
, 4UL);
1646 * A zone's size might be changed by hot-add, so it is not possible to determine
1647 * a suitable size for its wait_table. So we use the maximum size now.
1649 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1651 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1652 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1653 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1655 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1656 * or more by the traditional way. (See above). It equals:
1658 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1659 * ia64(16K page size) : = ( 8G + 4M)byte.
1660 * powerpc (64K page size) : = (32G +16M)byte.
1662 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1669 * This is an integer logarithm so that shifts can be used later
1670 * to extract the more random high bits from the multiplicative
1671 * hash function before the remainder is taken.
1673 static inline unsigned long wait_table_bits(unsigned long size
)
1678 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1681 * Initially all pages are reserved - free ones are freed
1682 * up by free_all_bootmem() once the early boot process is
1683 * done. Non-atomic initialization, single-pass.
1685 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1686 unsigned long start_pfn
)
1689 unsigned long end_pfn
= start_pfn
+ size
;
1692 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1693 if (!early_pfn_valid(pfn
))
1695 if (!early_pfn_in_nid(pfn
, nid
))
1697 page
= pfn_to_page(pfn
);
1698 set_page_links(page
, zone
, nid
, pfn
);
1699 init_page_count(page
);
1700 reset_page_mapcount(page
);
1701 SetPageReserved(page
);
1702 INIT_LIST_HEAD(&page
->lru
);
1703 #ifdef WANT_PAGE_VIRTUAL
1704 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1705 if (!is_highmem_idx(zone
))
1706 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1711 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1715 for (order
= 0; order
< MAX_ORDER
; order
++) {
1716 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1717 zone
->free_area
[order
].nr_free
= 0;
1721 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1722 void zonetable_add(struct zone
*zone
, int nid
, enum zone_type zid
,
1723 unsigned long pfn
, unsigned long size
)
1725 unsigned long snum
= pfn_to_section_nr(pfn
);
1726 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1729 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1731 for (; snum
<= end
; snum
++)
1732 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1735 #ifndef __HAVE_ARCH_MEMMAP_INIT
1736 #define memmap_init(size, nid, zone, start_pfn) \
1737 memmap_init_zone((size), (nid), (zone), (start_pfn))
1740 static int __cpuinit
zone_batchsize(struct zone
*zone
)
1745 * The per-cpu-pages pools are set to around 1000th of the
1746 * size of the zone. But no more than 1/2 of a meg.
1748 * OK, so we don't know how big the cache is. So guess.
1750 batch
= zone
->present_pages
/ 1024;
1751 if (batch
* PAGE_SIZE
> 512 * 1024)
1752 batch
= (512 * 1024) / PAGE_SIZE
;
1753 batch
/= 4; /* We effectively *= 4 below */
1758 * Clamp the batch to a 2^n - 1 value. Having a power
1759 * of 2 value was found to be more likely to have
1760 * suboptimal cache aliasing properties in some cases.
1762 * For example if 2 tasks are alternately allocating
1763 * batches of pages, one task can end up with a lot
1764 * of pages of one half of the possible page colors
1765 * and the other with pages of the other colors.
1767 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1772 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1774 struct per_cpu_pages
*pcp
;
1776 memset(p
, 0, sizeof(*p
));
1778 pcp
= &p
->pcp
[0]; /* hot */
1780 pcp
->high
= 6 * batch
;
1781 pcp
->batch
= max(1UL, 1 * batch
);
1782 INIT_LIST_HEAD(&pcp
->list
);
1784 pcp
= &p
->pcp
[1]; /* cold*/
1786 pcp
->high
= 2 * batch
;
1787 pcp
->batch
= max(1UL, batch
/2);
1788 INIT_LIST_HEAD(&pcp
->list
);
1792 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1793 * to the value high for the pageset p.
1796 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
1799 struct per_cpu_pages
*pcp
;
1801 pcp
= &p
->pcp
[0]; /* hot list */
1803 pcp
->batch
= max(1UL, high
/4);
1804 if ((high
/4) > (PAGE_SHIFT
* 8))
1805 pcp
->batch
= PAGE_SHIFT
* 8;
1811 * Boot pageset table. One per cpu which is going to be used for all
1812 * zones and all nodes. The parameters will be set in such a way
1813 * that an item put on a list will immediately be handed over to
1814 * the buddy list. This is safe since pageset manipulation is done
1815 * with interrupts disabled.
1817 * Some NUMA counter updates may also be caught by the boot pagesets.
1819 * The boot_pagesets must be kept even after bootup is complete for
1820 * unused processors and/or zones. They do play a role for bootstrapping
1821 * hotplugged processors.
1823 * zoneinfo_show() and maybe other functions do
1824 * not check if the processor is online before following the pageset pointer.
1825 * Other parts of the kernel may not check if the zone is available.
1827 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
1830 * Dynamically allocate memory for the
1831 * per cpu pageset array in struct zone.
1833 static int __cpuinit
process_zones(int cpu
)
1835 struct zone
*zone
, *dzone
;
1837 for_each_zone(zone
) {
1839 if (!populated_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 /* Free per_cpu_pageset if it is slab allocated */
1873 if (pset
!= &boot_pageset
[cpu
])
1875 zone_pcp(zone
, cpu
) = NULL
;
1879 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1880 unsigned long action
,
1883 int cpu
= (long)hcpu
;
1884 int ret
= NOTIFY_OK
;
1887 case CPU_UP_PREPARE
:
1888 if (process_zones(cpu
))
1891 case CPU_UP_CANCELED
:
1893 free_zone_pagesets(cpu
);
1901 static struct notifier_block __cpuinitdata pageset_notifier
=
1902 { &pageset_cpuup_callback
, NULL
, 0 };
1904 void __init
setup_per_cpu_pageset(void)
1908 /* Initialize per_cpu_pageset for cpu 0.
1909 * A cpuup callback will do this for every cpu
1910 * as it comes online
1912 err
= process_zones(smp_processor_id());
1914 register_cpu_notifier(&pageset_notifier
);
1920 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1923 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1927 * The per-page waitqueue mechanism uses hashed waitqueues
1930 zone
->wait_table_hash_nr_entries
=
1931 wait_table_hash_nr_entries(zone_size_pages
);
1932 zone
->wait_table_bits
=
1933 wait_table_bits(zone
->wait_table_hash_nr_entries
);
1934 alloc_size
= zone
->wait_table_hash_nr_entries
1935 * sizeof(wait_queue_head_t
);
1937 if (system_state
== SYSTEM_BOOTING
) {
1938 zone
->wait_table
= (wait_queue_head_t
*)
1939 alloc_bootmem_node(pgdat
, alloc_size
);
1942 * This case means that a zone whose size was 0 gets new memory
1943 * via memory hot-add.
1944 * But it may be the case that a new node was hot-added. In
1945 * this case vmalloc() will not be able to use this new node's
1946 * memory - this wait_table must be initialized to use this new
1947 * node itself as well.
1948 * To use this new node's memory, further consideration will be
1951 zone
->wait_table
= (wait_queue_head_t
*)vmalloc(alloc_size
);
1953 if (!zone
->wait_table
)
1956 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
1957 init_waitqueue_head(zone
->wait_table
+ i
);
1962 static __meminit
void zone_pcp_init(struct zone
*zone
)
1965 unsigned long batch
= zone_batchsize(zone
);
1967 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1969 /* Early boot. Slab allocator not functional yet */
1970 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
1971 setup_pageset(&boot_pageset
[cpu
],0);
1973 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1976 if (zone
->present_pages
)
1977 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1978 zone
->name
, zone
->present_pages
, batch
);
1981 __meminit
int init_currently_empty_zone(struct zone
*zone
,
1982 unsigned long zone_start_pfn
,
1985 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1987 ret
= zone_wait_table_init(zone
, size
);
1990 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1992 zone
->zone_start_pfn
= zone_start_pfn
;
1994 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1996 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2001 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2003 * Basic iterator support. Return the first range of PFNs for a node
2004 * Note: nid == MAX_NUMNODES returns first region regardless of node
2006 static int __init
first_active_region_index_in_nid(int nid
)
2010 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2011 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2018 * Basic iterator support. Return the next active range of PFNs for a node
2019 * Note: nid == MAX_NUMNODES returns next region regardles of node
2021 static int __init
next_active_region_index_in_nid(int index
, int nid
)
2023 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2024 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2030 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2032 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2033 * Architectures may implement their own version but if add_active_range()
2034 * was used and there are no special requirements, this is a convenient
2037 int __init
early_pfn_to_nid(unsigned long pfn
)
2041 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2042 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2043 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2045 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2046 return early_node_map
[i
].nid
;
2051 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2053 /* Basic iterator support to walk early_node_map[] */
2054 #define for_each_active_range_index_in_nid(i, nid) \
2055 for (i = first_active_region_index_in_nid(nid); i != -1; \
2056 i = next_active_region_index_in_nid(i, nid))
2059 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2060 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2061 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2063 * If an architecture guarantees that all ranges registered with
2064 * add_active_ranges() contain no holes and may be freed, this
2065 * this function may be used instead of calling free_bootmem() manually.
2067 void __init
free_bootmem_with_active_regions(int nid
,
2068 unsigned long max_low_pfn
)
2072 for_each_active_range_index_in_nid(i
, nid
) {
2073 unsigned long size_pages
= 0;
2074 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2076 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2079 if (end_pfn
> max_low_pfn
)
2080 end_pfn
= max_low_pfn
;
2082 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2083 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2084 PFN_PHYS(early_node_map
[i
].start_pfn
),
2085 size_pages
<< PAGE_SHIFT
);
2090 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2091 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2093 * If an architecture guarantees that all ranges registered with
2094 * add_active_ranges() contain no holes and may be freed, this
2095 * function may be used instead of calling memory_present() manually.
2097 void __init
sparse_memory_present_with_active_regions(int nid
)
2101 for_each_active_range_index_in_nid(i
, nid
)
2102 memory_present(early_node_map
[i
].nid
,
2103 early_node_map
[i
].start_pfn
,
2104 early_node_map
[i
].end_pfn
);
2108 * push_node_boundaries - Push node boundaries to at least the requested boundary
2109 * @nid: The nid of the node to push the boundary for
2110 * @start_pfn: The start pfn of the node
2111 * @end_pfn: The end pfn of the node
2113 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2114 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2115 * be hotplugged even though no physical memory exists. This function allows
2116 * an arch to push out the node boundaries so mem_map is allocated that can
2119 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2120 void __init
push_node_boundaries(unsigned int nid
,
2121 unsigned long start_pfn
, unsigned long end_pfn
)
2123 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2124 nid
, start_pfn
, end_pfn
);
2126 /* Initialise the boundary for this node if necessary */
2127 if (node_boundary_end_pfn
[nid
] == 0)
2128 node_boundary_start_pfn
[nid
] = -1UL;
2130 /* Update the boundaries */
2131 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2132 node_boundary_start_pfn
[nid
] = start_pfn
;
2133 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2134 node_boundary_end_pfn
[nid
] = end_pfn
;
2137 /* If necessary, push the node boundary out for reserve hotadd */
2138 static void __init
account_node_boundary(unsigned int nid
,
2139 unsigned long *start_pfn
, unsigned long *end_pfn
)
2141 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2142 nid
, *start_pfn
, *end_pfn
);
2144 /* Return if boundary information has not been provided */
2145 if (node_boundary_end_pfn
[nid
] == 0)
2148 /* Check the boundaries and update if necessary */
2149 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
2150 *start_pfn
= node_boundary_start_pfn
[nid
];
2151 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
2152 *end_pfn
= node_boundary_end_pfn
[nid
];
2155 void __init
push_node_boundaries(unsigned int nid
,
2156 unsigned long start_pfn
, unsigned long end_pfn
) {}
2158 static void __init
account_node_boundary(unsigned int nid
,
2159 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
2164 * get_pfn_range_for_nid - Return the start and end page frames for a node
2165 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2166 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2167 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2169 * It returns the start and end page frame of a node based on information
2170 * provided by an arch calling add_active_range(). If called for a node
2171 * with no available memory, a warning is printed and the start and end
2174 void __init
get_pfn_range_for_nid(unsigned int nid
,
2175 unsigned long *start_pfn
, unsigned long *end_pfn
)
2181 for_each_active_range_index_in_nid(i
, nid
) {
2182 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
2183 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
2186 if (*start_pfn
== -1UL) {
2187 printk(KERN_WARNING
"Node %u active with no memory\n", nid
);
2191 /* Push the node boundaries out if requested */
2192 account_node_boundary(nid
, start_pfn
, end_pfn
);
2196 * Return the number of pages a zone spans in a node, including holes
2197 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2199 unsigned long __init
zone_spanned_pages_in_node(int nid
,
2200 unsigned long zone_type
,
2201 unsigned long *ignored
)
2203 unsigned long node_start_pfn
, node_end_pfn
;
2204 unsigned long zone_start_pfn
, zone_end_pfn
;
2206 /* Get the start and end of the node and zone */
2207 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
2208 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
2209 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
2211 /* Check that this node has pages within the zone's required range */
2212 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
2215 /* Move the zone boundaries inside the node if necessary */
2216 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
2217 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
2219 /* Return the spanned pages */
2220 return zone_end_pfn
- zone_start_pfn
;
2224 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
2225 * then all holes in the requested range will be accounted for.
2227 unsigned long __init
__absent_pages_in_range(int nid
,
2228 unsigned long range_start_pfn
,
2229 unsigned long range_end_pfn
)
2232 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
2233 unsigned long start_pfn
;
2235 /* Find the end_pfn of the first active range of pfns in the node */
2236 i
= first_active_region_index_in_nid(nid
);
2240 /* Account for ranges before physical memory on this node */
2241 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
2242 hole_pages
= early_node_map
[i
].start_pfn
- range_start_pfn
;
2244 prev_end_pfn
= early_node_map
[i
].start_pfn
;
2246 /* Find all holes for the zone within the node */
2247 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
2249 /* No need to continue if prev_end_pfn is outside the zone */
2250 if (prev_end_pfn
>= range_end_pfn
)
2253 /* Make sure the end of the zone is not within the hole */
2254 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
2255 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
2257 /* Update the hole size cound and move on */
2258 if (start_pfn
> range_start_pfn
) {
2259 BUG_ON(prev_end_pfn
> start_pfn
);
2260 hole_pages
+= start_pfn
- prev_end_pfn
;
2262 prev_end_pfn
= early_node_map
[i
].end_pfn
;
2265 /* Account for ranges past physical memory on this node */
2266 if (range_end_pfn
> prev_end_pfn
)
2267 hole_pages
+= range_end_pfn
-
2268 max(range_start_pfn
, prev_end_pfn
);
2274 * absent_pages_in_range - Return number of page frames in holes within a range
2275 * @start_pfn: The start PFN to start searching for holes
2276 * @end_pfn: The end PFN to stop searching for holes
2278 * It returns the number of pages frames in memory holes within a range.
2280 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
2281 unsigned long end_pfn
)
2283 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
2286 /* Return the number of page frames in holes in a zone on a node */
2287 unsigned long __init
zone_absent_pages_in_node(int nid
,
2288 unsigned long zone_type
,
2289 unsigned long *ignored
)
2291 unsigned long node_start_pfn
, node_end_pfn
;
2292 unsigned long zone_start_pfn
, zone_end_pfn
;
2294 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
2295 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
2297 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
2300 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
2304 static inline unsigned long zone_spanned_pages_in_node(int nid
,
2305 unsigned long zone_type
,
2306 unsigned long *zones_size
)
2308 return zones_size
[zone_type
];
2311 static inline unsigned long zone_absent_pages_in_node(int nid
,
2312 unsigned long zone_type
,
2313 unsigned long *zholes_size
)
2318 return zholes_size
[zone_type
];
2323 static void __init
calculate_node_totalpages(struct pglist_data
*pgdat
,
2324 unsigned long *zones_size
, unsigned long *zholes_size
)
2326 unsigned long realtotalpages
, totalpages
= 0;
2329 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2330 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
2332 pgdat
->node_spanned_pages
= totalpages
;
2334 realtotalpages
= totalpages
;
2335 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2337 zone_absent_pages_in_node(pgdat
->node_id
, i
,
2339 pgdat
->node_present_pages
= realtotalpages
;
2340 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
2345 * Set up the zone data structures:
2346 * - mark all pages reserved
2347 * - mark all memory queues empty
2348 * - clear the memory bitmaps
2350 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
2351 unsigned long *zones_size
, unsigned long *zholes_size
)
2354 int nid
= pgdat
->node_id
;
2355 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
2358 pgdat_resize_init(pgdat
);
2359 pgdat
->nr_zones
= 0;
2360 init_waitqueue_head(&pgdat
->kswapd_wait
);
2361 pgdat
->kswapd_max_order
= 0;
2363 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2364 struct zone
*zone
= pgdat
->node_zones
+ j
;
2365 unsigned long size
, realsize
, memmap_pages
;
2367 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
2368 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
2372 * Adjust realsize so that it accounts for how much memory
2373 * is used by this zone for memmap. This affects the watermark
2374 * and per-cpu initialisations
2376 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
2377 if (realsize
>= memmap_pages
) {
2378 realsize
-= memmap_pages
;
2380 " %s zone: %lu pages used for memmap\n",
2381 zone_names
[j
], memmap_pages
);
2384 " %s zone: %lu pages exceeds realsize %lu\n",
2385 zone_names
[j
], memmap_pages
, realsize
);
2387 /* Account for reserved DMA pages */
2388 if (j
== ZONE_DMA
&& realsize
> dma_reserve
) {
2389 realsize
-= dma_reserve
;
2390 printk(KERN_DEBUG
" DMA zone: %lu pages reserved\n",
2394 if (!is_highmem_idx(j
))
2395 nr_kernel_pages
+= realsize
;
2396 nr_all_pages
+= realsize
;
2398 zone
->spanned_pages
= size
;
2399 zone
->present_pages
= realsize
;
2402 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
2404 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
2406 zone
->name
= zone_names
[j
];
2407 spin_lock_init(&zone
->lock
);
2408 spin_lock_init(&zone
->lru_lock
);
2409 zone_seqlock_init(zone
);
2410 zone
->zone_pgdat
= pgdat
;
2411 zone
->free_pages
= 0;
2413 zone
->prev_priority
= DEF_PRIORITY
;
2415 zone_pcp_init(zone
);
2416 INIT_LIST_HEAD(&zone
->active_list
);
2417 INIT_LIST_HEAD(&zone
->inactive_list
);
2418 zone
->nr_scan_active
= 0;
2419 zone
->nr_scan_inactive
= 0;
2420 zone
->nr_active
= 0;
2421 zone
->nr_inactive
= 0;
2422 zap_zone_vm_stats(zone
);
2423 atomic_set(&zone
->reclaim_in_progress
, 0);
2427 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2428 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2430 zone_start_pfn
+= size
;
2434 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2436 /* Skip empty nodes */
2437 if (!pgdat
->node_spanned_pages
)
2440 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2441 /* ia64 gets its own node_mem_map, before this, without bootmem */
2442 if (!pgdat
->node_mem_map
) {
2443 unsigned long size
, start
, end
;
2447 * The zone's endpoints aren't required to be MAX_ORDER
2448 * aligned but the node_mem_map endpoints must be in order
2449 * for the buddy allocator to function correctly.
2451 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
2452 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
2453 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
2454 size
= (end
- start
) * sizeof(struct page
);
2455 map
= alloc_remap(pgdat
->node_id
, size
);
2457 map
= alloc_bootmem_node(pgdat
, size
);
2458 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
2460 #ifdef CONFIG_FLATMEM
2462 * With no DISCONTIG, the global mem_map is just set as node 0's
2464 if (pgdat
== NODE_DATA(0)) {
2465 mem_map
= NODE_DATA(0)->node_mem_map
;
2466 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2467 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
2468 mem_map
-= pgdat
->node_start_pfn
;
2469 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2472 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2475 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2476 unsigned long *zones_size
, unsigned long node_start_pfn
,
2477 unsigned long *zholes_size
)
2479 pgdat
->node_id
= nid
;
2480 pgdat
->node_start_pfn
= node_start_pfn
;
2481 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
2483 alloc_node_mem_map(pgdat
);
2485 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2488 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2490 * add_active_range - Register a range of PFNs backed by physical memory
2491 * @nid: The node ID the range resides on
2492 * @start_pfn: The start PFN of the available physical memory
2493 * @end_pfn: The end PFN of the available physical memory
2495 * These ranges are stored in an early_node_map[] and later used by
2496 * free_area_init_nodes() to calculate zone sizes and holes. If the
2497 * range spans a memory hole, it is up to the architecture to ensure
2498 * the memory is not freed by the bootmem allocator. If possible
2499 * the range being registered will be merged with existing ranges.
2501 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
2502 unsigned long end_pfn
)
2506 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
2507 "%d entries of %d used\n",
2508 nid
, start_pfn
, end_pfn
,
2509 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
2511 /* Merge with existing active regions if possible */
2512 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2513 if (early_node_map
[i
].nid
!= nid
)
2516 /* Skip if an existing region covers this new one */
2517 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
2518 end_pfn
<= early_node_map
[i
].end_pfn
)
2521 /* Merge forward if suitable */
2522 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
2523 end_pfn
> early_node_map
[i
].end_pfn
) {
2524 early_node_map
[i
].end_pfn
= end_pfn
;
2528 /* Merge backward if suitable */
2529 if (start_pfn
< early_node_map
[i
].end_pfn
&&
2530 end_pfn
>= early_node_map
[i
].start_pfn
) {
2531 early_node_map
[i
].start_pfn
= start_pfn
;
2536 /* Check that early_node_map is large enough */
2537 if (i
>= MAX_ACTIVE_REGIONS
) {
2538 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
2539 MAX_ACTIVE_REGIONS
);
2543 early_node_map
[i
].nid
= nid
;
2544 early_node_map
[i
].start_pfn
= start_pfn
;
2545 early_node_map
[i
].end_pfn
= end_pfn
;
2546 nr_nodemap_entries
= i
+ 1;
2550 * shrink_active_range - Shrink an existing registered range of PFNs
2551 * @nid: The node id the range is on that should be shrunk
2552 * @old_end_pfn: The old end PFN of the range
2553 * @new_end_pfn: The new PFN of the range
2555 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2556 * The map is kept at the end physical page range that has already been
2557 * registered with add_active_range(). This function allows an arch to shrink
2558 * an existing registered range.
2560 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
2561 unsigned long new_end_pfn
)
2565 /* Find the old active region end and shrink */
2566 for_each_active_range_index_in_nid(i
, nid
)
2567 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
2568 early_node_map
[i
].end_pfn
= new_end_pfn
;
2574 * remove_all_active_ranges - Remove all currently registered regions
2576 * During discovery, it may be found that a table like SRAT is invalid
2577 * and an alternative discovery method must be used. This function removes
2578 * all currently registered regions.
2580 void __init
remove_all_active_ranges(void)
2582 memset(early_node_map
, 0, sizeof(early_node_map
));
2583 nr_nodemap_entries
= 0;
2584 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2585 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
2586 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
2587 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
2590 /* Compare two active node_active_regions */
2591 static int __init
cmp_node_active_region(const void *a
, const void *b
)
2593 struct node_active_region
*arange
= (struct node_active_region
*)a
;
2594 struct node_active_region
*brange
= (struct node_active_region
*)b
;
2596 /* Done this way to avoid overflows */
2597 if (arange
->start_pfn
> brange
->start_pfn
)
2599 if (arange
->start_pfn
< brange
->start_pfn
)
2605 /* sort the node_map by start_pfn */
2606 static void __init
sort_node_map(void)
2608 sort(early_node_map
, (size_t)nr_nodemap_entries
,
2609 sizeof(struct node_active_region
),
2610 cmp_node_active_region
, NULL
);
2613 /* Find the lowest pfn for a node. This depends on a sorted early_node_map */
2614 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
2618 /* Regions in the early_node_map can be in any order */
2621 /* Assuming a sorted map, the first range found has the starting pfn */
2622 for_each_active_range_index_in_nid(i
, nid
)
2623 return early_node_map
[i
].start_pfn
;
2625 printk(KERN_WARNING
"Could not find start_pfn for node %lu\n", nid
);
2630 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2632 * It returns the minimum PFN based on information provided via
2633 * add_active_range().
2635 unsigned long __init
find_min_pfn_with_active_regions(void)
2637 return find_min_pfn_for_node(MAX_NUMNODES
);
2641 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2643 * It returns the maximum PFN based on information provided via
2644 * add_active_range().
2646 unsigned long __init
find_max_pfn_with_active_regions(void)
2649 unsigned long max_pfn
= 0;
2651 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2652 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
2658 * free_area_init_nodes - Initialise all pg_data_t and zone data
2659 * @max_zone_pfn: an array of max PFNs for each zone
2661 * This will call free_area_init_node() for each active node in the system.
2662 * Using the page ranges provided by add_active_range(), the size of each
2663 * zone in each node and their holes is calculated. If the maximum PFN
2664 * between two adjacent zones match, it is assumed that the zone is empty.
2665 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2666 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2667 * starts where the previous one ended. For example, ZONE_DMA32 starts
2668 * at arch_max_dma_pfn.
2670 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
2675 /* Record where the zone boundaries are */
2676 memset(arch_zone_lowest_possible_pfn
, 0,
2677 sizeof(arch_zone_lowest_possible_pfn
));
2678 memset(arch_zone_highest_possible_pfn
, 0,
2679 sizeof(arch_zone_highest_possible_pfn
));
2680 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
2681 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
2682 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
2683 arch_zone_lowest_possible_pfn
[i
] =
2684 arch_zone_highest_possible_pfn
[i
-1];
2685 arch_zone_highest_possible_pfn
[i
] =
2686 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
2689 /* Print out the zone ranges */
2690 printk("Zone PFN ranges:\n");
2691 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2692 printk(" %-8s %8lu -> %8lu\n",
2694 arch_zone_lowest_possible_pfn
[i
],
2695 arch_zone_highest_possible_pfn
[i
]);
2697 /* Print out the early_node_map[] */
2698 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
2699 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2700 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
2701 early_node_map
[i
].start_pfn
,
2702 early_node_map
[i
].end_pfn
);
2704 /* Initialise every node */
2705 for_each_online_node(nid
) {
2706 pg_data_t
*pgdat
= NODE_DATA(nid
);
2707 free_area_init_node(nid
, pgdat
, NULL
,
2708 find_min_pfn_for_node(nid
), NULL
);
2711 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2714 * set_dma_reserve - set the specified number of pages reserved in the first zone
2715 * @new_dma_reserve: The number of pages to mark reserved
2717 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2718 * In the DMA zone, a significant percentage may be consumed by kernel image
2719 * and other unfreeable allocations which can skew the watermarks badly. This
2720 * function may optionally be used to account for unfreeable pages in the
2721 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2722 * smaller per-cpu batchsize.
2724 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
2726 dma_reserve
= new_dma_reserve
;
2729 #ifndef CONFIG_NEED_MULTIPLE_NODES
2730 static bootmem_data_t contig_bootmem_data
;
2731 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2733 EXPORT_SYMBOL(contig_page_data
);
2736 void __init
free_area_init(unsigned long *zones_size
)
2738 free_area_init_node(0, NODE_DATA(0), zones_size
,
2739 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2742 #ifdef CONFIG_HOTPLUG_CPU
2743 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2744 unsigned long action
, void *hcpu
)
2746 int cpu
= (unsigned long)hcpu
;
2748 if (action
== CPU_DEAD
) {
2749 local_irq_disable();
2751 vm_events_fold_cpu(cpu
);
2753 refresh_cpu_vm_stats(cpu
);
2757 #endif /* CONFIG_HOTPLUG_CPU */
2759 void __init
page_alloc_init(void)
2761 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2765 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2766 * or min_free_kbytes changes.
2768 static void calculate_totalreserve_pages(void)
2770 struct pglist_data
*pgdat
;
2771 unsigned long reserve_pages
= 0;
2772 enum zone_type i
, j
;
2774 for_each_online_pgdat(pgdat
) {
2775 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2776 struct zone
*zone
= pgdat
->node_zones
+ i
;
2777 unsigned long max
= 0;
2779 /* Find valid and maximum lowmem_reserve in the zone */
2780 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
2781 if (zone
->lowmem_reserve
[j
] > max
)
2782 max
= zone
->lowmem_reserve
[j
];
2785 /* we treat pages_high as reserved pages. */
2786 max
+= zone
->pages_high
;
2788 if (max
> zone
->present_pages
)
2789 max
= zone
->present_pages
;
2790 reserve_pages
+= max
;
2793 totalreserve_pages
= reserve_pages
;
2797 * setup_per_zone_lowmem_reserve - called whenever
2798 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2799 * has a correct pages reserved value, so an adequate number of
2800 * pages are left in the zone after a successful __alloc_pages().
2802 static void setup_per_zone_lowmem_reserve(void)
2804 struct pglist_data
*pgdat
;
2805 enum zone_type j
, idx
;
2807 for_each_online_pgdat(pgdat
) {
2808 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2809 struct zone
*zone
= pgdat
->node_zones
+ j
;
2810 unsigned long present_pages
= zone
->present_pages
;
2812 zone
->lowmem_reserve
[j
] = 0;
2816 struct zone
*lower_zone
;
2820 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2821 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2823 lower_zone
= pgdat
->node_zones
+ idx
;
2824 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2825 sysctl_lowmem_reserve_ratio
[idx
];
2826 present_pages
+= lower_zone
->present_pages
;
2831 /* update totalreserve_pages */
2832 calculate_totalreserve_pages();
2836 * setup_per_zone_pages_min - called when min_free_kbytes changes.
2838 * Ensures that the pages_{min,low,high} values for each zone are set correctly
2839 * with respect to min_free_kbytes.
2841 void setup_per_zone_pages_min(void)
2843 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2844 unsigned long lowmem_pages
= 0;
2846 unsigned long flags
;
2848 /* Calculate total number of !ZONE_HIGHMEM pages */
2849 for_each_zone(zone
) {
2850 if (!is_highmem(zone
))
2851 lowmem_pages
+= zone
->present_pages
;
2854 for_each_zone(zone
) {
2857 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2858 tmp
= (u64
)pages_min
* zone
->present_pages
;
2859 do_div(tmp
, lowmem_pages
);
2860 if (is_highmem(zone
)) {
2862 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2863 * need highmem pages, so cap pages_min to a small
2866 * The (pages_high-pages_low) and (pages_low-pages_min)
2867 * deltas controls asynch page reclaim, and so should
2868 * not be capped for highmem.
2872 min_pages
= zone
->present_pages
/ 1024;
2873 if (min_pages
< SWAP_CLUSTER_MAX
)
2874 min_pages
= SWAP_CLUSTER_MAX
;
2875 if (min_pages
> 128)
2877 zone
->pages_min
= min_pages
;
2880 * If it's a lowmem zone, reserve a number of pages
2881 * proportionate to the zone's size.
2883 zone
->pages_min
= tmp
;
2886 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
2887 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
2888 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2891 /* update totalreserve_pages */
2892 calculate_totalreserve_pages();
2896 * Initialise min_free_kbytes.
2898 * For small machines we want it small (128k min). For large machines
2899 * we want it large (64MB max). But it is not linear, because network
2900 * bandwidth does not increase linearly with machine size. We use
2902 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2903 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2919 static int __init
init_per_zone_pages_min(void)
2921 unsigned long lowmem_kbytes
;
2923 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2925 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2926 if (min_free_kbytes
< 128)
2927 min_free_kbytes
= 128;
2928 if (min_free_kbytes
> 65536)
2929 min_free_kbytes
= 65536;
2930 setup_per_zone_pages_min();
2931 setup_per_zone_lowmem_reserve();
2934 module_init(init_per_zone_pages_min
)
2937 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2938 * that we can call two helper functions whenever min_free_kbytes
2941 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2942 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2944 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2945 setup_per_zone_pages_min();
2950 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
2951 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2956 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2961 zone
->min_unmapped_pages
= (zone
->present_pages
*
2962 sysctl_min_unmapped_ratio
) / 100;
2966 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
2967 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2972 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2977 zone
->min_slab_pages
= (zone
->present_pages
*
2978 sysctl_min_slab_ratio
) / 100;
2984 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2985 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2986 * whenever sysctl_lowmem_reserve_ratio changes.
2988 * The reserve ratio obviously has absolutely no relation with the
2989 * pages_min watermarks. The lowmem reserve ratio can only make sense
2990 * if in function of the boot time zone sizes.
2992 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2993 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2995 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2996 setup_per_zone_lowmem_reserve();
3001 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3002 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3003 * can have before it gets flushed back to buddy allocator.
3006 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
3007 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3013 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3014 if (!write
|| (ret
== -EINVAL
))
3016 for_each_zone(zone
) {
3017 for_each_online_cpu(cpu
) {
3019 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
3020 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
3026 int hashdist
= HASHDIST_DEFAULT
;
3029 static int __init
set_hashdist(char *str
)
3033 hashdist
= simple_strtoul(str
, &str
, 0);
3036 __setup("hashdist=", set_hashdist
);
3040 * allocate a large system hash table from bootmem
3041 * - it is assumed that the hash table must contain an exact power-of-2
3042 * quantity of entries
3043 * - limit is the number of hash buckets, not the total allocation size
3045 void *__init
alloc_large_system_hash(const char *tablename
,
3046 unsigned long bucketsize
,
3047 unsigned long numentries
,
3050 unsigned int *_hash_shift
,
3051 unsigned int *_hash_mask
,
3052 unsigned long limit
)
3054 unsigned long long max
= limit
;
3055 unsigned long log2qty
, size
;
3058 /* allow the kernel cmdline to have a say */
3060 /* round applicable memory size up to nearest megabyte */
3061 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
3062 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
3063 numentries
>>= 20 - PAGE_SHIFT
;
3064 numentries
<<= 20 - PAGE_SHIFT
;
3066 /* limit to 1 bucket per 2^scale bytes of low memory */
3067 if (scale
> PAGE_SHIFT
)
3068 numentries
>>= (scale
- PAGE_SHIFT
);
3070 numentries
<<= (PAGE_SHIFT
- scale
);
3072 numentries
= roundup_pow_of_two(numentries
);
3074 /* limit allocation size to 1/16 total memory by default */
3076 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
3077 do_div(max
, bucketsize
);
3080 if (numentries
> max
)
3083 log2qty
= long_log2(numentries
);
3086 size
= bucketsize
<< log2qty
;
3087 if (flags
& HASH_EARLY
)
3088 table
= alloc_bootmem(size
);
3090 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
3092 unsigned long order
;
3093 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
3095 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
3097 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
3100 panic("Failed to allocate %s hash table\n", tablename
);
3102 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3105 long_log2(size
) - PAGE_SHIFT
,
3109 *_hash_shift
= log2qty
;
3111 *_hash_mask
= (1 << log2qty
) - 1;
3116 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
3117 struct page
*pfn_to_page(unsigned long pfn
)
3119 return __pfn_to_page(pfn
);
3121 unsigned long page_to_pfn(struct page
*page
)
3123 return __page_to_pfn(page
);
3125 EXPORT_SYMBOL(pfn_to_page
);
3126 EXPORT_SYMBOL(page_to_pfn
);
3127 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
3129 #if MAX_NUMNODES > 1
3131 * Find the highest possible node id.
3133 int highest_possible_node_id(void)
3136 unsigned int highest
= 0;
3138 for_each_node_mask(node
, node_possible_map
)
3142 EXPORT_SYMBOL(highest_possible_node_id
);