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>
43 #include <asm/tlbflush.h>
44 #include <asm/div64.h>
48 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
51 nodemask_t node_online_map __read_mostly
= { { [0] = 1UL } };
52 EXPORT_SYMBOL(node_online_map
);
53 nodemask_t node_possible_map __read_mostly
= NODE_MASK_ALL
;
54 EXPORT_SYMBOL(node_possible_map
);
55 unsigned long totalram_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] = {
75 #ifdef CONFIG_ZONE_DMA32
83 EXPORT_SYMBOL(totalram_pages
);
86 * Used by page_zone() to look up the address of the struct zone whose
87 * id is encoded in the upper bits of page->flags
89 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
90 EXPORT_SYMBOL(zone_table
);
92 static char *zone_names
[MAX_NR_ZONES
] = {
94 #ifdef CONFIG_ZONE_DMA32
103 int min_free_kbytes
= 1024;
105 unsigned long __meminitdata nr_kernel_pages
;
106 unsigned long __meminitdata nr_all_pages
;
107 static unsigned long __initdata dma_reserve
;
109 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
111 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
112 * ranges of memory (RAM) that may be registered with add_active_range().
113 * Ranges passed to add_active_range() will be merged if possible
114 * so the number of times add_active_range() can be called is
115 * related to the number of nodes and the number of holes
117 #ifdef CONFIG_MAX_ACTIVE_REGIONS
118 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
119 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
121 #if MAX_NUMNODES >= 32
122 /* If there can be many nodes, allow up to 50 holes per node */
123 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
125 /* By default, allow up to 256 distinct regions */
126 #define MAX_ACTIVE_REGIONS 256
130 struct node_active_region __initdata early_node_map
[MAX_ACTIVE_REGIONS
];
131 int __initdata nr_nodemap_entries
;
132 unsigned long __initdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
133 unsigned long __initdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
134 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
136 #ifdef CONFIG_DEBUG_VM
137 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
141 unsigned long pfn
= page_to_pfn(page
);
144 seq
= zone_span_seqbegin(zone
);
145 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
147 else if (pfn
< zone
->zone_start_pfn
)
149 } while (zone_span_seqretry(zone
, seq
));
154 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
156 #ifdef CONFIG_HOLES_IN_ZONE
157 if (!pfn_valid(page_to_pfn(page
)))
160 if (zone
!= page_zone(page
))
166 * Temporary debugging check for pages not lying within a given zone.
168 static int bad_range(struct zone
*zone
, struct page
*page
)
170 if (page_outside_zone_boundaries(zone
, page
))
172 if (!page_is_consistent(zone
, page
))
178 static inline int bad_range(struct zone
*zone
, struct page
*page
)
184 static void bad_page(struct page
*page
)
186 printk(KERN_EMERG
"Bad page state in process '%s'\n"
187 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
188 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
189 KERN_EMERG
"Backtrace:\n",
190 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
191 (unsigned long)page
->flags
, page
->mapping
,
192 page_mapcount(page
), page_count(page
));
194 page
->flags
&= ~(1 << PG_lru
|
204 set_page_count(page
, 0);
205 reset_page_mapcount(page
);
206 page
->mapping
= NULL
;
207 add_taint(TAINT_BAD_PAGE
);
211 * Higher-order pages are called "compound pages". They are structured thusly:
213 * The first PAGE_SIZE page is called the "head page".
215 * The remaining PAGE_SIZE pages are called "tail pages".
217 * All pages have PG_compound set. All pages have their ->private pointing at
218 * the head page (even the head page has this).
220 * The first tail page's ->lru.next holds the address of the compound page's
221 * put_page() function. Its ->lru.prev holds the order of allocation.
222 * This usage means that zero-order pages may not be compound.
225 static void free_compound_page(struct page
*page
)
227 __free_pages_ok(page
, (unsigned long)page
[1].lru
.prev
);
230 static void prep_compound_page(struct page
*page
, unsigned long order
)
233 int nr_pages
= 1 << order
;
235 page
[1].lru
.next
= (void *)free_compound_page
; /* set dtor */
236 page
[1].lru
.prev
= (void *)order
;
237 for (i
= 0; i
< nr_pages
; i
++) {
238 struct page
*p
= page
+ i
;
240 __SetPageCompound(p
);
241 set_page_private(p
, (unsigned long)page
);
245 static void destroy_compound_page(struct page
*page
, unsigned long order
)
248 int nr_pages
= 1 << order
;
250 if (unlikely((unsigned long)page
[1].lru
.prev
!= order
))
253 for (i
= 0; i
< nr_pages
; i
++) {
254 struct page
*p
= page
+ i
;
256 if (unlikely(!PageCompound(p
) |
257 (page_private(p
) != (unsigned long)page
)))
259 __ClearPageCompound(p
);
263 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
267 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
269 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
270 * and __GFP_HIGHMEM from hard or soft interrupt context.
272 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
273 for (i
= 0; i
< (1 << order
); i
++)
274 clear_highpage(page
+ i
);
278 * function for dealing with page's order in buddy system.
279 * zone->lock is already acquired when we use these.
280 * So, we don't need atomic page->flags operations here.
282 static inline unsigned long page_order(struct page
*page
)
284 return page_private(page
);
287 static inline void set_page_order(struct page
*page
, int order
)
289 set_page_private(page
, order
);
290 __SetPageBuddy(page
);
293 static inline void rmv_page_order(struct page
*page
)
295 __ClearPageBuddy(page
);
296 set_page_private(page
, 0);
300 * Locate the struct page for both the matching buddy in our
301 * pair (buddy1) and the combined O(n+1) page they form (page).
303 * 1) Any buddy B1 will have an order O twin B2 which satisfies
304 * the following equation:
306 * For example, if the starting buddy (buddy2) is #8 its order
308 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
310 * 2) Any buddy B will have an order O+1 parent P which
311 * satisfies the following equation:
314 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
316 static inline struct page
*
317 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
319 unsigned long buddy_idx
= page_idx
^ (1 << order
);
321 return page
+ (buddy_idx
- page_idx
);
324 static inline unsigned long
325 __find_combined_index(unsigned long page_idx
, unsigned int order
)
327 return (page_idx
& ~(1 << order
));
331 * This function checks whether a page is free && is the buddy
332 * we can do coalesce a page and its buddy if
333 * (a) the buddy is not in a hole &&
334 * (b) the buddy is in the buddy system &&
335 * (c) a page and its buddy have the same order &&
336 * (d) a page and its buddy are in the same zone.
338 * For recording whether a page is in the buddy system, we use PG_buddy.
339 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
341 * For recording page's order, we use page_private(page).
343 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
346 #ifdef CONFIG_HOLES_IN_ZONE
347 if (!pfn_valid(page_to_pfn(buddy
)))
351 if (page_zone_id(page
) != page_zone_id(buddy
))
354 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
355 BUG_ON(page_count(buddy
) != 0);
362 * Freeing function for a buddy system allocator.
364 * The concept of a buddy system is to maintain direct-mapped table
365 * (containing bit values) for memory blocks of various "orders".
366 * The bottom level table contains the map for the smallest allocatable
367 * units of memory (here, pages), and each level above it describes
368 * pairs of units from the levels below, hence, "buddies".
369 * At a high level, all that happens here is marking the table entry
370 * at the bottom level available, and propagating the changes upward
371 * as necessary, plus some accounting needed to play nicely with other
372 * parts of the VM system.
373 * At each level, we keep a list of pages, which are heads of continuous
374 * free pages of length of (1 << order) and marked with PG_buddy. Page's
375 * order is recorded in page_private(page) field.
376 * So when we are allocating or freeing one, we can derive the state of the
377 * other. That is, if we allocate a small block, and both were
378 * free, the remainder of the region must be split into blocks.
379 * If a block is freed, and its buddy is also free, then this
380 * triggers coalescing into a block of larger size.
385 static inline void __free_one_page(struct page
*page
,
386 struct zone
*zone
, unsigned int order
)
388 unsigned long page_idx
;
389 int order_size
= 1 << order
;
391 if (unlikely(PageCompound(page
)))
392 destroy_compound_page(page
, order
);
394 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
396 VM_BUG_ON(page_idx
& (order_size
- 1));
397 VM_BUG_ON(bad_range(zone
, page
));
399 zone
->free_pages
+= order_size
;
400 while (order
< MAX_ORDER
-1) {
401 unsigned long combined_idx
;
402 struct free_area
*area
;
405 buddy
= __page_find_buddy(page
, page_idx
, order
);
406 if (!page_is_buddy(page
, buddy
, order
))
407 break; /* Move the buddy up one level. */
409 list_del(&buddy
->lru
);
410 area
= zone
->free_area
+ order
;
412 rmv_page_order(buddy
);
413 combined_idx
= __find_combined_index(page_idx
, order
);
414 page
= page
+ (combined_idx
- page_idx
);
415 page_idx
= combined_idx
;
418 set_page_order(page
, order
);
419 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
420 zone
->free_area
[order
].nr_free
++;
423 static inline int free_pages_check(struct page
*page
)
425 if (unlikely(page_mapcount(page
) |
426 (page
->mapping
!= NULL
) |
427 (page_count(page
) != 0) |
441 __ClearPageDirty(page
);
443 * For now, we report if PG_reserved was found set, but do not
444 * clear it, and do not free the page. But we shall soon need
445 * to do more, for when the ZERO_PAGE count wraps negative.
447 return PageReserved(page
);
451 * Frees a list of pages.
452 * Assumes all pages on list are in same zone, and of same order.
453 * count is the number of pages to free.
455 * If the zone was previously in an "all pages pinned" state then look to
456 * see if this freeing clears that state.
458 * And clear the zone's pages_scanned counter, to hold off the "all pages are
459 * pinned" detection logic.
461 static void free_pages_bulk(struct zone
*zone
, int count
,
462 struct list_head
*list
, int order
)
464 spin_lock(&zone
->lock
);
465 zone
->all_unreclaimable
= 0;
466 zone
->pages_scanned
= 0;
470 VM_BUG_ON(list_empty(list
));
471 page
= list_entry(list
->prev
, struct page
, lru
);
472 /* have to delete it as __free_one_page list manipulates */
473 list_del(&page
->lru
);
474 __free_one_page(page
, zone
, order
);
476 spin_unlock(&zone
->lock
);
479 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
481 spin_lock(&zone
->lock
);
482 zone
->all_unreclaimable
= 0;
483 zone
->pages_scanned
= 0;
484 __free_one_page(page
, zone
,order
);
485 spin_unlock(&zone
->lock
);
488 static void __free_pages_ok(struct page
*page
, unsigned int order
)
494 arch_free_page(page
, order
);
495 if (!PageHighMem(page
))
496 debug_check_no_locks_freed(page_address(page
),
499 for (i
= 0 ; i
< (1 << order
) ; ++i
)
500 reserved
+= free_pages_check(page
+ i
);
504 kernel_map_pages(page
, 1 << order
, 0);
505 local_irq_save(flags
);
506 __count_vm_events(PGFREE
, 1 << order
);
507 free_one_page(page_zone(page
), page
, order
);
508 local_irq_restore(flags
);
512 * permit the bootmem allocator to evade page validation on high-order frees
514 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
517 __ClearPageReserved(page
);
518 set_page_count(page
, 0);
519 set_page_refcounted(page
);
525 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
526 struct page
*p
= &page
[loop
];
528 if (loop
+ 1 < BITS_PER_LONG
)
530 __ClearPageReserved(p
);
531 set_page_count(p
, 0);
534 set_page_refcounted(page
);
535 __free_pages(page
, order
);
541 * The order of subdivision here is critical for the IO subsystem.
542 * Please do not alter this order without good reasons and regression
543 * testing. Specifically, as large blocks of memory are subdivided,
544 * the order in which smaller blocks are delivered depends on the order
545 * they're subdivided in this function. This is the primary factor
546 * influencing the order in which pages are delivered to the IO
547 * subsystem according to empirical testing, and this is also justified
548 * by considering the behavior of a buddy system containing a single
549 * large block of memory acted on by a series of small allocations.
550 * This behavior is a critical factor in sglist merging's success.
554 static inline void expand(struct zone
*zone
, struct page
*page
,
555 int low
, int high
, struct free_area
*area
)
557 unsigned long size
= 1 << high
;
563 VM_BUG_ON(bad_range(zone
, &page
[size
]));
564 list_add(&page
[size
].lru
, &area
->free_list
);
566 set_page_order(&page
[size
], high
);
571 * This page is about to be returned from the page allocator
573 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
575 if (unlikely(page_mapcount(page
) |
576 (page
->mapping
!= NULL
) |
577 (page_count(page
) != 0) |
593 * For now, we report if PG_reserved was found set, but do not
594 * clear it, and do not allocate the page: as a safety net.
596 if (PageReserved(page
))
599 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
600 1 << PG_referenced
| 1 << PG_arch_1
|
601 1 << PG_checked
| 1 << PG_mappedtodisk
);
602 set_page_private(page
, 0);
603 set_page_refcounted(page
);
604 kernel_map_pages(page
, 1 << order
, 1);
606 if (gfp_flags
& __GFP_ZERO
)
607 prep_zero_page(page
, order
, gfp_flags
);
609 if (order
&& (gfp_flags
& __GFP_COMP
))
610 prep_compound_page(page
, order
);
616 * Do the hard work of removing an element from the buddy allocator.
617 * Call me with the zone->lock already held.
619 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
621 struct free_area
* area
;
622 unsigned int current_order
;
625 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
626 area
= zone
->free_area
+ current_order
;
627 if (list_empty(&area
->free_list
))
630 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
631 list_del(&page
->lru
);
632 rmv_page_order(page
);
634 zone
->free_pages
-= 1UL << order
;
635 expand(zone
, page
, order
, current_order
, area
);
643 * Obtain a specified number of elements from the buddy allocator, all under
644 * a single hold of the lock, for efficiency. Add them to the supplied list.
645 * Returns the number of new pages which were placed at *list.
647 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
648 unsigned long count
, struct list_head
*list
)
652 spin_lock(&zone
->lock
);
653 for (i
= 0; i
< count
; ++i
) {
654 struct page
*page
= __rmqueue(zone
, order
);
655 if (unlikely(page
== NULL
))
657 list_add_tail(&page
->lru
, list
);
659 spin_unlock(&zone
->lock
);
665 * Called from the slab reaper to drain pagesets on a particular node that
666 * belongs to the currently executing processor.
667 * Note that this function must be called with the thread pinned to
668 * a single processor.
670 void drain_node_pages(int nodeid
)
676 for (z
= 0; z
< MAX_NR_ZONES
; z
++) {
677 struct zone
*zone
= NODE_DATA(nodeid
)->node_zones
+ z
;
678 struct per_cpu_pageset
*pset
;
680 if (!populated_zone(zone
))
683 pset
= zone_pcp(zone
, smp_processor_id());
684 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
685 struct per_cpu_pages
*pcp
;
689 local_irq_save(flags
);
690 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
692 local_irq_restore(flags
);
699 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
700 static void __drain_pages(unsigned int cpu
)
706 for_each_zone(zone
) {
707 struct per_cpu_pageset
*pset
;
709 pset
= zone_pcp(zone
, cpu
);
710 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
711 struct per_cpu_pages
*pcp
;
714 local_irq_save(flags
);
715 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
717 local_irq_restore(flags
);
721 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
725 void mark_free_pages(struct zone
*zone
)
727 unsigned long pfn
, max_zone_pfn
;
730 struct list_head
*curr
;
732 if (!zone
->spanned_pages
)
735 spin_lock_irqsave(&zone
->lock
, flags
);
737 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
738 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
739 if (pfn_valid(pfn
)) {
740 struct page
*page
= pfn_to_page(pfn
);
742 if (!PageNosave(page
))
743 ClearPageNosaveFree(page
);
746 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
747 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
750 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
751 for (i
= 0; i
< (1UL << order
); i
++)
752 SetPageNosaveFree(pfn_to_page(pfn
+ i
));
755 spin_unlock_irqrestore(&zone
->lock
, flags
);
759 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
761 void drain_local_pages(void)
765 local_irq_save(flags
);
766 __drain_pages(smp_processor_id());
767 local_irq_restore(flags
);
769 #endif /* CONFIG_PM */
772 * Free a 0-order page
774 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
776 struct zone
*zone
= page_zone(page
);
777 struct per_cpu_pages
*pcp
;
780 arch_free_page(page
, 0);
783 page
->mapping
= NULL
;
784 if (free_pages_check(page
))
787 kernel_map_pages(page
, 1, 0);
789 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
790 local_irq_save(flags
);
791 __count_vm_event(PGFREE
);
792 list_add(&page
->lru
, &pcp
->list
);
794 if (pcp
->count
>= pcp
->high
) {
795 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
796 pcp
->count
-= pcp
->batch
;
798 local_irq_restore(flags
);
802 void fastcall
free_hot_page(struct page
*page
)
804 free_hot_cold_page(page
, 0);
807 void fastcall
free_cold_page(struct page
*page
)
809 free_hot_cold_page(page
, 1);
813 * split_page takes a non-compound higher-order page, and splits it into
814 * n (1<<order) sub-pages: page[0..n]
815 * Each sub-page must be freed individually.
817 * Note: this is probably too low level an operation for use in drivers.
818 * Please consult with lkml before using this in your driver.
820 void split_page(struct page
*page
, unsigned int order
)
824 VM_BUG_ON(PageCompound(page
));
825 VM_BUG_ON(!page_count(page
));
826 for (i
= 1; i
< (1 << order
); i
++)
827 set_page_refcounted(page
+ i
);
831 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
832 * we cheat by calling it from here, in the order > 0 path. Saves a branch
835 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
836 struct zone
*zone
, int order
, gfp_t gfp_flags
)
840 int cold
= !!(gfp_flags
& __GFP_COLD
);
845 if (likely(order
== 0)) {
846 struct per_cpu_pages
*pcp
;
848 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
849 local_irq_save(flags
);
851 pcp
->count
+= rmqueue_bulk(zone
, 0,
852 pcp
->batch
, &pcp
->list
);
853 if (unlikely(!pcp
->count
))
856 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
857 list_del(&page
->lru
);
860 spin_lock_irqsave(&zone
->lock
, flags
);
861 page
= __rmqueue(zone
, order
);
862 spin_unlock(&zone
->lock
);
867 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
868 zone_statistics(zonelist
, zone
);
869 local_irq_restore(flags
);
872 VM_BUG_ON(bad_range(zone
, page
));
873 if (prep_new_page(page
, order
, gfp_flags
))
878 local_irq_restore(flags
);
883 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
884 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
885 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
886 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
887 #define ALLOC_HARDER 0x10 /* try to alloc harder */
888 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
889 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
892 * Return 1 if free pages are above 'mark'. This takes into account the order
895 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
896 int classzone_idx
, int alloc_flags
)
898 /* free_pages my go negative - that's OK */
899 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
902 if (alloc_flags
& ALLOC_HIGH
)
904 if (alloc_flags
& ALLOC_HARDER
)
907 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
909 for (o
= 0; o
< order
; o
++) {
910 /* At the next order, this order's pages become unavailable */
911 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
913 /* Require fewer higher order pages to be free */
916 if (free_pages
<= min
)
923 * get_page_from_freeliest goes through the zonelist trying to allocate
927 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
928 struct zonelist
*zonelist
, int alloc_flags
)
930 struct zone
**z
= zonelist
->zones
;
931 struct page
*page
= NULL
;
932 int classzone_idx
= zone_idx(*z
);
936 * Go through the zonelist once, looking for a zone with enough free.
937 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
941 if (unlikely((gfp_mask
& __GFP_THISNODE
) &&
942 zone
->zone_pgdat
!= zonelist
->zones
[0]->zone_pgdat
))
944 if ((alloc_flags
& ALLOC_CPUSET
) &&
945 !cpuset_zone_allowed(zone
, gfp_mask
))
948 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
950 if (alloc_flags
& ALLOC_WMARK_MIN
)
951 mark
= zone
->pages_min
;
952 else if (alloc_flags
& ALLOC_WMARK_LOW
)
953 mark
= zone
->pages_low
;
955 mark
= zone
->pages_high
;
956 if (!zone_watermark_ok(zone
, order
, mark
,
957 classzone_idx
, alloc_flags
))
958 if (!zone_reclaim_mode
||
959 !zone_reclaim(zone
, gfp_mask
, order
))
963 page
= buffered_rmqueue(zonelist
, zone
, order
, gfp_mask
);
967 } while (*(++z
) != NULL
);
972 * This is the 'heart' of the zoned buddy allocator.
974 struct page
* fastcall
975 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
976 struct zonelist
*zonelist
)
978 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
981 struct reclaim_state reclaim_state
;
982 struct task_struct
*p
= current
;
985 int did_some_progress
;
987 might_sleep_if(wait
);
990 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
992 if (unlikely(*z
== NULL
)) {
993 /* Should this ever happen?? */
997 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
998 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1003 wakeup_kswapd(*z
, order
);
1007 * OK, we're below the kswapd watermark and have kicked background
1008 * reclaim. Now things get more complex, so set up alloc_flags according
1009 * to how we want to proceed.
1011 * The caller may dip into page reserves a bit more if the caller
1012 * cannot run direct reclaim, or if the caller has realtime scheduling
1013 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1014 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1016 alloc_flags
= ALLOC_WMARK_MIN
;
1017 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1018 alloc_flags
|= ALLOC_HARDER
;
1019 if (gfp_mask
& __GFP_HIGH
)
1020 alloc_flags
|= ALLOC_HIGH
;
1022 alloc_flags
|= ALLOC_CPUSET
;
1025 * Go through the zonelist again. Let __GFP_HIGH and allocations
1026 * coming from realtime tasks go deeper into reserves.
1028 * This is the last chance, in general, before the goto nopage.
1029 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1030 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1032 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1036 /* This allocation should allow future memory freeing. */
1038 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1039 && !in_interrupt()) {
1040 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1042 /* go through the zonelist yet again, ignoring mins */
1043 page
= get_page_from_freelist(gfp_mask
, order
,
1044 zonelist
, ALLOC_NO_WATERMARKS
);
1047 if (gfp_mask
& __GFP_NOFAIL
) {
1048 blk_congestion_wait(WRITE
, HZ
/50);
1055 /* Atomic allocations - we can't balance anything */
1062 /* We now go into synchronous reclaim */
1063 cpuset_memory_pressure_bump();
1064 p
->flags
|= PF_MEMALLOC
;
1065 reclaim_state
.reclaimed_slab
= 0;
1066 p
->reclaim_state
= &reclaim_state
;
1068 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
1070 p
->reclaim_state
= NULL
;
1071 p
->flags
&= ~PF_MEMALLOC
;
1075 if (likely(did_some_progress
)) {
1076 page
= get_page_from_freelist(gfp_mask
, order
,
1077 zonelist
, alloc_flags
);
1080 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1082 * Go through the zonelist yet one more time, keep
1083 * very high watermark here, this is only to catch
1084 * a parallel oom killing, we must fail if we're still
1085 * under heavy pressure.
1087 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1088 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1092 out_of_memory(zonelist
, gfp_mask
, order
);
1097 * Don't let big-order allocations loop unless the caller explicitly
1098 * requests that. Wait for some write requests to complete then retry.
1100 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1101 * <= 3, but that may not be true in other implementations.
1104 if (!(gfp_mask
& __GFP_NORETRY
)) {
1105 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
1107 if (gfp_mask
& __GFP_NOFAIL
)
1111 blk_congestion_wait(WRITE
, HZ
/50);
1116 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1117 printk(KERN_WARNING
"%s: page allocation failure."
1118 " order:%d, mode:0x%x\n",
1119 p
->comm
, order
, gfp_mask
);
1127 EXPORT_SYMBOL(__alloc_pages
);
1130 * Common helper functions.
1132 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1135 page
= alloc_pages(gfp_mask
, order
);
1138 return (unsigned long) page_address(page
);
1141 EXPORT_SYMBOL(__get_free_pages
);
1143 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1148 * get_zeroed_page() returns a 32-bit address, which cannot represent
1151 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1153 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1155 return (unsigned long) page_address(page
);
1159 EXPORT_SYMBOL(get_zeroed_page
);
1161 void __pagevec_free(struct pagevec
*pvec
)
1163 int i
= pagevec_count(pvec
);
1166 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1169 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1171 if (put_page_testzero(page
)) {
1173 free_hot_page(page
);
1175 __free_pages_ok(page
, order
);
1179 EXPORT_SYMBOL(__free_pages
);
1181 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1184 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1185 __free_pages(virt_to_page((void *)addr
), order
);
1189 EXPORT_SYMBOL(free_pages
);
1192 * Total amount of free (allocatable) RAM:
1194 unsigned int nr_free_pages(void)
1196 unsigned int sum
= 0;
1200 sum
+= zone
->free_pages
;
1205 EXPORT_SYMBOL(nr_free_pages
);
1208 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1210 unsigned int sum
= 0;
1213 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1214 sum
+= pgdat
->node_zones
[i
].free_pages
;
1220 static unsigned int nr_free_zone_pages(int offset
)
1222 /* Just pick one node, since fallback list is circular */
1223 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1224 unsigned int sum
= 0;
1226 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1227 struct zone
**zonep
= zonelist
->zones
;
1230 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1231 unsigned long size
= zone
->present_pages
;
1232 unsigned long high
= zone
->pages_high
;
1241 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1243 unsigned int nr_free_buffer_pages(void)
1245 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1249 * Amount of free RAM allocatable within all zones
1251 unsigned int nr_free_pagecache_pages(void)
1253 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1256 static void show_node(struct zone
*zone
)
1258 printk("Node %ld ", zone_to_nid(zone
));
1261 #define show_node(zone) do { } while (0)
1264 void si_meminfo(struct sysinfo
*val
)
1266 val
->totalram
= totalram_pages
;
1268 val
->freeram
= nr_free_pages();
1269 val
->bufferram
= nr_blockdev_pages();
1270 val
->totalhigh
= totalhigh_pages
;
1271 val
->freehigh
= nr_free_highpages();
1272 val
->mem_unit
= PAGE_SIZE
;
1275 EXPORT_SYMBOL(si_meminfo
);
1278 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1280 pg_data_t
*pgdat
= NODE_DATA(nid
);
1282 val
->totalram
= pgdat
->node_present_pages
;
1283 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1284 #ifdef CONFIG_HIGHMEM
1285 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1286 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1291 val
->mem_unit
= PAGE_SIZE
;
1295 #define K(x) ((x) << (PAGE_SHIFT-10))
1298 * Show free area list (used inside shift_scroll-lock stuff)
1299 * We also calculate the percentage fragmentation. We do this by counting the
1300 * memory on each free list with the exception of the first item on the list.
1302 void show_free_areas(void)
1304 int cpu
, temperature
;
1305 unsigned long active
;
1306 unsigned long inactive
;
1310 for_each_zone(zone
) {
1312 printk("%s per-cpu:", zone
->name
);
1314 if (!populated_zone(zone
)) {
1320 for_each_online_cpu(cpu
) {
1321 struct per_cpu_pageset
*pageset
;
1323 pageset
= zone_pcp(zone
, cpu
);
1325 for (temperature
= 0; temperature
< 2; temperature
++)
1326 printk("cpu %d %s: high %d, batch %d used:%d\n",
1328 temperature
? "cold" : "hot",
1329 pageset
->pcp
[temperature
].high
,
1330 pageset
->pcp
[temperature
].batch
,
1331 pageset
->pcp
[temperature
].count
);
1335 get_zone_counts(&active
, &inactive
, &free
);
1337 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1338 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1341 global_page_state(NR_FILE_DIRTY
),
1342 global_page_state(NR_WRITEBACK
),
1343 global_page_state(NR_UNSTABLE_NFS
),
1345 global_page_state(NR_SLAB_RECLAIMABLE
) +
1346 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1347 global_page_state(NR_FILE_MAPPED
),
1348 global_page_state(NR_PAGETABLE
));
1350 for_each_zone(zone
) {
1362 " pages_scanned:%lu"
1363 " all_unreclaimable? %s"
1366 K(zone
->free_pages
),
1369 K(zone
->pages_high
),
1371 K(zone
->nr_inactive
),
1372 K(zone
->present_pages
),
1373 zone
->pages_scanned
,
1374 (zone
->all_unreclaimable
? "yes" : "no")
1376 printk("lowmem_reserve[]:");
1377 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1378 printk(" %lu", zone
->lowmem_reserve
[i
]);
1382 for_each_zone(zone
) {
1383 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1386 printk("%s: ", zone
->name
);
1387 if (!populated_zone(zone
)) {
1392 spin_lock_irqsave(&zone
->lock
, flags
);
1393 for (order
= 0; order
< MAX_ORDER
; order
++) {
1394 nr
[order
] = zone
->free_area
[order
].nr_free
;
1395 total
+= nr
[order
] << order
;
1397 spin_unlock_irqrestore(&zone
->lock
, flags
);
1398 for (order
= 0; order
< MAX_ORDER
; order
++)
1399 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1400 printk("= %lukB\n", K(total
));
1403 show_swap_cache_info();
1407 * Builds allocation fallback zone lists.
1409 * Add all populated zones of a node to the zonelist.
1411 static int __meminit
build_zonelists_node(pg_data_t
*pgdat
,
1412 struct zonelist
*zonelist
, int nr_zones
, enum zone_type zone_type
)
1416 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1421 zone
= pgdat
->node_zones
+ zone_type
;
1422 if (populated_zone(zone
)) {
1423 zonelist
->zones
[nr_zones
++] = zone
;
1424 check_highest_zone(zone_type
);
1427 } while (zone_type
);
1432 #define MAX_NODE_LOAD (num_online_nodes())
1433 static int __meminitdata node_load
[MAX_NUMNODES
];
1435 * find_next_best_node - find the next node that should appear in a given node's fallback list
1436 * @node: node whose fallback list we're appending
1437 * @used_node_mask: nodemask_t of already used nodes
1439 * We use a number of factors to determine which is the next node that should
1440 * appear on a given node's fallback list. The node should not have appeared
1441 * already in @node's fallback list, and it should be the next closest node
1442 * according to the distance array (which contains arbitrary distance values
1443 * from each node to each node in the system), and should also prefer nodes
1444 * with no CPUs, since presumably they'll have very little allocation pressure
1445 * on them otherwise.
1446 * It returns -1 if no node is found.
1448 static int __meminit
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1451 int min_val
= INT_MAX
;
1454 /* Use the local node if we haven't already */
1455 if (!node_isset(node
, *used_node_mask
)) {
1456 node_set(node
, *used_node_mask
);
1460 for_each_online_node(n
) {
1463 /* Don't want a node to appear more than once */
1464 if (node_isset(n
, *used_node_mask
))
1467 /* Use the distance array to find the distance */
1468 val
= node_distance(node
, n
);
1470 /* Penalize nodes under us ("prefer the next node") */
1473 /* Give preference to headless and unused nodes */
1474 tmp
= node_to_cpumask(n
);
1475 if (!cpus_empty(tmp
))
1476 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1478 /* Slight preference for less loaded node */
1479 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1480 val
+= node_load
[n
];
1482 if (val
< min_val
) {
1489 node_set(best_node
, *used_node_mask
);
1494 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1496 int j
, node
, local_node
;
1498 int prev_node
, load
;
1499 struct zonelist
*zonelist
;
1500 nodemask_t used_mask
;
1502 /* initialize zonelists */
1503 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1504 zonelist
= pgdat
->node_zonelists
+ i
;
1505 zonelist
->zones
[0] = NULL
;
1508 /* NUMA-aware ordering of nodes */
1509 local_node
= pgdat
->node_id
;
1510 load
= num_online_nodes();
1511 prev_node
= local_node
;
1512 nodes_clear(used_mask
);
1513 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1514 int distance
= node_distance(local_node
, node
);
1517 * If another node is sufficiently far away then it is better
1518 * to reclaim pages in a zone before going off node.
1520 if (distance
> RECLAIM_DISTANCE
)
1521 zone_reclaim_mode
= 1;
1524 * We don't want to pressure a particular node.
1525 * So adding penalty to the first node in same
1526 * distance group to make it round-robin.
1529 if (distance
!= node_distance(local_node
, prev_node
))
1530 node_load
[node
] += load
;
1533 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1534 zonelist
= pgdat
->node_zonelists
+ i
;
1535 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1537 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1538 zonelist
->zones
[j
] = NULL
;
1543 #else /* CONFIG_NUMA */
1545 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1547 int node
, local_node
;
1550 local_node
= pgdat
->node_id
;
1551 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1552 struct zonelist
*zonelist
;
1554 zonelist
= pgdat
->node_zonelists
+ i
;
1556 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
1558 * Now we build the zonelist so that it contains the zones
1559 * of all the other nodes.
1560 * We don't want to pressure a particular node, so when
1561 * building the zones for node N, we make sure that the
1562 * zones coming right after the local ones are those from
1563 * node N+1 (modulo N)
1565 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1566 if (!node_online(node
))
1568 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1570 for (node
= 0; node
< local_node
; node
++) {
1571 if (!node_online(node
))
1573 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1576 zonelist
->zones
[j
] = NULL
;
1580 #endif /* CONFIG_NUMA */
1582 /* return values int ....just for stop_machine_run() */
1583 static int __meminit
__build_all_zonelists(void *dummy
)
1586 for_each_online_node(nid
)
1587 build_zonelists(NODE_DATA(nid
));
1591 void __meminit
build_all_zonelists(void)
1593 if (system_state
== SYSTEM_BOOTING
) {
1594 __build_all_zonelists(0);
1595 cpuset_init_current_mems_allowed();
1597 /* we have to stop all cpus to guaranntee there is no user
1599 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
1600 /* cpuset refresh routine should be here */
1602 vm_total_pages
= nr_free_pagecache_pages();
1603 printk("Built %i zonelists. Total pages: %ld\n",
1604 num_online_nodes(), vm_total_pages
);
1608 * Helper functions to size the waitqueue hash table.
1609 * Essentially these want to choose hash table sizes sufficiently
1610 * large so that collisions trying to wait on pages are rare.
1611 * But in fact, the number of active page waitqueues on typical
1612 * systems is ridiculously low, less than 200. So this is even
1613 * conservative, even though it seems large.
1615 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1616 * waitqueues, i.e. the size of the waitq table given the number of pages.
1618 #define PAGES_PER_WAITQUEUE 256
1620 #ifndef CONFIG_MEMORY_HOTPLUG
1621 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1623 unsigned long size
= 1;
1625 pages
/= PAGES_PER_WAITQUEUE
;
1627 while (size
< pages
)
1631 * Once we have dozens or even hundreds of threads sleeping
1632 * on IO we've got bigger problems than wait queue collision.
1633 * Limit the size of the wait table to a reasonable size.
1635 size
= min(size
, 4096UL);
1637 return max(size
, 4UL);
1641 * A zone's size might be changed by hot-add, so it is not possible to determine
1642 * a suitable size for its wait_table. So we use the maximum size now.
1644 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1646 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1647 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1648 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1650 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1651 * or more by the traditional way. (See above). It equals:
1653 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1654 * ia64(16K page size) : = ( 8G + 4M)byte.
1655 * powerpc (64K page size) : = (32G +16M)byte.
1657 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1664 * This is an integer logarithm so that shifts can be used later
1665 * to extract the more random high bits from the multiplicative
1666 * hash function before the remainder is taken.
1668 static inline unsigned long wait_table_bits(unsigned long size
)
1673 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1676 * Initially all pages are reserved - free ones are freed
1677 * up by free_all_bootmem() once the early boot process is
1678 * done. Non-atomic initialization, single-pass.
1680 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1681 unsigned long start_pfn
)
1684 unsigned long end_pfn
= start_pfn
+ size
;
1687 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1688 if (!early_pfn_valid(pfn
))
1690 page
= pfn_to_page(pfn
);
1691 set_page_links(page
, zone
, nid
, pfn
);
1692 init_page_count(page
);
1693 reset_page_mapcount(page
);
1694 SetPageReserved(page
);
1695 INIT_LIST_HEAD(&page
->lru
);
1696 #ifdef WANT_PAGE_VIRTUAL
1697 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1698 if (!is_highmem_idx(zone
))
1699 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1704 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1708 for (order
= 0; order
< MAX_ORDER
; order
++) {
1709 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1710 zone
->free_area
[order
].nr_free
= 0;
1714 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1715 void zonetable_add(struct zone
*zone
, int nid
, enum zone_type zid
,
1716 unsigned long pfn
, unsigned long size
)
1718 unsigned long snum
= pfn_to_section_nr(pfn
);
1719 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1722 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1724 for (; snum
<= end
; snum
++)
1725 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1728 #ifndef __HAVE_ARCH_MEMMAP_INIT
1729 #define memmap_init(size, nid, zone, start_pfn) \
1730 memmap_init_zone((size), (nid), (zone), (start_pfn))
1733 static int __cpuinit
zone_batchsize(struct zone
*zone
)
1738 * The per-cpu-pages pools are set to around 1000th of the
1739 * size of the zone. But no more than 1/2 of a meg.
1741 * OK, so we don't know how big the cache is. So guess.
1743 batch
= zone
->present_pages
/ 1024;
1744 if (batch
* PAGE_SIZE
> 512 * 1024)
1745 batch
= (512 * 1024) / PAGE_SIZE
;
1746 batch
/= 4; /* We effectively *= 4 below */
1751 * Clamp the batch to a 2^n - 1 value. Having a power
1752 * of 2 value was found to be more likely to have
1753 * suboptimal cache aliasing properties in some cases.
1755 * For example if 2 tasks are alternately allocating
1756 * batches of pages, one task can end up with a lot
1757 * of pages of one half of the possible page colors
1758 * and the other with pages of the other colors.
1760 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1765 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1767 struct per_cpu_pages
*pcp
;
1769 memset(p
, 0, sizeof(*p
));
1771 pcp
= &p
->pcp
[0]; /* hot */
1773 pcp
->high
= 6 * batch
;
1774 pcp
->batch
= max(1UL, 1 * batch
);
1775 INIT_LIST_HEAD(&pcp
->list
);
1777 pcp
= &p
->pcp
[1]; /* cold*/
1779 pcp
->high
= 2 * batch
;
1780 pcp
->batch
= max(1UL, batch
/2);
1781 INIT_LIST_HEAD(&pcp
->list
);
1785 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1786 * to the value high for the pageset p.
1789 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
1792 struct per_cpu_pages
*pcp
;
1794 pcp
= &p
->pcp
[0]; /* hot list */
1796 pcp
->batch
= max(1UL, high
/4);
1797 if ((high
/4) > (PAGE_SHIFT
* 8))
1798 pcp
->batch
= PAGE_SHIFT
* 8;
1804 * Boot pageset table. One per cpu which is going to be used for all
1805 * zones and all nodes. The parameters will be set in such a way
1806 * that an item put on a list will immediately be handed over to
1807 * the buddy list. This is safe since pageset manipulation is done
1808 * with interrupts disabled.
1810 * Some NUMA counter updates may also be caught by the boot pagesets.
1812 * The boot_pagesets must be kept even after bootup is complete for
1813 * unused processors and/or zones. They do play a role for bootstrapping
1814 * hotplugged processors.
1816 * zoneinfo_show() and maybe other functions do
1817 * not check if the processor is online before following the pageset pointer.
1818 * Other parts of the kernel may not check if the zone is available.
1820 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
1823 * Dynamically allocate memory for the
1824 * per cpu pageset array in struct zone.
1826 static int __cpuinit
process_zones(int cpu
)
1828 struct zone
*zone
, *dzone
;
1830 for_each_zone(zone
) {
1832 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
1833 GFP_KERNEL
, cpu_to_node(cpu
));
1834 if (!zone_pcp(zone
, cpu
))
1837 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
1839 if (percpu_pagelist_fraction
)
1840 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
1841 (zone
->present_pages
/ percpu_pagelist_fraction
));
1846 for_each_zone(dzone
) {
1849 kfree(zone_pcp(dzone
, cpu
));
1850 zone_pcp(dzone
, cpu
) = NULL
;
1855 static inline void free_zone_pagesets(int cpu
)
1859 for_each_zone(zone
) {
1860 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1862 /* Free per_cpu_pageset if it is slab allocated */
1863 if (pset
!= &boot_pageset
[cpu
])
1865 zone_pcp(zone
, cpu
) = NULL
;
1869 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1870 unsigned long action
,
1873 int cpu
= (long)hcpu
;
1874 int ret
= NOTIFY_OK
;
1877 case CPU_UP_PREPARE
:
1878 if (process_zones(cpu
))
1881 case CPU_UP_CANCELED
:
1883 free_zone_pagesets(cpu
);
1891 static struct notifier_block __cpuinitdata pageset_notifier
=
1892 { &pageset_cpuup_callback
, NULL
, 0 };
1894 void __init
setup_per_cpu_pageset(void)
1898 /* Initialize per_cpu_pageset for cpu 0.
1899 * A cpuup callback will do this for every cpu
1900 * as it comes online
1902 err
= process_zones(smp_processor_id());
1904 register_cpu_notifier(&pageset_notifier
);
1910 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1913 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1917 * The per-page waitqueue mechanism uses hashed waitqueues
1920 zone
->wait_table_hash_nr_entries
=
1921 wait_table_hash_nr_entries(zone_size_pages
);
1922 zone
->wait_table_bits
=
1923 wait_table_bits(zone
->wait_table_hash_nr_entries
);
1924 alloc_size
= zone
->wait_table_hash_nr_entries
1925 * sizeof(wait_queue_head_t
);
1927 if (system_state
== SYSTEM_BOOTING
) {
1928 zone
->wait_table
= (wait_queue_head_t
*)
1929 alloc_bootmem_node(pgdat
, alloc_size
);
1932 * This case means that a zone whose size was 0 gets new memory
1933 * via memory hot-add.
1934 * But it may be the case that a new node was hot-added. In
1935 * this case vmalloc() will not be able to use this new node's
1936 * memory - this wait_table must be initialized to use this new
1937 * node itself as well.
1938 * To use this new node's memory, further consideration will be
1941 zone
->wait_table
= (wait_queue_head_t
*)vmalloc(alloc_size
);
1943 if (!zone
->wait_table
)
1946 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
1947 init_waitqueue_head(zone
->wait_table
+ i
);
1952 static __meminit
void zone_pcp_init(struct zone
*zone
)
1955 unsigned long batch
= zone_batchsize(zone
);
1957 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1959 /* Early boot. Slab allocator not functional yet */
1960 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
1961 setup_pageset(&boot_pageset
[cpu
],0);
1963 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1966 if (zone
->present_pages
)
1967 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1968 zone
->name
, zone
->present_pages
, batch
);
1971 __meminit
int init_currently_empty_zone(struct zone
*zone
,
1972 unsigned long zone_start_pfn
,
1975 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1977 ret
= zone_wait_table_init(zone
, size
);
1980 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1982 zone
->zone_start_pfn
= zone_start_pfn
;
1984 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1986 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
1991 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
1993 * Basic iterator support. Return the first range of PFNs for a node
1994 * Note: nid == MAX_NUMNODES returns first region regardless of node
1996 static int __init
first_active_region_index_in_nid(int nid
)
2000 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2001 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2008 * Basic iterator support. Return the next active range of PFNs for a node
2009 * Note: nid == MAX_NUMNODES returns next region regardles of node
2011 static int __init
next_active_region_index_in_nid(int index
, int nid
)
2013 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2014 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2020 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2022 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2023 * Architectures may implement their own version but if add_active_range()
2024 * was used and there are no special requirements, this is a convenient
2027 int __init
early_pfn_to_nid(unsigned long pfn
)
2031 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2032 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2033 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2035 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2036 return early_node_map
[i
].nid
;
2041 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2043 /* Basic iterator support to walk early_node_map[] */
2044 #define for_each_active_range_index_in_nid(i, nid) \
2045 for (i = first_active_region_index_in_nid(nid); i != -1; \
2046 i = next_active_region_index_in_nid(i, nid))
2049 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2050 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed
2051 * @max_low_pfn: The highest PFN that till be passed to free_bootmem_node
2053 * If an architecture guarantees that all ranges registered with
2054 * add_active_ranges() contain no holes and may be freed, this
2055 * this function may be used instead of calling free_bootmem() manually.
2057 void __init
free_bootmem_with_active_regions(int nid
,
2058 unsigned long max_low_pfn
)
2062 for_each_active_range_index_in_nid(i
, nid
) {
2063 unsigned long size_pages
= 0;
2064 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2066 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2069 if (end_pfn
> max_low_pfn
)
2070 end_pfn
= max_low_pfn
;
2072 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2073 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2074 PFN_PHYS(early_node_map
[i
].start_pfn
),
2075 size_pages
<< PAGE_SHIFT
);
2080 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2081 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used
2083 * If an architecture guarantees that all ranges registered with
2084 * add_active_ranges() contain no holes and may be freed, this
2085 * this function may be used instead of calling memory_present() manually.
2087 void __init
sparse_memory_present_with_active_regions(int nid
)
2091 for_each_active_range_index_in_nid(i
, nid
)
2092 memory_present(early_node_map
[i
].nid
,
2093 early_node_map
[i
].start_pfn
,
2094 early_node_map
[i
].end_pfn
);
2098 * get_pfn_range_for_nid - Return the start and end page frames for a node
2099 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned
2100 * @start_pfn: Passed by reference. On return, it will have the node start_pfn
2101 * @end_pfn: Passed by reference. On return, it will have the node end_pfn
2103 * It returns the start and end page frame of a node based on information
2104 * provided by an arch calling add_active_range(). If called for a node
2105 * with no available memory, a warning is printed and the start and end
2108 void __init
get_pfn_range_for_nid(unsigned int nid
,
2109 unsigned long *start_pfn
, unsigned long *end_pfn
)
2115 for_each_active_range_index_in_nid(i
, nid
) {
2116 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
2117 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
2120 if (*start_pfn
== -1UL) {
2121 printk(KERN_WARNING
"Node %u active with no memory\n", nid
);
2127 * Return the number of pages a zone spans in a node, including holes
2128 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2130 unsigned long __init
zone_spanned_pages_in_node(int nid
,
2131 unsigned long zone_type
,
2132 unsigned long *ignored
)
2134 unsigned long node_start_pfn
, node_end_pfn
;
2135 unsigned long zone_start_pfn
, zone_end_pfn
;
2137 /* Get the start and end of the node and zone */
2138 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
2139 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
2140 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
2142 /* Check that this node has pages within the zone's required range */
2143 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
2146 /* Move the zone boundaries inside the node if necessary */
2147 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
2148 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
2150 /* Return the spanned pages */
2151 return zone_end_pfn
- zone_start_pfn
;
2155 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
2156 * then all holes in the requested range will be accounted for
2158 unsigned long __init
__absent_pages_in_range(int nid
,
2159 unsigned long range_start_pfn
,
2160 unsigned long range_end_pfn
)
2163 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
2164 unsigned long start_pfn
;
2166 /* Find the end_pfn of the first active range of pfns in the node */
2167 i
= first_active_region_index_in_nid(nid
);
2171 prev_end_pfn
= early_node_map
[i
].start_pfn
;
2173 /* Find all holes for the zone within the node */
2174 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
2176 /* No need to continue if prev_end_pfn is outside the zone */
2177 if (prev_end_pfn
>= range_end_pfn
)
2180 /* Make sure the end of the zone is not within the hole */
2181 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
2182 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
2184 /* Update the hole size cound and move on */
2185 if (start_pfn
> range_start_pfn
) {
2186 BUG_ON(prev_end_pfn
> start_pfn
);
2187 hole_pages
+= start_pfn
- prev_end_pfn
;
2189 prev_end_pfn
= early_node_map
[i
].end_pfn
;
2196 * absent_pages_in_range - Return number of page frames in holes within a range
2197 * @start_pfn: The start PFN to start searching for holes
2198 * @end_pfn: The end PFN to stop searching for holes
2200 * It returns the number of pages frames in memory holes within a range
2202 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
2203 unsigned long end_pfn
)
2205 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
2208 /* Return the number of page frames in holes in a zone on a node */
2209 unsigned long __init
zone_absent_pages_in_node(int nid
,
2210 unsigned long zone_type
,
2211 unsigned long *ignored
)
2213 return __absent_pages_in_range(nid
,
2214 arch_zone_lowest_possible_pfn
[zone_type
],
2215 arch_zone_highest_possible_pfn
[zone_type
]);
2218 /* Return the zone index a PFN is in */
2219 int memmap_zone_idx(struct page
*lmem_map
)
2222 unsigned long phys_addr
= virt_to_phys(lmem_map
);
2223 unsigned long pfn
= phys_addr
>> PAGE_SHIFT
;
2225 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2226 if (pfn
< arch_zone_highest_possible_pfn
[i
])
2232 static inline unsigned long zone_spanned_pages_in_node(int nid
,
2233 unsigned long zone_type
,
2234 unsigned long *zones_size
)
2236 return zones_size
[zone_type
];
2239 static inline unsigned long zone_absent_pages_in_node(int nid
,
2240 unsigned long zone_type
,
2241 unsigned long *zholes_size
)
2246 return zholes_size
[zone_type
];
2249 static inline int memmap_zone_idx(struct page
*lmem_map
)
2251 return MAX_NR_ZONES
;
2255 static void __init
calculate_node_totalpages(struct pglist_data
*pgdat
,
2256 unsigned long *zones_size
, unsigned long *zholes_size
)
2258 unsigned long realtotalpages
, totalpages
= 0;
2261 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2262 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
2264 pgdat
->node_spanned_pages
= totalpages
;
2266 realtotalpages
= totalpages
;
2267 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2269 zone_absent_pages_in_node(pgdat
->node_id
, i
,
2271 pgdat
->node_present_pages
= realtotalpages
;
2272 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
2277 * Set up the zone data structures:
2278 * - mark all pages reserved
2279 * - mark all memory queues empty
2280 * - clear the memory bitmaps
2282 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
2283 unsigned long *zones_size
, unsigned long *zholes_size
)
2286 int nid
= pgdat
->node_id
;
2287 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
2290 pgdat_resize_init(pgdat
);
2291 pgdat
->nr_zones
= 0;
2292 init_waitqueue_head(&pgdat
->kswapd_wait
);
2293 pgdat
->kswapd_max_order
= 0;
2295 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2296 struct zone
*zone
= pgdat
->node_zones
+ j
;
2297 unsigned long size
, realsize
, memmap_pages
;
2299 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
2300 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
2304 * Adjust realsize so that it accounts for how much memory
2305 * is used by this zone for memmap. This affects the watermark
2306 * and per-cpu initialisations
2308 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
2309 if (realsize
>= memmap_pages
) {
2310 realsize
-= memmap_pages
;
2312 " %s zone: %lu pages used for memmap\n",
2313 zone_names
[j
], memmap_pages
);
2316 " %s zone: %lu pages exceeds realsize %lu\n",
2317 zone_names
[j
], memmap_pages
, realsize
);
2319 /* Account for reserved DMA pages */
2320 if (j
== ZONE_DMA
&& realsize
> dma_reserve
) {
2321 realsize
-= dma_reserve
;
2322 printk(KERN_DEBUG
" DMA zone: %lu pages reserved\n",
2326 if (!is_highmem_idx(j
))
2327 nr_kernel_pages
+= realsize
;
2328 nr_all_pages
+= realsize
;
2330 zone
->spanned_pages
= size
;
2331 zone
->present_pages
= realsize
;
2333 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
2335 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
2337 zone
->name
= zone_names
[j
];
2338 spin_lock_init(&zone
->lock
);
2339 spin_lock_init(&zone
->lru_lock
);
2340 zone_seqlock_init(zone
);
2341 zone
->zone_pgdat
= pgdat
;
2342 zone
->free_pages
= 0;
2344 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
2346 zone_pcp_init(zone
);
2347 INIT_LIST_HEAD(&zone
->active_list
);
2348 INIT_LIST_HEAD(&zone
->inactive_list
);
2349 zone
->nr_scan_active
= 0;
2350 zone
->nr_scan_inactive
= 0;
2351 zone
->nr_active
= 0;
2352 zone
->nr_inactive
= 0;
2353 zap_zone_vm_stats(zone
);
2354 atomic_set(&zone
->reclaim_in_progress
, 0);
2358 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2359 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2361 zone_start_pfn
+= size
;
2365 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2367 /* Skip empty nodes */
2368 if (!pgdat
->node_spanned_pages
)
2371 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2372 /* ia64 gets its own node_mem_map, before this, without bootmem */
2373 if (!pgdat
->node_mem_map
) {
2374 unsigned long size
, start
, end
;
2378 * The zone's endpoints aren't required to be MAX_ORDER
2379 * aligned but the node_mem_map endpoints must be in order
2380 * for the buddy allocator to function correctly.
2382 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
2383 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
2384 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
2385 size
= (end
- start
) * sizeof(struct page
);
2386 map
= alloc_remap(pgdat
->node_id
, size
);
2388 map
= alloc_bootmem_node(pgdat
, size
);
2389 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
2391 #ifdef CONFIG_FLATMEM
2393 * With no DISCONTIG, the global mem_map is just set as node 0's
2395 if (pgdat
== NODE_DATA(0)) {
2396 mem_map
= NODE_DATA(0)->node_mem_map
;
2397 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2398 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
2399 mem_map
-= pgdat
->node_start_pfn
;
2400 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2403 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2406 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2407 unsigned long *zones_size
, unsigned long node_start_pfn
,
2408 unsigned long *zholes_size
)
2410 pgdat
->node_id
= nid
;
2411 pgdat
->node_start_pfn
= node_start_pfn
;
2412 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
2414 alloc_node_mem_map(pgdat
);
2416 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2419 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2421 * add_active_range - Register a range of PFNs backed by physical memory
2422 * @nid: The node ID the range resides on
2423 * @start_pfn: The start PFN of the available physical memory
2424 * @end_pfn: The end PFN of the available physical memory
2426 * These ranges are stored in an early_node_map[] and later used by
2427 * free_area_init_nodes() to calculate zone sizes and holes. If the
2428 * range spans a memory hole, it is up to the architecture to ensure
2429 * the memory is not freed by the bootmem allocator. If possible
2430 * the range being registered will be merged with existing ranges.
2432 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
2433 unsigned long end_pfn
)
2437 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
2438 "%d entries of %d used\n",
2439 nid
, start_pfn
, end_pfn
,
2440 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
2442 /* Merge with existing active regions if possible */
2443 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2444 if (early_node_map
[i
].nid
!= nid
)
2447 /* Skip if an existing region covers this new one */
2448 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
2449 end_pfn
<= early_node_map
[i
].end_pfn
)
2452 /* Merge forward if suitable */
2453 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
2454 end_pfn
> early_node_map
[i
].end_pfn
) {
2455 early_node_map
[i
].end_pfn
= end_pfn
;
2459 /* Merge backward if suitable */
2460 if (start_pfn
< early_node_map
[i
].end_pfn
&&
2461 end_pfn
>= early_node_map
[i
].start_pfn
) {
2462 early_node_map
[i
].start_pfn
= start_pfn
;
2467 /* Check that early_node_map is large enough */
2468 if (i
>= MAX_ACTIVE_REGIONS
) {
2469 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
2470 MAX_ACTIVE_REGIONS
);
2474 early_node_map
[i
].nid
= nid
;
2475 early_node_map
[i
].start_pfn
= start_pfn
;
2476 early_node_map
[i
].end_pfn
= end_pfn
;
2477 nr_nodemap_entries
= i
+ 1;
2481 * shrink_active_range - Shrink an existing registered range of PFNs
2482 * @nid: The node id the range is on that should be shrunk
2483 * @old_end_pfn: The old end PFN of the range
2484 * @new_end_pfn: The new PFN of the range
2486 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2487 * The map is kept at the end physical page range that has already been
2488 * registered with add_active_range(). This function allows an arch to shrink
2489 * an existing registered range.
2491 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
2492 unsigned long new_end_pfn
)
2496 /* Find the old active region end and shrink */
2497 for_each_active_range_index_in_nid(i
, nid
)
2498 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
2499 early_node_map
[i
].end_pfn
= new_end_pfn
;
2505 * remove_all_active_ranges - Remove all currently registered regions
2506 * During discovery, it may be found that a table like SRAT is invalid
2507 * and an alternative discovery method must be used. This function removes
2508 * all currently registered regions.
2510 void __init
remove_all_active_ranges()
2512 memset(early_node_map
, 0, sizeof(early_node_map
));
2513 nr_nodemap_entries
= 0;
2516 /* Compare two active node_active_regions */
2517 static int __init
cmp_node_active_region(const void *a
, const void *b
)
2519 struct node_active_region
*arange
= (struct node_active_region
*)a
;
2520 struct node_active_region
*brange
= (struct node_active_region
*)b
;
2522 /* Done this way to avoid overflows */
2523 if (arange
->start_pfn
> brange
->start_pfn
)
2525 if (arange
->start_pfn
< brange
->start_pfn
)
2531 /* sort the node_map by start_pfn */
2532 static void __init
sort_node_map(void)
2534 sort(early_node_map
, (size_t)nr_nodemap_entries
,
2535 sizeof(struct node_active_region
),
2536 cmp_node_active_region
, NULL
);
2539 /* Find the lowest pfn for a node. This depends on a sorted early_node_map */
2540 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
2544 /* Assuming a sorted map, the first range found has the starting pfn */
2545 for_each_active_range_index_in_nid(i
, nid
)
2546 return early_node_map
[i
].start_pfn
;
2548 printk(KERN_WARNING
"Could not find start_pfn for node %lu\n", nid
);
2553 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2555 * It returns the minimum PFN based on information provided via
2556 * add_active_range()
2558 unsigned long __init
find_min_pfn_with_active_regions(void)
2560 return find_min_pfn_for_node(MAX_NUMNODES
);
2564 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2566 * It returns the maximum PFN based on information provided via
2567 * add_active_range()
2569 unsigned long __init
find_max_pfn_with_active_regions(void)
2572 unsigned long max_pfn
= 0;
2574 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2575 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
2581 * free_area_init_nodes - Initialise all pg_data_t and zone data
2582 * @arch_max_dma_pfn: The maximum PFN usable for ZONE_DMA
2583 * @arch_max_dma32_pfn: The maximum PFN usable for ZONE_DMA32
2584 * @arch_max_low_pfn: The maximum PFN usable for ZONE_NORMAL
2585 * @arch_max_high_pfn: The maximum PFN usable for ZONE_HIGHMEM
2587 * This will call free_area_init_node() for each active node in the system.
2588 * Using the page ranges provided by add_active_range(), the size of each
2589 * zone in each node and their holes is calculated. If the maximum PFN
2590 * between two adjacent zones match, it is assumed that the zone is empty.
2591 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2592 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2593 * starts where the previous one ended. For example, ZONE_DMA32 starts
2594 * at arch_max_dma_pfn.
2596 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
2601 /* Record where the zone boundaries are */
2602 memset(arch_zone_lowest_possible_pfn
, 0,
2603 sizeof(arch_zone_lowest_possible_pfn
));
2604 memset(arch_zone_highest_possible_pfn
, 0,
2605 sizeof(arch_zone_highest_possible_pfn
));
2606 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
2607 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
2608 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
2609 arch_zone_lowest_possible_pfn
[i
] =
2610 arch_zone_highest_possible_pfn
[i
-1];
2611 arch_zone_highest_possible_pfn
[i
] =
2612 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
2615 /* Regions in the early_node_map can be in any order */
2618 /* Print out the zone ranges */
2619 printk("Zone PFN ranges:\n");
2620 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2621 printk(" %-8s %8lu -> %8lu\n",
2623 arch_zone_lowest_possible_pfn
[i
],
2624 arch_zone_highest_possible_pfn
[i
]);
2626 /* Print out the early_node_map[] */
2627 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
2628 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2629 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
2630 early_node_map
[i
].start_pfn
,
2631 early_node_map
[i
].end_pfn
);
2633 /* Initialise every node */
2634 for_each_online_node(nid
) {
2635 pg_data_t
*pgdat
= NODE_DATA(nid
);
2636 free_area_init_node(nid
, pgdat
, NULL
,
2637 find_min_pfn_for_node(nid
), NULL
);
2640 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2643 * set_dma_reserve - Account the specified number of pages reserved in ZONE_DMA
2644 * @new_dma_reserve - The number of pages to mark reserved
2646 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2647 * In the DMA zone, a significant percentage may be consumed by kernel image
2648 * and other unfreeable allocations which can skew the watermarks badly. This
2649 * function may optionally be used to account for unfreeable pages in
2650 * ZONE_DMA. The effect will be lower watermarks and smaller per-cpu batchsize
2652 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
2654 dma_reserve
= new_dma_reserve
;
2657 #ifndef CONFIG_NEED_MULTIPLE_NODES
2658 static bootmem_data_t contig_bootmem_data
;
2659 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2661 EXPORT_SYMBOL(contig_page_data
);
2664 void __init
free_area_init(unsigned long *zones_size
)
2666 free_area_init_node(0, NODE_DATA(0), zones_size
,
2667 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2670 #ifdef CONFIG_HOTPLUG_CPU
2671 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2672 unsigned long action
, void *hcpu
)
2674 int cpu
= (unsigned long)hcpu
;
2676 if (action
== CPU_DEAD
) {
2677 local_irq_disable();
2679 vm_events_fold_cpu(cpu
);
2681 refresh_cpu_vm_stats(cpu
);
2685 #endif /* CONFIG_HOTPLUG_CPU */
2687 void __init
page_alloc_init(void)
2689 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2693 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2694 * or min_free_kbytes changes.
2696 static void calculate_totalreserve_pages(void)
2698 struct pglist_data
*pgdat
;
2699 unsigned long reserve_pages
= 0;
2700 enum zone_type i
, j
;
2702 for_each_online_pgdat(pgdat
) {
2703 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2704 struct zone
*zone
= pgdat
->node_zones
+ i
;
2705 unsigned long max
= 0;
2707 /* Find valid and maximum lowmem_reserve in the zone */
2708 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
2709 if (zone
->lowmem_reserve
[j
] > max
)
2710 max
= zone
->lowmem_reserve
[j
];
2713 /* we treat pages_high as reserved pages. */
2714 max
+= zone
->pages_high
;
2716 if (max
> zone
->present_pages
)
2717 max
= zone
->present_pages
;
2718 reserve_pages
+= max
;
2721 totalreserve_pages
= reserve_pages
;
2725 * setup_per_zone_lowmem_reserve - called whenever
2726 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2727 * has a correct pages reserved value, so an adequate number of
2728 * pages are left in the zone after a successful __alloc_pages().
2730 static void setup_per_zone_lowmem_reserve(void)
2732 struct pglist_data
*pgdat
;
2733 enum zone_type j
, idx
;
2735 for_each_online_pgdat(pgdat
) {
2736 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2737 struct zone
*zone
= pgdat
->node_zones
+ j
;
2738 unsigned long present_pages
= zone
->present_pages
;
2740 zone
->lowmem_reserve
[j
] = 0;
2744 struct zone
*lower_zone
;
2748 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2749 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2751 lower_zone
= pgdat
->node_zones
+ idx
;
2752 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2753 sysctl_lowmem_reserve_ratio
[idx
];
2754 present_pages
+= lower_zone
->present_pages
;
2759 /* update totalreserve_pages */
2760 calculate_totalreserve_pages();
2764 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2765 * that the pages_{min,low,high} values for each zone are set correctly
2766 * with respect to min_free_kbytes.
2768 void setup_per_zone_pages_min(void)
2770 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2771 unsigned long lowmem_pages
= 0;
2773 unsigned long flags
;
2775 /* Calculate total number of !ZONE_HIGHMEM pages */
2776 for_each_zone(zone
) {
2777 if (!is_highmem(zone
))
2778 lowmem_pages
+= zone
->present_pages
;
2781 for_each_zone(zone
) {
2784 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2785 tmp
= (u64
)pages_min
* zone
->present_pages
;
2786 do_div(tmp
, lowmem_pages
);
2787 if (is_highmem(zone
)) {
2789 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2790 * need highmem pages, so cap pages_min to a small
2793 * The (pages_high-pages_low) and (pages_low-pages_min)
2794 * deltas controls asynch page reclaim, and so should
2795 * not be capped for highmem.
2799 min_pages
= zone
->present_pages
/ 1024;
2800 if (min_pages
< SWAP_CLUSTER_MAX
)
2801 min_pages
= SWAP_CLUSTER_MAX
;
2802 if (min_pages
> 128)
2804 zone
->pages_min
= min_pages
;
2807 * If it's a lowmem zone, reserve a number of pages
2808 * proportionate to the zone's size.
2810 zone
->pages_min
= tmp
;
2813 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
2814 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
2815 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2818 /* update totalreserve_pages */
2819 calculate_totalreserve_pages();
2823 * Initialise min_free_kbytes.
2825 * For small machines we want it small (128k min). For large machines
2826 * we want it large (64MB max). But it is not linear, because network
2827 * bandwidth does not increase linearly with machine size. We use
2829 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2830 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2846 static int __init
init_per_zone_pages_min(void)
2848 unsigned long lowmem_kbytes
;
2850 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2852 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2853 if (min_free_kbytes
< 128)
2854 min_free_kbytes
= 128;
2855 if (min_free_kbytes
> 65536)
2856 min_free_kbytes
= 65536;
2857 setup_per_zone_pages_min();
2858 setup_per_zone_lowmem_reserve();
2861 module_init(init_per_zone_pages_min
)
2864 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2865 * that we can call two helper functions whenever min_free_kbytes
2868 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2869 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2871 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2872 setup_per_zone_pages_min();
2877 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
2878 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2883 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2888 zone
->min_unmapped_pages
= (zone
->present_pages
*
2889 sysctl_min_unmapped_ratio
) / 100;
2893 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
2894 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2899 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2904 zone
->min_slab_pages
= (zone
->present_pages
*
2905 sysctl_min_slab_ratio
) / 100;
2911 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2912 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2913 * whenever sysctl_lowmem_reserve_ratio changes.
2915 * The reserve ratio obviously has absolutely no relation with the
2916 * pages_min watermarks. The lowmem reserve ratio can only make sense
2917 * if in function of the boot time zone sizes.
2919 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2920 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2922 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2923 setup_per_zone_lowmem_reserve();
2928 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2929 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2930 * can have before it gets flushed back to buddy allocator.
2933 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
2934 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2940 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2941 if (!write
|| (ret
== -EINVAL
))
2943 for_each_zone(zone
) {
2944 for_each_online_cpu(cpu
) {
2946 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
2947 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
2953 int hashdist
= HASHDIST_DEFAULT
;
2956 static int __init
set_hashdist(char *str
)
2960 hashdist
= simple_strtoul(str
, &str
, 0);
2963 __setup("hashdist=", set_hashdist
);
2967 * allocate a large system hash table from bootmem
2968 * - it is assumed that the hash table must contain an exact power-of-2
2969 * quantity of entries
2970 * - limit is the number of hash buckets, not the total allocation size
2972 void *__init
alloc_large_system_hash(const char *tablename
,
2973 unsigned long bucketsize
,
2974 unsigned long numentries
,
2977 unsigned int *_hash_shift
,
2978 unsigned int *_hash_mask
,
2979 unsigned long limit
)
2981 unsigned long long max
= limit
;
2982 unsigned long log2qty
, size
;
2985 /* allow the kernel cmdline to have a say */
2987 /* round applicable memory size up to nearest megabyte */
2988 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2989 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2990 numentries
>>= 20 - PAGE_SHIFT
;
2991 numentries
<<= 20 - PAGE_SHIFT
;
2993 /* limit to 1 bucket per 2^scale bytes of low memory */
2994 if (scale
> PAGE_SHIFT
)
2995 numentries
>>= (scale
- PAGE_SHIFT
);
2997 numentries
<<= (PAGE_SHIFT
- scale
);
2999 numentries
= roundup_pow_of_two(numentries
);
3001 /* limit allocation size to 1/16 total memory by default */
3003 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
3004 do_div(max
, bucketsize
);
3007 if (numentries
> max
)
3010 log2qty
= long_log2(numentries
);
3013 size
= bucketsize
<< log2qty
;
3014 if (flags
& HASH_EARLY
)
3015 table
= alloc_bootmem(size
);
3017 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
3019 unsigned long order
;
3020 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
3022 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
3024 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
3027 panic("Failed to allocate %s hash table\n", tablename
);
3029 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3032 long_log2(size
) - PAGE_SHIFT
,
3036 *_hash_shift
= log2qty
;
3038 *_hash_mask
= (1 << log2qty
) - 1;
3043 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
3044 struct page
*pfn_to_page(unsigned long pfn
)
3046 return __pfn_to_page(pfn
);
3048 unsigned long page_to_pfn(struct page
*page
)
3050 return __page_to_pfn(page
);
3052 EXPORT_SYMBOL(pfn_to_page
);
3053 EXPORT_SYMBOL(page_to_pfn
);
3054 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */