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>
43 #include <linux/fault-inject.h>
45 #include <asm/tlbflush.h>
46 #include <asm/div64.h>
50 * Array of node states.
52 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
53 [N_POSSIBLE
] = NODE_MASK_ALL
,
54 [N_ONLINE
] = { { [0] = 1UL } },
56 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
58 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
60 [N_CPU
] = { { [0] = 1UL } },
63 EXPORT_SYMBOL(node_states
);
65 unsigned long totalram_pages __read_mostly
;
66 unsigned long totalreserve_pages __read_mostly
;
68 int percpu_pagelist_fraction
;
70 static void __free_pages_ok(struct page
*page
, unsigned int order
);
73 * results with 256, 32 in the lowmem_reserve sysctl:
74 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
75 * 1G machine -> (16M dma, 784M normal, 224M high)
76 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
77 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
78 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
80 * TBD: should special case ZONE_DMA32 machines here - in those we normally
81 * don't need any ZONE_NORMAL reservation
83 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
84 #ifdef CONFIG_ZONE_DMA
87 #ifdef CONFIG_ZONE_DMA32
96 EXPORT_SYMBOL(totalram_pages
);
98 static char * const zone_names
[MAX_NR_ZONES
] = {
99 #ifdef CONFIG_ZONE_DMA
102 #ifdef CONFIG_ZONE_DMA32
106 #ifdef CONFIG_HIGHMEM
112 int min_free_kbytes
= 1024;
114 unsigned long __meminitdata nr_kernel_pages
;
115 unsigned long __meminitdata nr_all_pages
;
116 static unsigned long __meminitdata dma_reserve
;
118 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
120 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
121 * ranges of memory (RAM) that may be registered with add_active_range().
122 * Ranges passed to add_active_range() will be merged if possible
123 * so the number of times add_active_range() can be called is
124 * related to the number of nodes and the number of holes
126 #ifdef CONFIG_MAX_ACTIVE_REGIONS
127 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
128 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
130 #if MAX_NUMNODES >= 32
131 /* If there can be many nodes, allow up to 50 holes per node */
132 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
134 /* By default, allow up to 256 distinct regions */
135 #define MAX_ACTIVE_REGIONS 256
139 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
140 static int __meminitdata nr_nodemap_entries
;
141 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
142 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
143 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
144 static unsigned long __meminitdata node_boundary_start_pfn
[MAX_NUMNODES
];
145 static unsigned long __meminitdata node_boundary_end_pfn
[MAX_NUMNODES
];
146 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
147 unsigned long __initdata required_kernelcore
;
148 unsigned long __initdata required_movablecore
;
149 unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
151 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
153 EXPORT_SYMBOL(movable_zone
);
154 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
157 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
158 EXPORT_SYMBOL(nr_node_ids
);
161 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
162 int page_group_by_mobility_disabled __read_mostly
;
164 static inline int get_pageblock_migratetype(struct page
*page
)
166 if (unlikely(page_group_by_mobility_disabled
))
167 return MIGRATE_UNMOVABLE
;
169 return get_pageblock_flags_group(page
, PB_migrate
, PB_migrate_end
);
172 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
174 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
175 PB_migrate
, PB_migrate_end
);
178 static inline int allocflags_to_migratetype(gfp_t gfp_flags
, int order
)
180 WARN_ON((gfp_flags
& GFP_MOVABLE_MASK
) == GFP_MOVABLE_MASK
);
182 if (unlikely(page_group_by_mobility_disabled
))
183 return MIGRATE_UNMOVABLE
;
185 /* Cluster high-order atomic allocations together */
186 if (unlikely(order
> 0) &&
187 (!(gfp_flags
& __GFP_WAIT
) || in_interrupt()))
188 return MIGRATE_HIGHATOMIC
;
190 /* Cluster based on mobility */
191 return (((gfp_flags
& __GFP_MOVABLE
) != 0) << 1) |
192 ((gfp_flags
& __GFP_RECLAIMABLE
) != 0);
196 static inline int get_pageblock_migratetype(struct page
*page
)
198 return MIGRATE_UNMOVABLE
;
201 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
205 static inline int allocflags_to_migratetype(gfp_t gfp_flags
, int order
)
207 return MIGRATE_UNMOVABLE
;
209 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
211 #ifdef CONFIG_DEBUG_VM
212 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
216 unsigned long pfn
= page_to_pfn(page
);
219 seq
= zone_span_seqbegin(zone
);
220 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
222 else if (pfn
< zone
->zone_start_pfn
)
224 } while (zone_span_seqretry(zone
, seq
));
229 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
231 if (!pfn_valid_within(page_to_pfn(page
)))
233 if (zone
!= page_zone(page
))
239 * Temporary debugging check for pages not lying within a given zone.
241 static int bad_range(struct zone
*zone
, struct page
*page
)
243 if (page_outside_zone_boundaries(zone
, page
))
245 if (!page_is_consistent(zone
, page
))
251 static inline int bad_range(struct zone
*zone
, struct page
*page
)
257 static void bad_page(struct page
*page
)
259 printk(KERN_EMERG
"Bad page state in process '%s'\n"
260 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
261 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
262 KERN_EMERG
"Backtrace:\n",
263 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
264 (unsigned long)page
->flags
, page
->mapping
,
265 page_mapcount(page
), page_count(page
));
267 page
->flags
&= ~(1 << PG_lru
|
277 set_page_count(page
, 0);
278 reset_page_mapcount(page
);
279 page
->mapping
= NULL
;
280 add_taint(TAINT_BAD_PAGE
);
284 * Higher-order pages are called "compound pages". They are structured thusly:
286 * The first PAGE_SIZE page is called the "head page".
288 * The remaining PAGE_SIZE pages are called "tail pages".
290 * All pages have PG_compound set. All pages have their ->private pointing at
291 * the head page (even the head page has this).
293 * The first tail page's ->lru.next holds the address of the compound page's
294 * put_page() function. Its ->lru.prev holds the order of allocation.
295 * This usage means that zero-order pages may not be compound.
298 static void free_compound_page(struct page
*page
)
300 __free_pages_ok(page
, compound_order(page
));
303 static void prep_compound_page(struct page
*page
, unsigned long order
)
306 int nr_pages
= 1 << order
;
308 set_compound_page_dtor(page
, free_compound_page
);
309 set_compound_order(page
, order
);
311 for (i
= 1; i
< nr_pages
; i
++) {
312 struct page
*p
= page
+ i
;
315 p
->first_page
= page
;
319 static void destroy_compound_page(struct page
*page
, unsigned long order
)
322 int nr_pages
= 1 << order
;
324 if (unlikely(compound_order(page
) != order
))
327 if (unlikely(!PageHead(page
)))
329 __ClearPageHead(page
);
330 for (i
= 1; i
< nr_pages
; i
++) {
331 struct page
*p
= page
+ i
;
333 if (unlikely(!PageTail(p
) |
334 (p
->first_page
!= page
)))
340 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
344 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
346 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
347 * and __GFP_HIGHMEM from hard or soft interrupt context.
349 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
350 for (i
= 0; i
< (1 << order
); i
++)
351 clear_highpage(page
+ i
);
355 * function for dealing with page's order in buddy system.
356 * zone->lock is already acquired when we use these.
357 * So, we don't need atomic page->flags operations here.
359 static inline unsigned long page_order(struct page
*page
)
361 return page_private(page
);
364 static inline void set_page_order(struct page
*page
, int order
)
366 set_page_private(page
, order
);
367 __SetPageBuddy(page
);
370 static inline void rmv_page_order(struct page
*page
)
372 __ClearPageBuddy(page
);
373 set_page_private(page
, 0);
377 * Locate the struct page for both the matching buddy in our
378 * pair (buddy1) and the combined O(n+1) page they form (page).
380 * 1) Any buddy B1 will have an order O twin B2 which satisfies
381 * the following equation:
383 * For example, if the starting buddy (buddy2) is #8 its order
385 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
387 * 2) Any buddy B will have an order O+1 parent P which
388 * satisfies the following equation:
391 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
393 static inline struct page
*
394 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
396 unsigned long buddy_idx
= page_idx
^ (1 << order
);
398 return page
+ (buddy_idx
- page_idx
);
401 static inline unsigned long
402 __find_combined_index(unsigned long page_idx
, unsigned int order
)
404 return (page_idx
& ~(1 << order
));
408 * This function checks whether a page is free && is the buddy
409 * we can do coalesce a page and its buddy if
410 * (a) the buddy is not in a hole &&
411 * (b) the buddy is in the buddy system &&
412 * (c) a page and its buddy have the same order &&
413 * (d) a page and its buddy are in the same zone.
415 * For recording whether a page is in the buddy system, we use PG_buddy.
416 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
418 * For recording page's order, we use page_private(page).
420 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
423 if (!pfn_valid_within(page_to_pfn(buddy
)))
426 if (page_zone_id(page
) != page_zone_id(buddy
))
429 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
430 BUG_ON(page_count(buddy
) != 0);
437 * Freeing function for a buddy system allocator.
439 * The concept of a buddy system is to maintain direct-mapped table
440 * (containing bit values) for memory blocks of various "orders".
441 * The bottom level table contains the map for the smallest allocatable
442 * units of memory (here, pages), and each level above it describes
443 * pairs of units from the levels below, hence, "buddies".
444 * At a high level, all that happens here is marking the table entry
445 * at the bottom level available, and propagating the changes upward
446 * as necessary, plus some accounting needed to play nicely with other
447 * parts of the VM system.
448 * At each level, we keep a list of pages, which are heads of continuous
449 * free pages of length of (1 << order) and marked with PG_buddy. Page's
450 * order is recorded in page_private(page) field.
451 * So when we are allocating or freeing one, we can derive the state of the
452 * other. That is, if we allocate a small block, and both were
453 * free, the remainder of the region must be split into blocks.
454 * If a block is freed, and its buddy is also free, then this
455 * triggers coalescing into a block of larger size.
460 static inline void __free_one_page(struct page
*page
,
461 struct zone
*zone
, unsigned int order
)
463 unsigned long page_idx
;
464 int order_size
= 1 << order
;
465 int migratetype
= get_pageblock_migratetype(page
);
467 if (unlikely(PageCompound(page
)))
468 destroy_compound_page(page
, order
);
470 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
472 VM_BUG_ON(page_idx
& (order_size
- 1));
473 VM_BUG_ON(bad_range(zone
, page
));
475 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
476 while (order
< MAX_ORDER
-1) {
477 unsigned long combined_idx
;
480 buddy
= __page_find_buddy(page
, page_idx
, order
);
481 if (!page_is_buddy(page
, buddy
, order
))
482 break; /* Move the buddy up one level. */
484 list_del(&buddy
->lru
);
485 zone
->free_area
[order
].nr_free
--;
486 rmv_page_order(buddy
);
487 combined_idx
= __find_combined_index(page_idx
, order
);
488 page
= page
+ (combined_idx
- page_idx
);
489 page_idx
= combined_idx
;
492 set_page_order(page
, order
);
494 &zone
->free_area
[order
].free_list
[migratetype
]);
495 zone
->free_area
[order
].nr_free
++;
498 static inline int free_pages_check(struct page
*page
)
500 if (unlikely(page_mapcount(page
) |
501 (page
->mapping
!= NULL
) |
502 (page_count(page
) != 0) |
515 __ClearPageDirty(page
);
517 * For now, we report if PG_reserved was found set, but do not
518 * clear it, and do not free the page. But we shall soon need
519 * to do more, for when the ZERO_PAGE count wraps negative.
521 return PageReserved(page
);
525 * Frees a list of pages.
526 * Assumes all pages on list are in same zone, and of same order.
527 * count is the number of pages to free.
529 * If the zone was previously in an "all pages pinned" state then look to
530 * see if this freeing clears that state.
532 * And clear the zone's pages_scanned counter, to hold off the "all pages are
533 * pinned" detection logic.
535 static void free_pages_bulk(struct zone
*zone
, int count
,
536 struct list_head
*list
, int order
)
538 spin_lock(&zone
->lock
);
539 zone
->all_unreclaimable
= 0;
540 zone
->pages_scanned
= 0;
544 VM_BUG_ON(list_empty(list
));
545 page
= list_entry(list
->prev
, struct page
, lru
);
546 /* have to delete it as __free_one_page list manipulates */
547 list_del(&page
->lru
);
548 __free_one_page(page
, zone
, order
);
550 spin_unlock(&zone
->lock
);
553 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
555 spin_lock(&zone
->lock
);
556 zone
->all_unreclaimable
= 0;
557 zone
->pages_scanned
= 0;
558 __free_one_page(page
, zone
, order
);
559 spin_unlock(&zone
->lock
);
562 static void __free_pages_ok(struct page
*page
, unsigned int order
)
568 for (i
= 0 ; i
< (1 << order
) ; ++i
)
569 reserved
+= free_pages_check(page
+ i
);
573 if (!PageHighMem(page
))
574 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
575 arch_free_page(page
, order
);
576 kernel_map_pages(page
, 1 << order
, 0);
578 local_irq_save(flags
);
579 __count_vm_events(PGFREE
, 1 << order
);
580 free_one_page(page_zone(page
), page
, order
);
581 local_irq_restore(flags
);
585 * permit the bootmem allocator to evade page validation on high-order frees
587 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
590 __ClearPageReserved(page
);
591 set_page_count(page
, 0);
592 set_page_refcounted(page
);
598 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
599 struct page
*p
= &page
[loop
];
601 if (loop
+ 1 < BITS_PER_LONG
)
603 __ClearPageReserved(p
);
604 set_page_count(p
, 0);
607 set_page_refcounted(page
);
608 __free_pages(page
, order
);
614 * The order of subdivision here is critical for the IO subsystem.
615 * Please do not alter this order without good reasons and regression
616 * testing. Specifically, as large blocks of memory are subdivided,
617 * the order in which smaller blocks are delivered depends on the order
618 * they're subdivided in this function. This is the primary factor
619 * influencing the order in which pages are delivered to the IO
620 * subsystem according to empirical testing, and this is also justified
621 * by considering the behavior of a buddy system containing a single
622 * large block of memory acted on by a series of small allocations.
623 * This behavior is a critical factor in sglist merging's success.
627 static inline void expand(struct zone
*zone
, struct page
*page
,
628 int low
, int high
, struct free_area
*area
,
631 unsigned long size
= 1 << high
;
637 VM_BUG_ON(bad_range(zone
, &page
[size
]));
638 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
640 set_page_order(&page
[size
], high
);
645 * This page is about to be returned from the page allocator
647 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
649 if (unlikely(page_mapcount(page
) |
650 (page
->mapping
!= NULL
) |
651 (page_count(page
) != 0) |
666 * For now, we report if PG_reserved was found set, but do not
667 * clear it, and do not allocate the page: as a safety net.
669 if (PageReserved(page
))
672 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_readahead
|
673 1 << PG_referenced
| 1 << PG_arch_1
|
674 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
);
675 set_page_private(page
, 0);
676 set_page_refcounted(page
);
678 arch_alloc_page(page
, order
);
679 kernel_map_pages(page
, 1 << order
, 1);
681 if (gfp_flags
& __GFP_ZERO
)
682 prep_zero_page(page
, order
, gfp_flags
);
684 if (order
&& (gfp_flags
& __GFP_COMP
))
685 prep_compound_page(page
, order
);
690 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
692 * This array describes the order lists are fallen back to when
693 * the free lists for the desirable migrate type are depleted
695 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
696 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_HIGHATOMIC
},
697 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_HIGHATOMIC
},
698 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
,MIGRATE_HIGHATOMIC
},
699 [MIGRATE_HIGHATOMIC
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
,MIGRATE_MOVABLE
},
703 * Move the free pages in a range to the free lists of the requested type.
704 * Note that start_page and end_pages are not aligned in a MAX_ORDER_NR_PAGES
705 * boundary. If alignment is required, use move_freepages_block()
707 int move_freepages(struct zone
*zone
,
708 struct page
*start_page
, struct page
*end_page
,
713 int blocks_moved
= 0;
715 #ifndef CONFIG_HOLES_IN_ZONE
717 * page_zone is not safe to call in this context when
718 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
719 * anyway as we check zone boundaries in move_freepages_block().
720 * Remove at a later date when no bug reports exist related to
721 * CONFIG_PAGE_GROUP_BY_MOBILITY
723 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
726 for (page
= start_page
; page
<= end_page
;) {
727 if (!pfn_valid_within(page_to_pfn(page
))) {
732 if (!PageBuddy(page
)) {
737 order
= page_order(page
);
738 list_del(&page
->lru
);
740 &zone
->free_area
[order
].free_list
[migratetype
]);
748 int move_freepages_block(struct zone
*zone
, struct page
*page
, int migratetype
)
750 unsigned long start_pfn
, end_pfn
;
751 struct page
*start_page
, *end_page
;
753 start_pfn
= page_to_pfn(page
);
754 start_pfn
= start_pfn
& ~(MAX_ORDER_NR_PAGES
-1);
755 start_page
= pfn_to_page(start_pfn
);
756 end_page
= start_page
+ MAX_ORDER_NR_PAGES
- 1;
757 end_pfn
= start_pfn
+ MAX_ORDER_NR_PAGES
- 1;
759 /* Do not cross zone boundaries */
760 if (start_pfn
< zone
->zone_start_pfn
)
762 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
765 return move_freepages(zone
, start_page
, end_page
, migratetype
);
768 /* Remove an element from the buddy allocator from the fallback list */
769 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
770 int start_migratetype
)
772 struct free_area
* area
;
776 int nonatomic_fallback_atomic
= 0;
779 /* Find the largest possible block of pages in the other list */
780 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
782 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
783 migratetype
= fallbacks
[start_migratetype
][i
];
786 * Make it hard to fallback to blocks used for
787 * high-order atomic allocations
789 if (migratetype
== MIGRATE_HIGHATOMIC
&&
790 start_migratetype
!= MIGRATE_UNMOVABLE
&&
791 !nonatomic_fallback_atomic
)
794 area
= &(zone
->free_area
[current_order
]);
795 if (list_empty(&area
->free_list
[migratetype
]))
798 page
= list_entry(area
->free_list
[migratetype
].next
,
803 * If breaking a large block of pages, move all free
804 * pages to the preferred allocation list
806 if (unlikely(current_order
>= MAX_ORDER
/ 2)) {
807 migratetype
= start_migratetype
;
808 move_freepages_block(zone
, page
, migratetype
);
811 /* Remove the page from the freelists */
812 list_del(&page
->lru
);
813 rmv_page_order(page
);
814 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
817 if (current_order
== MAX_ORDER
- 1)
818 set_pageblock_migratetype(page
,
821 expand(zone
, page
, order
, current_order
, area
, migratetype
);
826 /* Allow fallback to high-order atomic blocks if memory is that low */
827 if (!nonatomic_fallback_atomic
) {
828 nonatomic_fallback_atomic
= 1;
835 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
836 int start_migratetype
)
840 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
843 * Do the hard work of removing an element from the buddy allocator.
844 * Call me with the zone->lock already held.
846 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
849 struct free_area
* area
;
850 unsigned int current_order
;
853 /* Find a page of the appropriate size in the preferred list */
854 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
855 area
= &(zone
->free_area
[current_order
]);
856 if (list_empty(&area
->free_list
[migratetype
]))
859 page
= list_entry(area
->free_list
[migratetype
].next
,
861 list_del(&page
->lru
);
862 rmv_page_order(page
);
864 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
865 expand(zone
, page
, order
, current_order
, area
, migratetype
);
869 page
= __rmqueue_fallback(zone
, order
, migratetype
);
877 * Obtain a specified number of elements from the buddy allocator, all under
878 * a single hold of the lock, for efficiency. Add them to the supplied list.
879 * Returns the number of new pages which were placed at *list.
881 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
882 unsigned long count
, struct list_head
*list
,
887 spin_lock(&zone
->lock
);
888 for (i
= 0; i
< count
; ++i
) {
889 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
890 if (unlikely(page
== NULL
))
892 list_add(&page
->lru
, list
);
893 set_page_private(page
, migratetype
);
895 spin_unlock(&zone
->lock
);
901 * Called from the vmstat counter updater to drain pagesets of this
902 * currently executing processor on remote nodes after they have
905 * Note that this function must be called with the thread pinned to
906 * a single processor.
908 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
913 local_irq_save(flags
);
914 if (pcp
->count
>= pcp
->batch
)
915 to_drain
= pcp
->batch
;
917 to_drain
= pcp
->count
;
918 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
919 pcp
->count
-= to_drain
;
920 local_irq_restore(flags
);
924 static void __drain_pages(unsigned int cpu
)
930 for_each_zone(zone
) {
931 struct per_cpu_pageset
*pset
;
933 if (!populated_zone(zone
))
936 pset
= zone_pcp(zone
, cpu
);
937 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
938 struct per_cpu_pages
*pcp
;
941 local_irq_save(flags
);
942 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
944 local_irq_restore(flags
);
949 #ifdef CONFIG_HIBERNATION
951 void mark_free_pages(struct zone
*zone
)
953 unsigned long pfn
, max_zone_pfn
;
956 struct list_head
*curr
;
958 if (!zone
->spanned_pages
)
961 spin_lock_irqsave(&zone
->lock
, flags
);
963 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
964 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
965 if (pfn_valid(pfn
)) {
966 struct page
*page
= pfn_to_page(pfn
);
968 if (!swsusp_page_is_forbidden(page
))
969 swsusp_unset_page_free(page
);
972 for_each_migratetype_order(order
, t
) {
973 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
976 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
977 for (i
= 0; i
< (1UL << order
); i
++)
978 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
981 spin_unlock_irqrestore(&zone
->lock
, flags
);
983 #endif /* CONFIG_PM */
985 #if defined(CONFIG_HIBERNATION) || defined(CONFIG_PAGE_GROUP_BY_MOBILITY)
987 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
989 void drain_local_pages(void)
993 local_irq_save(flags
);
994 __drain_pages(smp_processor_id());
995 local_irq_restore(flags
);
998 void smp_drain_local_pages(void *arg
)
1000 drain_local_pages();
1004 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1006 void drain_all_local_pages(void)
1008 unsigned long flags
;
1010 local_irq_save(flags
);
1011 __drain_pages(smp_processor_id());
1012 local_irq_restore(flags
);
1014 smp_call_function(smp_drain_local_pages
, NULL
, 0, 1);
1017 void drain_all_local_pages(void) {}
1018 #endif /* CONFIG_HIBERNATION || CONFIG_PAGE_GROUP_BY_MOBILITY */
1021 * Free a 0-order page
1023 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
1025 struct zone
*zone
= page_zone(page
);
1026 struct per_cpu_pages
*pcp
;
1027 unsigned long flags
;
1030 page
->mapping
= NULL
;
1031 if (free_pages_check(page
))
1034 if (!PageHighMem(page
))
1035 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1036 arch_free_page(page
, 0);
1037 kernel_map_pages(page
, 1, 0);
1039 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
1040 local_irq_save(flags
);
1041 __count_vm_event(PGFREE
);
1042 list_add(&page
->lru
, &pcp
->list
);
1043 set_page_private(page
, get_pageblock_migratetype(page
));
1045 if (pcp
->count
>= pcp
->high
) {
1046 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1047 pcp
->count
-= pcp
->batch
;
1049 local_irq_restore(flags
);
1053 void fastcall
free_hot_page(struct page
*page
)
1055 free_hot_cold_page(page
, 0);
1058 void fastcall
free_cold_page(struct page
*page
)
1060 free_hot_cold_page(page
, 1);
1064 * split_page takes a non-compound higher-order page, and splits it into
1065 * n (1<<order) sub-pages: page[0..n]
1066 * Each sub-page must be freed individually.
1068 * Note: this is probably too low level an operation for use in drivers.
1069 * Please consult with lkml before using this in your driver.
1071 void split_page(struct page
*page
, unsigned int order
)
1075 VM_BUG_ON(PageCompound(page
));
1076 VM_BUG_ON(!page_count(page
));
1077 for (i
= 1; i
< (1 << order
); i
++)
1078 set_page_refcounted(page
+ i
);
1082 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1083 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1086 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
1087 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1089 unsigned long flags
;
1091 int cold
= !!(gfp_flags
& __GFP_COLD
);
1093 int migratetype
= allocflags_to_migratetype(gfp_flags
, order
);
1097 if (likely(order
== 0)) {
1098 struct per_cpu_pages
*pcp
;
1100 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
1101 local_irq_save(flags
);
1103 pcp
->count
= rmqueue_bulk(zone
, 0,
1104 pcp
->batch
, &pcp
->list
, migratetype
);
1105 if (unlikely(!pcp
->count
))
1109 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
1110 /* Find a page of the appropriate migrate type */
1111 list_for_each_entry(page
, &pcp
->list
, lru
)
1112 if (page_private(page
) == migratetype
)
1115 /* Allocate more to the pcp list if necessary */
1116 if (unlikely(&page
->lru
== &pcp
->list
)) {
1117 pcp
->count
+= rmqueue_bulk(zone
, 0,
1118 pcp
->batch
, &pcp
->list
, migratetype
);
1119 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1122 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1123 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
1125 list_del(&page
->lru
);
1128 spin_lock_irqsave(&zone
->lock
, flags
);
1129 page
= __rmqueue(zone
, order
, migratetype
);
1130 spin_unlock(&zone
->lock
);
1135 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1136 zone_statistics(zonelist
, zone
);
1137 local_irq_restore(flags
);
1140 VM_BUG_ON(bad_range(zone
, page
));
1141 if (prep_new_page(page
, order
, gfp_flags
))
1146 local_irq_restore(flags
);
1151 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1152 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1153 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1154 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1155 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1156 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1157 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1159 #ifdef CONFIG_FAIL_PAGE_ALLOC
1161 static struct fail_page_alloc_attr
{
1162 struct fault_attr attr
;
1164 u32 ignore_gfp_highmem
;
1165 u32 ignore_gfp_wait
;
1168 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1170 struct dentry
*ignore_gfp_highmem_file
;
1171 struct dentry
*ignore_gfp_wait_file
;
1172 struct dentry
*min_order_file
;
1174 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1176 } fail_page_alloc
= {
1177 .attr
= FAULT_ATTR_INITIALIZER
,
1178 .ignore_gfp_wait
= 1,
1179 .ignore_gfp_highmem
= 1,
1183 static int __init
setup_fail_page_alloc(char *str
)
1185 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1187 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1189 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1191 if (order
< fail_page_alloc
.min_order
)
1193 if (gfp_mask
& __GFP_NOFAIL
)
1195 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1197 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1200 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1203 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1205 static int __init
fail_page_alloc_debugfs(void)
1207 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1211 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1215 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1217 fail_page_alloc
.ignore_gfp_wait_file
=
1218 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1219 &fail_page_alloc
.ignore_gfp_wait
);
1221 fail_page_alloc
.ignore_gfp_highmem_file
=
1222 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1223 &fail_page_alloc
.ignore_gfp_highmem
);
1224 fail_page_alloc
.min_order_file
=
1225 debugfs_create_u32("min-order", mode
, dir
,
1226 &fail_page_alloc
.min_order
);
1228 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1229 !fail_page_alloc
.ignore_gfp_highmem_file
||
1230 !fail_page_alloc
.min_order_file
) {
1232 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1233 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1234 debugfs_remove(fail_page_alloc
.min_order_file
);
1235 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1241 late_initcall(fail_page_alloc_debugfs
);
1243 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1245 #else /* CONFIG_FAIL_PAGE_ALLOC */
1247 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1252 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1255 * Return 1 if free pages are above 'mark'. This takes into account the order
1256 * of the allocation.
1258 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1259 int classzone_idx
, int alloc_flags
)
1261 /* free_pages my go negative - that's OK */
1263 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1266 if (alloc_flags
& ALLOC_HIGH
)
1268 if (alloc_flags
& ALLOC_HARDER
)
1271 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1273 for (o
= 0; o
< order
; o
++) {
1274 /* At the next order, this order's pages become unavailable */
1275 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1277 /* Require fewer higher order pages to be free */
1280 if (free_pages
<= min
)
1288 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1289 * skip over zones that are not allowed by the cpuset, or that have
1290 * been recently (in last second) found to be nearly full. See further
1291 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1292 * that have to skip over alot of full or unallowed zones.
1294 * If the zonelist cache is present in the passed in zonelist, then
1295 * returns a pointer to the allowed node mask (either the current
1296 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1298 * If the zonelist cache is not available for this zonelist, does
1299 * nothing and returns NULL.
1301 * If the fullzones BITMAP in the zonelist cache is stale (more than
1302 * a second since last zap'd) then we zap it out (clear its bits.)
1304 * We hold off even calling zlc_setup, until after we've checked the
1305 * first zone in the zonelist, on the theory that most allocations will
1306 * be satisfied from that first zone, so best to examine that zone as
1307 * quickly as we can.
1309 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1311 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1312 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1314 zlc
= zonelist
->zlcache_ptr
;
1318 if (jiffies
- zlc
->last_full_zap
> 1 * HZ
) {
1319 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1320 zlc
->last_full_zap
= jiffies
;
1323 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1324 &cpuset_current_mems_allowed
:
1325 &node_states
[N_HIGH_MEMORY
];
1326 return allowednodes
;
1330 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1331 * if it is worth looking at further for free memory:
1332 * 1) Check that the zone isn't thought to be full (doesn't have its
1333 * bit set in the zonelist_cache fullzones BITMAP).
1334 * 2) Check that the zones node (obtained from the zonelist_cache
1335 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1336 * Return true (non-zero) if zone is worth looking at further, or
1337 * else return false (zero) if it is not.
1339 * This check -ignores- the distinction between various watermarks,
1340 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1341 * found to be full for any variation of these watermarks, it will
1342 * be considered full for up to one second by all requests, unless
1343 * we are so low on memory on all allowed nodes that we are forced
1344 * into the second scan of the zonelist.
1346 * In the second scan we ignore this zonelist cache and exactly
1347 * apply the watermarks to all zones, even it is slower to do so.
1348 * We are low on memory in the second scan, and should leave no stone
1349 * unturned looking for a free page.
1351 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1352 nodemask_t
*allowednodes
)
1354 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1355 int i
; /* index of *z in zonelist zones */
1356 int n
; /* node that zone *z is on */
1358 zlc
= zonelist
->zlcache_ptr
;
1362 i
= z
- zonelist
->zones
;
1365 /* This zone is worth trying if it is allowed but not full */
1366 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1370 * Given 'z' scanning a zonelist, set the corresponding bit in
1371 * zlc->fullzones, so that subsequent attempts to allocate a page
1372 * from that zone don't waste time re-examining it.
1374 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1376 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1377 int i
; /* index of *z in zonelist zones */
1379 zlc
= zonelist
->zlcache_ptr
;
1383 i
= z
- zonelist
->zones
;
1385 set_bit(i
, zlc
->fullzones
);
1388 #else /* CONFIG_NUMA */
1390 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1395 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1396 nodemask_t
*allowednodes
)
1401 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1404 #endif /* CONFIG_NUMA */
1407 * get_page_from_freelist goes through the zonelist trying to allocate
1410 static struct page
*
1411 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
1412 struct zonelist
*zonelist
, int alloc_flags
)
1415 struct page
*page
= NULL
;
1416 int classzone_idx
= zone_idx(zonelist
->zones
[0]);
1418 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1419 int zlc_active
= 0; /* set if using zonelist_cache */
1420 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1421 enum zone_type highest_zoneidx
= -1; /* Gets set for policy zonelists */
1425 * Scan zonelist, looking for a zone with enough free.
1426 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1428 z
= zonelist
->zones
;
1432 * In NUMA, this could be a policy zonelist which contains
1433 * zones that may not be allowed by the current gfp_mask.
1434 * Check the zone is allowed by the current flags
1436 if (unlikely(alloc_should_filter_zonelist(zonelist
))) {
1437 if (highest_zoneidx
== -1)
1438 highest_zoneidx
= gfp_zone(gfp_mask
);
1439 if (zone_idx(*z
) > highest_zoneidx
)
1443 if (NUMA_BUILD
&& zlc_active
&&
1444 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1447 if ((alloc_flags
& ALLOC_CPUSET
) &&
1448 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1451 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1453 if (alloc_flags
& ALLOC_WMARK_MIN
)
1454 mark
= zone
->pages_min
;
1455 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1456 mark
= zone
->pages_low
;
1458 mark
= zone
->pages_high
;
1459 if (!zone_watermark_ok(zone
, order
, mark
,
1460 classzone_idx
, alloc_flags
)) {
1461 if (!zone_reclaim_mode
||
1462 !zone_reclaim(zone
, gfp_mask
, order
))
1463 goto this_zone_full
;
1467 page
= buffered_rmqueue(zonelist
, zone
, order
, gfp_mask
);
1472 zlc_mark_zone_full(zonelist
, z
);
1474 if (NUMA_BUILD
&& !did_zlc_setup
) {
1475 /* we do zlc_setup after the first zone is tried */
1476 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1480 } while (*(++z
) != NULL
);
1482 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1483 /* Disable zlc cache for second zonelist scan */
1491 * This is the 'heart' of the zoned buddy allocator.
1493 struct page
* fastcall
1494 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
1495 struct zonelist
*zonelist
)
1497 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1500 struct reclaim_state reclaim_state
;
1501 struct task_struct
*p
= current
;
1504 int did_some_progress
;
1506 might_sleep_if(wait
);
1508 if (should_fail_alloc_page(gfp_mask
, order
))
1512 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
1514 if (unlikely(*z
== NULL
)) {
1516 * Happens if we have an empty zonelist as a result of
1517 * GFP_THISNODE being used on a memoryless node
1522 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1523 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1528 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1529 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1530 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1531 * using a larger set of nodes after it has established that the
1532 * allowed per node queues are empty and that nodes are
1535 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1538 for (z
= zonelist
->zones
; *z
; z
++)
1539 wakeup_kswapd(*z
, order
);
1542 * OK, we're below the kswapd watermark and have kicked background
1543 * reclaim. Now things get more complex, so set up alloc_flags according
1544 * to how we want to proceed.
1546 * The caller may dip into page reserves a bit more if the caller
1547 * cannot run direct reclaim, or if the caller has realtime scheduling
1548 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1549 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1551 alloc_flags
= ALLOC_WMARK_MIN
;
1552 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1553 alloc_flags
|= ALLOC_HARDER
;
1554 if (gfp_mask
& __GFP_HIGH
)
1555 alloc_flags
|= ALLOC_HIGH
;
1557 alloc_flags
|= ALLOC_CPUSET
;
1560 * Go through the zonelist again. Let __GFP_HIGH and allocations
1561 * coming from realtime tasks go deeper into reserves.
1563 * This is the last chance, in general, before the goto nopage.
1564 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1565 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1567 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1571 /* This allocation should allow future memory freeing. */
1574 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1575 && !in_interrupt()) {
1576 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1578 /* go through the zonelist yet again, ignoring mins */
1579 page
= get_page_from_freelist(gfp_mask
, order
,
1580 zonelist
, ALLOC_NO_WATERMARKS
);
1583 if (gfp_mask
& __GFP_NOFAIL
) {
1584 congestion_wait(WRITE
, HZ
/50);
1591 /* Atomic allocations - we can't balance anything */
1597 /* We now go into synchronous reclaim */
1598 cpuset_memory_pressure_bump();
1599 p
->flags
|= PF_MEMALLOC
;
1600 reclaim_state
.reclaimed_slab
= 0;
1601 p
->reclaim_state
= &reclaim_state
;
1603 did_some_progress
= try_to_free_pages(zonelist
->zones
, order
, gfp_mask
);
1605 p
->reclaim_state
= NULL
;
1606 p
->flags
&= ~PF_MEMALLOC
;
1611 drain_all_local_pages();
1613 if (likely(did_some_progress
)) {
1614 page
= get_page_from_freelist(gfp_mask
, order
,
1615 zonelist
, alloc_flags
);
1618 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1620 * Go through the zonelist yet one more time, keep
1621 * very high watermark here, this is only to catch
1622 * a parallel oom killing, we must fail if we're still
1623 * under heavy pressure.
1625 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1626 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1630 /* The OOM killer will not help higher order allocs so fail */
1631 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1634 out_of_memory(zonelist
, gfp_mask
, order
);
1639 * Don't let big-order allocations loop unless the caller explicitly
1640 * requests that. Wait for some write requests to complete then retry.
1642 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1643 * <= 3, but that may not be true in other implementations.
1646 if (!(gfp_mask
& __GFP_NORETRY
)) {
1647 if ((order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
1648 (gfp_mask
& __GFP_REPEAT
))
1650 if (gfp_mask
& __GFP_NOFAIL
)
1654 congestion_wait(WRITE
, HZ
/50);
1659 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1660 printk(KERN_WARNING
"%s: page allocation failure."
1661 " order:%d, mode:0x%x\n",
1662 p
->comm
, order
, gfp_mask
);
1670 EXPORT_SYMBOL(__alloc_pages
);
1673 * Common helper functions.
1675 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1678 page
= alloc_pages(gfp_mask
, order
);
1681 return (unsigned long) page_address(page
);
1684 EXPORT_SYMBOL(__get_free_pages
);
1686 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1691 * get_zeroed_page() returns a 32-bit address, which cannot represent
1694 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1696 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1698 return (unsigned long) page_address(page
);
1702 EXPORT_SYMBOL(get_zeroed_page
);
1704 void __pagevec_free(struct pagevec
*pvec
)
1706 int i
= pagevec_count(pvec
);
1709 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1712 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1714 if (put_page_testzero(page
)) {
1716 free_hot_page(page
);
1718 __free_pages_ok(page
, order
);
1722 EXPORT_SYMBOL(__free_pages
);
1724 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1727 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1728 __free_pages(virt_to_page((void *)addr
), order
);
1732 EXPORT_SYMBOL(free_pages
);
1734 static unsigned int nr_free_zone_pages(int offset
)
1736 /* Just pick one node, since fallback list is circular */
1737 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1738 unsigned int sum
= 0;
1740 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1741 struct zone
**zonep
= zonelist
->zones
;
1744 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1745 unsigned long size
= zone
->present_pages
;
1746 unsigned long high
= zone
->pages_high
;
1755 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1757 unsigned int nr_free_buffer_pages(void)
1759 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1761 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1764 * Amount of free RAM allocatable within all zones
1766 unsigned int nr_free_pagecache_pages(void)
1768 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1771 static inline void show_node(struct zone
*zone
)
1774 printk("Node %d ", zone_to_nid(zone
));
1777 void si_meminfo(struct sysinfo
*val
)
1779 val
->totalram
= totalram_pages
;
1781 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1782 val
->bufferram
= nr_blockdev_pages();
1783 val
->totalhigh
= totalhigh_pages
;
1784 val
->freehigh
= nr_free_highpages();
1785 val
->mem_unit
= PAGE_SIZE
;
1788 EXPORT_SYMBOL(si_meminfo
);
1791 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1793 pg_data_t
*pgdat
= NODE_DATA(nid
);
1795 val
->totalram
= pgdat
->node_present_pages
;
1796 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1797 #ifdef CONFIG_HIGHMEM
1798 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1799 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1805 val
->mem_unit
= PAGE_SIZE
;
1809 #define K(x) ((x) << (PAGE_SHIFT-10))
1812 * Show free area list (used inside shift_scroll-lock stuff)
1813 * We also calculate the percentage fragmentation. We do this by counting the
1814 * memory on each free list with the exception of the first item on the list.
1816 void show_free_areas(void)
1821 for_each_zone(zone
) {
1822 if (!populated_zone(zone
))
1826 printk("%s per-cpu:\n", zone
->name
);
1828 for_each_online_cpu(cpu
) {
1829 struct per_cpu_pageset
*pageset
;
1831 pageset
= zone_pcp(zone
, cpu
);
1833 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1834 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1835 cpu
, pageset
->pcp
[0].high
,
1836 pageset
->pcp
[0].batch
, pageset
->pcp
[0].count
,
1837 pageset
->pcp
[1].high
, pageset
->pcp
[1].batch
,
1838 pageset
->pcp
[1].count
);
1842 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1843 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1844 global_page_state(NR_ACTIVE
),
1845 global_page_state(NR_INACTIVE
),
1846 global_page_state(NR_FILE_DIRTY
),
1847 global_page_state(NR_WRITEBACK
),
1848 global_page_state(NR_UNSTABLE_NFS
),
1849 global_page_state(NR_FREE_PAGES
),
1850 global_page_state(NR_SLAB_RECLAIMABLE
) +
1851 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1852 global_page_state(NR_FILE_MAPPED
),
1853 global_page_state(NR_PAGETABLE
),
1854 global_page_state(NR_BOUNCE
));
1856 for_each_zone(zone
) {
1859 if (!populated_zone(zone
))
1871 " pages_scanned:%lu"
1872 " all_unreclaimable? %s"
1875 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1878 K(zone
->pages_high
),
1879 K(zone_page_state(zone
, NR_ACTIVE
)),
1880 K(zone_page_state(zone
, NR_INACTIVE
)),
1881 K(zone
->present_pages
),
1882 zone
->pages_scanned
,
1883 (zone
->all_unreclaimable
? "yes" : "no")
1885 printk("lowmem_reserve[]:");
1886 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1887 printk(" %lu", zone
->lowmem_reserve
[i
]);
1891 for_each_zone(zone
) {
1892 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1894 if (!populated_zone(zone
))
1898 printk("%s: ", zone
->name
);
1900 spin_lock_irqsave(&zone
->lock
, flags
);
1901 for (order
= 0; order
< MAX_ORDER
; order
++) {
1902 nr
[order
] = zone
->free_area
[order
].nr_free
;
1903 total
+= nr
[order
] << order
;
1905 spin_unlock_irqrestore(&zone
->lock
, flags
);
1906 for (order
= 0; order
< MAX_ORDER
; order
++)
1907 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1908 printk("= %lukB\n", K(total
));
1911 show_swap_cache_info();
1915 * Builds allocation fallback zone lists.
1917 * Add all populated zones of a node to the zonelist.
1919 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1920 int nr_zones
, enum zone_type zone_type
)
1924 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1929 zone
= pgdat
->node_zones
+ zone_type
;
1930 if (populated_zone(zone
)) {
1931 zonelist
->zones
[nr_zones
++] = zone
;
1932 check_highest_zone(zone_type
);
1935 } while (zone_type
);
1942 * 0 = automatic detection of better ordering.
1943 * 1 = order by ([node] distance, -zonetype)
1944 * 2 = order by (-zonetype, [node] distance)
1946 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1947 * the same zonelist. So only NUMA can configure this param.
1949 #define ZONELIST_ORDER_DEFAULT 0
1950 #define ZONELIST_ORDER_NODE 1
1951 #define ZONELIST_ORDER_ZONE 2
1953 /* zonelist order in the kernel.
1954 * set_zonelist_order() will set this to NODE or ZONE.
1956 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1957 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
1961 /* The value user specified ....changed by config */
1962 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1963 /* string for sysctl */
1964 #define NUMA_ZONELIST_ORDER_LEN 16
1965 char numa_zonelist_order
[16] = "default";
1968 * interface for configure zonelist ordering.
1969 * command line option "numa_zonelist_order"
1970 * = "[dD]efault - default, automatic configuration.
1971 * = "[nN]ode - order by node locality, then by zone within node
1972 * = "[zZ]one - order by zone, then by locality within zone
1975 static int __parse_numa_zonelist_order(char *s
)
1977 if (*s
== 'd' || *s
== 'D') {
1978 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1979 } else if (*s
== 'n' || *s
== 'N') {
1980 user_zonelist_order
= ZONELIST_ORDER_NODE
;
1981 } else if (*s
== 'z' || *s
== 'Z') {
1982 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
1985 "Ignoring invalid numa_zonelist_order value: "
1992 static __init
int setup_numa_zonelist_order(char *s
)
1995 return __parse_numa_zonelist_order(s
);
1998 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2001 * sysctl handler for numa_zonelist_order
2003 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2004 struct file
*file
, void __user
*buffer
, size_t *length
,
2007 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2011 strncpy(saved_string
, (char*)table
->data
,
2012 NUMA_ZONELIST_ORDER_LEN
);
2013 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2017 int oldval
= user_zonelist_order
;
2018 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2020 * bogus value. restore saved string
2022 strncpy((char*)table
->data
, saved_string
,
2023 NUMA_ZONELIST_ORDER_LEN
);
2024 user_zonelist_order
= oldval
;
2025 } else if (oldval
!= user_zonelist_order
)
2026 build_all_zonelists();
2032 #define MAX_NODE_LOAD (num_online_nodes())
2033 static int node_load
[MAX_NUMNODES
];
2036 * find_next_best_node - find the next node that should appear in a given node's fallback list
2037 * @node: node whose fallback list we're appending
2038 * @used_node_mask: nodemask_t of already used nodes
2040 * We use a number of factors to determine which is the next node that should
2041 * appear on a given node's fallback list. The node should not have appeared
2042 * already in @node's fallback list, and it should be the next closest node
2043 * according to the distance array (which contains arbitrary distance values
2044 * from each node to each node in the system), and should also prefer nodes
2045 * with no CPUs, since presumably they'll have very little allocation pressure
2046 * on them otherwise.
2047 * It returns -1 if no node is found.
2049 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2052 int min_val
= INT_MAX
;
2055 /* Use the local node if we haven't already */
2056 if (!node_isset(node
, *used_node_mask
)) {
2057 node_set(node
, *used_node_mask
);
2061 for_each_node_state(n
, N_HIGH_MEMORY
) {
2064 /* Don't want a node to appear more than once */
2065 if (node_isset(n
, *used_node_mask
))
2068 /* Use the distance array to find the distance */
2069 val
= node_distance(node
, n
);
2071 /* Penalize nodes under us ("prefer the next node") */
2074 /* Give preference to headless and unused nodes */
2075 tmp
= node_to_cpumask(n
);
2076 if (!cpus_empty(tmp
))
2077 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2079 /* Slight preference for less loaded node */
2080 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2081 val
+= node_load
[n
];
2083 if (val
< min_val
) {
2090 node_set(best_node
, *used_node_mask
);
2097 * Build zonelists ordered by node and zones within node.
2098 * This results in maximum locality--normal zone overflows into local
2099 * DMA zone, if any--but risks exhausting DMA zone.
2101 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2105 struct zonelist
*zonelist
;
2107 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2108 zonelist
= pgdat
->node_zonelists
+ i
;
2109 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++)
2111 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2112 zonelist
->zones
[j
] = NULL
;
2117 * Build gfp_thisnode zonelists
2119 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2123 struct zonelist
*zonelist
;
2125 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2126 zonelist
= pgdat
->node_zonelists
+ MAX_NR_ZONES
+ i
;
2127 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2128 zonelist
->zones
[j
] = NULL
;
2133 * Build zonelists ordered by zone and nodes within zones.
2134 * This results in conserving DMA zone[s] until all Normal memory is
2135 * exhausted, but results in overflowing to remote node while memory
2136 * may still exist in local DMA zone.
2138 static int node_order
[MAX_NUMNODES
];
2140 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2144 int zone_type
; /* needs to be signed */
2146 struct zonelist
*zonelist
;
2148 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2149 zonelist
= pgdat
->node_zonelists
+ i
;
2151 for (zone_type
= i
; zone_type
>= 0; zone_type
--) {
2152 for (j
= 0; j
< nr_nodes
; j
++) {
2153 node
= node_order
[j
];
2154 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2155 if (populated_zone(z
)) {
2156 zonelist
->zones
[pos
++] = z
;
2157 check_highest_zone(zone_type
);
2161 zonelist
->zones
[pos
] = NULL
;
2165 static int default_zonelist_order(void)
2168 unsigned long low_kmem_size
,total_size
;
2172 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2173 * If they are really small and used heavily, the system can fall
2174 * into OOM very easily.
2175 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2177 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2180 for_each_online_node(nid
) {
2181 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2182 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2183 if (populated_zone(z
)) {
2184 if (zone_type
< ZONE_NORMAL
)
2185 low_kmem_size
+= z
->present_pages
;
2186 total_size
+= z
->present_pages
;
2190 if (!low_kmem_size
|| /* there are no DMA area. */
2191 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2192 return ZONELIST_ORDER_NODE
;
2194 * look into each node's config.
2195 * If there is a node whose DMA/DMA32 memory is very big area on
2196 * local memory, NODE_ORDER may be suitable.
2198 average_size
= total_size
/
2199 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2200 for_each_online_node(nid
) {
2203 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2204 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2205 if (populated_zone(z
)) {
2206 if (zone_type
< ZONE_NORMAL
)
2207 low_kmem_size
+= z
->present_pages
;
2208 total_size
+= z
->present_pages
;
2211 if (low_kmem_size
&&
2212 total_size
> average_size
&& /* ignore small node */
2213 low_kmem_size
> total_size
* 70/100)
2214 return ZONELIST_ORDER_NODE
;
2216 return ZONELIST_ORDER_ZONE
;
2219 static void set_zonelist_order(void)
2221 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2222 current_zonelist_order
= default_zonelist_order();
2224 current_zonelist_order
= user_zonelist_order
;
2227 static void build_zonelists(pg_data_t
*pgdat
)
2231 nodemask_t used_mask
;
2232 int local_node
, prev_node
;
2233 struct zonelist
*zonelist
;
2234 int order
= current_zonelist_order
;
2236 /* initialize zonelists */
2237 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2238 zonelist
= pgdat
->node_zonelists
+ i
;
2239 zonelist
->zones
[0] = NULL
;
2242 /* NUMA-aware ordering of nodes */
2243 local_node
= pgdat
->node_id
;
2244 load
= num_online_nodes();
2245 prev_node
= local_node
;
2246 nodes_clear(used_mask
);
2248 memset(node_load
, 0, sizeof(node_load
));
2249 memset(node_order
, 0, sizeof(node_order
));
2252 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2253 int distance
= node_distance(local_node
, node
);
2256 * If another node is sufficiently far away then it is better
2257 * to reclaim pages in a zone before going off node.
2259 if (distance
> RECLAIM_DISTANCE
)
2260 zone_reclaim_mode
= 1;
2263 * We don't want to pressure a particular node.
2264 * So adding penalty to the first node in same
2265 * distance group to make it round-robin.
2267 if (distance
!= node_distance(local_node
, prev_node
))
2268 node_load
[node
] = load
;
2272 if (order
== ZONELIST_ORDER_NODE
)
2273 build_zonelists_in_node_order(pgdat
, node
);
2275 node_order
[j
++] = node
; /* remember order */
2278 if (order
== ZONELIST_ORDER_ZONE
) {
2279 /* calculate node order -- i.e., DMA last! */
2280 build_zonelists_in_zone_order(pgdat
, j
);
2283 build_thisnode_zonelists(pgdat
);
2286 /* Construct the zonelist performance cache - see further mmzone.h */
2287 static void build_zonelist_cache(pg_data_t
*pgdat
)
2291 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2292 struct zonelist
*zonelist
;
2293 struct zonelist_cache
*zlc
;
2296 zonelist
= pgdat
->node_zonelists
+ i
;
2297 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2298 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2299 for (z
= zonelist
->zones
; *z
; z
++)
2300 zlc
->z_to_n
[z
- zonelist
->zones
] = zone_to_nid(*z
);
2305 #else /* CONFIG_NUMA */
2307 static void set_zonelist_order(void)
2309 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2312 static void build_zonelists(pg_data_t
*pgdat
)
2314 int node
, local_node
;
2317 local_node
= pgdat
->node_id
;
2318 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2319 struct zonelist
*zonelist
;
2321 zonelist
= pgdat
->node_zonelists
+ i
;
2323 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2325 * Now we build the zonelist so that it contains the zones
2326 * of all the other nodes.
2327 * We don't want to pressure a particular node, so when
2328 * building the zones for node N, we make sure that the
2329 * zones coming right after the local ones are those from
2330 * node N+1 (modulo N)
2332 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2333 if (!node_online(node
))
2335 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2337 for (node
= 0; node
< local_node
; node
++) {
2338 if (!node_online(node
))
2340 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2343 zonelist
->zones
[j
] = NULL
;
2347 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2348 static void build_zonelist_cache(pg_data_t
*pgdat
)
2352 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2353 pgdat
->node_zonelists
[i
].zlcache_ptr
= NULL
;
2356 #endif /* CONFIG_NUMA */
2358 /* return values int ....just for stop_machine_run() */
2359 static int __build_all_zonelists(void *dummy
)
2363 for_each_online_node(nid
) {
2364 pg_data_t
*pgdat
= NODE_DATA(nid
);
2366 build_zonelists(pgdat
);
2367 build_zonelist_cache(pgdat
);
2372 void build_all_zonelists(void)
2374 set_zonelist_order();
2376 if (system_state
== SYSTEM_BOOTING
) {
2377 __build_all_zonelists(NULL
);
2378 cpuset_init_current_mems_allowed();
2380 /* we have to stop all cpus to guaranntee there is no user
2382 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
2383 /* cpuset refresh routine should be here */
2385 vm_total_pages
= nr_free_pagecache_pages();
2387 * Disable grouping by mobility if the number of pages in the
2388 * system is too low to allow the mechanism to work. It would be
2389 * more accurate, but expensive to check per-zone. This check is
2390 * made on memory-hotadd so a system can start with mobility
2391 * disabled and enable it later
2393 if (vm_total_pages
< (MAX_ORDER_NR_PAGES
* MIGRATE_TYPES
))
2394 page_group_by_mobility_disabled
= 1;
2396 page_group_by_mobility_disabled
= 0;
2398 printk("Built %i zonelists in %s order, mobility grouping %s. "
2399 "Total pages: %ld\n",
2401 zonelist_order_name
[current_zonelist_order
],
2402 page_group_by_mobility_disabled
? "off" : "on",
2405 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2410 * Helper functions to size the waitqueue hash table.
2411 * Essentially these want to choose hash table sizes sufficiently
2412 * large so that collisions trying to wait on pages are rare.
2413 * But in fact, the number of active page waitqueues on typical
2414 * systems is ridiculously low, less than 200. So this is even
2415 * conservative, even though it seems large.
2417 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2418 * waitqueues, i.e. the size of the waitq table given the number of pages.
2420 #define PAGES_PER_WAITQUEUE 256
2422 #ifndef CONFIG_MEMORY_HOTPLUG
2423 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2425 unsigned long size
= 1;
2427 pages
/= PAGES_PER_WAITQUEUE
;
2429 while (size
< pages
)
2433 * Once we have dozens or even hundreds of threads sleeping
2434 * on IO we've got bigger problems than wait queue collision.
2435 * Limit the size of the wait table to a reasonable size.
2437 size
= min(size
, 4096UL);
2439 return max(size
, 4UL);
2443 * A zone's size might be changed by hot-add, so it is not possible to determine
2444 * a suitable size for its wait_table. So we use the maximum size now.
2446 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2448 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2449 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2450 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2452 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2453 * or more by the traditional way. (See above). It equals:
2455 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2456 * ia64(16K page size) : = ( 8G + 4M)byte.
2457 * powerpc (64K page size) : = (32G +16M)byte.
2459 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2466 * This is an integer logarithm so that shifts can be used later
2467 * to extract the more random high bits from the multiplicative
2468 * hash function before the remainder is taken.
2470 static inline unsigned long wait_table_bits(unsigned long size
)
2475 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2478 * Initially all pages are reserved - free ones are freed
2479 * up by free_all_bootmem() once the early boot process is
2480 * done. Non-atomic initialization, single-pass.
2482 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2483 unsigned long start_pfn
, enum memmap_context context
)
2486 unsigned long end_pfn
= start_pfn
+ size
;
2489 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2491 * There can be holes in boot-time mem_map[]s
2492 * handed to this function. They do not
2493 * exist on hotplugged memory.
2495 if (context
== MEMMAP_EARLY
) {
2496 if (!early_pfn_valid(pfn
))
2498 if (!early_pfn_in_nid(pfn
, nid
))
2501 page
= pfn_to_page(pfn
);
2502 set_page_links(page
, zone
, nid
, pfn
);
2503 init_page_count(page
);
2504 reset_page_mapcount(page
);
2505 SetPageReserved(page
);
2508 * Mark the block movable so that blocks are reserved for
2509 * movable at startup. This will force kernel allocations
2510 * to reserve their blocks rather than leaking throughout
2511 * the address space during boot when many long-lived
2512 * kernel allocations are made
2514 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2516 INIT_LIST_HEAD(&page
->lru
);
2517 #ifdef WANT_PAGE_VIRTUAL
2518 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2519 if (!is_highmem_idx(zone
))
2520 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2525 static void __meminit
zone_init_free_lists(struct pglist_data
*pgdat
,
2526 struct zone
*zone
, unsigned long size
)
2529 for_each_migratetype_order(order
, t
) {
2530 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2531 zone
->free_area
[order
].nr_free
= 0;
2535 #ifndef __HAVE_ARCH_MEMMAP_INIT
2536 #define memmap_init(size, nid, zone, start_pfn) \
2537 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2540 static int __devinit
zone_batchsize(struct zone
*zone
)
2545 * The per-cpu-pages pools are set to around 1000th of the
2546 * size of the zone. But no more than 1/2 of a meg.
2548 * OK, so we don't know how big the cache is. So guess.
2550 batch
= zone
->present_pages
/ 1024;
2551 if (batch
* PAGE_SIZE
> 512 * 1024)
2552 batch
= (512 * 1024) / PAGE_SIZE
;
2553 batch
/= 4; /* We effectively *= 4 below */
2558 * Clamp the batch to a 2^n - 1 value. Having a power
2559 * of 2 value was found to be more likely to have
2560 * suboptimal cache aliasing properties in some cases.
2562 * For example if 2 tasks are alternately allocating
2563 * batches of pages, one task can end up with a lot
2564 * of pages of one half of the possible page colors
2565 * and the other with pages of the other colors.
2567 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2572 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2574 struct per_cpu_pages
*pcp
;
2576 memset(p
, 0, sizeof(*p
));
2578 pcp
= &p
->pcp
[0]; /* hot */
2580 pcp
->high
= 6 * batch
;
2581 pcp
->batch
= max(1UL, 1 * batch
);
2582 INIT_LIST_HEAD(&pcp
->list
);
2584 pcp
= &p
->pcp
[1]; /* cold*/
2586 pcp
->high
= 2 * batch
;
2587 pcp
->batch
= max(1UL, batch
/2);
2588 INIT_LIST_HEAD(&pcp
->list
);
2592 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2593 * to the value high for the pageset p.
2596 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2599 struct per_cpu_pages
*pcp
;
2601 pcp
= &p
->pcp
[0]; /* hot list */
2603 pcp
->batch
= max(1UL, high
/4);
2604 if ((high
/4) > (PAGE_SHIFT
* 8))
2605 pcp
->batch
= PAGE_SHIFT
* 8;
2611 * Boot pageset table. One per cpu which is going to be used for all
2612 * zones and all nodes. The parameters will be set in such a way
2613 * that an item put on a list will immediately be handed over to
2614 * the buddy list. This is safe since pageset manipulation is done
2615 * with interrupts disabled.
2617 * Some NUMA counter updates may also be caught by the boot pagesets.
2619 * The boot_pagesets must be kept even after bootup is complete for
2620 * unused processors and/or zones. They do play a role for bootstrapping
2621 * hotplugged processors.
2623 * zoneinfo_show() and maybe other functions do
2624 * not check if the processor is online before following the pageset pointer.
2625 * Other parts of the kernel may not check if the zone is available.
2627 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2630 * Dynamically allocate memory for the
2631 * per cpu pageset array in struct zone.
2633 static int __cpuinit
process_zones(int cpu
)
2635 struct zone
*zone
, *dzone
;
2636 int node
= cpu_to_node(cpu
);
2638 node_set_state(node
, N_CPU
); /* this node has a cpu */
2640 for_each_zone(zone
) {
2642 if (!populated_zone(zone
))
2645 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2647 if (!zone_pcp(zone
, cpu
))
2650 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2652 if (percpu_pagelist_fraction
)
2653 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2654 (zone
->present_pages
/ percpu_pagelist_fraction
));
2659 for_each_zone(dzone
) {
2660 if (!populated_zone(dzone
))
2664 kfree(zone_pcp(dzone
, cpu
));
2665 zone_pcp(dzone
, cpu
) = NULL
;
2670 static inline void free_zone_pagesets(int cpu
)
2674 for_each_zone(zone
) {
2675 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2677 /* Free per_cpu_pageset if it is slab allocated */
2678 if (pset
!= &boot_pageset
[cpu
])
2680 zone_pcp(zone
, cpu
) = NULL
;
2684 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2685 unsigned long action
,
2688 int cpu
= (long)hcpu
;
2689 int ret
= NOTIFY_OK
;
2692 case CPU_UP_PREPARE
:
2693 case CPU_UP_PREPARE_FROZEN
:
2694 if (process_zones(cpu
))
2697 case CPU_UP_CANCELED
:
2698 case CPU_UP_CANCELED_FROZEN
:
2700 case CPU_DEAD_FROZEN
:
2701 free_zone_pagesets(cpu
);
2709 static struct notifier_block __cpuinitdata pageset_notifier
=
2710 { &pageset_cpuup_callback
, NULL
, 0 };
2712 void __init
setup_per_cpu_pageset(void)
2716 /* Initialize per_cpu_pageset for cpu 0.
2717 * A cpuup callback will do this for every cpu
2718 * as it comes online
2720 err
= process_zones(smp_processor_id());
2722 register_cpu_notifier(&pageset_notifier
);
2727 static noinline __init_refok
2728 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2731 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2735 * The per-page waitqueue mechanism uses hashed waitqueues
2738 zone
->wait_table_hash_nr_entries
=
2739 wait_table_hash_nr_entries(zone_size_pages
);
2740 zone
->wait_table_bits
=
2741 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2742 alloc_size
= zone
->wait_table_hash_nr_entries
2743 * sizeof(wait_queue_head_t
);
2745 if (system_state
== SYSTEM_BOOTING
) {
2746 zone
->wait_table
= (wait_queue_head_t
*)
2747 alloc_bootmem_node(pgdat
, alloc_size
);
2750 * This case means that a zone whose size was 0 gets new memory
2751 * via memory hot-add.
2752 * But it may be the case that a new node was hot-added. In
2753 * this case vmalloc() will not be able to use this new node's
2754 * memory - this wait_table must be initialized to use this new
2755 * node itself as well.
2756 * To use this new node's memory, further consideration will be
2759 zone
->wait_table
= vmalloc(alloc_size
);
2761 if (!zone
->wait_table
)
2764 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2765 init_waitqueue_head(zone
->wait_table
+ i
);
2770 static __meminit
void zone_pcp_init(struct zone
*zone
)
2773 unsigned long batch
= zone_batchsize(zone
);
2775 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2777 /* Early boot. Slab allocator not functional yet */
2778 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2779 setup_pageset(&boot_pageset
[cpu
],0);
2781 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2784 if (zone
->present_pages
)
2785 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2786 zone
->name
, zone
->present_pages
, batch
);
2789 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2790 unsigned long zone_start_pfn
,
2792 enum memmap_context context
)
2794 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2796 ret
= zone_wait_table_init(zone
, size
);
2799 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2801 zone
->zone_start_pfn
= zone_start_pfn
;
2803 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
2805 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2810 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2812 * Basic iterator support. Return the first range of PFNs for a node
2813 * Note: nid == MAX_NUMNODES returns first region regardless of node
2815 static int __meminit
first_active_region_index_in_nid(int nid
)
2819 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2820 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2827 * Basic iterator support. Return the next active range of PFNs for a node
2828 * Note: nid == MAX_NUMNODES returns next region regardles of node
2830 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2832 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2833 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2839 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2841 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2842 * Architectures may implement their own version but if add_active_range()
2843 * was used and there are no special requirements, this is a convenient
2846 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2850 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2851 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2852 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2854 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2855 return early_node_map
[i
].nid
;
2860 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2862 /* Basic iterator support to walk early_node_map[] */
2863 #define for_each_active_range_index_in_nid(i, nid) \
2864 for (i = first_active_region_index_in_nid(nid); i != -1; \
2865 i = next_active_region_index_in_nid(i, nid))
2868 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2869 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2870 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2872 * If an architecture guarantees that all ranges registered with
2873 * add_active_ranges() contain no holes and may be freed, this
2874 * this function may be used instead of calling free_bootmem() manually.
2876 void __init
free_bootmem_with_active_regions(int nid
,
2877 unsigned long max_low_pfn
)
2881 for_each_active_range_index_in_nid(i
, nid
) {
2882 unsigned long size_pages
= 0;
2883 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2885 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2888 if (end_pfn
> max_low_pfn
)
2889 end_pfn
= max_low_pfn
;
2891 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2892 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2893 PFN_PHYS(early_node_map
[i
].start_pfn
),
2894 size_pages
<< PAGE_SHIFT
);
2899 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2900 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2902 * If an architecture guarantees that all ranges registered with
2903 * add_active_ranges() contain no holes and may be freed, this
2904 * function may be used instead of calling memory_present() manually.
2906 void __init
sparse_memory_present_with_active_regions(int nid
)
2910 for_each_active_range_index_in_nid(i
, nid
)
2911 memory_present(early_node_map
[i
].nid
,
2912 early_node_map
[i
].start_pfn
,
2913 early_node_map
[i
].end_pfn
);
2917 * push_node_boundaries - Push node boundaries to at least the requested boundary
2918 * @nid: The nid of the node to push the boundary for
2919 * @start_pfn: The start pfn of the node
2920 * @end_pfn: The end pfn of the node
2922 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2923 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2924 * be hotplugged even though no physical memory exists. This function allows
2925 * an arch to push out the node boundaries so mem_map is allocated that can
2928 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2929 void __init
push_node_boundaries(unsigned int nid
,
2930 unsigned long start_pfn
, unsigned long end_pfn
)
2932 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2933 nid
, start_pfn
, end_pfn
);
2935 /* Initialise the boundary for this node if necessary */
2936 if (node_boundary_end_pfn
[nid
] == 0)
2937 node_boundary_start_pfn
[nid
] = -1UL;
2939 /* Update the boundaries */
2940 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2941 node_boundary_start_pfn
[nid
] = start_pfn
;
2942 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2943 node_boundary_end_pfn
[nid
] = end_pfn
;
2946 /* If necessary, push the node boundary out for reserve hotadd */
2947 static void __meminit
account_node_boundary(unsigned int nid
,
2948 unsigned long *start_pfn
, unsigned long *end_pfn
)
2950 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2951 nid
, *start_pfn
, *end_pfn
);
2953 /* Return if boundary information has not been provided */
2954 if (node_boundary_end_pfn
[nid
] == 0)
2957 /* Check the boundaries and update if necessary */
2958 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
2959 *start_pfn
= node_boundary_start_pfn
[nid
];
2960 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
2961 *end_pfn
= node_boundary_end_pfn
[nid
];
2964 void __init
push_node_boundaries(unsigned int nid
,
2965 unsigned long start_pfn
, unsigned long end_pfn
) {}
2967 static void __meminit
account_node_boundary(unsigned int nid
,
2968 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
2973 * get_pfn_range_for_nid - Return the start and end page frames for a node
2974 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2975 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2976 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2978 * It returns the start and end page frame of a node based on information
2979 * provided by an arch calling add_active_range(). If called for a node
2980 * with no available memory, a warning is printed and the start and end
2983 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
2984 unsigned long *start_pfn
, unsigned long *end_pfn
)
2990 for_each_active_range_index_in_nid(i
, nid
) {
2991 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
2992 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
2995 if (*start_pfn
== -1UL)
2998 /* Push the node boundaries out if requested */
2999 account_node_boundary(nid
, start_pfn
, end_pfn
);
3003 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3004 * assumption is made that zones within a node are ordered in monotonic
3005 * increasing memory addresses so that the "highest" populated zone is used
3007 void __init
find_usable_zone_for_movable(void)
3010 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3011 if (zone_index
== ZONE_MOVABLE
)
3014 if (arch_zone_highest_possible_pfn
[zone_index
] >
3015 arch_zone_lowest_possible_pfn
[zone_index
])
3019 VM_BUG_ON(zone_index
== -1);
3020 movable_zone
= zone_index
;
3024 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3025 * because it is sized independant of architecture. Unlike the other zones,
3026 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3027 * in each node depending on the size of each node and how evenly kernelcore
3028 * is distributed. This helper function adjusts the zone ranges
3029 * provided by the architecture for a given node by using the end of the
3030 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3031 * zones within a node are in order of monotonic increases memory addresses
3033 void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3034 unsigned long zone_type
,
3035 unsigned long node_start_pfn
,
3036 unsigned long node_end_pfn
,
3037 unsigned long *zone_start_pfn
,
3038 unsigned long *zone_end_pfn
)
3040 /* Only adjust if ZONE_MOVABLE is on this node */
3041 if (zone_movable_pfn
[nid
]) {
3042 /* Size ZONE_MOVABLE */
3043 if (zone_type
== ZONE_MOVABLE
) {
3044 *zone_start_pfn
= zone_movable_pfn
[nid
];
3045 *zone_end_pfn
= min(node_end_pfn
,
3046 arch_zone_highest_possible_pfn
[movable_zone
]);
3048 /* Adjust for ZONE_MOVABLE starting within this range */
3049 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3050 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3051 *zone_end_pfn
= zone_movable_pfn
[nid
];
3053 /* Check if this whole range is within ZONE_MOVABLE */
3054 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3055 *zone_start_pfn
= *zone_end_pfn
;
3060 * Return the number of pages a zone spans in a node, including holes
3061 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3063 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3064 unsigned long zone_type
,
3065 unsigned long *ignored
)
3067 unsigned long node_start_pfn
, node_end_pfn
;
3068 unsigned long zone_start_pfn
, zone_end_pfn
;
3070 /* Get the start and end of the node and zone */
3071 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3072 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3073 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3074 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3075 node_start_pfn
, node_end_pfn
,
3076 &zone_start_pfn
, &zone_end_pfn
);
3078 /* Check that this node has pages within the zone's required range */
3079 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3082 /* Move the zone boundaries inside the node if necessary */
3083 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3084 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3086 /* Return the spanned pages */
3087 return zone_end_pfn
- zone_start_pfn
;
3091 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3092 * then all holes in the requested range will be accounted for.
3094 unsigned long __meminit
__absent_pages_in_range(int nid
,
3095 unsigned long range_start_pfn
,
3096 unsigned long range_end_pfn
)
3099 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3100 unsigned long start_pfn
;
3102 /* Find the end_pfn of the first active range of pfns in the node */
3103 i
= first_active_region_index_in_nid(nid
);
3107 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3109 /* Account for ranges before physical memory on this node */
3110 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3111 hole_pages
= prev_end_pfn
- range_start_pfn
;
3113 /* Find all holes for the zone within the node */
3114 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3116 /* No need to continue if prev_end_pfn is outside the zone */
3117 if (prev_end_pfn
>= range_end_pfn
)
3120 /* Make sure the end of the zone is not within the hole */
3121 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3122 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3124 /* Update the hole size cound and move on */
3125 if (start_pfn
> range_start_pfn
) {
3126 BUG_ON(prev_end_pfn
> start_pfn
);
3127 hole_pages
+= start_pfn
- prev_end_pfn
;
3129 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3132 /* Account for ranges past physical memory on this node */
3133 if (range_end_pfn
> prev_end_pfn
)
3134 hole_pages
+= range_end_pfn
-
3135 max(range_start_pfn
, prev_end_pfn
);
3141 * absent_pages_in_range - Return number of page frames in holes within a range
3142 * @start_pfn: The start PFN to start searching for holes
3143 * @end_pfn: The end PFN to stop searching for holes
3145 * It returns the number of pages frames in memory holes within a range.
3147 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3148 unsigned long end_pfn
)
3150 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3153 /* Return the number of page frames in holes in a zone on a node */
3154 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3155 unsigned long zone_type
,
3156 unsigned long *ignored
)
3158 unsigned long node_start_pfn
, node_end_pfn
;
3159 unsigned long zone_start_pfn
, zone_end_pfn
;
3161 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3162 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3164 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3167 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3168 node_start_pfn
, node_end_pfn
,
3169 &zone_start_pfn
, &zone_end_pfn
);
3170 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3174 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3175 unsigned long zone_type
,
3176 unsigned long *zones_size
)
3178 return zones_size
[zone_type
];
3181 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3182 unsigned long zone_type
,
3183 unsigned long *zholes_size
)
3188 return zholes_size
[zone_type
];
3193 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3194 unsigned long *zones_size
, unsigned long *zholes_size
)
3196 unsigned long realtotalpages
, totalpages
= 0;
3199 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3200 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3202 pgdat
->node_spanned_pages
= totalpages
;
3204 realtotalpages
= totalpages
;
3205 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3207 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3209 pgdat
->node_present_pages
= realtotalpages
;
3210 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3214 #ifndef CONFIG_SPARSEMEM
3216 * Calculate the size of the zone->blockflags rounded to an unsigned long
3217 * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up
3218 * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally
3219 * round what is now in bits to nearest long in bits, then return it in
3222 static unsigned long __init
usemap_size(unsigned long zonesize
)
3224 unsigned long usemapsize
;
3226 usemapsize
= roundup(zonesize
, MAX_ORDER_NR_PAGES
);
3227 usemapsize
= usemapsize
>> (MAX_ORDER
-1);
3228 usemapsize
*= NR_PAGEBLOCK_BITS
;
3229 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3231 return usemapsize
/ 8;
3234 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3235 struct zone
*zone
, unsigned long zonesize
)
3237 unsigned long usemapsize
= usemap_size(zonesize
);
3238 zone
->pageblock_flags
= NULL
;
3240 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3241 memset(zone
->pageblock_flags
, 0, usemapsize
);
3245 static void inline setup_usemap(struct pglist_data
*pgdat
,
3246 struct zone
*zone
, unsigned long zonesize
) {}
3247 #endif /* CONFIG_SPARSEMEM */
3250 * Set up the zone data structures:
3251 * - mark all pages reserved
3252 * - mark all memory queues empty
3253 * - clear the memory bitmaps
3255 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
3256 unsigned long *zones_size
, unsigned long *zholes_size
)
3259 int nid
= pgdat
->node_id
;
3260 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3263 pgdat_resize_init(pgdat
);
3264 pgdat
->nr_zones
= 0;
3265 init_waitqueue_head(&pgdat
->kswapd_wait
);
3266 pgdat
->kswapd_max_order
= 0;
3268 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3269 struct zone
*zone
= pgdat
->node_zones
+ j
;
3270 unsigned long size
, realsize
, memmap_pages
;
3272 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3273 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3277 * Adjust realsize so that it accounts for how much memory
3278 * is used by this zone for memmap. This affects the watermark
3279 * and per-cpu initialisations
3281 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3282 if (realsize
>= memmap_pages
) {
3283 realsize
-= memmap_pages
;
3285 " %s zone: %lu pages used for memmap\n",
3286 zone_names
[j
], memmap_pages
);
3289 " %s zone: %lu pages exceeds realsize %lu\n",
3290 zone_names
[j
], memmap_pages
, realsize
);
3292 /* Account for reserved pages */
3293 if (j
== 0 && realsize
> dma_reserve
) {
3294 realsize
-= dma_reserve
;
3295 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3296 zone_names
[0], dma_reserve
);
3299 if (!is_highmem_idx(j
))
3300 nr_kernel_pages
+= realsize
;
3301 nr_all_pages
+= realsize
;
3303 zone
->spanned_pages
= size
;
3304 zone
->present_pages
= realsize
;
3307 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3309 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3311 zone
->name
= zone_names
[j
];
3312 spin_lock_init(&zone
->lock
);
3313 spin_lock_init(&zone
->lru_lock
);
3314 zone_seqlock_init(zone
);
3315 zone
->zone_pgdat
= pgdat
;
3317 zone
->prev_priority
= DEF_PRIORITY
;
3319 zone_pcp_init(zone
);
3320 INIT_LIST_HEAD(&zone
->active_list
);
3321 INIT_LIST_HEAD(&zone
->inactive_list
);
3322 zone
->nr_scan_active
= 0;
3323 zone
->nr_scan_inactive
= 0;
3324 zap_zone_vm_stats(zone
);
3325 atomic_set(&zone
->reclaim_in_progress
, 0);
3329 setup_usemap(pgdat
, zone
, size
);
3330 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3331 size
, MEMMAP_EARLY
);
3333 zone_start_pfn
+= size
;
3337 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3339 /* Skip empty nodes */
3340 if (!pgdat
->node_spanned_pages
)
3343 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3344 /* ia64 gets its own node_mem_map, before this, without bootmem */
3345 if (!pgdat
->node_mem_map
) {
3346 unsigned long size
, start
, end
;
3350 * The zone's endpoints aren't required to be MAX_ORDER
3351 * aligned but the node_mem_map endpoints must be in order
3352 * for the buddy allocator to function correctly.
3354 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3355 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3356 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3357 size
= (end
- start
) * sizeof(struct page
);
3358 map
= alloc_remap(pgdat
->node_id
, size
);
3360 map
= alloc_bootmem_node(pgdat
, size
);
3361 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3363 #ifndef CONFIG_NEED_MULTIPLE_NODES
3365 * With no DISCONTIG, the global mem_map is just set as node 0's
3367 if (pgdat
== NODE_DATA(0)) {
3368 mem_map
= NODE_DATA(0)->node_mem_map
;
3369 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3370 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3371 mem_map
-= pgdat
->node_start_pfn
;
3372 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3375 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3378 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
3379 unsigned long *zones_size
, unsigned long node_start_pfn
,
3380 unsigned long *zholes_size
)
3382 pgdat
->node_id
= nid
;
3383 pgdat
->node_start_pfn
= node_start_pfn
;
3384 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3386 alloc_node_mem_map(pgdat
);
3388 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3391 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3393 #if MAX_NUMNODES > 1
3395 * Figure out the number of possible node ids.
3397 static void __init
setup_nr_node_ids(void)
3400 unsigned int highest
= 0;
3402 for_each_node_mask(node
, node_possible_map
)
3404 nr_node_ids
= highest
+ 1;
3407 static inline void setup_nr_node_ids(void)
3413 * add_active_range - Register a range of PFNs backed by physical memory
3414 * @nid: The node ID the range resides on
3415 * @start_pfn: The start PFN of the available physical memory
3416 * @end_pfn: The end PFN of the available physical memory
3418 * These ranges are stored in an early_node_map[] and later used by
3419 * free_area_init_nodes() to calculate zone sizes and holes. If the
3420 * range spans a memory hole, it is up to the architecture to ensure
3421 * the memory is not freed by the bootmem allocator. If possible
3422 * the range being registered will be merged with existing ranges.
3424 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3425 unsigned long end_pfn
)
3429 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
3430 "%d entries of %d used\n",
3431 nid
, start_pfn
, end_pfn
,
3432 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3434 /* Merge with existing active regions if possible */
3435 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3436 if (early_node_map
[i
].nid
!= nid
)
3439 /* Skip if an existing region covers this new one */
3440 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3441 end_pfn
<= early_node_map
[i
].end_pfn
)
3444 /* Merge forward if suitable */
3445 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3446 end_pfn
> early_node_map
[i
].end_pfn
) {
3447 early_node_map
[i
].end_pfn
= end_pfn
;
3451 /* Merge backward if suitable */
3452 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3453 end_pfn
>= early_node_map
[i
].start_pfn
) {
3454 early_node_map
[i
].start_pfn
= start_pfn
;
3459 /* Check that early_node_map is large enough */
3460 if (i
>= MAX_ACTIVE_REGIONS
) {
3461 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3462 MAX_ACTIVE_REGIONS
);
3466 early_node_map
[i
].nid
= nid
;
3467 early_node_map
[i
].start_pfn
= start_pfn
;
3468 early_node_map
[i
].end_pfn
= end_pfn
;
3469 nr_nodemap_entries
= i
+ 1;
3473 * shrink_active_range - Shrink an existing registered range of PFNs
3474 * @nid: The node id the range is on that should be shrunk
3475 * @old_end_pfn: The old end PFN of the range
3476 * @new_end_pfn: The new PFN of the range
3478 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3479 * The map is kept at the end physical page range that has already been
3480 * registered with add_active_range(). This function allows an arch to shrink
3481 * an existing registered range.
3483 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
3484 unsigned long new_end_pfn
)
3488 /* Find the old active region end and shrink */
3489 for_each_active_range_index_in_nid(i
, nid
)
3490 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
3491 early_node_map
[i
].end_pfn
= new_end_pfn
;
3497 * remove_all_active_ranges - Remove all currently registered regions
3499 * During discovery, it may be found that a table like SRAT is invalid
3500 * and an alternative discovery method must be used. This function removes
3501 * all currently registered regions.
3503 void __init
remove_all_active_ranges(void)
3505 memset(early_node_map
, 0, sizeof(early_node_map
));
3506 nr_nodemap_entries
= 0;
3507 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3508 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3509 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3510 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3513 /* Compare two active node_active_regions */
3514 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3516 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3517 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3519 /* Done this way to avoid overflows */
3520 if (arange
->start_pfn
> brange
->start_pfn
)
3522 if (arange
->start_pfn
< brange
->start_pfn
)
3528 /* sort the node_map by start_pfn */
3529 static void __init
sort_node_map(void)
3531 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3532 sizeof(struct node_active_region
),
3533 cmp_node_active_region
, NULL
);
3536 /* Find the lowest pfn for a node */
3537 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
3540 unsigned long min_pfn
= ULONG_MAX
;
3542 /* Assuming a sorted map, the first range found has the starting pfn */
3543 for_each_active_range_index_in_nid(i
, nid
)
3544 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3546 if (min_pfn
== ULONG_MAX
) {
3548 "Could not find start_pfn for node %lu\n", nid
);
3556 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3558 * It returns the minimum PFN based on information provided via
3559 * add_active_range().
3561 unsigned long __init
find_min_pfn_with_active_regions(void)
3563 return find_min_pfn_for_node(MAX_NUMNODES
);
3567 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3569 * It returns the maximum PFN based on information provided via
3570 * add_active_range().
3572 unsigned long __init
find_max_pfn_with_active_regions(void)
3575 unsigned long max_pfn
= 0;
3577 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3578 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
3584 * early_calculate_totalpages()
3585 * Sum pages in active regions for movable zone.
3586 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3588 unsigned long __init
early_calculate_totalpages(void)
3591 unsigned long totalpages
= 0;
3593 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3594 unsigned long pages
= early_node_map
[i
].end_pfn
-
3595 early_node_map
[i
].start_pfn
;
3596 totalpages
+= pages
;
3598 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3604 * Find the PFN the Movable zone begins in each node. Kernel memory
3605 * is spread evenly between nodes as long as the nodes have enough
3606 * memory. When they don't, some nodes will have more kernelcore than
3609 void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3612 unsigned long usable_startpfn
;
3613 unsigned long kernelcore_node
, kernelcore_remaining
;
3614 unsigned long totalpages
= early_calculate_totalpages();
3615 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3618 * If movablecore was specified, calculate what size of
3619 * kernelcore that corresponds so that memory usable for
3620 * any allocation type is evenly spread. If both kernelcore
3621 * and movablecore are specified, then the value of kernelcore
3622 * will be used for required_kernelcore if it's greater than
3623 * what movablecore would have allowed.
3625 if (required_movablecore
) {
3626 unsigned long corepages
;
3629 * Round-up so that ZONE_MOVABLE is at least as large as what
3630 * was requested by the user
3632 required_movablecore
=
3633 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3634 corepages
= totalpages
- required_movablecore
;
3636 required_kernelcore
= max(required_kernelcore
, corepages
);
3639 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3640 if (!required_kernelcore
)
3643 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3644 find_usable_zone_for_movable();
3645 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3648 /* Spread kernelcore memory as evenly as possible throughout nodes */
3649 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3650 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3652 * Recalculate kernelcore_node if the division per node
3653 * now exceeds what is necessary to satisfy the requested
3654 * amount of memory for the kernel
3656 if (required_kernelcore
< kernelcore_node
)
3657 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3660 * As the map is walked, we track how much memory is usable
3661 * by the kernel using kernelcore_remaining. When it is
3662 * 0, the rest of the node is usable by ZONE_MOVABLE
3664 kernelcore_remaining
= kernelcore_node
;
3666 /* Go through each range of PFNs within this node */
3667 for_each_active_range_index_in_nid(i
, nid
) {
3668 unsigned long start_pfn
, end_pfn
;
3669 unsigned long size_pages
;
3671 start_pfn
= max(early_node_map
[i
].start_pfn
,
3672 zone_movable_pfn
[nid
]);
3673 end_pfn
= early_node_map
[i
].end_pfn
;
3674 if (start_pfn
>= end_pfn
)
3677 /* Account for what is only usable for kernelcore */
3678 if (start_pfn
< usable_startpfn
) {
3679 unsigned long kernel_pages
;
3680 kernel_pages
= min(end_pfn
, usable_startpfn
)
3683 kernelcore_remaining
-= min(kernel_pages
,
3684 kernelcore_remaining
);
3685 required_kernelcore
-= min(kernel_pages
,
3686 required_kernelcore
);
3688 /* Continue if range is now fully accounted */
3689 if (end_pfn
<= usable_startpfn
) {
3692 * Push zone_movable_pfn to the end so
3693 * that if we have to rebalance
3694 * kernelcore across nodes, we will
3695 * not double account here
3697 zone_movable_pfn
[nid
] = end_pfn
;
3700 start_pfn
= usable_startpfn
;
3704 * The usable PFN range for ZONE_MOVABLE is from
3705 * start_pfn->end_pfn. Calculate size_pages as the
3706 * number of pages used as kernelcore
3708 size_pages
= end_pfn
- start_pfn
;
3709 if (size_pages
> kernelcore_remaining
)
3710 size_pages
= kernelcore_remaining
;
3711 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3714 * Some kernelcore has been met, update counts and
3715 * break if the kernelcore for this node has been
3718 required_kernelcore
-= min(required_kernelcore
,
3720 kernelcore_remaining
-= size_pages
;
3721 if (!kernelcore_remaining
)
3727 * If there is still required_kernelcore, we do another pass with one
3728 * less node in the count. This will push zone_movable_pfn[nid] further
3729 * along on the nodes that still have memory until kernelcore is
3733 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3736 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3737 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3738 zone_movable_pfn
[nid
] =
3739 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3742 /* Any regular memory on that node ? */
3743 static void check_for_regular_memory(pg_data_t
*pgdat
)
3745 #ifdef CONFIG_HIGHMEM
3746 enum zone_type zone_type
;
3748 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3749 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3750 if (zone
->present_pages
)
3751 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3757 * free_area_init_nodes - Initialise all pg_data_t and zone data
3758 * @max_zone_pfn: an array of max PFNs for each zone
3760 * This will call free_area_init_node() for each active node in the system.
3761 * Using the page ranges provided by add_active_range(), the size of each
3762 * zone in each node and their holes is calculated. If the maximum PFN
3763 * between two adjacent zones match, it is assumed that the zone is empty.
3764 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3765 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3766 * starts where the previous one ended. For example, ZONE_DMA32 starts
3767 * at arch_max_dma_pfn.
3769 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3774 /* Sort early_node_map as initialisation assumes it is sorted */
3777 /* Record where the zone boundaries are */
3778 memset(arch_zone_lowest_possible_pfn
, 0,
3779 sizeof(arch_zone_lowest_possible_pfn
));
3780 memset(arch_zone_highest_possible_pfn
, 0,
3781 sizeof(arch_zone_highest_possible_pfn
));
3782 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3783 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3784 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3785 if (i
== ZONE_MOVABLE
)
3787 arch_zone_lowest_possible_pfn
[i
] =
3788 arch_zone_highest_possible_pfn
[i
-1];
3789 arch_zone_highest_possible_pfn
[i
] =
3790 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3792 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3793 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3795 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3796 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3797 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3799 /* Print out the zone ranges */
3800 printk("Zone PFN ranges:\n");
3801 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3802 if (i
== ZONE_MOVABLE
)
3804 printk(" %-8s %8lu -> %8lu\n",
3806 arch_zone_lowest_possible_pfn
[i
],
3807 arch_zone_highest_possible_pfn
[i
]);
3810 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3811 printk("Movable zone start PFN for each node\n");
3812 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
3813 if (zone_movable_pfn
[i
])
3814 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
3817 /* Print out the early_node_map[] */
3818 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
3819 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3820 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
3821 early_node_map
[i
].start_pfn
,
3822 early_node_map
[i
].end_pfn
);
3824 /* Initialise every node */
3825 setup_nr_node_ids();
3826 for_each_online_node(nid
) {
3827 pg_data_t
*pgdat
= NODE_DATA(nid
);
3828 free_area_init_node(nid
, pgdat
, NULL
,
3829 find_min_pfn_for_node(nid
), NULL
);
3831 /* Any memory on that node */
3832 if (pgdat
->node_present_pages
)
3833 node_set_state(nid
, N_HIGH_MEMORY
);
3834 check_for_regular_memory(pgdat
);
3838 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
3840 unsigned long long coremem
;
3844 coremem
= memparse(p
, &p
);
3845 *core
= coremem
>> PAGE_SHIFT
;
3847 /* Paranoid check that UL is enough for the coremem value */
3848 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
3854 * kernelcore=size sets the amount of memory for use for allocations that
3855 * cannot be reclaimed or migrated.
3857 static int __init
cmdline_parse_kernelcore(char *p
)
3859 return cmdline_parse_core(p
, &required_kernelcore
);
3863 * movablecore=size sets the amount of memory for use for allocations that
3864 * can be reclaimed or migrated.
3866 static int __init
cmdline_parse_movablecore(char *p
)
3868 return cmdline_parse_core(p
, &required_movablecore
);
3871 early_param("kernelcore", cmdline_parse_kernelcore
);
3872 early_param("movablecore", cmdline_parse_movablecore
);
3874 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3877 * set_dma_reserve - set the specified number of pages reserved in the first zone
3878 * @new_dma_reserve: The number of pages to mark reserved
3880 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3881 * In the DMA zone, a significant percentage may be consumed by kernel image
3882 * and other unfreeable allocations which can skew the watermarks badly. This
3883 * function may optionally be used to account for unfreeable pages in the
3884 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3885 * smaller per-cpu batchsize.
3887 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
3889 dma_reserve
= new_dma_reserve
;
3892 #ifndef CONFIG_NEED_MULTIPLE_NODES
3893 static bootmem_data_t contig_bootmem_data
;
3894 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
3896 EXPORT_SYMBOL(contig_page_data
);
3899 void __init
free_area_init(unsigned long *zones_size
)
3901 free_area_init_node(0, NODE_DATA(0), zones_size
,
3902 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
3905 static int page_alloc_cpu_notify(struct notifier_block
*self
,
3906 unsigned long action
, void *hcpu
)
3908 int cpu
= (unsigned long)hcpu
;
3910 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3911 local_irq_disable();
3913 vm_events_fold_cpu(cpu
);
3915 refresh_cpu_vm_stats(cpu
);
3920 void __init
page_alloc_init(void)
3922 hotcpu_notifier(page_alloc_cpu_notify
, 0);
3926 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3927 * or min_free_kbytes changes.
3929 static void calculate_totalreserve_pages(void)
3931 struct pglist_data
*pgdat
;
3932 unsigned long reserve_pages
= 0;
3933 enum zone_type i
, j
;
3935 for_each_online_pgdat(pgdat
) {
3936 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3937 struct zone
*zone
= pgdat
->node_zones
+ i
;
3938 unsigned long max
= 0;
3940 /* Find valid and maximum lowmem_reserve in the zone */
3941 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
3942 if (zone
->lowmem_reserve
[j
] > max
)
3943 max
= zone
->lowmem_reserve
[j
];
3946 /* we treat pages_high as reserved pages. */
3947 max
+= zone
->pages_high
;
3949 if (max
> zone
->present_pages
)
3950 max
= zone
->present_pages
;
3951 reserve_pages
+= max
;
3954 totalreserve_pages
= reserve_pages
;
3958 * setup_per_zone_lowmem_reserve - called whenever
3959 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3960 * has a correct pages reserved value, so an adequate number of
3961 * pages are left in the zone after a successful __alloc_pages().
3963 static void setup_per_zone_lowmem_reserve(void)
3965 struct pglist_data
*pgdat
;
3966 enum zone_type j
, idx
;
3968 for_each_online_pgdat(pgdat
) {
3969 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3970 struct zone
*zone
= pgdat
->node_zones
+ j
;
3971 unsigned long present_pages
= zone
->present_pages
;
3973 zone
->lowmem_reserve
[j
] = 0;
3977 struct zone
*lower_zone
;
3981 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
3982 sysctl_lowmem_reserve_ratio
[idx
] = 1;
3984 lower_zone
= pgdat
->node_zones
+ idx
;
3985 lower_zone
->lowmem_reserve
[j
] = present_pages
/
3986 sysctl_lowmem_reserve_ratio
[idx
];
3987 present_pages
+= lower_zone
->present_pages
;
3992 /* update totalreserve_pages */
3993 calculate_totalreserve_pages();
3997 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3999 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4000 * with respect to min_free_kbytes.
4002 void setup_per_zone_pages_min(void)
4004 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4005 unsigned long lowmem_pages
= 0;
4007 unsigned long flags
;
4009 /* Calculate total number of !ZONE_HIGHMEM pages */
4010 for_each_zone(zone
) {
4011 if (!is_highmem(zone
))
4012 lowmem_pages
+= zone
->present_pages
;
4015 for_each_zone(zone
) {
4018 spin_lock_irqsave(&zone
->lru_lock
, flags
);
4019 tmp
= (u64
)pages_min
* zone
->present_pages
;
4020 do_div(tmp
, lowmem_pages
);
4021 if (is_highmem(zone
)) {
4023 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4024 * need highmem pages, so cap pages_min to a small
4027 * The (pages_high-pages_low) and (pages_low-pages_min)
4028 * deltas controls asynch page reclaim, and so should
4029 * not be capped for highmem.
4033 min_pages
= zone
->present_pages
/ 1024;
4034 if (min_pages
< SWAP_CLUSTER_MAX
)
4035 min_pages
= SWAP_CLUSTER_MAX
;
4036 if (min_pages
> 128)
4038 zone
->pages_min
= min_pages
;
4041 * If it's a lowmem zone, reserve a number of pages
4042 * proportionate to the zone's size.
4044 zone
->pages_min
= tmp
;
4047 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4048 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4049 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4052 /* update totalreserve_pages */
4053 calculate_totalreserve_pages();
4057 * Initialise min_free_kbytes.
4059 * For small machines we want it small (128k min). For large machines
4060 * we want it large (64MB max). But it is not linear, because network
4061 * bandwidth does not increase linearly with machine size. We use
4063 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4064 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4080 static int __init
init_per_zone_pages_min(void)
4082 unsigned long lowmem_kbytes
;
4084 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4086 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4087 if (min_free_kbytes
< 128)
4088 min_free_kbytes
= 128;
4089 if (min_free_kbytes
> 65536)
4090 min_free_kbytes
= 65536;
4091 setup_per_zone_pages_min();
4092 setup_per_zone_lowmem_reserve();
4095 module_init(init_per_zone_pages_min
)
4098 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4099 * that we can call two helper functions whenever min_free_kbytes
4102 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4103 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4105 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4107 setup_per_zone_pages_min();
4112 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4113 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4118 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4123 zone
->min_unmapped_pages
= (zone
->present_pages
*
4124 sysctl_min_unmapped_ratio
) / 100;
4128 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4129 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4134 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4139 zone
->min_slab_pages
= (zone
->present_pages
*
4140 sysctl_min_slab_ratio
) / 100;
4146 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4147 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4148 * whenever sysctl_lowmem_reserve_ratio changes.
4150 * The reserve ratio obviously has absolutely no relation with the
4151 * pages_min watermarks. The lowmem reserve ratio can only make sense
4152 * if in function of the boot time zone sizes.
4154 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4155 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4157 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4158 setup_per_zone_lowmem_reserve();
4163 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4164 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4165 * can have before it gets flushed back to buddy allocator.
4168 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4169 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4175 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4176 if (!write
|| (ret
== -EINVAL
))
4178 for_each_zone(zone
) {
4179 for_each_online_cpu(cpu
) {
4181 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4182 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4188 int hashdist
= HASHDIST_DEFAULT
;
4191 static int __init
set_hashdist(char *str
)
4195 hashdist
= simple_strtoul(str
, &str
, 0);
4198 __setup("hashdist=", set_hashdist
);
4202 * allocate a large system hash table from bootmem
4203 * - it is assumed that the hash table must contain an exact power-of-2
4204 * quantity of entries
4205 * - limit is the number of hash buckets, not the total allocation size
4207 void *__init
alloc_large_system_hash(const char *tablename
,
4208 unsigned long bucketsize
,
4209 unsigned long numentries
,
4212 unsigned int *_hash_shift
,
4213 unsigned int *_hash_mask
,
4214 unsigned long limit
)
4216 unsigned long long max
= limit
;
4217 unsigned long log2qty
, size
;
4220 /* allow the kernel cmdline to have a say */
4222 /* round applicable memory size up to nearest megabyte */
4223 numentries
= nr_kernel_pages
;
4224 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4225 numentries
>>= 20 - PAGE_SHIFT
;
4226 numentries
<<= 20 - PAGE_SHIFT
;
4228 /* limit to 1 bucket per 2^scale bytes of low memory */
4229 if (scale
> PAGE_SHIFT
)
4230 numentries
>>= (scale
- PAGE_SHIFT
);
4232 numentries
<<= (PAGE_SHIFT
- scale
);
4234 /* Make sure we've got at least a 0-order allocation.. */
4235 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4236 numentries
= PAGE_SIZE
/ bucketsize
;
4238 numentries
= roundup_pow_of_two(numentries
);
4240 /* limit allocation size to 1/16 total memory by default */
4242 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4243 do_div(max
, bucketsize
);
4246 if (numentries
> max
)
4249 log2qty
= ilog2(numentries
);
4252 size
= bucketsize
<< log2qty
;
4253 if (flags
& HASH_EARLY
)
4254 table
= alloc_bootmem(size
);
4256 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4258 unsigned long order
;
4259 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
4261 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4263 * If bucketsize is not a power-of-two, we may free
4264 * some pages at the end of hash table.
4267 unsigned long alloc_end
= (unsigned long)table
+
4268 (PAGE_SIZE
<< order
);
4269 unsigned long used
= (unsigned long)table
+
4271 split_page(virt_to_page(table
), order
);
4272 while (used
< alloc_end
) {
4278 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4281 panic("Failed to allocate %s hash table\n", tablename
);
4283 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4286 ilog2(size
) - PAGE_SHIFT
,
4290 *_hash_shift
= log2qty
;
4292 *_hash_mask
= (1 << log2qty
) - 1;
4297 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4298 struct page
*pfn_to_page(unsigned long pfn
)
4300 return __pfn_to_page(pfn
);
4302 unsigned long page_to_pfn(struct page
*page
)
4304 return __page_to_pfn(page
);
4306 EXPORT_SYMBOL(pfn_to_page
);
4307 EXPORT_SYMBOL(page_to_pfn
);
4308 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4310 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4311 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4314 #ifdef CONFIG_SPARSEMEM
4315 return __pfn_to_section(pfn
)->pageblock_flags
;
4317 return zone
->pageblock_flags
;
4318 #endif /* CONFIG_SPARSEMEM */
4321 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4323 #ifdef CONFIG_SPARSEMEM
4324 pfn
&= (PAGES_PER_SECTION
-1);
4325 return (pfn
>> (MAX_ORDER
-1)) * NR_PAGEBLOCK_BITS
;
4327 pfn
= pfn
- zone
->zone_start_pfn
;
4328 return (pfn
>> (MAX_ORDER
-1)) * NR_PAGEBLOCK_BITS
;
4329 #endif /* CONFIG_SPARSEMEM */
4333 * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages
4334 * @page: The page within the block of interest
4335 * @start_bitidx: The first bit of interest to retrieve
4336 * @end_bitidx: The last bit of interest
4337 * returns pageblock_bits flags
4339 unsigned long get_pageblock_flags_group(struct page
*page
,
4340 int start_bitidx
, int end_bitidx
)
4343 unsigned long *bitmap
;
4344 unsigned long pfn
, bitidx
;
4345 unsigned long flags
= 0;
4346 unsigned long value
= 1;
4348 zone
= page_zone(page
);
4349 pfn
= page_to_pfn(page
);
4350 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4351 bitidx
= pfn_to_bitidx(zone
, pfn
);
4353 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4354 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4361 * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages
4362 * @page: The page within the block of interest
4363 * @start_bitidx: The first bit of interest
4364 * @end_bitidx: The last bit of interest
4365 * @flags: The flags to set
4367 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4368 int start_bitidx
, int end_bitidx
)
4371 unsigned long *bitmap
;
4372 unsigned long pfn
, bitidx
;
4373 unsigned long value
= 1;
4375 zone
= page_zone(page
);
4376 pfn
= page_to_pfn(page
);
4377 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4378 bitidx
= pfn_to_bitidx(zone
, pfn
);
4380 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4382 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4384 __clear_bit(bitidx
+ start_bitidx
, bitmap
);