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/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/oom.h>
32 #include <linux/notifier.h>
33 #include <linux/topology.h>
34 #include <linux/sysctl.h>
35 #include <linux/cpu.h>
36 #include <linux/cpuset.h>
37 #include <linux/memory_hotplug.h>
38 #include <linux/nodemask.h>
39 #include <linux/vmalloc.h>
40 #include <linux/mempolicy.h>
41 #include <linux/stop_machine.h>
42 #include <linux/sort.h>
43 #include <linux/pfn.h>
44 #include <linux/backing-dev.h>
45 #include <linux/fault-inject.h>
46 #include <linux/page-isolation.h>
47 #include <linux/page_cgroup.h>
48 #include <linux/debugobjects.h>
50 #include <asm/tlbflush.h>
51 #include <asm/div64.h>
55 * Array of node states.
57 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
58 [N_POSSIBLE
] = NODE_MASK_ALL
,
59 [N_ONLINE
] = { { [0] = 1UL } },
61 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
63 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
65 [N_CPU
] = { { [0] = 1UL } },
68 EXPORT_SYMBOL(node_states
);
70 unsigned long totalram_pages __read_mostly
;
71 unsigned long totalreserve_pages __read_mostly
;
72 unsigned long highest_memmap_pfn __read_mostly
;
73 int percpu_pagelist_fraction
;
75 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
76 int pageblock_order __read_mostly
;
79 static void __free_pages_ok(struct page
*page
, unsigned int order
);
82 * results with 256, 32 in the lowmem_reserve sysctl:
83 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
84 * 1G machine -> (16M dma, 784M normal, 224M high)
85 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
86 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
87 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
89 * TBD: should special case ZONE_DMA32 machines here - in those we normally
90 * don't need any ZONE_NORMAL reservation
92 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
93 #ifdef CONFIG_ZONE_DMA
96 #ifdef CONFIG_ZONE_DMA32
105 EXPORT_SYMBOL(totalram_pages
);
107 static char * const zone_names
[MAX_NR_ZONES
] = {
108 #ifdef CONFIG_ZONE_DMA
111 #ifdef CONFIG_ZONE_DMA32
115 #ifdef CONFIG_HIGHMEM
121 int min_free_kbytes
= 1024;
123 unsigned long __meminitdata nr_kernel_pages
;
124 unsigned long __meminitdata nr_all_pages
;
125 static unsigned long __meminitdata dma_reserve
;
127 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
129 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
130 * ranges of memory (RAM) that may be registered with add_active_range().
131 * Ranges passed to add_active_range() will be merged if possible
132 * so the number of times add_active_range() can be called is
133 * related to the number of nodes and the number of holes
135 #ifdef CONFIG_MAX_ACTIVE_REGIONS
136 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
137 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
139 #if MAX_NUMNODES >= 32
140 /* If there can be many nodes, allow up to 50 holes per node */
141 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
143 /* By default, allow up to 256 distinct regions */
144 #define MAX_ACTIVE_REGIONS 256
148 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
149 static int __meminitdata nr_nodemap_entries
;
150 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
151 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
152 static unsigned long __initdata required_kernelcore
;
153 static unsigned long __initdata required_movablecore
;
154 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
156 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
158 EXPORT_SYMBOL(movable_zone
);
159 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
162 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
163 EXPORT_SYMBOL(nr_node_ids
);
166 int page_group_by_mobility_disabled __read_mostly
;
168 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
170 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
171 PB_migrate
, PB_migrate_end
);
174 #ifdef CONFIG_DEBUG_VM
175 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
179 unsigned long pfn
= page_to_pfn(page
);
182 seq
= zone_span_seqbegin(zone
);
183 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
185 else if (pfn
< zone
->zone_start_pfn
)
187 } while (zone_span_seqretry(zone
, seq
));
192 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
194 if (!pfn_valid_within(page_to_pfn(page
)))
196 if (zone
!= page_zone(page
))
202 * Temporary debugging check for pages not lying within a given zone.
204 static int bad_range(struct zone
*zone
, struct page
*page
)
206 if (page_outside_zone_boundaries(zone
, page
))
208 if (!page_is_consistent(zone
, page
))
214 static inline int bad_range(struct zone
*zone
, struct page
*page
)
220 static void bad_page(struct page
*page
)
222 static unsigned long resume
;
223 static unsigned long nr_shown
;
224 static unsigned long nr_unshown
;
227 * Allow a burst of 60 reports, then keep quiet for that minute;
228 * or allow a steady drip of one report per second.
230 if (nr_shown
== 60) {
231 if (time_before(jiffies
, resume
)) {
237 "BUG: Bad page state: %lu messages suppressed\n",
244 resume
= jiffies
+ 60 * HZ
;
246 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
247 current
->comm
, page_to_pfn(page
));
249 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
250 page
, (void *)page
->flags
, page_count(page
),
251 page_mapcount(page
), page
->mapping
, page
->index
);
255 /* Leave bad fields for debug, except PageBuddy could make trouble */
256 __ClearPageBuddy(page
);
257 add_taint(TAINT_BAD_PAGE
);
261 * Higher-order pages are called "compound pages". They are structured thusly:
263 * The first PAGE_SIZE page is called the "head page".
265 * The remaining PAGE_SIZE pages are called "tail pages".
267 * All pages have PG_compound set. All pages have their ->private pointing at
268 * the head page (even the head page has this).
270 * The first tail page's ->lru.next holds the address of the compound page's
271 * put_page() function. Its ->lru.prev holds the order of allocation.
272 * This usage means that zero-order pages may not be compound.
275 static void free_compound_page(struct page
*page
)
277 __free_pages_ok(page
, compound_order(page
));
280 void prep_compound_page(struct page
*page
, unsigned long order
)
283 int nr_pages
= 1 << order
;
285 set_compound_page_dtor(page
, free_compound_page
);
286 set_compound_order(page
, order
);
288 for (i
= 1; i
< nr_pages
; i
++) {
289 struct page
*p
= page
+ i
;
292 p
->first_page
= page
;
296 #ifdef CONFIG_HUGETLBFS
297 void prep_compound_gigantic_page(struct page
*page
, unsigned long order
)
300 int nr_pages
= 1 << order
;
301 struct page
*p
= page
+ 1;
303 set_compound_page_dtor(page
, free_compound_page
);
304 set_compound_order(page
, order
);
306 for (i
= 1; i
< nr_pages
; i
++, p
= mem_map_next(p
, page
, i
)) {
308 p
->first_page
= page
;
313 static int destroy_compound_page(struct page
*page
, unsigned long order
)
316 int nr_pages
= 1 << order
;
319 if (unlikely(compound_order(page
) != order
) ||
320 unlikely(!PageHead(page
))) {
325 __ClearPageHead(page
);
327 for (i
= 1; i
< nr_pages
; i
++) {
328 struct page
*p
= page
+ i
;
330 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
340 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
345 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
346 * and __GFP_HIGHMEM from hard or soft interrupt context.
348 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
349 for (i
= 0; i
< (1 << order
); i
++)
350 clear_highpage(page
+ i
);
353 static inline void set_page_order(struct page
*page
, int order
)
355 set_page_private(page
, order
);
356 __SetPageBuddy(page
);
359 static inline void rmv_page_order(struct page
*page
)
361 __ClearPageBuddy(page
);
362 set_page_private(page
, 0);
366 * Locate the struct page for both the matching buddy in our
367 * pair (buddy1) and the combined O(n+1) page they form (page).
369 * 1) Any buddy B1 will have an order O twin B2 which satisfies
370 * the following equation:
372 * For example, if the starting buddy (buddy2) is #8 its order
374 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
376 * 2) Any buddy B will have an order O+1 parent P which
377 * satisfies the following equation:
380 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
382 static inline struct page
*
383 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
385 unsigned long buddy_idx
= page_idx
^ (1 << order
);
387 return page
+ (buddy_idx
- page_idx
);
390 static inline unsigned long
391 __find_combined_index(unsigned long page_idx
, unsigned int order
)
393 return (page_idx
& ~(1 << order
));
397 * This function checks whether a page is free && is the buddy
398 * we can do coalesce a page and its buddy if
399 * (a) the buddy is not in a hole &&
400 * (b) the buddy is in the buddy system &&
401 * (c) a page and its buddy have the same order &&
402 * (d) a page and its buddy are in the same zone.
404 * For recording whether a page is in the buddy system, we use PG_buddy.
405 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
407 * For recording page's order, we use page_private(page).
409 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
412 if (!pfn_valid_within(page_to_pfn(buddy
)))
415 if (page_zone_id(page
) != page_zone_id(buddy
))
418 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
419 BUG_ON(page_count(buddy
) != 0);
426 * Freeing function for a buddy system allocator.
428 * The concept of a buddy system is to maintain direct-mapped table
429 * (containing bit values) for memory blocks of various "orders".
430 * The bottom level table contains the map for the smallest allocatable
431 * units of memory (here, pages), and each level above it describes
432 * pairs of units from the levels below, hence, "buddies".
433 * At a high level, all that happens here is marking the table entry
434 * at the bottom level available, and propagating the changes upward
435 * as necessary, plus some accounting needed to play nicely with other
436 * parts of the VM system.
437 * At each level, we keep a list of pages, which are heads of continuous
438 * free pages of length of (1 << order) and marked with PG_buddy. Page's
439 * order is recorded in page_private(page) field.
440 * So when we are allocating or freeing one, we can derive the state of the
441 * other. That is, if we allocate a small block, and both were
442 * free, the remainder of the region must be split into blocks.
443 * If a block is freed, and its buddy is also free, then this
444 * triggers coalescing into a block of larger size.
449 static inline void __free_one_page(struct page
*page
,
450 struct zone
*zone
, unsigned int order
)
452 unsigned long page_idx
;
453 int order_size
= 1 << order
;
454 int migratetype
= get_pageblock_migratetype(page
);
456 if (unlikely(PageCompound(page
)))
457 if (unlikely(destroy_compound_page(page
, order
)))
460 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
462 VM_BUG_ON(page_idx
& (order_size
- 1));
463 VM_BUG_ON(bad_range(zone
, page
));
465 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
466 while (order
< MAX_ORDER
-1) {
467 unsigned long combined_idx
;
470 buddy
= __page_find_buddy(page
, page_idx
, order
);
471 if (!page_is_buddy(page
, buddy
, order
))
474 /* Our buddy is free, merge with it and move up one order. */
475 list_del(&buddy
->lru
);
476 zone
->free_area
[order
].nr_free
--;
477 rmv_page_order(buddy
);
478 combined_idx
= __find_combined_index(page_idx
, order
);
479 page
= page
+ (combined_idx
- page_idx
);
480 page_idx
= combined_idx
;
483 set_page_order(page
, order
);
485 &zone
->free_area
[order
].free_list
[migratetype
]);
486 zone
->free_area
[order
].nr_free
++;
489 static inline int free_pages_check(struct page
*page
)
491 free_page_mlock(page
);
492 if (unlikely(page_mapcount(page
) |
493 (page
->mapping
!= NULL
) |
494 (page_count(page
) != 0) |
495 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
499 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
500 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
505 * Frees a list of pages.
506 * Assumes all pages on list are in same zone, and of same order.
507 * count is the number of pages to free.
509 * If the zone was previously in an "all pages pinned" state then look to
510 * see if this freeing clears that state.
512 * And clear the zone's pages_scanned counter, to hold off the "all pages are
513 * pinned" detection logic.
515 static void free_pages_bulk(struct zone
*zone
, int count
,
516 struct list_head
*list
, int order
)
518 spin_lock(&zone
->lock
);
519 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
520 zone
->pages_scanned
= 0;
524 VM_BUG_ON(list_empty(list
));
525 page
= list_entry(list
->prev
, struct page
, lru
);
526 /* have to delete it as __free_one_page list manipulates */
527 list_del(&page
->lru
);
528 __free_one_page(page
, zone
, order
);
530 spin_unlock(&zone
->lock
);
533 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
535 spin_lock(&zone
->lock
);
536 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
537 zone
->pages_scanned
= 0;
538 __free_one_page(page
, zone
, order
);
539 spin_unlock(&zone
->lock
);
542 static void __free_pages_ok(struct page
*page
, unsigned int order
)
548 for (i
= 0 ; i
< (1 << order
) ; ++i
)
549 bad
+= free_pages_check(page
+ i
);
553 if (!PageHighMem(page
)) {
554 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
555 debug_check_no_obj_freed(page_address(page
),
558 arch_free_page(page
, order
);
559 kernel_map_pages(page
, 1 << order
, 0);
561 local_irq_save(flags
);
562 __count_vm_events(PGFREE
, 1 << order
);
563 free_one_page(page_zone(page
), page
, order
);
564 local_irq_restore(flags
);
568 * permit the bootmem allocator to evade page validation on high-order frees
570 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
573 __ClearPageReserved(page
);
574 set_page_count(page
, 0);
575 set_page_refcounted(page
);
581 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
582 struct page
*p
= &page
[loop
];
584 if (loop
+ 1 < BITS_PER_LONG
)
586 __ClearPageReserved(p
);
587 set_page_count(p
, 0);
590 set_page_refcounted(page
);
591 __free_pages(page
, order
);
597 * The order of subdivision here is critical for the IO subsystem.
598 * Please do not alter this order without good reasons and regression
599 * testing. Specifically, as large blocks of memory are subdivided,
600 * the order in which smaller blocks are delivered depends on the order
601 * they're subdivided in this function. This is the primary factor
602 * influencing the order in which pages are delivered to the IO
603 * subsystem according to empirical testing, and this is also justified
604 * by considering the behavior of a buddy system containing a single
605 * large block of memory acted on by a series of small allocations.
606 * This behavior is a critical factor in sglist merging's success.
610 static inline void expand(struct zone
*zone
, struct page
*page
,
611 int low
, int high
, struct free_area
*area
,
614 unsigned long size
= 1 << high
;
620 VM_BUG_ON(bad_range(zone
, &page
[size
]));
621 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
623 set_page_order(&page
[size
], high
);
628 * This page is about to be returned from the page allocator
630 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
632 if (unlikely(page_mapcount(page
) |
633 (page
->mapping
!= NULL
) |
634 (page_count(page
) != 0) |
635 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
640 set_page_private(page
, 0);
641 set_page_refcounted(page
);
643 arch_alloc_page(page
, order
);
644 kernel_map_pages(page
, 1 << order
, 1);
646 if (gfp_flags
& __GFP_ZERO
)
647 prep_zero_page(page
, order
, gfp_flags
);
649 if (order
&& (gfp_flags
& __GFP_COMP
))
650 prep_compound_page(page
, order
);
656 * Go through the free lists for the given migratetype and remove
657 * the smallest available page from the freelists
659 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
662 unsigned int current_order
;
663 struct free_area
* area
;
666 /* Find a page of the appropriate size in the preferred list */
667 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
668 area
= &(zone
->free_area
[current_order
]);
669 if (list_empty(&area
->free_list
[migratetype
]))
672 page
= list_entry(area
->free_list
[migratetype
].next
,
674 list_del(&page
->lru
);
675 rmv_page_order(page
);
677 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
678 expand(zone
, page
, order
, current_order
, area
, migratetype
);
687 * This array describes the order lists are fallen back to when
688 * the free lists for the desirable migrate type are depleted
690 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
691 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
692 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
693 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
694 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
698 * Move the free pages in a range to the free lists of the requested type.
699 * Note that start_page and end_pages are not aligned on a pageblock
700 * boundary. If alignment is required, use move_freepages_block()
702 static int move_freepages(struct zone
*zone
,
703 struct page
*start_page
, struct page
*end_page
,
710 #ifndef CONFIG_HOLES_IN_ZONE
712 * page_zone is not safe to call in this context when
713 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
714 * anyway as we check zone boundaries in move_freepages_block().
715 * Remove at a later date when no bug reports exist related to
716 * grouping pages by mobility
718 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
721 for (page
= start_page
; page
<= end_page
;) {
722 /* Make sure we are not inadvertently changing nodes */
723 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
725 if (!pfn_valid_within(page_to_pfn(page
))) {
730 if (!PageBuddy(page
)) {
735 order
= page_order(page
);
736 list_del(&page
->lru
);
738 &zone
->free_area
[order
].free_list
[migratetype
]);
740 pages_moved
+= 1 << order
;
746 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
749 unsigned long start_pfn
, end_pfn
;
750 struct page
*start_page
, *end_page
;
752 start_pfn
= page_to_pfn(page
);
753 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
754 start_page
= pfn_to_page(start_pfn
);
755 end_page
= start_page
+ pageblock_nr_pages
- 1;
756 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
758 /* Do not cross zone boundaries */
759 if (start_pfn
< zone
->zone_start_pfn
)
761 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
764 return move_freepages(zone
, start_page
, end_page
, migratetype
);
767 /* Remove an element from the buddy allocator from the fallback list */
768 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
769 int start_migratetype
)
771 struct free_area
* area
;
776 /* Find the largest possible block of pages in the other list */
777 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
779 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
780 migratetype
= fallbacks
[start_migratetype
][i
];
782 /* MIGRATE_RESERVE handled later if necessary */
783 if (migratetype
== MIGRATE_RESERVE
)
786 area
= &(zone
->free_area
[current_order
]);
787 if (list_empty(&area
->free_list
[migratetype
]))
790 page
= list_entry(area
->free_list
[migratetype
].next
,
795 * If breaking a large block of pages, move all free
796 * pages to the preferred allocation list. If falling
797 * back for a reclaimable kernel allocation, be more
798 * agressive about taking ownership of free pages
800 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
801 start_migratetype
== MIGRATE_RECLAIMABLE
) {
803 pages
= move_freepages_block(zone
, page
,
806 /* Claim the whole block if over half of it is free */
807 if (pages
>= (1 << (pageblock_order
-1)))
808 set_pageblock_migratetype(page
,
811 migratetype
= start_migratetype
;
814 /* Remove the page from the freelists */
815 list_del(&page
->lru
);
816 rmv_page_order(page
);
817 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
820 if (current_order
== pageblock_order
)
821 set_pageblock_migratetype(page
,
824 expand(zone
, page
, order
, current_order
, area
, migratetype
);
829 /* Use MIGRATE_RESERVE rather than fail an allocation */
830 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
834 * Do the hard work of removing an element from the buddy allocator.
835 * Call me with the zone->lock already held.
837 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
842 page
= __rmqueue_smallest(zone
, order
, migratetype
);
845 page
= __rmqueue_fallback(zone
, order
, migratetype
);
851 * Obtain a specified number of elements from the buddy allocator, all under
852 * a single hold of the lock, for efficiency. Add them to the supplied list.
853 * Returns the number of new pages which were placed at *list.
855 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
856 unsigned long count
, struct list_head
*list
,
861 spin_lock(&zone
->lock
);
862 for (i
= 0; i
< count
; ++i
) {
863 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
864 if (unlikely(page
== NULL
))
868 * Split buddy pages returned by expand() are received here
869 * in physical page order. The page is added to the callers and
870 * list and the list head then moves forward. From the callers
871 * perspective, the linked list is ordered by page number in
872 * some conditions. This is useful for IO devices that can
873 * merge IO requests if the physical pages are ordered
876 list_add(&page
->lru
, list
);
877 set_page_private(page
, migratetype
);
880 spin_unlock(&zone
->lock
);
886 * Called from the vmstat counter updater to drain pagesets of this
887 * currently executing processor on remote nodes after they have
890 * Note that this function must be called with the thread pinned to
891 * a single processor.
893 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
898 local_irq_save(flags
);
899 if (pcp
->count
>= pcp
->batch
)
900 to_drain
= pcp
->batch
;
902 to_drain
= pcp
->count
;
903 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
904 pcp
->count
-= to_drain
;
905 local_irq_restore(flags
);
910 * Drain pages of the indicated processor.
912 * The processor must either be the current processor and the
913 * thread pinned to the current processor or a processor that
916 static void drain_pages(unsigned int cpu
)
921 for_each_populated_zone(zone
) {
922 struct per_cpu_pageset
*pset
;
923 struct per_cpu_pages
*pcp
;
925 pset
= zone_pcp(zone
, cpu
);
928 local_irq_save(flags
);
929 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
931 local_irq_restore(flags
);
936 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
938 void drain_local_pages(void *arg
)
940 drain_pages(smp_processor_id());
944 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
946 void drain_all_pages(void)
948 on_each_cpu(drain_local_pages
, NULL
, 1);
951 #ifdef CONFIG_HIBERNATION
953 void mark_free_pages(struct zone
*zone
)
955 unsigned long pfn
, max_zone_pfn
;
958 struct list_head
*curr
;
960 if (!zone
->spanned_pages
)
963 spin_lock_irqsave(&zone
->lock
, flags
);
965 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
966 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
967 if (pfn_valid(pfn
)) {
968 struct page
*page
= pfn_to_page(pfn
);
970 if (!swsusp_page_is_forbidden(page
))
971 swsusp_unset_page_free(page
);
974 for_each_migratetype_order(order
, t
) {
975 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
978 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
979 for (i
= 0; i
< (1UL << order
); i
++)
980 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
983 spin_unlock_irqrestore(&zone
->lock
, flags
);
985 #endif /* CONFIG_PM */
988 * Free a 0-order page
990 static void free_hot_cold_page(struct page
*page
, int cold
)
992 struct zone
*zone
= page_zone(page
);
993 struct per_cpu_pages
*pcp
;
997 page
->mapping
= NULL
;
998 if (free_pages_check(page
))
1001 if (!PageHighMem(page
)) {
1002 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1003 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1005 arch_free_page(page
, 0);
1006 kernel_map_pages(page
, 1, 0);
1008 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1009 local_irq_save(flags
);
1010 __count_vm_event(PGFREE
);
1012 list_add_tail(&page
->lru
, &pcp
->list
);
1014 list_add(&page
->lru
, &pcp
->list
);
1015 set_page_private(page
, get_pageblock_migratetype(page
));
1017 if (pcp
->count
>= pcp
->high
) {
1018 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1019 pcp
->count
-= pcp
->batch
;
1021 local_irq_restore(flags
);
1025 void free_hot_page(struct page
*page
)
1027 free_hot_cold_page(page
, 0);
1030 void free_cold_page(struct page
*page
)
1032 free_hot_cold_page(page
, 1);
1036 * split_page takes a non-compound higher-order page, and splits it into
1037 * n (1<<order) sub-pages: page[0..n]
1038 * Each sub-page must be freed individually.
1040 * Note: this is probably too low level an operation for use in drivers.
1041 * Please consult with lkml before using this in your driver.
1043 void split_page(struct page
*page
, unsigned int order
)
1047 VM_BUG_ON(PageCompound(page
));
1048 VM_BUG_ON(!page_count(page
));
1049 for (i
= 1; i
< (1 << order
); i
++)
1050 set_page_refcounted(page
+ i
);
1054 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1055 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1058 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1059 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1061 unsigned long flags
;
1063 int cold
= !!(gfp_flags
& __GFP_COLD
);
1065 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1069 if (likely(order
== 0)) {
1070 struct per_cpu_pages
*pcp
;
1072 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1073 local_irq_save(flags
);
1075 pcp
->count
= rmqueue_bulk(zone
, 0,
1076 pcp
->batch
, &pcp
->list
, migratetype
);
1077 if (unlikely(!pcp
->count
))
1081 /* Find a page of the appropriate migrate type */
1083 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1084 if (page_private(page
) == migratetype
)
1087 list_for_each_entry(page
, &pcp
->list
, lru
)
1088 if (page_private(page
) == migratetype
)
1092 /* Allocate more to the pcp list if necessary */
1093 if (unlikely(&page
->lru
== &pcp
->list
)) {
1094 pcp
->count
+= rmqueue_bulk(zone
, 0,
1095 pcp
->batch
, &pcp
->list
, migratetype
);
1096 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1099 list_del(&page
->lru
);
1102 spin_lock_irqsave(&zone
->lock
, flags
);
1103 page
= __rmqueue(zone
, order
, migratetype
);
1104 spin_unlock(&zone
->lock
);
1109 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1110 zone_statistics(preferred_zone
, zone
);
1111 local_irq_restore(flags
);
1114 VM_BUG_ON(bad_range(zone
, page
));
1115 if (prep_new_page(page
, order
, gfp_flags
))
1120 local_irq_restore(flags
);
1125 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1126 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1127 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1128 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1129 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1130 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1131 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1133 #ifdef CONFIG_FAIL_PAGE_ALLOC
1135 static struct fail_page_alloc_attr
{
1136 struct fault_attr attr
;
1138 u32 ignore_gfp_highmem
;
1139 u32 ignore_gfp_wait
;
1142 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1144 struct dentry
*ignore_gfp_highmem_file
;
1145 struct dentry
*ignore_gfp_wait_file
;
1146 struct dentry
*min_order_file
;
1148 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1150 } fail_page_alloc
= {
1151 .attr
= FAULT_ATTR_INITIALIZER
,
1152 .ignore_gfp_wait
= 1,
1153 .ignore_gfp_highmem
= 1,
1157 static int __init
setup_fail_page_alloc(char *str
)
1159 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1161 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1163 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1165 if (order
< fail_page_alloc
.min_order
)
1167 if (gfp_mask
& __GFP_NOFAIL
)
1169 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1171 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1174 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1177 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1179 static int __init
fail_page_alloc_debugfs(void)
1181 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1185 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1189 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1191 fail_page_alloc
.ignore_gfp_wait_file
=
1192 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1193 &fail_page_alloc
.ignore_gfp_wait
);
1195 fail_page_alloc
.ignore_gfp_highmem_file
=
1196 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1197 &fail_page_alloc
.ignore_gfp_highmem
);
1198 fail_page_alloc
.min_order_file
=
1199 debugfs_create_u32("min-order", mode
, dir
,
1200 &fail_page_alloc
.min_order
);
1202 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1203 !fail_page_alloc
.ignore_gfp_highmem_file
||
1204 !fail_page_alloc
.min_order_file
) {
1206 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1207 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1208 debugfs_remove(fail_page_alloc
.min_order_file
);
1209 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1215 late_initcall(fail_page_alloc_debugfs
);
1217 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1219 #else /* CONFIG_FAIL_PAGE_ALLOC */
1221 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1226 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1229 * Return 1 if free pages are above 'mark'. This takes into account the order
1230 * of the allocation.
1232 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1233 int classzone_idx
, int alloc_flags
)
1235 /* free_pages my go negative - that's OK */
1237 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1240 if (alloc_flags
& ALLOC_HIGH
)
1242 if (alloc_flags
& ALLOC_HARDER
)
1245 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1247 for (o
= 0; o
< order
; o
++) {
1248 /* At the next order, this order's pages become unavailable */
1249 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1251 /* Require fewer higher order pages to be free */
1254 if (free_pages
<= min
)
1262 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1263 * skip over zones that are not allowed by the cpuset, or that have
1264 * been recently (in last second) found to be nearly full. See further
1265 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1266 * that have to skip over a lot of full or unallowed zones.
1268 * If the zonelist cache is present in the passed in zonelist, then
1269 * returns a pointer to the allowed node mask (either the current
1270 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1272 * If the zonelist cache is not available for this zonelist, does
1273 * nothing and returns NULL.
1275 * If the fullzones BITMAP in the zonelist cache is stale (more than
1276 * a second since last zap'd) then we zap it out (clear its bits.)
1278 * We hold off even calling zlc_setup, until after we've checked the
1279 * first zone in the zonelist, on the theory that most allocations will
1280 * be satisfied from that first zone, so best to examine that zone as
1281 * quickly as we can.
1283 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1285 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1286 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1288 zlc
= zonelist
->zlcache_ptr
;
1292 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1293 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1294 zlc
->last_full_zap
= jiffies
;
1297 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1298 &cpuset_current_mems_allowed
:
1299 &node_states
[N_HIGH_MEMORY
];
1300 return allowednodes
;
1304 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1305 * if it is worth looking at further for free memory:
1306 * 1) Check that the zone isn't thought to be full (doesn't have its
1307 * bit set in the zonelist_cache fullzones BITMAP).
1308 * 2) Check that the zones node (obtained from the zonelist_cache
1309 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1310 * Return true (non-zero) if zone is worth looking at further, or
1311 * else return false (zero) if it is not.
1313 * This check -ignores- the distinction between various watermarks,
1314 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1315 * found to be full for any variation of these watermarks, it will
1316 * be considered full for up to one second by all requests, unless
1317 * we are so low on memory on all allowed nodes that we are forced
1318 * into the second scan of the zonelist.
1320 * In the second scan we ignore this zonelist cache and exactly
1321 * apply the watermarks to all zones, even it is slower to do so.
1322 * We are low on memory in the second scan, and should leave no stone
1323 * unturned looking for a free page.
1325 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1326 nodemask_t
*allowednodes
)
1328 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1329 int i
; /* index of *z in zonelist zones */
1330 int n
; /* node that zone *z is on */
1332 zlc
= zonelist
->zlcache_ptr
;
1336 i
= z
- zonelist
->_zonerefs
;
1339 /* This zone is worth trying if it is allowed but not full */
1340 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1344 * Given 'z' scanning a zonelist, set the corresponding bit in
1345 * zlc->fullzones, so that subsequent attempts to allocate a page
1346 * from that zone don't waste time re-examining it.
1348 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1350 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1351 int i
; /* index of *z in zonelist zones */
1353 zlc
= zonelist
->zlcache_ptr
;
1357 i
= z
- zonelist
->_zonerefs
;
1359 set_bit(i
, zlc
->fullzones
);
1362 #else /* CONFIG_NUMA */
1364 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1369 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1370 nodemask_t
*allowednodes
)
1375 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1378 #endif /* CONFIG_NUMA */
1381 * get_page_from_freelist goes through the zonelist trying to allocate
1384 static struct page
*
1385 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1386 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1389 struct page
*page
= NULL
;
1391 struct zone
*zone
, *preferred_zone
;
1392 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1393 int zlc_active
= 0; /* set if using zonelist_cache */
1394 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1396 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1398 if (!preferred_zone
)
1401 classzone_idx
= zone_idx(preferred_zone
);
1405 * Scan zonelist, looking for a zone with enough free.
1406 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1408 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1409 high_zoneidx
, nodemask
) {
1410 if (NUMA_BUILD
&& zlc_active
&&
1411 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1413 if ((alloc_flags
& ALLOC_CPUSET
) &&
1414 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1417 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1419 if (alloc_flags
& ALLOC_WMARK_MIN
)
1420 mark
= zone
->pages_min
;
1421 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1422 mark
= zone
->pages_low
;
1424 mark
= zone
->pages_high
;
1425 if (!zone_watermark_ok(zone
, order
, mark
,
1426 classzone_idx
, alloc_flags
)) {
1427 if (!zone_reclaim_mode
||
1428 !zone_reclaim(zone
, gfp_mask
, order
))
1429 goto this_zone_full
;
1433 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1438 zlc_mark_zone_full(zonelist
, z
);
1440 if (NUMA_BUILD
&& !did_zlc_setup
) {
1441 /* we do zlc_setup after the first zone is tried */
1442 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1448 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1449 /* Disable zlc cache for second zonelist scan */
1457 * This is the 'heart' of the zoned buddy allocator.
1460 __alloc_pages_internal(gfp_t gfp_mask
, unsigned int order
,
1461 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1463 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1464 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1468 struct reclaim_state reclaim_state
;
1469 struct task_struct
*p
= current
;
1472 unsigned long did_some_progress
;
1473 unsigned long pages_reclaimed
= 0;
1475 lockdep_trace_alloc(gfp_mask
);
1477 might_sleep_if(wait
);
1479 if (should_fail_alloc_page(gfp_mask
, order
))
1483 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1485 if (unlikely(!z
->zone
)) {
1487 * Happens if we have an empty zonelist as a result of
1488 * GFP_THISNODE being used on a memoryless node
1493 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1494 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1499 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1500 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1501 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1502 * using a larger set of nodes after it has established that the
1503 * allowed per node queues are empty and that nodes are
1506 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1509 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1510 wakeup_kswapd(zone
, order
);
1513 * OK, we're below the kswapd watermark and have kicked background
1514 * reclaim. Now things get more complex, so set up alloc_flags according
1515 * to how we want to proceed.
1517 * The caller may dip into page reserves a bit more if the caller
1518 * cannot run direct reclaim, or if the caller has realtime scheduling
1519 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1520 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1522 alloc_flags
= ALLOC_WMARK_MIN
;
1523 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1524 alloc_flags
|= ALLOC_HARDER
;
1525 if (gfp_mask
& __GFP_HIGH
)
1526 alloc_flags
|= ALLOC_HIGH
;
1528 alloc_flags
|= ALLOC_CPUSET
;
1531 * Go through the zonelist again. Let __GFP_HIGH and allocations
1532 * coming from realtime tasks go deeper into reserves.
1534 * This is the last chance, in general, before the goto nopage.
1535 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1536 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1538 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1539 high_zoneidx
, alloc_flags
);
1543 /* This allocation should allow future memory freeing. */
1546 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1547 && !in_interrupt()) {
1548 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1550 /* go through the zonelist yet again, ignoring mins */
1551 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1552 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1555 if (gfp_mask
& __GFP_NOFAIL
) {
1556 congestion_wait(WRITE
, HZ
/50);
1563 /* Atomic allocations - we can't balance anything */
1569 /* We now go into synchronous reclaim */
1570 cpuset_memory_pressure_bump();
1572 * The task's cpuset might have expanded its set of allowable nodes
1574 cpuset_update_task_memory_state();
1575 p
->flags
|= PF_MEMALLOC
;
1577 lockdep_set_current_reclaim_state(gfp_mask
);
1578 reclaim_state
.reclaimed_slab
= 0;
1579 p
->reclaim_state
= &reclaim_state
;
1581 did_some_progress
= try_to_free_pages(zonelist
, order
,
1582 gfp_mask
, nodemask
);
1584 p
->reclaim_state
= NULL
;
1585 lockdep_clear_current_reclaim_state();
1586 p
->flags
&= ~PF_MEMALLOC
;
1593 if (likely(did_some_progress
)) {
1594 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1595 zonelist
, high_zoneidx
, alloc_flags
);
1598 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1599 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1600 schedule_timeout_uninterruptible(1);
1605 * Go through the zonelist yet one more time, keep
1606 * very high watermark here, this is only to catch
1607 * a parallel oom killing, we must fail if we're still
1608 * under heavy pressure.
1610 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1611 order
, zonelist
, high_zoneidx
,
1612 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1614 clear_zonelist_oom(zonelist
, gfp_mask
);
1618 /* The OOM killer will not help higher order allocs so fail */
1619 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1620 clear_zonelist_oom(zonelist
, gfp_mask
);
1624 out_of_memory(zonelist
, gfp_mask
, order
);
1625 clear_zonelist_oom(zonelist
, gfp_mask
);
1630 * Don't let big-order allocations loop unless the caller explicitly
1631 * requests that. Wait for some write requests to complete then retry.
1633 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1634 * means __GFP_NOFAIL, but that may not be true in other
1637 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1638 * specified, then we retry until we no longer reclaim any pages
1639 * (above), or we've reclaimed an order of pages at least as
1640 * large as the allocation's order. In both cases, if the
1641 * allocation still fails, we stop retrying.
1643 pages_reclaimed
+= did_some_progress
;
1645 if (!(gfp_mask
& __GFP_NORETRY
)) {
1646 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1649 if (gfp_mask
& __GFP_REPEAT
&&
1650 pages_reclaimed
< (1 << order
))
1653 if (gfp_mask
& __GFP_NOFAIL
)
1657 congestion_wait(WRITE
, HZ
/50);
1662 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1663 printk(KERN_WARNING
"%s: page allocation failure."
1664 " order:%d, mode:0x%x\n",
1665 p
->comm
, order
, gfp_mask
);
1672 EXPORT_SYMBOL(__alloc_pages_internal
);
1675 * Common helper functions.
1677 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1680 page
= alloc_pages(gfp_mask
, order
);
1683 return (unsigned long) page_address(page
);
1686 EXPORT_SYMBOL(__get_free_pages
);
1688 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1693 * get_zeroed_page() returns a 32-bit address, which cannot represent
1696 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1698 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1700 return (unsigned long) page_address(page
);
1704 EXPORT_SYMBOL(get_zeroed_page
);
1706 void __pagevec_free(struct pagevec
*pvec
)
1708 int i
= pagevec_count(pvec
);
1711 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1714 void __free_pages(struct page
*page
, unsigned int order
)
1716 if (put_page_testzero(page
)) {
1718 free_hot_page(page
);
1720 __free_pages_ok(page
, order
);
1724 EXPORT_SYMBOL(__free_pages
);
1726 void free_pages(unsigned long addr
, unsigned int order
)
1729 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1730 __free_pages(virt_to_page((void *)addr
), order
);
1734 EXPORT_SYMBOL(free_pages
);
1737 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1738 * @size: the number of bytes to allocate
1739 * @gfp_mask: GFP flags for the allocation
1741 * This function is similar to alloc_pages(), except that it allocates the
1742 * minimum number of pages to satisfy the request. alloc_pages() can only
1743 * allocate memory in power-of-two pages.
1745 * This function is also limited by MAX_ORDER.
1747 * Memory allocated by this function must be released by free_pages_exact().
1749 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1751 unsigned int order
= get_order(size
);
1754 addr
= __get_free_pages(gfp_mask
, order
);
1756 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1757 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1759 split_page(virt_to_page(addr
), order
);
1760 while (used
< alloc_end
) {
1766 return (void *)addr
;
1768 EXPORT_SYMBOL(alloc_pages_exact
);
1771 * free_pages_exact - release memory allocated via alloc_pages_exact()
1772 * @virt: the value returned by alloc_pages_exact.
1773 * @size: size of allocation, same value as passed to alloc_pages_exact().
1775 * Release the memory allocated by a previous call to alloc_pages_exact.
1777 void free_pages_exact(void *virt
, size_t size
)
1779 unsigned long addr
= (unsigned long)virt
;
1780 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1782 while (addr
< end
) {
1787 EXPORT_SYMBOL(free_pages_exact
);
1789 static unsigned int nr_free_zone_pages(int offset
)
1794 /* Just pick one node, since fallback list is circular */
1795 unsigned int sum
= 0;
1797 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1799 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1800 unsigned long size
= zone
->present_pages
;
1801 unsigned long high
= zone
->pages_high
;
1810 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1812 unsigned int nr_free_buffer_pages(void)
1814 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1816 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1819 * Amount of free RAM allocatable within all zones
1821 unsigned int nr_free_pagecache_pages(void)
1823 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1826 static inline void show_node(struct zone
*zone
)
1829 printk("Node %d ", zone_to_nid(zone
));
1832 void si_meminfo(struct sysinfo
*val
)
1834 val
->totalram
= totalram_pages
;
1836 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1837 val
->bufferram
= nr_blockdev_pages();
1838 val
->totalhigh
= totalhigh_pages
;
1839 val
->freehigh
= nr_free_highpages();
1840 val
->mem_unit
= PAGE_SIZE
;
1843 EXPORT_SYMBOL(si_meminfo
);
1846 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1848 pg_data_t
*pgdat
= NODE_DATA(nid
);
1850 val
->totalram
= pgdat
->node_present_pages
;
1851 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1852 #ifdef CONFIG_HIGHMEM
1853 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1854 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1860 val
->mem_unit
= PAGE_SIZE
;
1864 #define K(x) ((x) << (PAGE_SHIFT-10))
1867 * Show free area list (used inside shift_scroll-lock stuff)
1868 * We also calculate the percentage fragmentation. We do this by counting the
1869 * memory on each free list with the exception of the first item on the list.
1871 void show_free_areas(void)
1876 for_each_populated_zone(zone
) {
1878 printk("%s per-cpu:\n", zone
->name
);
1880 for_each_online_cpu(cpu
) {
1881 struct per_cpu_pageset
*pageset
;
1883 pageset
= zone_pcp(zone
, cpu
);
1885 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1886 cpu
, pageset
->pcp
.high
,
1887 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1891 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1892 " inactive_file:%lu"
1893 //TODO: check/adjust line lengths
1894 #ifdef CONFIG_UNEVICTABLE_LRU
1897 " dirty:%lu writeback:%lu unstable:%lu\n"
1898 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1899 global_page_state(NR_ACTIVE_ANON
),
1900 global_page_state(NR_ACTIVE_FILE
),
1901 global_page_state(NR_INACTIVE_ANON
),
1902 global_page_state(NR_INACTIVE_FILE
),
1903 #ifdef CONFIG_UNEVICTABLE_LRU
1904 global_page_state(NR_UNEVICTABLE
),
1906 global_page_state(NR_FILE_DIRTY
),
1907 global_page_state(NR_WRITEBACK
),
1908 global_page_state(NR_UNSTABLE_NFS
),
1909 global_page_state(NR_FREE_PAGES
),
1910 global_page_state(NR_SLAB_RECLAIMABLE
) +
1911 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1912 global_page_state(NR_FILE_MAPPED
),
1913 global_page_state(NR_PAGETABLE
),
1914 global_page_state(NR_BOUNCE
));
1916 for_each_populated_zone(zone
) {
1925 " active_anon:%lukB"
1926 " inactive_anon:%lukB"
1927 " active_file:%lukB"
1928 " inactive_file:%lukB"
1929 #ifdef CONFIG_UNEVICTABLE_LRU
1930 " unevictable:%lukB"
1933 " pages_scanned:%lu"
1934 " all_unreclaimable? %s"
1937 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1940 K(zone
->pages_high
),
1941 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
1942 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
1943 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
1944 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
1945 #ifdef CONFIG_UNEVICTABLE_LRU
1946 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
1948 K(zone
->present_pages
),
1949 zone
->pages_scanned
,
1950 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1952 printk("lowmem_reserve[]:");
1953 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1954 printk(" %lu", zone
->lowmem_reserve
[i
]);
1958 for_each_populated_zone(zone
) {
1959 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1962 printk("%s: ", zone
->name
);
1964 spin_lock_irqsave(&zone
->lock
, flags
);
1965 for (order
= 0; order
< MAX_ORDER
; order
++) {
1966 nr
[order
] = zone
->free_area
[order
].nr_free
;
1967 total
+= nr
[order
] << order
;
1969 spin_unlock_irqrestore(&zone
->lock
, flags
);
1970 for (order
= 0; order
< MAX_ORDER
; order
++)
1971 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1972 printk("= %lukB\n", K(total
));
1975 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
1977 show_swap_cache_info();
1980 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
1982 zoneref
->zone
= zone
;
1983 zoneref
->zone_idx
= zone_idx(zone
);
1987 * Builds allocation fallback zone lists.
1989 * Add all populated zones of a node to the zonelist.
1991 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1992 int nr_zones
, enum zone_type zone_type
)
1996 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2001 zone
= pgdat
->node_zones
+ zone_type
;
2002 if (populated_zone(zone
)) {
2003 zoneref_set_zone(zone
,
2004 &zonelist
->_zonerefs
[nr_zones
++]);
2005 check_highest_zone(zone_type
);
2008 } while (zone_type
);
2015 * 0 = automatic detection of better ordering.
2016 * 1 = order by ([node] distance, -zonetype)
2017 * 2 = order by (-zonetype, [node] distance)
2019 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2020 * the same zonelist. So only NUMA can configure this param.
2022 #define ZONELIST_ORDER_DEFAULT 0
2023 #define ZONELIST_ORDER_NODE 1
2024 #define ZONELIST_ORDER_ZONE 2
2026 /* zonelist order in the kernel.
2027 * set_zonelist_order() will set this to NODE or ZONE.
2029 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2030 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2034 /* The value user specified ....changed by config */
2035 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2036 /* string for sysctl */
2037 #define NUMA_ZONELIST_ORDER_LEN 16
2038 char numa_zonelist_order
[16] = "default";
2041 * interface for configure zonelist ordering.
2042 * command line option "numa_zonelist_order"
2043 * = "[dD]efault - default, automatic configuration.
2044 * = "[nN]ode - order by node locality, then by zone within node
2045 * = "[zZ]one - order by zone, then by locality within zone
2048 static int __parse_numa_zonelist_order(char *s
)
2050 if (*s
== 'd' || *s
== 'D') {
2051 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2052 } else if (*s
== 'n' || *s
== 'N') {
2053 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2054 } else if (*s
== 'z' || *s
== 'Z') {
2055 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2058 "Ignoring invalid numa_zonelist_order value: "
2065 static __init
int setup_numa_zonelist_order(char *s
)
2068 return __parse_numa_zonelist_order(s
);
2071 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2074 * sysctl handler for numa_zonelist_order
2076 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2077 struct file
*file
, void __user
*buffer
, size_t *length
,
2080 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2084 strncpy(saved_string
, (char*)table
->data
,
2085 NUMA_ZONELIST_ORDER_LEN
);
2086 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2090 int oldval
= user_zonelist_order
;
2091 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2093 * bogus value. restore saved string
2095 strncpy((char*)table
->data
, saved_string
,
2096 NUMA_ZONELIST_ORDER_LEN
);
2097 user_zonelist_order
= oldval
;
2098 } else if (oldval
!= user_zonelist_order
)
2099 build_all_zonelists();
2105 #define MAX_NODE_LOAD (num_online_nodes())
2106 static int node_load
[MAX_NUMNODES
];
2109 * find_next_best_node - find the next node that should appear in a given node's fallback list
2110 * @node: node whose fallback list we're appending
2111 * @used_node_mask: nodemask_t of already used nodes
2113 * We use a number of factors to determine which is the next node that should
2114 * appear on a given node's fallback list. The node should not have appeared
2115 * already in @node's fallback list, and it should be the next closest node
2116 * according to the distance array (which contains arbitrary distance values
2117 * from each node to each node in the system), and should also prefer nodes
2118 * with no CPUs, since presumably they'll have very little allocation pressure
2119 * on them otherwise.
2120 * It returns -1 if no node is found.
2122 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2125 int min_val
= INT_MAX
;
2127 const struct cpumask
*tmp
= cpumask_of_node(0);
2129 /* Use the local node if we haven't already */
2130 if (!node_isset(node
, *used_node_mask
)) {
2131 node_set(node
, *used_node_mask
);
2135 for_each_node_state(n
, N_HIGH_MEMORY
) {
2137 /* Don't want a node to appear more than once */
2138 if (node_isset(n
, *used_node_mask
))
2141 /* Use the distance array to find the distance */
2142 val
= node_distance(node
, n
);
2144 /* Penalize nodes under us ("prefer the next node") */
2147 /* Give preference to headless and unused nodes */
2148 tmp
= cpumask_of_node(n
);
2149 if (!cpumask_empty(tmp
))
2150 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2152 /* Slight preference for less loaded node */
2153 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2154 val
+= node_load
[n
];
2156 if (val
< min_val
) {
2163 node_set(best_node
, *used_node_mask
);
2170 * Build zonelists ordered by node and zones within node.
2171 * This results in maximum locality--normal zone overflows into local
2172 * DMA zone, if any--but risks exhausting DMA zone.
2174 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2177 struct zonelist
*zonelist
;
2179 zonelist
= &pgdat
->node_zonelists
[0];
2180 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2182 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2184 zonelist
->_zonerefs
[j
].zone
= NULL
;
2185 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2189 * Build gfp_thisnode zonelists
2191 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2194 struct zonelist
*zonelist
;
2196 zonelist
= &pgdat
->node_zonelists
[1];
2197 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2198 zonelist
->_zonerefs
[j
].zone
= NULL
;
2199 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2203 * Build zonelists ordered by zone and nodes within zones.
2204 * This results in conserving DMA zone[s] until all Normal memory is
2205 * exhausted, but results in overflowing to remote node while memory
2206 * may still exist in local DMA zone.
2208 static int node_order
[MAX_NUMNODES
];
2210 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2213 int zone_type
; /* needs to be signed */
2215 struct zonelist
*zonelist
;
2217 zonelist
= &pgdat
->node_zonelists
[0];
2219 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2220 for (j
= 0; j
< nr_nodes
; j
++) {
2221 node
= node_order
[j
];
2222 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2223 if (populated_zone(z
)) {
2225 &zonelist
->_zonerefs
[pos
++]);
2226 check_highest_zone(zone_type
);
2230 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2231 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2234 static int default_zonelist_order(void)
2237 unsigned long low_kmem_size
,total_size
;
2241 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2242 * If they are really small and used heavily, the system can fall
2243 * into OOM very easily.
2244 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2246 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2249 for_each_online_node(nid
) {
2250 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2251 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2252 if (populated_zone(z
)) {
2253 if (zone_type
< ZONE_NORMAL
)
2254 low_kmem_size
+= z
->present_pages
;
2255 total_size
+= z
->present_pages
;
2259 if (!low_kmem_size
|| /* there are no DMA area. */
2260 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2261 return ZONELIST_ORDER_NODE
;
2263 * look into each node's config.
2264 * If there is a node whose DMA/DMA32 memory is very big area on
2265 * local memory, NODE_ORDER may be suitable.
2267 average_size
= total_size
/
2268 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2269 for_each_online_node(nid
) {
2272 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2273 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2274 if (populated_zone(z
)) {
2275 if (zone_type
< ZONE_NORMAL
)
2276 low_kmem_size
+= z
->present_pages
;
2277 total_size
+= z
->present_pages
;
2280 if (low_kmem_size
&&
2281 total_size
> average_size
&& /* ignore small node */
2282 low_kmem_size
> total_size
* 70/100)
2283 return ZONELIST_ORDER_NODE
;
2285 return ZONELIST_ORDER_ZONE
;
2288 static void set_zonelist_order(void)
2290 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2291 current_zonelist_order
= default_zonelist_order();
2293 current_zonelist_order
= user_zonelist_order
;
2296 static void build_zonelists(pg_data_t
*pgdat
)
2300 nodemask_t used_mask
;
2301 int local_node
, prev_node
;
2302 struct zonelist
*zonelist
;
2303 int order
= current_zonelist_order
;
2305 /* initialize zonelists */
2306 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2307 zonelist
= pgdat
->node_zonelists
+ i
;
2308 zonelist
->_zonerefs
[0].zone
= NULL
;
2309 zonelist
->_zonerefs
[0].zone_idx
= 0;
2312 /* NUMA-aware ordering of nodes */
2313 local_node
= pgdat
->node_id
;
2314 load
= num_online_nodes();
2315 prev_node
= local_node
;
2316 nodes_clear(used_mask
);
2318 memset(node_load
, 0, sizeof(node_load
));
2319 memset(node_order
, 0, sizeof(node_order
));
2322 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2323 int distance
= node_distance(local_node
, node
);
2326 * If another node is sufficiently far away then it is better
2327 * to reclaim pages in a zone before going off node.
2329 if (distance
> RECLAIM_DISTANCE
)
2330 zone_reclaim_mode
= 1;
2333 * We don't want to pressure a particular node.
2334 * So adding penalty to the first node in same
2335 * distance group to make it round-robin.
2337 if (distance
!= node_distance(local_node
, prev_node
))
2338 node_load
[node
] = load
;
2342 if (order
== ZONELIST_ORDER_NODE
)
2343 build_zonelists_in_node_order(pgdat
, node
);
2345 node_order
[j
++] = node
; /* remember order */
2348 if (order
== ZONELIST_ORDER_ZONE
) {
2349 /* calculate node order -- i.e., DMA last! */
2350 build_zonelists_in_zone_order(pgdat
, j
);
2353 build_thisnode_zonelists(pgdat
);
2356 /* Construct the zonelist performance cache - see further mmzone.h */
2357 static void build_zonelist_cache(pg_data_t
*pgdat
)
2359 struct zonelist
*zonelist
;
2360 struct zonelist_cache
*zlc
;
2363 zonelist
= &pgdat
->node_zonelists
[0];
2364 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2365 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2366 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2367 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2371 #else /* CONFIG_NUMA */
2373 static void set_zonelist_order(void)
2375 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2378 static void build_zonelists(pg_data_t
*pgdat
)
2380 int node
, local_node
;
2382 struct zonelist
*zonelist
;
2384 local_node
= pgdat
->node_id
;
2386 zonelist
= &pgdat
->node_zonelists
[0];
2387 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2390 * Now we build the zonelist so that it contains the zones
2391 * of all the other nodes.
2392 * We don't want to pressure a particular node, so when
2393 * building the zones for node N, we make sure that the
2394 * zones coming right after the local ones are those from
2395 * node N+1 (modulo N)
2397 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2398 if (!node_online(node
))
2400 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2403 for (node
= 0; node
< local_node
; node
++) {
2404 if (!node_online(node
))
2406 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2410 zonelist
->_zonerefs
[j
].zone
= NULL
;
2411 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2414 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2415 static void build_zonelist_cache(pg_data_t
*pgdat
)
2417 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2420 #endif /* CONFIG_NUMA */
2422 /* return values int ....just for stop_machine() */
2423 static int __build_all_zonelists(void *dummy
)
2427 for_each_online_node(nid
) {
2428 pg_data_t
*pgdat
= NODE_DATA(nid
);
2430 build_zonelists(pgdat
);
2431 build_zonelist_cache(pgdat
);
2436 void build_all_zonelists(void)
2438 set_zonelist_order();
2440 if (system_state
== SYSTEM_BOOTING
) {
2441 __build_all_zonelists(NULL
);
2442 mminit_verify_zonelist();
2443 cpuset_init_current_mems_allowed();
2445 /* we have to stop all cpus to guarantee there is no user
2447 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2448 /* cpuset refresh routine should be here */
2450 vm_total_pages
= nr_free_pagecache_pages();
2452 * Disable grouping by mobility if the number of pages in the
2453 * system is too low to allow the mechanism to work. It would be
2454 * more accurate, but expensive to check per-zone. This check is
2455 * made on memory-hotadd so a system can start with mobility
2456 * disabled and enable it later
2458 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2459 page_group_by_mobility_disabled
= 1;
2461 page_group_by_mobility_disabled
= 0;
2463 printk("Built %i zonelists in %s order, mobility grouping %s. "
2464 "Total pages: %ld\n",
2466 zonelist_order_name
[current_zonelist_order
],
2467 page_group_by_mobility_disabled
? "off" : "on",
2470 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2475 * Helper functions to size the waitqueue hash table.
2476 * Essentially these want to choose hash table sizes sufficiently
2477 * large so that collisions trying to wait on pages are rare.
2478 * But in fact, the number of active page waitqueues on typical
2479 * systems is ridiculously low, less than 200. So this is even
2480 * conservative, even though it seems large.
2482 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2483 * waitqueues, i.e. the size of the waitq table given the number of pages.
2485 #define PAGES_PER_WAITQUEUE 256
2487 #ifndef CONFIG_MEMORY_HOTPLUG
2488 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2490 unsigned long size
= 1;
2492 pages
/= PAGES_PER_WAITQUEUE
;
2494 while (size
< pages
)
2498 * Once we have dozens or even hundreds of threads sleeping
2499 * on IO we've got bigger problems than wait queue collision.
2500 * Limit the size of the wait table to a reasonable size.
2502 size
= min(size
, 4096UL);
2504 return max(size
, 4UL);
2508 * A zone's size might be changed by hot-add, so it is not possible to determine
2509 * a suitable size for its wait_table. So we use the maximum size now.
2511 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2513 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2514 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2515 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2517 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2518 * or more by the traditional way. (See above). It equals:
2520 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2521 * ia64(16K page size) : = ( 8G + 4M)byte.
2522 * powerpc (64K page size) : = (32G +16M)byte.
2524 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2531 * This is an integer logarithm so that shifts can be used later
2532 * to extract the more random high bits from the multiplicative
2533 * hash function before the remainder is taken.
2535 static inline unsigned long wait_table_bits(unsigned long size
)
2540 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2543 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2544 * of blocks reserved is based on zone->pages_min. The memory within the
2545 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2546 * higher will lead to a bigger reserve which will get freed as contiguous
2547 * blocks as reclaim kicks in
2549 static void setup_zone_migrate_reserve(struct zone
*zone
)
2551 unsigned long start_pfn
, pfn
, end_pfn
;
2553 unsigned long reserve
, block_migratetype
;
2555 /* Get the start pfn, end pfn and the number of blocks to reserve */
2556 start_pfn
= zone
->zone_start_pfn
;
2557 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2558 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2561 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2562 if (!pfn_valid(pfn
))
2564 page
= pfn_to_page(pfn
);
2566 /* Watch out for overlapping nodes */
2567 if (page_to_nid(page
) != zone_to_nid(zone
))
2570 /* Blocks with reserved pages will never free, skip them. */
2571 if (PageReserved(page
))
2574 block_migratetype
= get_pageblock_migratetype(page
);
2576 /* If this block is reserved, account for it */
2577 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2582 /* Suitable for reserving if this block is movable */
2583 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2584 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2585 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2591 * If the reserve is met and this is a previous reserved block,
2594 if (block_migratetype
== MIGRATE_RESERVE
) {
2595 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2596 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2602 * Initially all pages are reserved - free ones are freed
2603 * up by free_all_bootmem() once the early boot process is
2604 * done. Non-atomic initialization, single-pass.
2606 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2607 unsigned long start_pfn
, enum memmap_context context
)
2610 unsigned long end_pfn
= start_pfn
+ size
;
2614 if (highest_memmap_pfn
< end_pfn
- 1)
2615 highest_memmap_pfn
= end_pfn
- 1;
2617 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2618 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2620 * There can be holes in boot-time mem_map[]s
2621 * handed to this function. They do not
2622 * exist on hotplugged memory.
2624 if (context
== MEMMAP_EARLY
) {
2625 if (!early_pfn_valid(pfn
))
2627 if (!early_pfn_in_nid(pfn
, nid
))
2630 page
= pfn_to_page(pfn
);
2631 set_page_links(page
, zone
, nid
, pfn
);
2632 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2633 init_page_count(page
);
2634 reset_page_mapcount(page
);
2635 SetPageReserved(page
);
2637 * Mark the block movable so that blocks are reserved for
2638 * movable at startup. This will force kernel allocations
2639 * to reserve their blocks rather than leaking throughout
2640 * the address space during boot when many long-lived
2641 * kernel allocations are made. Later some blocks near
2642 * the start are marked MIGRATE_RESERVE by
2643 * setup_zone_migrate_reserve()
2645 * bitmap is created for zone's valid pfn range. but memmap
2646 * can be created for invalid pages (for alignment)
2647 * check here not to call set_pageblock_migratetype() against
2650 if ((z
->zone_start_pfn
<= pfn
)
2651 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2652 && !(pfn
& (pageblock_nr_pages
- 1)))
2653 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2655 INIT_LIST_HEAD(&page
->lru
);
2656 #ifdef WANT_PAGE_VIRTUAL
2657 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2658 if (!is_highmem_idx(zone
))
2659 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2664 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2667 for_each_migratetype_order(order
, t
) {
2668 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2669 zone
->free_area
[order
].nr_free
= 0;
2673 #ifndef __HAVE_ARCH_MEMMAP_INIT
2674 #define memmap_init(size, nid, zone, start_pfn) \
2675 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2678 static int zone_batchsize(struct zone
*zone
)
2684 * The per-cpu-pages pools are set to around 1000th of the
2685 * size of the zone. But no more than 1/2 of a meg.
2687 * OK, so we don't know how big the cache is. So guess.
2689 batch
= zone
->present_pages
/ 1024;
2690 if (batch
* PAGE_SIZE
> 512 * 1024)
2691 batch
= (512 * 1024) / PAGE_SIZE
;
2692 batch
/= 4; /* We effectively *= 4 below */
2697 * Clamp the batch to a 2^n - 1 value. Having a power
2698 * of 2 value was found to be more likely to have
2699 * suboptimal cache aliasing properties in some cases.
2701 * For example if 2 tasks are alternately allocating
2702 * batches of pages, one task can end up with a lot
2703 * of pages of one half of the possible page colors
2704 * and the other with pages of the other colors.
2706 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
2711 /* The deferral and batching of frees should be suppressed under NOMMU
2714 * The problem is that NOMMU needs to be able to allocate large chunks
2715 * of contiguous memory as there's no hardware page translation to
2716 * assemble apparent contiguous memory from discontiguous pages.
2718 * Queueing large contiguous runs of pages for batching, however,
2719 * causes the pages to actually be freed in smaller chunks. As there
2720 * can be a significant delay between the individual batches being
2721 * recycled, this leads to the once large chunks of space being
2722 * fragmented and becoming unavailable for high-order allocations.
2728 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2730 struct per_cpu_pages
*pcp
;
2732 memset(p
, 0, sizeof(*p
));
2736 pcp
->high
= 6 * batch
;
2737 pcp
->batch
= max(1UL, 1 * batch
);
2738 INIT_LIST_HEAD(&pcp
->list
);
2742 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2743 * to the value high for the pageset p.
2746 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2749 struct per_cpu_pages
*pcp
;
2753 pcp
->batch
= max(1UL, high
/4);
2754 if ((high
/4) > (PAGE_SHIFT
* 8))
2755 pcp
->batch
= PAGE_SHIFT
* 8;
2761 * Boot pageset table. One per cpu which is going to be used for all
2762 * zones and all nodes. The parameters will be set in such a way
2763 * that an item put on a list will immediately be handed over to
2764 * the buddy list. This is safe since pageset manipulation is done
2765 * with interrupts disabled.
2767 * Some NUMA counter updates may also be caught by the boot pagesets.
2769 * The boot_pagesets must be kept even after bootup is complete for
2770 * unused processors and/or zones. They do play a role for bootstrapping
2771 * hotplugged processors.
2773 * zoneinfo_show() and maybe other functions do
2774 * not check if the processor is online before following the pageset pointer.
2775 * Other parts of the kernel may not check if the zone is available.
2777 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2780 * Dynamically allocate memory for the
2781 * per cpu pageset array in struct zone.
2783 static int __cpuinit
process_zones(int cpu
)
2785 struct zone
*zone
, *dzone
;
2786 int node
= cpu_to_node(cpu
);
2788 node_set_state(node
, N_CPU
); /* this node has a cpu */
2790 for_each_populated_zone(zone
) {
2791 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2793 if (!zone_pcp(zone
, cpu
))
2796 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2798 if (percpu_pagelist_fraction
)
2799 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2800 (zone
->present_pages
/ percpu_pagelist_fraction
));
2805 for_each_zone(dzone
) {
2806 if (!populated_zone(dzone
))
2810 kfree(zone_pcp(dzone
, cpu
));
2811 zone_pcp(dzone
, cpu
) = NULL
;
2816 static inline void free_zone_pagesets(int cpu
)
2820 for_each_zone(zone
) {
2821 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2823 /* Free per_cpu_pageset if it is slab allocated */
2824 if (pset
!= &boot_pageset
[cpu
])
2826 zone_pcp(zone
, cpu
) = NULL
;
2830 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2831 unsigned long action
,
2834 int cpu
= (long)hcpu
;
2835 int ret
= NOTIFY_OK
;
2838 case CPU_UP_PREPARE
:
2839 case CPU_UP_PREPARE_FROZEN
:
2840 if (process_zones(cpu
))
2843 case CPU_UP_CANCELED
:
2844 case CPU_UP_CANCELED_FROZEN
:
2846 case CPU_DEAD_FROZEN
:
2847 free_zone_pagesets(cpu
);
2855 static struct notifier_block __cpuinitdata pageset_notifier
=
2856 { &pageset_cpuup_callback
, NULL
, 0 };
2858 void __init
setup_per_cpu_pageset(void)
2862 /* Initialize per_cpu_pageset for cpu 0.
2863 * A cpuup callback will do this for every cpu
2864 * as it comes online
2866 err
= process_zones(smp_processor_id());
2868 register_cpu_notifier(&pageset_notifier
);
2873 static noinline __init_refok
2874 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2877 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2881 * The per-page waitqueue mechanism uses hashed waitqueues
2884 zone
->wait_table_hash_nr_entries
=
2885 wait_table_hash_nr_entries(zone_size_pages
);
2886 zone
->wait_table_bits
=
2887 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2888 alloc_size
= zone
->wait_table_hash_nr_entries
2889 * sizeof(wait_queue_head_t
);
2891 if (!slab_is_available()) {
2892 zone
->wait_table
= (wait_queue_head_t
*)
2893 alloc_bootmem_node(pgdat
, alloc_size
);
2896 * This case means that a zone whose size was 0 gets new memory
2897 * via memory hot-add.
2898 * But it may be the case that a new node was hot-added. In
2899 * this case vmalloc() will not be able to use this new node's
2900 * memory - this wait_table must be initialized to use this new
2901 * node itself as well.
2902 * To use this new node's memory, further consideration will be
2905 zone
->wait_table
= vmalloc(alloc_size
);
2907 if (!zone
->wait_table
)
2910 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2911 init_waitqueue_head(zone
->wait_table
+ i
);
2916 static __meminit
void zone_pcp_init(struct zone
*zone
)
2919 unsigned long batch
= zone_batchsize(zone
);
2921 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2923 /* Early boot. Slab allocator not functional yet */
2924 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2925 setup_pageset(&boot_pageset
[cpu
],0);
2927 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2930 if (zone
->present_pages
)
2931 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2932 zone
->name
, zone
->present_pages
, batch
);
2935 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2936 unsigned long zone_start_pfn
,
2938 enum memmap_context context
)
2940 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2942 ret
= zone_wait_table_init(zone
, size
);
2945 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2947 zone
->zone_start_pfn
= zone_start_pfn
;
2949 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
2950 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2952 (unsigned long)zone_idx(zone
),
2953 zone_start_pfn
, (zone_start_pfn
+ size
));
2955 zone_init_free_lists(zone
);
2960 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2962 * Basic iterator support. Return the first range of PFNs for a node
2963 * Note: nid == MAX_NUMNODES returns first region regardless of node
2965 static int __meminit
first_active_region_index_in_nid(int nid
)
2969 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2970 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2977 * Basic iterator support. Return the next active range of PFNs for a node
2978 * Note: nid == MAX_NUMNODES returns next region regardless of node
2980 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2982 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2983 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2989 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2991 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2992 * Architectures may implement their own version but if add_active_range()
2993 * was used and there are no special requirements, this is a convenient
2996 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3000 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3001 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3002 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3004 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3005 return early_node_map
[i
].nid
;
3007 /* This is a memory hole */
3010 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3012 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3016 nid
= __early_pfn_to_nid(pfn
);
3019 /* just returns 0 */
3023 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3024 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3028 nid
= __early_pfn_to_nid(pfn
);
3029 if (nid
>= 0 && nid
!= node
)
3035 /* Basic iterator support to walk early_node_map[] */
3036 #define for_each_active_range_index_in_nid(i, nid) \
3037 for (i = first_active_region_index_in_nid(nid); i != -1; \
3038 i = next_active_region_index_in_nid(i, nid))
3041 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3042 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3043 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3045 * If an architecture guarantees that all ranges registered with
3046 * add_active_ranges() contain no holes and may be freed, this
3047 * this function may be used instead of calling free_bootmem() manually.
3049 void __init
free_bootmem_with_active_regions(int nid
,
3050 unsigned long max_low_pfn
)
3054 for_each_active_range_index_in_nid(i
, nid
) {
3055 unsigned long size_pages
= 0;
3056 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3058 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3061 if (end_pfn
> max_low_pfn
)
3062 end_pfn
= max_low_pfn
;
3064 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3065 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3066 PFN_PHYS(early_node_map
[i
].start_pfn
),
3067 size_pages
<< PAGE_SHIFT
);
3071 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3076 for_each_active_range_index_in_nid(i
, nid
) {
3077 ret
= work_fn(early_node_map
[i
].start_pfn
,
3078 early_node_map
[i
].end_pfn
, data
);
3084 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3085 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3087 * If an architecture guarantees that all ranges registered with
3088 * add_active_ranges() contain no holes and may be freed, this
3089 * function may be used instead of calling memory_present() manually.
3091 void __init
sparse_memory_present_with_active_regions(int nid
)
3095 for_each_active_range_index_in_nid(i
, nid
)
3096 memory_present(early_node_map
[i
].nid
,
3097 early_node_map
[i
].start_pfn
,
3098 early_node_map
[i
].end_pfn
);
3102 * get_pfn_range_for_nid - Return the start and end page frames for a node
3103 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3104 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3105 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3107 * It returns the start and end page frame of a node based on information
3108 * provided by an arch calling add_active_range(). If called for a node
3109 * with no available memory, a warning is printed and the start and end
3112 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3113 unsigned long *start_pfn
, unsigned long *end_pfn
)
3119 for_each_active_range_index_in_nid(i
, nid
) {
3120 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3121 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3124 if (*start_pfn
== -1UL)
3129 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3130 * assumption is made that zones within a node are ordered in monotonic
3131 * increasing memory addresses so that the "highest" populated zone is used
3133 static void __init
find_usable_zone_for_movable(void)
3136 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3137 if (zone_index
== ZONE_MOVABLE
)
3140 if (arch_zone_highest_possible_pfn
[zone_index
] >
3141 arch_zone_lowest_possible_pfn
[zone_index
])
3145 VM_BUG_ON(zone_index
== -1);
3146 movable_zone
= zone_index
;
3150 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3151 * because it is sized independant of architecture. Unlike the other zones,
3152 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3153 * in each node depending on the size of each node and how evenly kernelcore
3154 * is distributed. This helper function adjusts the zone ranges
3155 * provided by the architecture for a given node by using the end of the
3156 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3157 * zones within a node are in order of monotonic increases memory addresses
3159 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3160 unsigned long zone_type
,
3161 unsigned long node_start_pfn
,
3162 unsigned long node_end_pfn
,
3163 unsigned long *zone_start_pfn
,
3164 unsigned long *zone_end_pfn
)
3166 /* Only adjust if ZONE_MOVABLE is on this node */
3167 if (zone_movable_pfn
[nid
]) {
3168 /* Size ZONE_MOVABLE */
3169 if (zone_type
== ZONE_MOVABLE
) {
3170 *zone_start_pfn
= zone_movable_pfn
[nid
];
3171 *zone_end_pfn
= min(node_end_pfn
,
3172 arch_zone_highest_possible_pfn
[movable_zone
]);
3174 /* Adjust for ZONE_MOVABLE starting within this range */
3175 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3176 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3177 *zone_end_pfn
= zone_movable_pfn
[nid
];
3179 /* Check if this whole range is within ZONE_MOVABLE */
3180 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3181 *zone_start_pfn
= *zone_end_pfn
;
3186 * Return the number of pages a zone spans in a node, including holes
3187 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3189 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3190 unsigned long zone_type
,
3191 unsigned long *ignored
)
3193 unsigned long node_start_pfn
, node_end_pfn
;
3194 unsigned long zone_start_pfn
, zone_end_pfn
;
3196 /* Get the start and end of the node and zone */
3197 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3198 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3199 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3200 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3201 node_start_pfn
, node_end_pfn
,
3202 &zone_start_pfn
, &zone_end_pfn
);
3204 /* Check that this node has pages within the zone's required range */
3205 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3208 /* Move the zone boundaries inside the node if necessary */
3209 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3210 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3212 /* Return the spanned pages */
3213 return zone_end_pfn
- zone_start_pfn
;
3217 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3218 * then all holes in the requested range will be accounted for.
3220 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3221 unsigned long range_start_pfn
,
3222 unsigned long range_end_pfn
)
3225 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3226 unsigned long start_pfn
;
3228 /* Find the end_pfn of the first active range of pfns in the node */
3229 i
= first_active_region_index_in_nid(nid
);
3233 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3235 /* Account for ranges before physical memory on this node */
3236 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3237 hole_pages
= prev_end_pfn
- range_start_pfn
;
3239 /* Find all holes for the zone within the node */
3240 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3242 /* No need to continue if prev_end_pfn is outside the zone */
3243 if (prev_end_pfn
>= range_end_pfn
)
3246 /* Make sure the end of the zone is not within the hole */
3247 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3248 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3250 /* Update the hole size cound and move on */
3251 if (start_pfn
> range_start_pfn
) {
3252 BUG_ON(prev_end_pfn
> start_pfn
);
3253 hole_pages
+= start_pfn
- prev_end_pfn
;
3255 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3258 /* Account for ranges past physical memory on this node */
3259 if (range_end_pfn
> prev_end_pfn
)
3260 hole_pages
+= range_end_pfn
-
3261 max(range_start_pfn
, prev_end_pfn
);
3267 * absent_pages_in_range - Return number of page frames in holes within a range
3268 * @start_pfn: The start PFN to start searching for holes
3269 * @end_pfn: The end PFN to stop searching for holes
3271 * It returns the number of pages frames in memory holes within a range.
3273 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3274 unsigned long end_pfn
)
3276 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3279 /* Return the number of page frames in holes in a zone on a node */
3280 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3281 unsigned long zone_type
,
3282 unsigned long *ignored
)
3284 unsigned long node_start_pfn
, node_end_pfn
;
3285 unsigned long zone_start_pfn
, zone_end_pfn
;
3287 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3288 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3290 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3293 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3294 node_start_pfn
, node_end_pfn
,
3295 &zone_start_pfn
, &zone_end_pfn
);
3296 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3300 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3301 unsigned long zone_type
,
3302 unsigned long *zones_size
)
3304 return zones_size
[zone_type
];
3307 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3308 unsigned long zone_type
,
3309 unsigned long *zholes_size
)
3314 return zholes_size
[zone_type
];
3319 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3320 unsigned long *zones_size
, unsigned long *zholes_size
)
3322 unsigned long realtotalpages
, totalpages
= 0;
3325 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3326 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3328 pgdat
->node_spanned_pages
= totalpages
;
3330 realtotalpages
= totalpages
;
3331 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3333 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3335 pgdat
->node_present_pages
= realtotalpages
;
3336 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3340 #ifndef CONFIG_SPARSEMEM
3342 * Calculate the size of the zone->blockflags rounded to an unsigned long
3343 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3344 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3345 * round what is now in bits to nearest long in bits, then return it in
3348 static unsigned long __init
usemap_size(unsigned long zonesize
)
3350 unsigned long usemapsize
;
3352 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3353 usemapsize
= usemapsize
>> pageblock_order
;
3354 usemapsize
*= NR_PAGEBLOCK_BITS
;
3355 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3357 return usemapsize
/ 8;
3360 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3361 struct zone
*zone
, unsigned long zonesize
)
3363 unsigned long usemapsize
= usemap_size(zonesize
);
3364 zone
->pageblock_flags
= NULL
;
3366 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3369 static void inline setup_usemap(struct pglist_data
*pgdat
,
3370 struct zone
*zone
, unsigned long zonesize
) {}
3371 #endif /* CONFIG_SPARSEMEM */
3373 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3375 /* Return a sensible default order for the pageblock size. */
3376 static inline int pageblock_default_order(void)
3378 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3379 return HUGETLB_PAGE_ORDER
;
3384 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3385 static inline void __init
set_pageblock_order(unsigned int order
)
3387 /* Check that pageblock_nr_pages has not already been setup */
3388 if (pageblock_order
)
3392 * Assume the largest contiguous order of interest is a huge page.
3393 * This value may be variable depending on boot parameters on IA64
3395 pageblock_order
= order
;
3397 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3400 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3401 * and pageblock_default_order() are unused as pageblock_order is set
3402 * at compile-time. See include/linux/pageblock-flags.h for the values of
3403 * pageblock_order based on the kernel config
3405 static inline int pageblock_default_order(unsigned int order
)
3409 #define set_pageblock_order(x) do {} while (0)
3411 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3414 * Set up the zone data structures:
3415 * - mark all pages reserved
3416 * - mark all memory queues empty
3417 * - clear the memory bitmaps
3419 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3420 unsigned long *zones_size
, unsigned long *zholes_size
)
3423 int nid
= pgdat
->node_id
;
3424 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3427 pgdat_resize_init(pgdat
);
3428 pgdat
->nr_zones
= 0;
3429 init_waitqueue_head(&pgdat
->kswapd_wait
);
3430 pgdat
->kswapd_max_order
= 0;
3431 pgdat_page_cgroup_init(pgdat
);
3433 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3434 struct zone
*zone
= pgdat
->node_zones
+ j
;
3435 unsigned long size
, realsize
, memmap_pages
;
3438 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3439 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3443 * Adjust realsize so that it accounts for how much memory
3444 * is used by this zone for memmap. This affects the watermark
3445 * and per-cpu initialisations
3448 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3449 if (realsize
>= memmap_pages
) {
3450 realsize
-= memmap_pages
;
3453 " %s zone: %lu pages used for memmap\n",
3454 zone_names
[j
], memmap_pages
);
3457 " %s zone: %lu pages exceeds realsize %lu\n",
3458 zone_names
[j
], memmap_pages
, realsize
);
3460 /* Account for reserved pages */
3461 if (j
== 0 && realsize
> dma_reserve
) {
3462 realsize
-= dma_reserve
;
3463 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3464 zone_names
[0], dma_reserve
);
3467 if (!is_highmem_idx(j
))
3468 nr_kernel_pages
+= realsize
;
3469 nr_all_pages
+= realsize
;
3471 zone
->spanned_pages
= size
;
3472 zone
->present_pages
= realsize
;
3475 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3477 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3479 zone
->name
= zone_names
[j
];
3480 spin_lock_init(&zone
->lock
);
3481 spin_lock_init(&zone
->lru_lock
);
3482 zone_seqlock_init(zone
);
3483 zone
->zone_pgdat
= pgdat
;
3485 zone
->prev_priority
= DEF_PRIORITY
;
3487 zone_pcp_init(zone
);
3489 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3490 zone
->lru
[l
].nr_scan
= 0;
3492 zone
->reclaim_stat
.recent_rotated
[0] = 0;
3493 zone
->reclaim_stat
.recent_rotated
[1] = 0;
3494 zone
->reclaim_stat
.recent_scanned
[0] = 0;
3495 zone
->reclaim_stat
.recent_scanned
[1] = 0;
3496 zap_zone_vm_stats(zone
);
3501 set_pageblock_order(pageblock_default_order());
3502 setup_usemap(pgdat
, zone
, size
);
3503 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3504 size
, MEMMAP_EARLY
);
3506 memmap_init(size
, nid
, j
, zone_start_pfn
);
3507 zone_start_pfn
+= size
;
3511 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3513 /* Skip empty nodes */
3514 if (!pgdat
->node_spanned_pages
)
3517 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3518 /* ia64 gets its own node_mem_map, before this, without bootmem */
3519 if (!pgdat
->node_mem_map
) {
3520 unsigned long size
, start
, end
;
3524 * The zone's endpoints aren't required to be MAX_ORDER
3525 * aligned but the node_mem_map endpoints must be in order
3526 * for the buddy allocator to function correctly.
3528 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3529 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3530 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3531 size
= (end
- start
) * sizeof(struct page
);
3532 map
= alloc_remap(pgdat
->node_id
, size
);
3534 map
= alloc_bootmem_node(pgdat
, size
);
3535 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3537 #ifndef CONFIG_NEED_MULTIPLE_NODES
3539 * With no DISCONTIG, the global mem_map is just set as node 0's
3541 if (pgdat
== NODE_DATA(0)) {
3542 mem_map
= NODE_DATA(0)->node_mem_map
;
3543 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3544 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3545 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3546 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3549 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3552 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3553 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3555 pg_data_t
*pgdat
= NODE_DATA(nid
);
3557 pgdat
->node_id
= nid
;
3558 pgdat
->node_start_pfn
= node_start_pfn
;
3559 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3561 alloc_node_mem_map(pgdat
);
3562 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3563 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3564 nid
, (unsigned long)pgdat
,
3565 (unsigned long)pgdat
->node_mem_map
);
3568 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3571 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3573 #if MAX_NUMNODES > 1
3575 * Figure out the number of possible node ids.
3577 static void __init
setup_nr_node_ids(void)
3580 unsigned int highest
= 0;
3582 for_each_node_mask(node
, node_possible_map
)
3584 nr_node_ids
= highest
+ 1;
3587 static inline void setup_nr_node_ids(void)
3593 * add_active_range - Register a range of PFNs backed by physical memory
3594 * @nid: The node ID the range resides on
3595 * @start_pfn: The start PFN of the available physical memory
3596 * @end_pfn: The end PFN of the available physical memory
3598 * These ranges are stored in an early_node_map[] and later used by
3599 * free_area_init_nodes() to calculate zone sizes and holes. If the
3600 * range spans a memory hole, it is up to the architecture to ensure
3601 * the memory is not freed by the bootmem allocator. If possible
3602 * the range being registered will be merged with existing ranges.
3604 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3605 unsigned long end_pfn
)
3609 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3610 "Entering add_active_range(%d, %#lx, %#lx) "
3611 "%d entries of %d used\n",
3612 nid
, start_pfn
, end_pfn
,
3613 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3615 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3617 /* Merge with existing active regions if possible */
3618 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3619 if (early_node_map
[i
].nid
!= nid
)
3622 /* Skip if an existing region covers this new one */
3623 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3624 end_pfn
<= early_node_map
[i
].end_pfn
)
3627 /* Merge forward if suitable */
3628 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3629 end_pfn
> early_node_map
[i
].end_pfn
) {
3630 early_node_map
[i
].end_pfn
= end_pfn
;
3634 /* Merge backward if suitable */
3635 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3636 end_pfn
>= early_node_map
[i
].start_pfn
) {
3637 early_node_map
[i
].start_pfn
= start_pfn
;
3642 /* Check that early_node_map is large enough */
3643 if (i
>= MAX_ACTIVE_REGIONS
) {
3644 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3645 MAX_ACTIVE_REGIONS
);
3649 early_node_map
[i
].nid
= nid
;
3650 early_node_map
[i
].start_pfn
= start_pfn
;
3651 early_node_map
[i
].end_pfn
= end_pfn
;
3652 nr_nodemap_entries
= i
+ 1;
3656 * remove_active_range - Shrink an existing registered range of PFNs
3657 * @nid: The node id the range is on that should be shrunk
3658 * @start_pfn: The new PFN of the range
3659 * @end_pfn: The new PFN of the range
3661 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3662 * The map is kept near the end physical page range that has already been
3663 * registered. This function allows an arch to shrink an existing registered
3666 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3667 unsigned long end_pfn
)
3672 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3673 nid
, start_pfn
, end_pfn
);
3675 /* Find the old active region end and shrink */
3676 for_each_active_range_index_in_nid(i
, nid
) {
3677 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3678 early_node_map
[i
].end_pfn
<= end_pfn
) {
3680 early_node_map
[i
].start_pfn
= 0;
3681 early_node_map
[i
].end_pfn
= 0;
3685 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3686 early_node_map
[i
].end_pfn
> start_pfn
) {
3687 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3688 early_node_map
[i
].end_pfn
= start_pfn
;
3689 if (temp_end_pfn
> end_pfn
)
3690 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3693 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3694 early_node_map
[i
].end_pfn
> end_pfn
&&
3695 early_node_map
[i
].start_pfn
< end_pfn
) {
3696 early_node_map
[i
].start_pfn
= end_pfn
;
3704 /* remove the blank ones */
3705 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3706 if (early_node_map
[i
].nid
!= nid
)
3708 if (early_node_map
[i
].end_pfn
)
3710 /* we found it, get rid of it */
3711 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3712 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3713 sizeof(early_node_map
[j
]));
3714 j
= nr_nodemap_entries
- 1;
3715 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3716 nr_nodemap_entries
--;
3721 * remove_all_active_ranges - Remove all currently registered regions
3723 * During discovery, it may be found that a table like SRAT is invalid
3724 * and an alternative discovery method must be used. This function removes
3725 * all currently registered regions.
3727 void __init
remove_all_active_ranges(void)
3729 memset(early_node_map
, 0, sizeof(early_node_map
));
3730 nr_nodemap_entries
= 0;
3733 /* Compare two active node_active_regions */
3734 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3736 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3737 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3739 /* Done this way to avoid overflows */
3740 if (arange
->start_pfn
> brange
->start_pfn
)
3742 if (arange
->start_pfn
< brange
->start_pfn
)
3748 /* sort the node_map by start_pfn */
3749 static void __init
sort_node_map(void)
3751 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3752 sizeof(struct node_active_region
),
3753 cmp_node_active_region
, NULL
);
3756 /* Find the lowest pfn for a node */
3757 static unsigned long __init
find_min_pfn_for_node(int nid
)
3760 unsigned long min_pfn
= ULONG_MAX
;
3762 /* Assuming a sorted map, the first range found has the starting pfn */
3763 for_each_active_range_index_in_nid(i
, nid
)
3764 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3766 if (min_pfn
== ULONG_MAX
) {
3768 "Could not find start_pfn for node %d\n", nid
);
3776 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3778 * It returns the minimum PFN based on information provided via
3779 * add_active_range().
3781 unsigned long __init
find_min_pfn_with_active_regions(void)
3783 return find_min_pfn_for_node(MAX_NUMNODES
);
3787 * early_calculate_totalpages()
3788 * Sum pages in active regions for movable zone.
3789 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3791 static unsigned long __init
early_calculate_totalpages(void)
3794 unsigned long totalpages
= 0;
3796 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3797 unsigned long pages
= early_node_map
[i
].end_pfn
-
3798 early_node_map
[i
].start_pfn
;
3799 totalpages
+= pages
;
3801 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3807 * Find the PFN the Movable zone begins in each node. Kernel memory
3808 * is spread evenly between nodes as long as the nodes have enough
3809 * memory. When they don't, some nodes will have more kernelcore than
3812 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3815 unsigned long usable_startpfn
;
3816 unsigned long kernelcore_node
, kernelcore_remaining
;
3817 unsigned long totalpages
= early_calculate_totalpages();
3818 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3821 * If movablecore was specified, calculate what size of
3822 * kernelcore that corresponds so that memory usable for
3823 * any allocation type is evenly spread. If both kernelcore
3824 * and movablecore are specified, then the value of kernelcore
3825 * will be used for required_kernelcore if it's greater than
3826 * what movablecore would have allowed.
3828 if (required_movablecore
) {
3829 unsigned long corepages
;
3832 * Round-up so that ZONE_MOVABLE is at least as large as what
3833 * was requested by the user
3835 required_movablecore
=
3836 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3837 corepages
= totalpages
- required_movablecore
;
3839 required_kernelcore
= max(required_kernelcore
, corepages
);
3842 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3843 if (!required_kernelcore
)
3846 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3847 find_usable_zone_for_movable();
3848 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3851 /* Spread kernelcore memory as evenly as possible throughout nodes */
3852 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3853 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3855 * Recalculate kernelcore_node if the division per node
3856 * now exceeds what is necessary to satisfy the requested
3857 * amount of memory for the kernel
3859 if (required_kernelcore
< kernelcore_node
)
3860 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3863 * As the map is walked, we track how much memory is usable
3864 * by the kernel using kernelcore_remaining. When it is
3865 * 0, the rest of the node is usable by ZONE_MOVABLE
3867 kernelcore_remaining
= kernelcore_node
;
3869 /* Go through each range of PFNs within this node */
3870 for_each_active_range_index_in_nid(i
, nid
) {
3871 unsigned long start_pfn
, end_pfn
;
3872 unsigned long size_pages
;
3874 start_pfn
= max(early_node_map
[i
].start_pfn
,
3875 zone_movable_pfn
[nid
]);
3876 end_pfn
= early_node_map
[i
].end_pfn
;
3877 if (start_pfn
>= end_pfn
)
3880 /* Account for what is only usable for kernelcore */
3881 if (start_pfn
< usable_startpfn
) {
3882 unsigned long kernel_pages
;
3883 kernel_pages
= min(end_pfn
, usable_startpfn
)
3886 kernelcore_remaining
-= min(kernel_pages
,
3887 kernelcore_remaining
);
3888 required_kernelcore
-= min(kernel_pages
,
3889 required_kernelcore
);
3891 /* Continue if range is now fully accounted */
3892 if (end_pfn
<= usable_startpfn
) {
3895 * Push zone_movable_pfn to the end so
3896 * that if we have to rebalance
3897 * kernelcore across nodes, we will
3898 * not double account here
3900 zone_movable_pfn
[nid
] = end_pfn
;
3903 start_pfn
= usable_startpfn
;
3907 * The usable PFN range for ZONE_MOVABLE is from
3908 * start_pfn->end_pfn. Calculate size_pages as the
3909 * number of pages used as kernelcore
3911 size_pages
= end_pfn
- start_pfn
;
3912 if (size_pages
> kernelcore_remaining
)
3913 size_pages
= kernelcore_remaining
;
3914 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3917 * Some kernelcore has been met, update counts and
3918 * break if the kernelcore for this node has been
3921 required_kernelcore
-= min(required_kernelcore
,
3923 kernelcore_remaining
-= size_pages
;
3924 if (!kernelcore_remaining
)
3930 * If there is still required_kernelcore, we do another pass with one
3931 * less node in the count. This will push zone_movable_pfn[nid] further
3932 * along on the nodes that still have memory until kernelcore is
3936 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3939 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3940 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3941 zone_movable_pfn
[nid
] =
3942 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3945 /* Any regular memory on that node ? */
3946 static void check_for_regular_memory(pg_data_t
*pgdat
)
3948 #ifdef CONFIG_HIGHMEM
3949 enum zone_type zone_type
;
3951 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3952 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3953 if (zone
->present_pages
)
3954 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3960 * free_area_init_nodes - Initialise all pg_data_t and zone data
3961 * @max_zone_pfn: an array of max PFNs for each zone
3963 * This will call free_area_init_node() for each active node in the system.
3964 * Using the page ranges provided by add_active_range(), the size of each
3965 * zone in each node and their holes is calculated. If the maximum PFN
3966 * between two adjacent zones match, it is assumed that the zone is empty.
3967 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3968 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3969 * starts where the previous one ended. For example, ZONE_DMA32 starts
3970 * at arch_max_dma_pfn.
3972 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3977 /* Sort early_node_map as initialisation assumes it is sorted */
3980 /* Record where the zone boundaries are */
3981 memset(arch_zone_lowest_possible_pfn
, 0,
3982 sizeof(arch_zone_lowest_possible_pfn
));
3983 memset(arch_zone_highest_possible_pfn
, 0,
3984 sizeof(arch_zone_highest_possible_pfn
));
3985 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3986 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3987 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3988 if (i
== ZONE_MOVABLE
)
3990 arch_zone_lowest_possible_pfn
[i
] =
3991 arch_zone_highest_possible_pfn
[i
-1];
3992 arch_zone_highest_possible_pfn
[i
] =
3993 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3995 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3996 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3998 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3999 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4000 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4002 /* Print out the zone ranges */
4003 printk("Zone PFN ranges:\n");
4004 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4005 if (i
== ZONE_MOVABLE
)
4007 printk(" %-8s %0#10lx -> %0#10lx\n",
4009 arch_zone_lowest_possible_pfn
[i
],
4010 arch_zone_highest_possible_pfn
[i
]);
4013 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4014 printk("Movable zone start PFN for each node\n");
4015 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4016 if (zone_movable_pfn
[i
])
4017 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4020 /* Print out the early_node_map[] */
4021 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4022 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4023 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4024 early_node_map
[i
].start_pfn
,
4025 early_node_map
[i
].end_pfn
);
4027 /* Initialise every node */
4028 mminit_verify_pageflags_layout();
4029 setup_nr_node_ids();
4030 for_each_online_node(nid
) {
4031 pg_data_t
*pgdat
= NODE_DATA(nid
);
4032 free_area_init_node(nid
, NULL
,
4033 find_min_pfn_for_node(nid
), NULL
);
4035 /* Any memory on that node */
4036 if (pgdat
->node_present_pages
)
4037 node_set_state(nid
, N_HIGH_MEMORY
);
4038 check_for_regular_memory(pgdat
);
4042 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4044 unsigned long long coremem
;
4048 coremem
= memparse(p
, &p
);
4049 *core
= coremem
>> PAGE_SHIFT
;
4051 /* Paranoid check that UL is enough for the coremem value */
4052 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4058 * kernelcore=size sets the amount of memory for use for allocations that
4059 * cannot be reclaimed or migrated.
4061 static int __init
cmdline_parse_kernelcore(char *p
)
4063 return cmdline_parse_core(p
, &required_kernelcore
);
4067 * movablecore=size sets the amount of memory for use for allocations that
4068 * can be reclaimed or migrated.
4070 static int __init
cmdline_parse_movablecore(char *p
)
4072 return cmdline_parse_core(p
, &required_movablecore
);
4075 early_param("kernelcore", cmdline_parse_kernelcore
);
4076 early_param("movablecore", cmdline_parse_movablecore
);
4078 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4081 * set_dma_reserve - set the specified number of pages reserved in the first zone
4082 * @new_dma_reserve: The number of pages to mark reserved
4084 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4085 * In the DMA zone, a significant percentage may be consumed by kernel image
4086 * and other unfreeable allocations which can skew the watermarks badly. This
4087 * function may optionally be used to account for unfreeable pages in the
4088 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4089 * smaller per-cpu batchsize.
4091 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4093 dma_reserve
= new_dma_reserve
;
4096 #ifndef CONFIG_NEED_MULTIPLE_NODES
4097 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4098 EXPORT_SYMBOL(contig_page_data
);
4101 void __init
free_area_init(unsigned long *zones_size
)
4103 free_area_init_node(0, zones_size
,
4104 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4107 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4108 unsigned long action
, void *hcpu
)
4110 int cpu
= (unsigned long)hcpu
;
4112 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4116 * Spill the event counters of the dead processor
4117 * into the current processors event counters.
4118 * This artificially elevates the count of the current
4121 vm_events_fold_cpu(cpu
);
4124 * Zero the differential counters of the dead processor
4125 * so that the vm statistics are consistent.
4127 * This is only okay since the processor is dead and cannot
4128 * race with what we are doing.
4130 refresh_cpu_vm_stats(cpu
);
4135 void __init
page_alloc_init(void)
4137 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4141 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4142 * or min_free_kbytes changes.
4144 static void calculate_totalreserve_pages(void)
4146 struct pglist_data
*pgdat
;
4147 unsigned long reserve_pages
= 0;
4148 enum zone_type i
, j
;
4150 for_each_online_pgdat(pgdat
) {
4151 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4152 struct zone
*zone
= pgdat
->node_zones
+ i
;
4153 unsigned long max
= 0;
4155 /* Find valid and maximum lowmem_reserve in the zone */
4156 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4157 if (zone
->lowmem_reserve
[j
] > max
)
4158 max
= zone
->lowmem_reserve
[j
];
4161 /* we treat pages_high as reserved pages. */
4162 max
+= zone
->pages_high
;
4164 if (max
> zone
->present_pages
)
4165 max
= zone
->present_pages
;
4166 reserve_pages
+= max
;
4169 totalreserve_pages
= reserve_pages
;
4173 * setup_per_zone_lowmem_reserve - called whenever
4174 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4175 * has a correct pages reserved value, so an adequate number of
4176 * pages are left in the zone after a successful __alloc_pages().
4178 static void setup_per_zone_lowmem_reserve(void)
4180 struct pglist_data
*pgdat
;
4181 enum zone_type j
, idx
;
4183 for_each_online_pgdat(pgdat
) {
4184 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4185 struct zone
*zone
= pgdat
->node_zones
+ j
;
4186 unsigned long present_pages
= zone
->present_pages
;
4188 zone
->lowmem_reserve
[j
] = 0;
4192 struct zone
*lower_zone
;
4196 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4197 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4199 lower_zone
= pgdat
->node_zones
+ idx
;
4200 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4201 sysctl_lowmem_reserve_ratio
[idx
];
4202 present_pages
+= lower_zone
->present_pages
;
4207 /* update totalreserve_pages */
4208 calculate_totalreserve_pages();
4212 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4214 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4215 * with respect to min_free_kbytes.
4217 void setup_per_zone_pages_min(void)
4219 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4220 unsigned long lowmem_pages
= 0;
4222 unsigned long flags
;
4224 /* Calculate total number of !ZONE_HIGHMEM pages */
4225 for_each_zone(zone
) {
4226 if (!is_highmem(zone
))
4227 lowmem_pages
+= zone
->present_pages
;
4230 for_each_zone(zone
) {
4233 spin_lock_irqsave(&zone
->lock
, flags
);
4234 tmp
= (u64
)pages_min
* zone
->present_pages
;
4235 do_div(tmp
, lowmem_pages
);
4236 if (is_highmem(zone
)) {
4238 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4239 * need highmem pages, so cap pages_min to a small
4242 * The (pages_high-pages_low) and (pages_low-pages_min)
4243 * deltas controls asynch page reclaim, and so should
4244 * not be capped for highmem.
4248 min_pages
= zone
->present_pages
/ 1024;
4249 if (min_pages
< SWAP_CLUSTER_MAX
)
4250 min_pages
= SWAP_CLUSTER_MAX
;
4251 if (min_pages
> 128)
4253 zone
->pages_min
= min_pages
;
4256 * If it's a lowmem zone, reserve a number of pages
4257 * proportionate to the zone's size.
4259 zone
->pages_min
= tmp
;
4262 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4263 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4264 setup_zone_migrate_reserve(zone
);
4265 spin_unlock_irqrestore(&zone
->lock
, flags
);
4268 /* update totalreserve_pages */
4269 calculate_totalreserve_pages();
4273 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4275 * The inactive anon list should be small enough that the VM never has to
4276 * do too much work, but large enough that each inactive page has a chance
4277 * to be referenced again before it is swapped out.
4279 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4280 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4281 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4282 * the anonymous pages are kept on the inactive list.
4285 * memory ratio inactive anon
4286 * -------------------------------------
4295 static void setup_per_zone_inactive_ratio(void)
4299 for_each_zone(zone
) {
4300 unsigned int gb
, ratio
;
4302 /* Zone size in gigabytes */
4303 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4304 ratio
= int_sqrt(10 * gb
);
4308 zone
->inactive_ratio
= ratio
;
4313 * Initialise min_free_kbytes.
4315 * For small machines we want it small (128k min). For large machines
4316 * we want it large (64MB max). But it is not linear, because network
4317 * bandwidth does not increase linearly with machine size. We use
4319 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4320 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4336 static int __init
init_per_zone_pages_min(void)
4338 unsigned long lowmem_kbytes
;
4340 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4342 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4343 if (min_free_kbytes
< 128)
4344 min_free_kbytes
= 128;
4345 if (min_free_kbytes
> 65536)
4346 min_free_kbytes
= 65536;
4347 setup_per_zone_pages_min();
4348 setup_per_zone_lowmem_reserve();
4349 setup_per_zone_inactive_ratio();
4352 module_init(init_per_zone_pages_min
)
4355 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4356 * that we can call two helper functions whenever min_free_kbytes
4359 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4360 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4362 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4364 setup_per_zone_pages_min();
4369 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4370 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4375 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4380 zone
->min_unmapped_pages
= (zone
->present_pages
*
4381 sysctl_min_unmapped_ratio
) / 100;
4385 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4386 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4391 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4396 zone
->min_slab_pages
= (zone
->present_pages
*
4397 sysctl_min_slab_ratio
) / 100;
4403 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4404 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4405 * whenever sysctl_lowmem_reserve_ratio changes.
4407 * The reserve ratio obviously has absolutely no relation with the
4408 * pages_min watermarks. The lowmem reserve ratio can only make sense
4409 * if in function of the boot time zone sizes.
4411 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4412 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4414 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4415 setup_per_zone_lowmem_reserve();
4420 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4421 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4422 * can have before it gets flushed back to buddy allocator.
4425 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4426 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4432 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4433 if (!write
|| (ret
== -EINVAL
))
4435 for_each_zone(zone
) {
4436 for_each_online_cpu(cpu
) {
4438 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4439 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4445 int hashdist
= HASHDIST_DEFAULT
;
4448 static int __init
set_hashdist(char *str
)
4452 hashdist
= simple_strtoul(str
, &str
, 0);
4455 __setup("hashdist=", set_hashdist
);
4459 * allocate a large system hash table from bootmem
4460 * - it is assumed that the hash table must contain an exact power-of-2
4461 * quantity of entries
4462 * - limit is the number of hash buckets, not the total allocation size
4464 void *__init
alloc_large_system_hash(const char *tablename
,
4465 unsigned long bucketsize
,
4466 unsigned long numentries
,
4469 unsigned int *_hash_shift
,
4470 unsigned int *_hash_mask
,
4471 unsigned long limit
)
4473 unsigned long long max
= limit
;
4474 unsigned long log2qty
, size
;
4477 /* allow the kernel cmdline to have a say */
4479 /* round applicable memory size up to nearest megabyte */
4480 numentries
= nr_kernel_pages
;
4481 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4482 numentries
>>= 20 - PAGE_SHIFT
;
4483 numentries
<<= 20 - PAGE_SHIFT
;
4485 /* limit to 1 bucket per 2^scale bytes of low memory */
4486 if (scale
> PAGE_SHIFT
)
4487 numentries
>>= (scale
- PAGE_SHIFT
);
4489 numentries
<<= (PAGE_SHIFT
- scale
);
4491 /* Make sure we've got at least a 0-order allocation.. */
4492 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4493 numentries
= PAGE_SIZE
/ bucketsize
;
4495 numentries
= roundup_pow_of_two(numentries
);
4497 /* limit allocation size to 1/16 total memory by default */
4499 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4500 do_div(max
, bucketsize
);
4503 if (numentries
> max
)
4506 log2qty
= ilog2(numentries
);
4509 size
= bucketsize
<< log2qty
;
4510 if (flags
& HASH_EARLY
)
4511 table
= alloc_bootmem_nopanic(size
);
4513 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4515 unsigned long order
= get_order(size
);
4516 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4518 * If bucketsize is not a power-of-two, we may free
4519 * some pages at the end of hash table.
4522 unsigned long alloc_end
= (unsigned long)table
+
4523 (PAGE_SIZE
<< order
);
4524 unsigned long used
= (unsigned long)table
+
4526 split_page(virt_to_page(table
), order
);
4527 while (used
< alloc_end
) {
4533 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4536 panic("Failed to allocate %s hash table\n", tablename
);
4538 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4541 ilog2(size
) - PAGE_SHIFT
,
4545 *_hash_shift
= log2qty
;
4547 *_hash_mask
= (1 << log2qty
) - 1;
4552 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4553 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4556 #ifdef CONFIG_SPARSEMEM
4557 return __pfn_to_section(pfn
)->pageblock_flags
;
4559 return zone
->pageblock_flags
;
4560 #endif /* CONFIG_SPARSEMEM */
4563 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4565 #ifdef CONFIG_SPARSEMEM
4566 pfn
&= (PAGES_PER_SECTION
-1);
4567 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4569 pfn
= pfn
- zone
->zone_start_pfn
;
4570 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4571 #endif /* CONFIG_SPARSEMEM */
4575 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4576 * @page: The page within the block of interest
4577 * @start_bitidx: The first bit of interest to retrieve
4578 * @end_bitidx: The last bit of interest
4579 * returns pageblock_bits flags
4581 unsigned long get_pageblock_flags_group(struct page
*page
,
4582 int start_bitidx
, int end_bitidx
)
4585 unsigned long *bitmap
;
4586 unsigned long pfn
, bitidx
;
4587 unsigned long flags
= 0;
4588 unsigned long value
= 1;
4590 zone
= page_zone(page
);
4591 pfn
= page_to_pfn(page
);
4592 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4593 bitidx
= pfn_to_bitidx(zone
, pfn
);
4595 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4596 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4603 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4604 * @page: The page within the block of interest
4605 * @start_bitidx: The first bit of interest
4606 * @end_bitidx: The last bit of interest
4607 * @flags: The flags to set
4609 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4610 int start_bitidx
, int end_bitidx
)
4613 unsigned long *bitmap
;
4614 unsigned long pfn
, bitidx
;
4615 unsigned long value
= 1;
4617 zone
= page_zone(page
);
4618 pfn
= page_to_pfn(page
);
4619 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4620 bitidx
= pfn_to_bitidx(zone
, pfn
);
4621 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4622 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4624 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4626 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4628 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4632 * This is designed as sub function...plz see page_isolation.c also.
4633 * set/clear page block's type to be ISOLATE.
4634 * page allocater never alloc memory from ISOLATE block.
4637 int set_migratetype_isolate(struct page
*page
)
4640 unsigned long flags
;
4643 zone
= page_zone(page
);
4644 spin_lock_irqsave(&zone
->lock
, flags
);
4646 * In future, more migrate types will be able to be isolation target.
4648 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4650 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4651 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4654 spin_unlock_irqrestore(&zone
->lock
, flags
);
4660 void unset_migratetype_isolate(struct page
*page
)
4663 unsigned long flags
;
4664 zone
= page_zone(page
);
4665 spin_lock_irqsave(&zone
->lock
, flags
);
4666 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4668 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4669 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4671 spin_unlock_irqrestore(&zone
->lock
, flags
);
4674 #ifdef CONFIG_MEMORY_HOTREMOVE
4676 * All pages in the range must be isolated before calling this.
4679 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4685 unsigned long flags
;
4686 /* find the first valid pfn */
4687 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4692 zone
= page_zone(pfn_to_page(pfn
));
4693 spin_lock_irqsave(&zone
->lock
, flags
);
4695 while (pfn
< end_pfn
) {
4696 if (!pfn_valid(pfn
)) {
4700 page
= pfn_to_page(pfn
);
4701 BUG_ON(page_count(page
));
4702 BUG_ON(!PageBuddy(page
));
4703 order
= page_order(page
);
4704 #ifdef CONFIG_DEBUG_VM
4705 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4706 pfn
, 1 << order
, end_pfn
);
4708 list_del(&page
->lru
);
4709 rmv_page_order(page
);
4710 zone
->free_area
[order
].nr_free
--;
4711 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4713 for (i
= 0; i
< (1 << order
); i
++)
4714 SetPageReserved((page
+i
));
4715 pfn
+= (1 << order
);
4717 spin_unlock_irqrestore(&zone
->lock
, flags
);