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>
49 #include <linux/kmemleak.h>
51 #include <asm/tlbflush.h>
52 #include <asm/div64.h>
56 * Array of node states.
58 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
59 [N_POSSIBLE
] = NODE_MASK_ALL
,
60 [N_ONLINE
] = { { [0] = 1UL } },
62 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
64 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
66 [N_CPU
] = { { [0] = 1UL } },
69 EXPORT_SYMBOL(node_states
);
71 unsigned long totalram_pages __read_mostly
;
72 unsigned long totalreserve_pages __read_mostly
;
73 unsigned long highest_memmap_pfn __read_mostly
;
74 int percpu_pagelist_fraction
;
76 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
77 int pageblock_order __read_mostly
;
80 static void __free_pages_ok(struct page
*page
, unsigned int order
);
83 * results with 256, 32 in the lowmem_reserve sysctl:
84 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
85 * 1G machine -> (16M dma, 784M normal, 224M high)
86 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
87 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
88 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
90 * TBD: should special case ZONE_DMA32 machines here - in those we normally
91 * don't need any ZONE_NORMAL reservation
93 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
94 #ifdef CONFIG_ZONE_DMA
97 #ifdef CONFIG_ZONE_DMA32
100 #ifdef CONFIG_HIGHMEM
106 EXPORT_SYMBOL(totalram_pages
);
108 static char * const zone_names
[MAX_NR_ZONES
] = {
109 #ifdef CONFIG_ZONE_DMA
112 #ifdef CONFIG_ZONE_DMA32
116 #ifdef CONFIG_HIGHMEM
122 int min_free_kbytes
= 1024;
124 unsigned long __meminitdata nr_kernel_pages
;
125 unsigned long __meminitdata nr_all_pages
;
126 static unsigned long __meminitdata dma_reserve
;
128 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
130 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
131 * ranges of memory (RAM) that may be registered with add_active_range().
132 * Ranges passed to add_active_range() will be merged if possible
133 * so the number of times add_active_range() can be called is
134 * related to the number of nodes and the number of holes
136 #ifdef CONFIG_MAX_ACTIVE_REGIONS
137 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
138 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
140 #if MAX_NUMNODES >= 32
141 /* If there can be many nodes, allow up to 50 holes per node */
142 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
144 /* By default, allow up to 256 distinct regions */
145 #define MAX_ACTIVE_REGIONS 256
149 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
150 static int __meminitdata nr_nodemap_entries
;
151 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
152 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
153 static unsigned long __initdata required_kernelcore
;
154 static unsigned long __initdata required_movablecore
;
155 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
157 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
159 EXPORT_SYMBOL(movable_zone
);
160 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
164 EXPORT_SYMBOL(nr_node_ids
);
167 int page_group_by_mobility_disabled __read_mostly
;
169 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
172 if (unlikely(page_group_by_mobility_disabled
))
173 migratetype
= MIGRATE_UNMOVABLE
;
175 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
176 PB_migrate
, PB_migrate_end
);
179 #ifdef CONFIG_DEBUG_VM
180 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
184 unsigned long pfn
= page_to_pfn(page
);
187 seq
= zone_span_seqbegin(zone
);
188 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
190 else if (pfn
< zone
->zone_start_pfn
)
192 } while (zone_span_seqretry(zone
, seq
));
197 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
199 if (!pfn_valid_within(page_to_pfn(page
)))
201 if (zone
!= page_zone(page
))
207 * Temporary debugging check for pages not lying within a given zone.
209 static int bad_range(struct zone
*zone
, struct page
*page
)
211 if (page_outside_zone_boundaries(zone
, page
))
213 if (!page_is_consistent(zone
, page
))
219 static inline int bad_range(struct zone
*zone
, struct page
*page
)
225 static void bad_page(struct page
*page
)
227 static unsigned long resume
;
228 static unsigned long nr_shown
;
229 static unsigned long nr_unshown
;
232 * Allow a burst of 60 reports, then keep quiet for that minute;
233 * or allow a steady drip of one report per second.
235 if (nr_shown
== 60) {
236 if (time_before(jiffies
, resume
)) {
242 "BUG: Bad page state: %lu messages suppressed\n",
249 resume
= jiffies
+ 60 * HZ
;
251 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
252 current
->comm
, page_to_pfn(page
));
254 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
255 page
, (void *)page
->flags
, page_count(page
),
256 page_mapcount(page
), page
->mapping
, page
->index
);
260 /* Leave bad fields for debug, except PageBuddy could make trouble */
261 __ClearPageBuddy(page
);
262 add_taint(TAINT_BAD_PAGE
);
266 * Higher-order pages are called "compound pages". They are structured thusly:
268 * The first PAGE_SIZE page is called the "head page".
270 * The remaining PAGE_SIZE pages are called "tail pages".
272 * All pages have PG_compound set. All pages have their ->private pointing at
273 * the head page (even the head page has this).
275 * The first tail page's ->lru.next holds the address of the compound page's
276 * put_page() function. Its ->lru.prev holds the order of allocation.
277 * This usage means that zero-order pages may not be compound.
280 static void free_compound_page(struct page
*page
)
282 __free_pages_ok(page
, compound_order(page
));
285 void prep_compound_page(struct page
*page
, unsigned long order
)
288 int nr_pages
= 1 << order
;
290 set_compound_page_dtor(page
, free_compound_page
);
291 set_compound_order(page
, order
);
293 for (i
= 1; i
< nr_pages
; i
++) {
294 struct page
*p
= page
+ i
;
297 p
->first_page
= page
;
301 #ifdef CONFIG_HUGETLBFS
302 void prep_compound_gigantic_page(struct page
*page
, unsigned long order
)
305 int nr_pages
= 1 << order
;
306 struct page
*p
= page
+ 1;
308 set_compound_page_dtor(page
, free_compound_page
);
309 set_compound_order(page
, order
);
311 for (i
= 1; i
< nr_pages
; i
++, p
= mem_map_next(p
, page
, i
)) {
313 p
->first_page
= page
;
318 static int destroy_compound_page(struct page
*page
, unsigned long order
)
321 int nr_pages
= 1 << order
;
324 if (unlikely(compound_order(page
) != order
) ||
325 unlikely(!PageHead(page
))) {
330 __ClearPageHead(page
);
332 for (i
= 1; i
< nr_pages
; i
++) {
333 struct page
*p
= page
+ i
;
335 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
345 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
350 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
351 * and __GFP_HIGHMEM from hard or soft interrupt context.
353 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
354 for (i
= 0; i
< (1 << order
); i
++)
355 clear_highpage(page
+ i
);
358 static inline void set_page_order(struct page
*page
, int order
)
360 set_page_private(page
, order
);
361 __SetPageBuddy(page
);
364 static inline void rmv_page_order(struct page
*page
)
366 __ClearPageBuddy(page
);
367 set_page_private(page
, 0);
371 * Locate the struct page for both the matching buddy in our
372 * pair (buddy1) and the combined O(n+1) page they form (page).
374 * 1) Any buddy B1 will have an order O twin B2 which satisfies
375 * the following equation:
377 * For example, if the starting buddy (buddy2) is #8 its order
379 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
381 * 2) Any buddy B will have an order O+1 parent P which
382 * satisfies the following equation:
385 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
387 static inline struct page
*
388 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
390 unsigned long buddy_idx
= page_idx
^ (1 << order
);
392 return page
+ (buddy_idx
- page_idx
);
395 static inline unsigned long
396 __find_combined_index(unsigned long page_idx
, unsigned int order
)
398 return (page_idx
& ~(1 << order
));
402 * This function checks whether a page is free && is the buddy
403 * we can do coalesce a page and its buddy if
404 * (a) the buddy is not in a hole &&
405 * (b) the buddy is in the buddy system &&
406 * (c) a page and its buddy have the same order &&
407 * (d) a page and its buddy are in the same zone.
409 * For recording whether a page is in the buddy system, we use PG_buddy.
410 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
412 * For recording page's order, we use page_private(page).
414 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
417 if (!pfn_valid_within(page_to_pfn(buddy
)))
420 if (page_zone_id(page
) != page_zone_id(buddy
))
423 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
424 BUG_ON(page_count(buddy
) != 0);
431 * Freeing function for a buddy system allocator.
433 * The concept of a buddy system is to maintain direct-mapped table
434 * (containing bit values) for memory blocks of various "orders".
435 * The bottom level table contains the map for the smallest allocatable
436 * units of memory (here, pages), and each level above it describes
437 * pairs of units from the levels below, hence, "buddies".
438 * At a high level, all that happens here is marking the table entry
439 * at the bottom level available, and propagating the changes upward
440 * as necessary, plus some accounting needed to play nicely with other
441 * parts of the VM system.
442 * At each level, we keep a list of pages, which are heads of continuous
443 * free pages of length of (1 << order) and marked with PG_buddy. Page's
444 * order is recorded in page_private(page) field.
445 * So when we are allocating or freeing one, we can derive the state of the
446 * other. That is, if we allocate a small block, and both were
447 * free, the remainder of the region must be split into blocks.
448 * If a block is freed, and its buddy is also free, then this
449 * triggers coalescing into a block of larger size.
454 static inline void __free_one_page(struct page
*page
,
455 struct zone
*zone
, unsigned int order
)
457 unsigned long page_idx
;
458 int order_size
= 1 << order
;
459 int migratetype
= get_pageblock_migratetype(page
);
461 if (unlikely(PageCompound(page
)))
462 if (unlikely(destroy_compound_page(page
, order
)))
465 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
467 VM_BUG_ON(page_idx
& (order_size
- 1));
468 VM_BUG_ON(bad_range(zone
, page
));
470 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
471 while (order
< MAX_ORDER
-1) {
472 unsigned long combined_idx
;
475 buddy
= __page_find_buddy(page
, page_idx
, order
);
476 if (!page_is_buddy(page
, buddy
, order
))
479 /* Our buddy is free, merge with it and move up one order. */
480 list_del(&buddy
->lru
);
481 zone
->free_area
[order
].nr_free
--;
482 rmv_page_order(buddy
);
483 combined_idx
= __find_combined_index(page_idx
, order
);
484 page
= page
+ (combined_idx
- page_idx
);
485 page_idx
= combined_idx
;
488 set_page_order(page
, order
);
490 &zone
->free_area
[order
].free_list
[migratetype
]);
491 zone
->free_area
[order
].nr_free
++;
494 static inline int free_pages_check(struct page
*page
)
496 free_page_mlock(page
);
497 if (unlikely(page_mapcount(page
) |
498 (page
->mapping
!= NULL
) |
499 (page_count(page
) != 0) |
500 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
504 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
505 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
510 * Frees a list of pages.
511 * Assumes all pages on list are in same zone, and of same order.
512 * count is the number of pages to free.
514 * If the zone was previously in an "all pages pinned" state then look to
515 * see if this freeing clears that state.
517 * And clear the zone's pages_scanned counter, to hold off the "all pages are
518 * pinned" detection logic.
520 static void free_pages_bulk(struct zone
*zone
, int count
,
521 struct list_head
*list
, int order
)
523 spin_lock(&zone
->lock
);
524 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
525 zone
->pages_scanned
= 0;
529 VM_BUG_ON(list_empty(list
));
530 page
= list_entry(list
->prev
, struct page
, lru
);
531 /* have to delete it as __free_one_page list manipulates */
532 list_del(&page
->lru
);
533 __free_one_page(page
, zone
, order
);
535 spin_unlock(&zone
->lock
);
538 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
540 spin_lock(&zone
->lock
);
541 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
542 zone
->pages_scanned
= 0;
543 __free_one_page(page
, zone
, order
);
544 spin_unlock(&zone
->lock
);
547 static void __free_pages_ok(struct page
*page
, unsigned int order
)
553 for (i
= 0 ; i
< (1 << order
) ; ++i
)
554 bad
+= free_pages_check(page
+ i
);
558 if (!PageHighMem(page
)) {
559 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
560 debug_check_no_obj_freed(page_address(page
),
563 arch_free_page(page
, order
);
564 kernel_map_pages(page
, 1 << order
, 0);
566 local_irq_save(flags
);
567 __count_vm_events(PGFREE
, 1 << order
);
568 free_one_page(page_zone(page
), page
, order
);
569 local_irq_restore(flags
);
573 * permit the bootmem allocator to evade page validation on high-order frees
575 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
578 __ClearPageReserved(page
);
579 set_page_count(page
, 0);
580 set_page_refcounted(page
);
586 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
587 struct page
*p
= &page
[loop
];
589 if (loop
+ 1 < BITS_PER_LONG
)
591 __ClearPageReserved(p
);
592 set_page_count(p
, 0);
595 set_page_refcounted(page
);
596 __free_pages(page
, order
);
602 * The order of subdivision here is critical for the IO subsystem.
603 * Please do not alter this order without good reasons and regression
604 * testing. Specifically, as large blocks of memory are subdivided,
605 * the order in which smaller blocks are delivered depends on the order
606 * they're subdivided in this function. This is the primary factor
607 * influencing the order in which pages are delivered to the IO
608 * subsystem according to empirical testing, and this is also justified
609 * by considering the behavior of a buddy system containing a single
610 * large block of memory acted on by a series of small allocations.
611 * This behavior is a critical factor in sglist merging's success.
615 static inline void expand(struct zone
*zone
, struct page
*page
,
616 int low
, int high
, struct free_area
*area
,
619 unsigned long size
= 1 << high
;
625 VM_BUG_ON(bad_range(zone
, &page
[size
]));
626 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
628 set_page_order(&page
[size
], high
);
633 * This page is about to be returned from the page allocator
635 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
637 if (unlikely(page_mapcount(page
) |
638 (page
->mapping
!= NULL
) |
639 (page_count(page
) != 0) |
640 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
645 set_page_private(page
, 0);
646 set_page_refcounted(page
);
648 arch_alloc_page(page
, order
);
649 kernel_map_pages(page
, 1 << order
, 1);
651 if (gfp_flags
& __GFP_ZERO
)
652 prep_zero_page(page
, order
, gfp_flags
);
654 if (order
&& (gfp_flags
& __GFP_COMP
))
655 prep_compound_page(page
, order
);
661 * Go through the free lists for the given migratetype and remove
662 * the smallest available page from the freelists
665 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
668 unsigned int current_order
;
669 struct free_area
* area
;
672 /* Find a page of the appropriate size in the preferred list */
673 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
674 area
= &(zone
->free_area
[current_order
]);
675 if (list_empty(&area
->free_list
[migratetype
]))
678 page
= list_entry(area
->free_list
[migratetype
].next
,
680 list_del(&page
->lru
);
681 rmv_page_order(page
);
683 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
684 expand(zone
, page
, order
, current_order
, area
, migratetype
);
693 * This array describes the order lists are fallen back to when
694 * the free lists for the desirable migrate type are depleted
696 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
697 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
698 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
699 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
700 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
704 * Move the free pages in a range to the free lists of the requested type.
705 * Note that start_page and end_pages are not aligned on a pageblock
706 * boundary. If alignment is required, use move_freepages_block()
708 static int move_freepages(struct zone
*zone
,
709 struct page
*start_page
, struct page
*end_page
,
716 #ifndef CONFIG_HOLES_IN_ZONE
718 * page_zone is not safe to call in this context when
719 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
720 * anyway as we check zone boundaries in move_freepages_block().
721 * Remove at a later date when no bug reports exist related to
722 * grouping pages by mobility
724 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
727 for (page
= start_page
; page
<= end_page
;) {
728 /* Make sure we are not inadvertently changing nodes */
729 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
731 if (!pfn_valid_within(page_to_pfn(page
))) {
736 if (!PageBuddy(page
)) {
741 order
= page_order(page
);
742 list_del(&page
->lru
);
744 &zone
->free_area
[order
].free_list
[migratetype
]);
746 pages_moved
+= 1 << order
;
752 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
755 unsigned long start_pfn
, end_pfn
;
756 struct page
*start_page
, *end_page
;
758 start_pfn
= page_to_pfn(page
);
759 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
760 start_page
= pfn_to_page(start_pfn
);
761 end_page
= start_page
+ pageblock_nr_pages
- 1;
762 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
764 /* Do not cross zone boundaries */
765 if (start_pfn
< zone
->zone_start_pfn
)
767 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
770 return move_freepages(zone
, start_page
, end_page
, migratetype
);
773 /* Remove an element from the buddy allocator from the fallback list */
774 static inline struct page
*
775 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
777 struct free_area
* area
;
782 /* Find the largest possible block of pages in the other list */
783 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
785 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
786 migratetype
= fallbacks
[start_migratetype
][i
];
788 /* MIGRATE_RESERVE handled later if necessary */
789 if (migratetype
== MIGRATE_RESERVE
)
792 area
= &(zone
->free_area
[current_order
]);
793 if (list_empty(&area
->free_list
[migratetype
]))
796 page
= list_entry(area
->free_list
[migratetype
].next
,
801 * If breaking a large block of pages, move all free
802 * pages to the preferred allocation list. If falling
803 * back for a reclaimable kernel allocation, be more
804 * agressive about taking ownership of free pages
806 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
807 start_migratetype
== MIGRATE_RECLAIMABLE
) {
809 pages
= move_freepages_block(zone
, page
,
812 /* Claim the whole block if over half of it is free */
813 if (pages
>= (1 << (pageblock_order
-1)))
814 set_pageblock_migratetype(page
,
817 migratetype
= start_migratetype
;
820 /* Remove the page from the freelists */
821 list_del(&page
->lru
);
822 rmv_page_order(page
);
823 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
826 if (current_order
== pageblock_order
)
827 set_pageblock_migratetype(page
,
830 expand(zone
, page
, order
, current_order
, area
, migratetype
);
839 * Do the hard work of removing an element from the buddy allocator.
840 * Call me with the zone->lock already held.
842 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
848 page
= __rmqueue_smallest(zone
, order
, migratetype
);
850 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
851 page
= __rmqueue_fallback(zone
, order
, migratetype
);
854 * Use MIGRATE_RESERVE rather than fail an allocation. goto
855 * is used because __rmqueue_smallest is an inline function
856 * and we want just one call site
859 migratetype
= MIGRATE_RESERVE
;
868 * Obtain a specified number of elements from the buddy allocator, all under
869 * a single hold of the lock, for efficiency. Add them to the supplied list.
870 * Returns the number of new pages which were placed at *list.
872 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
873 unsigned long count
, struct list_head
*list
,
878 spin_lock(&zone
->lock
);
879 for (i
= 0; i
< count
; ++i
) {
880 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
881 if (unlikely(page
== NULL
))
885 * Split buddy pages returned by expand() are received here
886 * in physical page order. The page is added to the callers and
887 * list and the list head then moves forward. From the callers
888 * perspective, the linked list is ordered by page number in
889 * some conditions. This is useful for IO devices that can
890 * merge IO requests if the physical pages are ordered
893 list_add(&page
->lru
, list
);
894 set_page_private(page
, migratetype
);
897 spin_unlock(&zone
->lock
);
903 * Called from the vmstat counter updater to drain pagesets of this
904 * currently executing processor on remote nodes after they have
907 * Note that this function must be called with the thread pinned to
908 * a single processor.
910 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
915 local_irq_save(flags
);
916 if (pcp
->count
>= pcp
->batch
)
917 to_drain
= pcp
->batch
;
919 to_drain
= pcp
->count
;
920 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
921 pcp
->count
-= to_drain
;
922 local_irq_restore(flags
);
927 * Drain pages of the indicated processor.
929 * The processor must either be the current processor and the
930 * thread pinned to the current processor or a processor that
933 static void drain_pages(unsigned int cpu
)
938 for_each_populated_zone(zone
) {
939 struct per_cpu_pageset
*pset
;
940 struct per_cpu_pages
*pcp
;
942 pset
= zone_pcp(zone
, cpu
);
945 local_irq_save(flags
);
946 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
948 local_irq_restore(flags
);
953 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
955 void drain_local_pages(void *arg
)
957 drain_pages(smp_processor_id());
961 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
963 void drain_all_pages(void)
965 on_each_cpu(drain_local_pages
, NULL
, 1);
968 #ifdef CONFIG_HIBERNATION
970 void mark_free_pages(struct zone
*zone
)
972 unsigned long pfn
, max_zone_pfn
;
975 struct list_head
*curr
;
977 if (!zone
->spanned_pages
)
980 spin_lock_irqsave(&zone
->lock
, flags
);
982 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
983 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
984 if (pfn_valid(pfn
)) {
985 struct page
*page
= pfn_to_page(pfn
);
987 if (!swsusp_page_is_forbidden(page
))
988 swsusp_unset_page_free(page
);
991 for_each_migratetype_order(order
, t
) {
992 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
995 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
996 for (i
= 0; i
< (1UL << order
); i
++)
997 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1000 spin_unlock_irqrestore(&zone
->lock
, flags
);
1002 #endif /* CONFIG_PM */
1005 * Free a 0-order page
1007 static void free_hot_cold_page(struct page
*page
, int cold
)
1009 struct zone
*zone
= page_zone(page
);
1010 struct per_cpu_pages
*pcp
;
1011 unsigned long flags
;
1014 page
->mapping
= NULL
;
1015 if (free_pages_check(page
))
1018 if (!PageHighMem(page
)) {
1019 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1020 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1022 arch_free_page(page
, 0);
1023 kernel_map_pages(page
, 1, 0);
1025 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1026 local_irq_save(flags
);
1027 __count_vm_event(PGFREE
);
1029 list_add_tail(&page
->lru
, &pcp
->list
);
1031 list_add(&page
->lru
, &pcp
->list
);
1032 set_page_private(page
, get_pageblock_migratetype(page
));
1034 if (pcp
->count
>= pcp
->high
) {
1035 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1036 pcp
->count
-= pcp
->batch
;
1038 local_irq_restore(flags
);
1042 void free_hot_page(struct page
*page
)
1044 free_hot_cold_page(page
, 0);
1047 void free_cold_page(struct page
*page
)
1049 free_hot_cold_page(page
, 1);
1053 * split_page takes a non-compound higher-order page, and splits it into
1054 * n (1<<order) sub-pages: page[0..n]
1055 * Each sub-page must be freed individually.
1057 * Note: this is probably too low level an operation for use in drivers.
1058 * Please consult with lkml before using this in your driver.
1060 void split_page(struct page
*page
, unsigned int order
)
1064 VM_BUG_ON(PageCompound(page
));
1065 VM_BUG_ON(!page_count(page
));
1066 for (i
= 1; i
< (1 << order
); i
++)
1067 set_page_refcounted(page
+ i
);
1071 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1072 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1076 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1077 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1080 unsigned long flags
;
1082 int cold
= !!(gfp_flags
& __GFP_COLD
);
1087 if (likely(order
== 0)) {
1088 struct per_cpu_pages
*pcp
;
1090 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1091 local_irq_save(flags
);
1093 pcp
->count
= rmqueue_bulk(zone
, 0,
1094 pcp
->batch
, &pcp
->list
, migratetype
);
1095 if (unlikely(!pcp
->count
))
1099 /* Find a page of the appropriate migrate type */
1101 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1102 if (page_private(page
) == migratetype
)
1105 list_for_each_entry(page
, &pcp
->list
, lru
)
1106 if (page_private(page
) == migratetype
)
1110 /* Allocate more to the pcp list if necessary */
1111 if (unlikely(&page
->lru
== &pcp
->list
)) {
1112 pcp
->count
+= rmqueue_bulk(zone
, 0,
1113 pcp
->batch
, &pcp
->list
, migratetype
);
1114 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1117 list_del(&page
->lru
);
1120 spin_lock_irqsave(&zone
->lock
, flags
);
1121 page
= __rmqueue(zone
, order
, migratetype
);
1122 spin_unlock(&zone
->lock
);
1127 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1128 zone_statistics(preferred_zone
, zone
);
1129 local_irq_restore(flags
);
1132 VM_BUG_ON(bad_range(zone
, page
));
1133 if (prep_new_page(page
, order
, gfp_flags
))
1138 local_irq_restore(flags
);
1143 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1144 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1145 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1146 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1147 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1148 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1149 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1151 #ifdef CONFIG_FAIL_PAGE_ALLOC
1153 static struct fail_page_alloc_attr
{
1154 struct fault_attr attr
;
1156 u32 ignore_gfp_highmem
;
1157 u32 ignore_gfp_wait
;
1160 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1162 struct dentry
*ignore_gfp_highmem_file
;
1163 struct dentry
*ignore_gfp_wait_file
;
1164 struct dentry
*min_order_file
;
1166 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1168 } fail_page_alloc
= {
1169 .attr
= FAULT_ATTR_INITIALIZER
,
1170 .ignore_gfp_wait
= 1,
1171 .ignore_gfp_highmem
= 1,
1175 static int __init
setup_fail_page_alloc(char *str
)
1177 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1179 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1181 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1183 if (order
< fail_page_alloc
.min_order
)
1185 if (gfp_mask
& __GFP_NOFAIL
)
1187 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1189 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1192 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1195 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1197 static int __init
fail_page_alloc_debugfs(void)
1199 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1203 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1207 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1209 fail_page_alloc
.ignore_gfp_wait_file
=
1210 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1211 &fail_page_alloc
.ignore_gfp_wait
);
1213 fail_page_alloc
.ignore_gfp_highmem_file
=
1214 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1215 &fail_page_alloc
.ignore_gfp_highmem
);
1216 fail_page_alloc
.min_order_file
=
1217 debugfs_create_u32("min-order", mode
, dir
,
1218 &fail_page_alloc
.min_order
);
1220 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1221 !fail_page_alloc
.ignore_gfp_highmem_file
||
1222 !fail_page_alloc
.min_order_file
) {
1224 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1225 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1226 debugfs_remove(fail_page_alloc
.min_order_file
);
1227 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1233 late_initcall(fail_page_alloc_debugfs
);
1235 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1237 #else /* CONFIG_FAIL_PAGE_ALLOC */
1239 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1244 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1247 * Return 1 if free pages are above 'mark'. This takes into account the order
1248 * of the allocation.
1250 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1251 int classzone_idx
, int alloc_flags
)
1253 /* free_pages my go negative - that's OK */
1255 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1258 if (alloc_flags
& ALLOC_HIGH
)
1260 if (alloc_flags
& ALLOC_HARDER
)
1263 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1265 for (o
= 0; o
< order
; o
++) {
1266 /* At the next order, this order's pages become unavailable */
1267 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1269 /* Require fewer higher order pages to be free */
1272 if (free_pages
<= min
)
1280 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1281 * skip over zones that are not allowed by the cpuset, or that have
1282 * been recently (in last second) found to be nearly full. See further
1283 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1284 * that have to skip over a lot of full or unallowed zones.
1286 * If the zonelist cache is present in the passed in zonelist, then
1287 * returns a pointer to the allowed node mask (either the current
1288 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1290 * If the zonelist cache is not available for this zonelist, does
1291 * nothing and returns NULL.
1293 * If the fullzones BITMAP in the zonelist cache is stale (more than
1294 * a second since last zap'd) then we zap it out (clear its bits.)
1296 * We hold off even calling zlc_setup, until after we've checked the
1297 * first zone in the zonelist, on the theory that most allocations will
1298 * be satisfied from that first zone, so best to examine that zone as
1299 * quickly as we can.
1301 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1303 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1304 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1306 zlc
= zonelist
->zlcache_ptr
;
1310 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1311 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1312 zlc
->last_full_zap
= jiffies
;
1315 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1316 &cpuset_current_mems_allowed
:
1317 &node_states
[N_HIGH_MEMORY
];
1318 return allowednodes
;
1322 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1323 * if it is worth looking at further for free memory:
1324 * 1) Check that the zone isn't thought to be full (doesn't have its
1325 * bit set in the zonelist_cache fullzones BITMAP).
1326 * 2) Check that the zones node (obtained from the zonelist_cache
1327 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1328 * Return true (non-zero) if zone is worth looking at further, or
1329 * else return false (zero) if it is not.
1331 * This check -ignores- the distinction between various watermarks,
1332 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1333 * found to be full for any variation of these watermarks, it will
1334 * be considered full for up to one second by all requests, unless
1335 * we are so low on memory on all allowed nodes that we are forced
1336 * into the second scan of the zonelist.
1338 * In the second scan we ignore this zonelist cache and exactly
1339 * apply the watermarks to all zones, even it is slower to do so.
1340 * We are low on memory in the second scan, and should leave no stone
1341 * unturned looking for a free page.
1343 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1344 nodemask_t
*allowednodes
)
1346 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1347 int i
; /* index of *z in zonelist zones */
1348 int n
; /* node that zone *z is on */
1350 zlc
= zonelist
->zlcache_ptr
;
1354 i
= z
- zonelist
->_zonerefs
;
1357 /* This zone is worth trying if it is allowed but not full */
1358 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1362 * Given 'z' scanning a zonelist, set the corresponding bit in
1363 * zlc->fullzones, so that subsequent attempts to allocate a page
1364 * from that zone don't waste time re-examining it.
1366 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1368 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1369 int i
; /* index of *z in zonelist zones */
1371 zlc
= zonelist
->zlcache_ptr
;
1375 i
= z
- zonelist
->_zonerefs
;
1377 set_bit(i
, zlc
->fullzones
);
1380 #else /* CONFIG_NUMA */
1382 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1387 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1388 nodemask_t
*allowednodes
)
1393 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1396 #endif /* CONFIG_NUMA */
1399 * get_page_from_freelist goes through the zonelist trying to allocate
1402 static struct page
*
1403 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1404 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1405 struct zone
*preferred_zone
, int migratetype
)
1408 struct page
*page
= NULL
;
1411 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1412 int zlc_active
= 0; /* set if using zonelist_cache */
1413 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1415 if (WARN_ON_ONCE(order
>= MAX_ORDER
))
1418 classzone_idx
= zone_idx(preferred_zone
);
1421 * Scan zonelist, looking for a zone with enough free.
1422 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1424 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1425 high_zoneidx
, nodemask
) {
1426 if (NUMA_BUILD
&& zlc_active
&&
1427 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1429 if ((alloc_flags
& ALLOC_CPUSET
) &&
1430 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1433 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1435 if (alloc_flags
& ALLOC_WMARK_MIN
)
1436 mark
= zone
->pages_min
;
1437 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1438 mark
= zone
->pages_low
;
1440 mark
= zone
->pages_high
;
1441 if (!zone_watermark_ok(zone
, order
, mark
,
1442 classzone_idx
, alloc_flags
)) {
1443 if (!zone_reclaim_mode
||
1444 !zone_reclaim(zone
, gfp_mask
, order
))
1445 goto this_zone_full
;
1449 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1450 gfp_mask
, migratetype
);
1455 zlc_mark_zone_full(zonelist
, z
);
1457 if (NUMA_BUILD
&& !did_zlc_setup
) {
1458 /* we do zlc_setup after the first zone is tried */
1459 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1465 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1466 /* Disable zlc cache for second zonelist scan */
1474 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1475 unsigned long pages_reclaimed
)
1477 /* Do not loop if specifically requested */
1478 if (gfp_mask
& __GFP_NORETRY
)
1482 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1483 * means __GFP_NOFAIL, but that may not be true in other
1486 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1490 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1491 * specified, then we retry until we no longer reclaim any pages
1492 * (above), or we've reclaimed an order of pages at least as
1493 * large as the allocation's order. In both cases, if the
1494 * allocation still fails, we stop retrying.
1496 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
1500 * Don't let big-order allocations loop unless the caller
1501 * explicitly requests that.
1503 if (gfp_mask
& __GFP_NOFAIL
)
1509 static inline struct page
*
1510 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
1511 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1512 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1517 /* Acquire the OOM killer lock for the zones in zonelist */
1518 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1519 schedule_timeout_uninterruptible(1);
1524 * Go through the zonelist yet one more time, keep very high watermark
1525 * here, this is only to catch a parallel oom killing, we must fail if
1526 * we're still under heavy pressure.
1528 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1529 order
, zonelist
, high_zoneidx
,
1530 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
1531 preferred_zone
, migratetype
);
1535 /* The OOM killer will not help higher order allocs */
1536 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1539 /* Exhausted what can be done so it's blamo time */
1540 out_of_memory(zonelist
, gfp_mask
, order
);
1543 clear_zonelist_oom(zonelist
, gfp_mask
);
1547 /* The really slow allocator path where we enter direct reclaim */
1548 static inline struct page
*
1549 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1550 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1551 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1552 int migratetype
, unsigned long *did_some_progress
)
1554 struct page
*page
= NULL
;
1555 struct reclaim_state reclaim_state
;
1556 struct task_struct
*p
= current
;
1560 /* We now go into synchronous reclaim */
1561 cpuset_memory_pressure_bump();
1564 * The task's cpuset might have expanded its set of allowable nodes
1566 p
->flags
|= PF_MEMALLOC
;
1567 lockdep_set_current_reclaim_state(gfp_mask
);
1568 reclaim_state
.reclaimed_slab
= 0;
1569 p
->reclaim_state
= &reclaim_state
;
1571 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1573 p
->reclaim_state
= NULL
;
1574 lockdep_clear_current_reclaim_state();
1575 p
->flags
&= ~PF_MEMALLOC
;
1582 if (likely(*did_some_progress
))
1583 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1584 zonelist
, high_zoneidx
,
1585 alloc_flags
, preferred_zone
,
1591 * This is called in the allocator slow-path if the allocation request is of
1592 * sufficient urgency to ignore watermarks and take other desperate measures
1594 static inline struct page
*
1595 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1596 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1597 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1603 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1604 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
1605 preferred_zone
, migratetype
);
1607 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1608 congestion_wait(WRITE
, HZ
/50);
1609 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
1615 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
1616 enum zone_type high_zoneidx
)
1621 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1622 wakeup_kswapd(zone
, order
);
1626 gfp_to_alloc_flags(gfp_t gfp_mask
)
1628 struct task_struct
*p
= current
;
1629 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
1630 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1632 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1633 BUILD_BUG_ON(__GFP_HIGH
!= ALLOC_HIGH
);
1636 * The caller may dip into page reserves a bit more if the caller
1637 * cannot run direct reclaim, or if the caller has realtime scheduling
1638 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1639 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1641 alloc_flags
|= (gfp_mask
& __GFP_HIGH
);
1644 alloc_flags
|= ALLOC_HARDER
;
1646 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1647 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1649 alloc_flags
&= ~ALLOC_CPUSET
;
1650 } else if (unlikely(rt_task(p
)))
1651 alloc_flags
|= ALLOC_HARDER
;
1653 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
1654 if (!in_interrupt() &&
1655 ((p
->flags
& PF_MEMALLOC
) ||
1656 unlikely(test_thread_flag(TIF_MEMDIE
))))
1657 alloc_flags
|= ALLOC_NO_WATERMARKS
;
1663 static inline struct page
*
1664 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
1665 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1666 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1669 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1670 struct page
*page
= NULL
;
1672 unsigned long pages_reclaimed
= 0;
1673 unsigned long did_some_progress
;
1674 struct task_struct
*p
= current
;
1677 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1678 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1679 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1680 * using a larger set of nodes after it has established that the
1681 * allowed per node queues are empty and that nodes are
1684 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1687 wake_all_kswapd(order
, zonelist
, high_zoneidx
);
1690 * OK, we're below the kswapd watermark and have kicked background
1691 * reclaim. Now things get more complex, so set up alloc_flags according
1692 * to how we want to proceed.
1694 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
1697 /* This is the last chance, in general, before the goto nopage. */
1698 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1699 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
1700 preferred_zone
, migratetype
);
1705 /* Allocate without watermarks if the context allows */
1706 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
1707 page
= __alloc_pages_high_priority(gfp_mask
, order
,
1708 zonelist
, high_zoneidx
, nodemask
,
1709 preferred_zone
, migratetype
);
1714 /* Atomic allocations - we can't balance anything */
1718 /* Avoid recursion of direct reclaim */
1719 if (p
->flags
& PF_MEMALLOC
)
1722 /* Try direct reclaim and then allocating */
1723 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
1724 zonelist
, high_zoneidx
,
1726 alloc_flags
, preferred_zone
,
1727 migratetype
, &did_some_progress
);
1732 * If we failed to make any progress reclaiming, then we are
1733 * running out of options and have to consider going OOM
1735 if (!did_some_progress
) {
1736 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1737 page
= __alloc_pages_may_oom(gfp_mask
, order
,
1738 zonelist
, high_zoneidx
,
1739 nodemask
, preferred_zone
,
1745 * The OOM killer does not trigger for high-order allocations
1746 * but if no progress is being made, there are no other
1747 * options and retrying is unlikely to help
1749 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1756 /* Check if we should retry the allocation */
1757 pages_reclaimed
+= did_some_progress
;
1758 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
1759 /* Wait for some write requests to complete then retry */
1760 congestion_wait(WRITE
, HZ
/50);
1765 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1766 printk(KERN_WARNING
"%s: page allocation failure."
1767 " order:%d, mode:0x%x\n",
1768 p
->comm
, order
, gfp_mask
);
1778 * This is the 'heart' of the zoned buddy allocator.
1781 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
1782 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1784 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1785 struct zone
*preferred_zone
;
1787 int migratetype
= allocflags_to_migratetype(gfp_mask
);
1789 lockdep_trace_alloc(gfp_mask
);
1791 might_sleep_if(gfp_mask
& __GFP_WAIT
);
1793 if (should_fail_alloc_page(gfp_mask
, order
))
1797 * Check the zones suitable for the gfp_mask contain at least one
1798 * valid zone. It's possible to have an empty zonelist as a result
1799 * of GFP_THISNODE and a memoryless node
1801 if (unlikely(!zonelist
->_zonerefs
->zone
))
1804 /* The preferred zone is used for statistics later */
1805 first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
, &preferred_zone
);
1806 if (!preferred_zone
)
1809 /* First allocation attempt */
1810 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1811 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
1812 preferred_zone
, migratetype
);
1813 if (unlikely(!page
))
1814 page
= __alloc_pages_slowpath(gfp_mask
, order
,
1815 zonelist
, high_zoneidx
, nodemask
,
1816 preferred_zone
, migratetype
);
1820 EXPORT_SYMBOL(__alloc_pages_nodemask
);
1823 * Common helper functions.
1825 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1828 page
= alloc_pages(gfp_mask
, order
);
1831 return (unsigned long) page_address(page
);
1834 EXPORT_SYMBOL(__get_free_pages
);
1836 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1841 * get_zeroed_page() returns a 32-bit address, which cannot represent
1844 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1846 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1848 return (unsigned long) page_address(page
);
1852 EXPORT_SYMBOL(get_zeroed_page
);
1854 void __pagevec_free(struct pagevec
*pvec
)
1856 int i
= pagevec_count(pvec
);
1859 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1862 void __free_pages(struct page
*page
, unsigned int order
)
1864 if (put_page_testzero(page
)) {
1866 free_hot_page(page
);
1868 __free_pages_ok(page
, order
);
1872 EXPORT_SYMBOL(__free_pages
);
1874 void free_pages(unsigned long addr
, unsigned int order
)
1877 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1878 __free_pages(virt_to_page((void *)addr
), order
);
1882 EXPORT_SYMBOL(free_pages
);
1885 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1886 * @size: the number of bytes to allocate
1887 * @gfp_mask: GFP flags for the allocation
1889 * This function is similar to alloc_pages(), except that it allocates the
1890 * minimum number of pages to satisfy the request. alloc_pages() can only
1891 * allocate memory in power-of-two pages.
1893 * This function is also limited by MAX_ORDER.
1895 * Memory allocated by this function must be released by free_pages_exact().
1897 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1899 unsigned int order
= get_order(size
);
1902 addr
= __get_free_pages(gfp_mask
, order
);
1904 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1905 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1907 split_page(virt_to_page(addr
), order
);
1908 while (used
< alloc_end
) {
1914 return (void *)addr
;
1916 EXPORT_SYMBOL(alloc_pages_exact
);
1919 * free_pages_exact - release memory allocated via alloc_pages_exact()
1920 * @virt: the value returned by alloc_pages_exact.
1921 * @size: size of allocation, same value as passed to alloc_pages_exact().
1923 * Release the memory allocated by a previous call to alloc_pages_exact.
1925 void free_pages_exact(void *virt
, size_t size
)
1927 unsigned long addr
= (unsigned long)virt
;
1928 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1930 while (addr
< end
) {
1935 EXPORT_SYMBOL(free_pages_exact
);
1937 static unsigned int nr_free_zone_pages(int offset
)
1942 /* Just pick one node, since fallback list is circular */
1943 unsigned int sum
= 0;
1945 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1947 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1948 unsigned long size
= zone
->present_pages
;
1949 unsigned long high
= zone
->pages_high
;
1958 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1960 unsigned int nr_free_buffer_pages(void)
1962 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1964 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1967 * Amount of free RAM allocatable within all zones
1969 unsigned int nr_free_pagecache_pages(void)
1971 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1974 static inline void show_node(struct zone
*zone
)
1977 printk("Node %d ", zone_to_nid(zone
));
1980 void si_meminfo(struct sysinfo
*val
)
1982 val
->totalram
= totalram_pages
;
1984 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1985 val
->bufferram
= nr_blockdev_pages();
1986 val
->totalhigh
= totalhigh_pages
;
1987 val
->freehigh
= nr_free_highpages();
1988 val
->mem_unit
= PAGE_SIZE
;
1991 EXPORT_SYMBOL(si_meminfo
);
1994 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1996 pg_data_t
*pgdat
= NODE_DATA(nid
);
1998 val
->totalram
= pgdat
->node_present_pages
;
1999 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2000 #ifdef CONFIG_HIGHMEM
2001 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2002 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2008 val
->mem_unit
= PAGE_SIZE
;
2012 #define K(x) ((x) << (PAGE_SHIFT-10))
2015 * Show free area list (used inside shift_scroll-lock stuff)
2016 * We also calculate the percentage fragmentation. We do this by counting the
2017 * memory on each free list with the exception of the first item on the list.
2019 void show_free_areas(void)
2024 for_each_populated_zone(zone
) {
2026 printk("%s per-cpu:\n", zone
->name
);
2028 for_each_online_cpu(cpu
) {
2029 struct per_cpu_pageset
*pageset
;
2031 pageset
= zone_pcp(zone
, cpu
);
2033 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2034 cpu
, pageset
->pcp
.high
,
2035 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2039 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2040 " inactive_file:%lu"
2041 //TODO: check/adjust line lengths
2042 #ifdef CONFIG_UNEVICTABLE_LRU
2045 " dirty:%lu writeback:%lu unstable:%lu\n"
2046 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2047 global_page_state(NR_ACTIVE_ANON
),
2048 global_page_state(NR_ACTIVE_FILE
),
2049 global_page_state(NR_INACTIVE_ANON
),
2050 global_page_state(NR_INACTIVE_FILE
),
2051 #ifdef CONFIG_UNEVICTABLE_LRU
2052 global_page_state(NR_UNEVICTABLE
),
2054 global_page_state(NR_FILE_DIRTY
),
2055 global_page_state(NR_WRITEBACK
),
2056 global_page_state(NR_UNSTABLE_NFS
),
2057 global_page_state(NR_FREE_PAGES
),
2058 global_page_state(NR_SLAB_RECLAIMABLE
) +
2059 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2060 global_page_state(NR_FILE_MAPPED
),
2061 global_page_state(NR_PAGETABLE
),
2062 global_page_state(NR_BOUNCE
));
2064 for_each_populated_zone(zone
) {
2073 " active_anon:%lukB"
2074 " inactive_anon:%lukB"
2075 " active_file:%lukB"
2076 " inactive_file:%lukB"
2077 #ifdef CONFIG_UNEVICTABLE_LRU
2078 " unevictable:%lukB"
2081 " pages_scanned:%lu"
2082 " all_unreclaimable? %s"
2085 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2088 K(zone
->pages_high
),
2089 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2090 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2091 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2092 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2093 #ifdef CONFIG_UNEVICTABLE_LRU
2094 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2096 K(zone
->present_pages
),
2097 zone
->pages_scanned
,
2098 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
2100 printk("lowmem_reserve[]:");
2101 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2102 printk(" %lu", zone
->lowmem_reserve
[i
]);
2106 for_each_populated_zone(zone
) {
2107 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2110 printk("%s: ", zone
->name
);
2112 spin_lock_irqsave(&zone
->lock
, flags
);
2113 for (order
= 0; order
< MAX_ORDER
; order
++) {
2114 nr
[order
] = zone
->free_area
[order
].nr_free
;
2115 total
+= nr
[order
] << order
;
2117 spin_unlock_irqrestore(&zone
->lock
, flags
);
2118 for (order
= 0; order
< MAX_ORDER
; order
++)
2119 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2120 printk("= %lukB\n", K(total
));
2123 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2125 show_swap_cache_info();
2128 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2130 zoneref
->zone
= zone
;
2131 zoneref
->zone_idx
= zone_idx(zone
);
2135 * Builds allocation fallback zone lists.
2137 * Add all populated zones of a node to the zonelist.
2139 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2140 int nr_zones
, enum zone_type zone_type
)
2144 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2149 zone
= pgdat
->node_zones
+ zone_type
;
2150 if (populated_zone(zone
)) {
2151 zoneref_set_zone(zone
,
2152 &zonelist
->_zonerefs
[nr_zones
++]);
2153 check_highest_zone(zone_type
);
2156 } while (zone_type
);
2163 * 0 = automatic detection of better ordering.
2164 * 1 = order by ([node] distance, -zonetype)
2165 * 2 = order by (-zonetype, [node] distance)
2167 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2168 * the same zonelist. So only NUMA can configure this param.
2170 #define ZONELIST_ORDER_DEFAULT 0
2171 #define ZONELIST_ORDER_NODE 1
2172 #define ZONELIST_ORDER_ZONE 2
2174 /* zonelist order in the kernel.
2175 * set_zonelist_order() will set this to NODE or ZONE.
2177 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2178 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2182 /* The value user specified ....changed by config */
2183 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2184 /* string for sysctl */
2185 #define NUMA_ZONELIST_ORDER_LEN 16
2186 char numa_zonelist_order
[16] = "default";
2189 * interface for configure zonelist ordering.
2190 * command line option "numa_zonelist_order"
2191 * = "[dD]efault - default, automatic configuration.
2192 * = "[nN]ode - order by node locality, then by zone within node
2193 * = "[zZ]one - order by zone, then by locality within zone
2196 static int __parse_numa_zonelist_order(char *s
)
2198 if (*s
== 'd' || *s
== 'D') {
2199 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2200 } else if (*s
== 'n' || *s
== 'N') {
2201 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2202 } else if (*s
== 'z' || *s
== 'Z') {
2203 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2206 "Ignoring invalid numa_zonelist_order value: "
2213 static __init
int setup_numa_zonelist_order(char *s
)
2216 return __parse_numa_zonelist_order(s
);
2219 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2222 * sysctl handler for numa_zonelist_order
2224 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2225 struct file
*file
, void __user
*buffer
, size_t *length
,
2228 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2232 strncpy(saved_string
, (char*)table
->data
,
2233 NUMA_ZONELIST_ORDER_LEN
);
2234 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2238 int oldval
= user_zonelist_order
;
2239 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2241 * bogus value. restore saved string
2243 strncpy((char*)table
->data
, saved_string
,
2244 NUMA_ZONELIST_ORDER_LEN
);
2245 user_zonelist_order
= oldval
;
2246 } else if (oldval
!= user_zonelist_order
)
2247 build_all_zonelists();
2253 #define MAX_NODE_LOAD (num_online_nodes())
2254 static int node_load
[MAX_NUMNODES
];
2257 * find_next_best_node - find the next node that should appear in a given node's fallback list
2258 * @node: node whose fallback list we're appending
2259 * @used_node_mask: nodemask_t of already used nodes
2261 * We use a number of factors to determine which is the next node that should
2262 * appear on a given node's fallback list. The node should not have appeared
2263 * already in @node's fallback list, and it should be the next closest node
2264 * according to the distance array (which contains arbitrary distance values
2265 * from each node to each node in the system), and should also prefer nodes
2266 * with no CPUs, since presumably they'll have very little allocation pressure
2267 * on them otherwise.
2268 * It returns -1 if no node is found.
2270 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2273 int min_val
= INT_MAX
;
2275 const struct cpumask
*tmp
= cpumask_of_node(0);
2277 /* Use the local node if we haven't already */
2278 if (!node_isset(node
, *used_node_mask
)) {
2279 node_set(node
, *used_node_mask
);
2283 for_each_node_state(n
, N_HIGH_MEMORY
) {
2285 /* Don't want a node to appear more than once */
2286 if (node_isset(n
, *used_node_mask
))
2289 /* Use the distance array to find the distance */
2290 val
= node_distance(node
, n
);
2292 /* Penalize nodes under us ("prefer the next node") */
2295 /* Give preference to headless and unused nodes */
2296 tmp
= cpumask_of_node(n
);
2297 if (!cpumask_empty(tmp
))
2298 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2300 /* Slight preference for less loaded node */
2301 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2302 val
+= node_load
[n
];
2304 if (val
< min_val
) {
2311 node_set(best_node
, *used_node_mask
);
2318 * Build zonelists ordered by node and zones within node.
2319 * This results in maximum locality--normal zone overflows into local
2320 * DMA zone, if any--but risks exhausting DMA zone.
2322 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2325 struct zonelist
*zonelist
;
2327 zonelist
= &pgdat
->node_zonelists
[0];
2328 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2330 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2332 zonelist
->_zonerefs
[j
].zone
= NULL
;
2333 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2337 * Build gfp_thisnode zonelists
2339 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2342 struct zonelist
*zonelist
;
2344 zonelist
= &pgdat
->node_zonelists
[1];
2345 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2346 zonelist
->_zonerefs
[j
].zone
= NULL
;
2347 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2351 * Build zonelists ordered by zone and nodes within zones.
2352 * This results in conserving DMA zone[s] until all Normal memory is
2353 * exhausted, but results in overflowing to remote node while memory
2354 * may still exist in local DMA zone.
2356 static int node_order
[MAX_NUMNODES
];
2358 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2361 int zone_type
; /* needs to be signed */
2363 struct zonelist
*zonelist
;
2365 zonelist
= &pgdat
->node_zonelists
[0];
2367 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2368 for (j
= 0; j
< nr_nodes
; j
++) {
2369 node
= node_order
[j
];
2370 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2371 if (populated_zone(z
)) {
2373 &zonelist
->_zonerefs
[pos
++]);
2374 check_highest_zone(zone_type
);
2378 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2379 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2382 static int default_zonelist_order(void)
2385 unsigned long low_kmem_size
,total_size
;
2389 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2390 * If they are really small and used heavily, the system can fall
2391 * into OOM very easily.
2392 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2394 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2397 for_each_online_node(nid
) {
2398 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2399 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2400 if (populated_zone(z
)) {
2401 if (zone_type
< ZONE_NORMAL
)
2402 low_kmem_size
+= z
->present_pages
;
2403 total_size
+= z
->present_pages
;
2407 if (!low_kmem_size
|| /* there are no DMA area. */
2408 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2409 return ZONELIST_ORDER_NODE
;
2411 * look into each node's config.
2412 * If there is a node whose DMA/DMA32 memory is very big area on
2413 * local memory, NODE_ORDER may be suitable.
2415 average_size
= total_size
/
2416 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2417 for_each_online_node(nid
) {
2420 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2421 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2422 if (populated_zone(z
)) {
2423 if (zone_type
< ZONE_NORMAL
)
2424 low_kmem_size
+= z
->present_pages
;
2425 total_size
+= z
->present_pages
;
2428 if (low_kmem_size
&&
2429 total_size
> average_size
&& /* ignore small node */
2430 low_kmem_size
> total_size
* 70/100)
2431 return ZONELIST_ORDER_NODE
;
2433 return ZONELIST_ORDER_ZONE
;
2436 static void set_zonelist_order(void)
2438 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2439 current_zonelist_order
= default_zonelist_order();
2441 current_zonelist_order
= user_zonelist_order
;
2444 static void build_zonelists(pg_data_t
*pgdat
)
2448 nodemask_t used_mask
;
2449 int local_node
, prev_node
;
2450 struct zonelist
*zonelist
;
2451 int order
= current_zonelist_order
;
2453 /* initialize zonelists */
2454 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2455 zonelist
= pgdat
->node_zonelists
+ i
;
2456 zonelist
->_zonerefs
[0].zone
= NULL
;
2457 zonelist
->_zonerefs
[0].zone_idx
= 0;
2460 /* NUMA-aware ordering of nodes */
2461 local_node
= pgdat
->node_id
;
2462 load
= num_online_nodes();
2463 prev_node
= local_node
;
2464 nodes_clear(used_mask
);
2466 memset(node_load
, 0, sizeof(node_load
));
2467 memset(node_order
, 0, sizeof(node_order
));
2470 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2471 int distance
= node_distance(local_node
, node
);
2474 * If another node is sufficiently far away then it is better
2475 * to reclaim pages in a zone before going off node.
2477 if (distance
> RECLAIM_DISTANCE
)
2478 zone_reclaim_mode
= 1;
2481 * We don't want to pressure a particular node.
2482 * So adding penalty to the first node in same
2483 * distance group to make it round-robin.
2485 if (distance
!= node_distance(local_node
, prev_node
))
2486 node_load
[node
] = load
;
2490 if (order
== ZONELIST_ORDER_NODE
)
2491 build_zonelists_in_node_order(pgdat
, node
);
2493 node_order
[j
++] = node
; /* remember order */
2496 if (order
== ZONELIST_ORDER_ZONE
) {
2497 /* calculate node order -- i.e., DMA last! */
2498 build_zonelists_in_zone_order(pgdat
, j
);
2501 build_thisnode_zonelists(pgdat
);
2504 /* Construct the zonelist performance cache - see further mmzone.h */
2505 static void build_zonelist_cache(pg_data_t
*pgdat
)
2507 struct zonelist
*zonelist
;
2508 struct zonelist_cache
*zlc
;
2511 zonelist
= &pgdat
->node_zonelists
[0];
2512 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2513 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2514 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2515 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2519 #else /* CONFIG_NUMA */
2521 static void set_zonelist_order(void)
2523 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2526 static void build_zonelists(pg_data_t
*pgdat
)
2528 int node
, local_node
;
2530 struct zonelist
*zonelist
;
2532 local_node
= pgdat
->node_id
;
2534 zonelist
= &pgdat
->node_zonelists
[0];
2535 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2538 * Now we build the zonelist so that it contains the zones
2539 * of all the other nodes.
2540 * We don't want to pressure a particular node, so when
2541 * building the zones for node N, we make sure that the
2542 * zones coming right after the local ones are those from
2543 * node N+1 (modulo N)
2545 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2546 if (!node_online(node
))
2548 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2551 for (node
= 0; node
< local_node
; node
++) {
2552 if (!node_online(node
))
2554 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2558 zonelist
->_zonerefs
[j
].zone
= NULL
;
2559 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2562 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2563 static void build_zonelist_cache(pg_data_t
*pgdat
)
2565 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2568 #endif /* CONFIG_NUMA */
2570 /* return values int ....just for stop_machine() */
2571 static int __build_all_zonelists(void *dummy
)
2575 for_each_online_node(nid
) {
2576 pg_data_t
*pgdat
= NODE_DATA(nid
);
2578 build_zonelists(pgdat
);
2579 build_zonelist_cache(pgdat
);
2584 void build_all_zonelists(void)
2586 set_zonelist_order();
2588 if (system_state
== SYSTEM_BOOTING
) {
2589 __build_all_zonelists(NULL
);
2590 mminit_verify_zonelist();
2591 cpuset_init_current_mems_allowed();
2593 /* we have to stop all cpus to guarantee there is no user
2595 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2596 /* cpuset refresh routine should be here */
2598 vm_total_pages
= nr_free_pagecache_pages();
2600 * Disable grouping by mobility if the number of pages in the
2601 * system is too low to allow the mechanism to work. It would be
2602 * more accurate, but expensive to check per-zone. This check is
2603 * made on memory-hotadd so a system can start with mobility
2604 * disabled and enable it later
2606 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2607 page_group_by_mobility_disabled
= 1;
2609 page_group_by_mobility_disabled
= 0;
2611 printk("Built %i zonelists in %s order, mobility grouping %s. "
2612 "Total pages: %ld\n",
2614 zonelist_order_name
[current_zonelist_order
],
2615 page_group_by_mobility_disabled
? "off" : "on",
2618 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2623 * Helper functions to size the waitqueue hash table.
2624 * Essentially these want to choose hash table sizes sufficiently
2625 * large so that collisions trying to wait on pages are rare.
2626 * But in fact, the number of active page waitqueues on typical
2627 * systems is ridiculously low, less than 200. So this is even
2628 * conservative, even though it seems large.
2630 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2631 * waitqueues, i.e. the size of the waitq table given the number of pages.
2633 #define PAGES_PER_WAITQUEUE 256
2635 #ifndef CONFIG_MEMORY_HOTPLUG
2636 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2638 unsigned long size
= 1;
2640 pages
/= PAGES_PER_WAITQUEUE
;
2642 while (size
< pages
)
2646 * Once we have dozens or even hundreds of threads sleeping
2647 * on IO we've got bigger problems than wait queue collision.
2648 * Limit the size of the wait table to a reasonable size.
2650 size
= min(size
, 4096UL);
2652 return max(size
, 4UL);
2656 * A zone's size might be changed by hot-add, so it is not possible to determine
2657 * a suitable size for its wait_table. So we use the maximum size now.
2659 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2661 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2662 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2663 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2665 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2666 * or more by the traditional way. (See above). It equals:
2668 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2669 * ia64(16K page size) : = ( 8G + 4M)byte.
2670 * powerpc (64K page size) : = (32G +16M)byte.
2672 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2679 * This is an integer logarithm so that shifts can be used later
2680 * to extract the more random high bits from the multiplicative
2681 * hash function before the remainder is taken.
2683 static inline unsigned long wait_table_bits(unsigned long size
)
2688 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2691 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2692 * of blocks reserved is based on zone->pages_min. The memory within the
2693 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2694 * higher will lead to a bigger reserve which will get freed as contiguous
2695 * blocks as reclaim kicks in
2697 static void setup_zone_migrate_reserve(struct zone
*zone
)
2699 unsigned long start_pfn
, pfn
, end_pfn
;
2701 unsigned long reserve
, block_migratetype
;
2703 /* Get the start pfn, end pfn and the number of blocks to reserve */
2704 start_pfn
= zone
->zone_start_pfn
;
2705 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2706 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2709 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2710 if (!pfn_valid(pfn
))
2712 page
= pfn_to_page(pfn
);
2714 /* Watch out for overlapping nodes */
2715 if (page_to_nid(page
) != zone_to_nid(zone
))
2718 /* Blocks with reserved pages will never free, skip them. */
2719 if (PageReserved(page
))
2722 block_migratetype
= get_pageblock_migratetype(page
);
2724 /* If this block is reserved, account for it */
2725 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2730 /* Suitable for reserving if this block is movable */
2731 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2732 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2733 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2739 * If the reserve is met and this is a previous reserved block,
2742 if (block_migratetype
== MIGRATE_RESERVE
) {
2743 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2744 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2750 * Initially all pages are reserved - free ones are freed
2751 * up by free_all_bootmem() once the early boot process is
2752 * done. Non-atomic initialization, single-pass.
2754 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2755 unsigned long start_pfn
, enum memmap_context context
)
2758 unsigned long end_pfn
= start_pfn
+ size
;
2762 if (highest_memmap_pfn
< end_pfn
- 1)
2763 highest_memmap_pfn
= end_pfn
- 1;
2765 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2766 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2768 * There can be holes in boot-time mem_map[]s
2769 * handed to this function. They do not
2770 * exist on hotplugged memory.
2772 if (context
== MEMMAP_EARLY
) {
2773 if (!early_pfn_valid(pfn
))
2775 if (!early_pfn_in_nid(pfn
, nid
))
2778 page
= pfn_to_page(pfn
);
2779 set_page_links(page
, zone
, nid
, pfn
);
2780 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2781 init_page_count(page
);
2782 reset_page_mapcount(page
);
2783 SetPageReserved(page
);
2785 * Mark the block movable so that blocks are reserved for
2786 * movable at startup. This will force kernel allocations
2787 * to reserve their blocks rather than leaking throughout
2788 * the address space during boot when many long-lived
2789 * kernel allocations are made. Later some blocks near
2790 * the start are marked MIGRATE_RESERVE by
2791 * setup_zone_migrate_reserve()
2793 * bitmap is created for zone's valid pfn range. but memmap
2794 * can be created for invalid pages (for alignment)
2795 * check here not to call set_pageblock_migratetype() against
2798 if ((z
->zone_start_pfn
<= pfn
)
2799 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2800 && !(pfn
& (pageblock_nr_pages
- 1)))
2801 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2803 INIT_LIST_HEAD(&page
->lru
);
2804 #ifdef WANT_PAGE_VIRTUAL
2805 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2806 if (!is_highmem_idx(zone
))
2807 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2812 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2815 for_each_migratetype_order(order
, t
) {
2816 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2817 zone
->free_area
[order
].nr_free
= 0;
2821 #ifndef __HAVE_ARCH_MEMMAP_INIT
2822 #define memmap_init(size, nid, zone, start_pfn) \
2823 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2826 static int zone_batchsize(struct zone
*zone
)
2832 * The per-cpu-pages pools are set to around 1000th of the
2833 * size of the zone. But no more than 1/2 of a meg.
2835 * OK, so we don't know how big the cache is. So guess.
2837 batch
= zone
->present_pages
/ 1024;
2838 if (batch
* PAGE_SIZE
> 512 * 1024)
2839 batch
= (512 * 1024) / PAGE_SIZE
;
2840 batch
/= 4; /* We effectively *= 4 below */
2845 * Clamp the batch to a 2^n - 1 value. Having a power
2846 * of 2 value was found to be more likely to have
2847 * suboptimal cache aliasing properties in some cases.
2849 * For example if 2 tasks are alternately allocating
2850 * batches of pages, one task can end up with a lot
2851 * of pages of one half of the possible page colors
2852 * and the other with pages of the other colors.
2854 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
2859 /* The deferral and batching of frees should be suppressed under NOMMU
2862 * The problem is that NOMMU needs to be able to allocate large chunks
2863 * of contiguous memory as there's no hardware page translation to
2864 * assemble apparent contiguous memory from discontiguous pages.
2866 * Queueing large contiguous runs of pages for batching, however,
2867 * causes the pages to actually be freed in smaller chunks. As there
2868 * can be a significant delay between the individual batches being
2869 * recycled, this leads to the once large chunks of space being
2870 * fragmented and becoming unavailable for high-order allocations.
2876 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2878 struct per_cpu_pages
*pcp
;
2880 memset(p
, 0, sizeof(*p
));
2884 pcp
->high
= 6 * batch
;
2885 pcp
->batch
= max(1UL, 1 * batch
);
2886 INIT_LIST_HEAD(&pcp
->list
);
2890 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2891 * to the value high for the pageset p.
2894 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2897 struct per_cpu_pages
*pcp
;
2901 pcp
->batch
= max(1UL, high
/4);
2902 if ((high
/4) > (PAGE_SHIFT
* 8))
2903 pcp
->batch
= PAGE_SHIFT
* 8;
2909 * Boot pageset table. One per cpu which is going to be used for all
2910 * zones and all nodes. The parameters will be set in such a way
2911 * that an item put on a list will immediately be handed over to
2912 * the buddy list. This is safe since pageset manipulation is done
2913 * with interrupts disabled.
2915 * Some NUMA counter updates may also be caught by the boot pagesets.
2917 * The boot_pagesets must be kept even after bootup is complete for
2918 * unused processors and/or zones. They do play a role for bootstrapping
2919 * hotplugged processors.
2921 * zoneinfo_show() and maybe other functions do
2922 * not check if the processor is online before following the pageset pointer.
2923 * Other parts of the kernel may not check if the zone is available.
2925 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2928 * Dynamically allocate memory for the
2929 * per cpu pageset array in struct zone.
2931 static int __cpuinit
process_zones(int cpu
)
2933 struct zone
*zone
, *dzone
;
2934 int node
= cpu_to_node(cpu
);
2936 node_set_state(node
, N_CPU
); /* this node has a cpu */
2938 for_each_populated_zone(zone
) {
2939 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2941 if (!zone_pcp(zone
, cpu
))
2944 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2946 if (percpu_pagelist_fraction
)
2947 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2948 (zone
->present_pages
/ percpu_pagelist_fraction
));
2953 for_each_zone(dzone
) {
2954 if (!populated_zone(dzone
))
2958 kfree(zone_pcp(dzone
, cpu
));
2959 zone_pcp(dzone
, cpu
) = NULL
;
2964 static inline void free_zone_pagesets(int cpu
)
2968 for_each_zone(zone
) {
2969 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2971 /* Free per_cpu_pageset if it is slab allocated */
2972 if (pset
!= &boot_pageset
[cpu
])
2974 zone_pcp(zone
, cpu
) = NULL
;
2978 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2979 unsigned long action
,
2982 int cpu
= (long)hcpu
;
2983 int ret
= NOTIFY_OK
;
2986 case CPU_UP_PREPARE
:
2987 case CPU_UP_PREPARE_FROZEN
:
2988 if (process_zones(cpu
))
2991 case CPU_UP_CANCELED
:
2992 case CPU_UP_CANCELED_FROZEN
:
2994 case CPU_DEAD_FROZEN
:
2995 free_zone_pagesets(cpu
);
3003 static struct notifier_block __cpuinitdata pageset_notifier
=
3004 { &pageset_cpuup_callback
, NULL
, 0 };
3006 void __init
setup_per_cpu_pageset(void)
3010 /* Initialize per_cpu_pageset for cpu 0.
3011 * A cpuup callback will do this for every cpu
3012 * as it comes online
3014 err
= process_zones(smp_processor_id());
3016 register_cpu_notifier(&pageset_notifier
);
3021 static noinline __init_refok
3022 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3025 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3029 * The per-page waitqueue mechanism uses hashed waitqueues
3032 zone
->wait_table_hash_nr_entries
=
3033 wait_table_hash_nr_entries(zone_size_pages
);
3034 zone
->wait_table_bits
=
3035 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3036 alloc_size
= zone
->wait_table_hash_nr_entries
3037 * sizeof(wait_queue_head_t
);
3039 if (!slab_is_available()) {
3040 zone
->wait_table
= (wait_queue_head_t
*)
3041 alloc_bootmem_node(pgdat
, alloc_size
);
3044 * This case means that a zone whose size was 0 gets new memory
3045 * via memory hot-add.
3046 * But it may be the case that a new node was hot-added. In
3047 * this case vmalloc() will not be able to use this new node's
3048 * memory - this wait_table must be initialized to use this new
3049 * node itself as well.
3050 * To use this new node's memory, further consideration will be
3053 zone
->wait_table
= vmalloc(alloc_size
);
3055 if (!zone
->wait_table
)
3058 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3059 init_waitqueue_head(zone
->wait_table
+ i
);
3064 static __meminit
void zone_pcp_init(struct zone
*zone
)
3067 unsigned long batch
= zone_batchsize(zone
);
3069 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
3071 /* Early boot. Slab allocator not functional yet */
3072 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
3073 setup_pageset(&boot_pageset
[cpu
],0);
3075 setup_pageset(zone_pcp(zone
,cpu
), batch
);
3078 if (zone
->present_pages
)
3079 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
3080 zone
->name
, zone
->present_pages
, batch
);
3083 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3084 unsigned long zone_start_pfn
,
3086 enum memmap_context context
)
3088 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3090 ret
= zone_wait_table_init(zone
, size
);
3093 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3095 zone
->zone_start_pfn
= zone_start_pfn
;
3097 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3098 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3100 (unsigned long)zone_idx(zone
),
3101 zone_start_pfn
, (zone_start_pfn
+ size
));
3103 zone_init_free_lists(zone
);
3108 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3110 * Basic iterator support. Return the first range of PFNs for a node
3111 * Note: nid == MAX_NUMNODES returns first region regardless of node
3113 static int __meminit
first_active_region_index_in_nid(int nid
)
3117 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3118 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
3125 * Basic iterator support. Return the next active range of PFNs for a node
3126 * Note: nid == MAX_NUMNODES returns next region regardless of node
3128 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
3130 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
3131 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
3137 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3139 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3140 * Architectures may implement their own version but if add_active_range()
3141 * was used and there are no special requirements, this is a convenient
3144 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3148 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3149 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3150 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3152 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3153 return early_node_map
[i
].nid
;
3155 /* This is a memory hole */
3158 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3160 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3164 nid
= __early_pfn_to_nid(pfn
);
3167 /* just returns 0 */
3171 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3172 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3176 nid
= __early_pfn_to_nid(pfn
);
3177 if (nid
>= 0 && nid
!= node
)
3183 /* Basic iterator support to walk early_node_map[] */
3184 #define for_each_active_range_index_in_nid(i, nid) \
3185 for (i = first_active_region_index_in_nid(nid); i != -1; \
3186 i = next_active_region_index_in_nid(i, nid))
3189 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3190 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3191 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3193 * If an architecture guarantees that all ranges registered with
3194 * add_active_ranges() contain no holes and may be freed, this
3195 * this function may be used instead of calling free_bootmem() manually.
3197 void __init
free_bootmem_with_active_regions(int nid
,
3198 unsigned long max_low_pfn
)
3202 for_each_active_range_index_in_nid(i
, nid
) {
3203 unsigned long size_pages
= 0;
3204 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3206 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3209 if (end_pfn
> max_low_pfn
)
3210 end_pfn
= max_low_pfn
;
3212 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3213 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3214 PFN_PHYS(early_node_map
[i
].start_pfn
),
3215 size_pages
<< PAGE_SHIFT
);
3219 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3224 for_each_active_range_index_in_nid(i
, nid
) {
3225 ret
= work_fn(early_node_map
[i
].start_pfn
,
3226 early_node_map
[i
].end_pfn
, data
);
3232 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3233 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3235 * If an architecture guarantees that all ranges registered with
3236 * add_active_ranges() contain no holes and may be freed, this
3237 * function may be used instead of calling memory_present() manually.
3239 void __init
sparse_memory_present_with_active_regions(int nid
)
3243 for_each_active_range_index_in_nid(i
, nid
)
3244 memory_present(early_node_map
[i
].nid
,
3245 early_node_map
[i
].start_pfn
,
3246 early_node_map
[i
].end_pfn
);
3250 * get_pfn_range_for_nid - Return the start and end page frames for a node
3251 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3252 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3253 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3255 * It returns the start and end page frame of a node based on information
3256 * provided by an arch calling add_active_range(). If called for a node
3257 * with no available memory, a warning is printed and the start and end
3260 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3261 unsigned long *start_pfn
, unsigned long *end_pfn
)
3267 for_each_active_range_index_in_nid(i
, nid
) {
3268 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3269 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3272 if (*start_pfn
== -1UL)
3277 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3278 * assumption is made that zones within a node are ordered in monotonic
3279 * increasing memory addresses so that the "highest" populated zone is used
3281 static void __init
find_usable_zone_for_movable(void)
3284 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3285 if (zone_index
== ZONE_MOVABLE
)
3288 if (arch_zone_highest_possible_pfn
[zone_index
] >
3289 arch_zone_lowest_possible_pfn
[zone_index
])
3293 VM_BUG_ON(zone_index
== -1);
3294 movable_zone
= zone_index
;
3298 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3299 * because it is sized independant of architecture. Unlike the other zones,
3300 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3301 * in each node depending on the size of each node and how evenly kernelcore
3302 * is distributed. This helper function adjusts the zone ranges
3303 * provided by the architecture for a given node by using the end of the
3304 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3305 * zones within a node are in order of monotonic increases memory addresses
3307 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3308 unsigned long zone_type
,
3309 unsigned long node_start_pfn
,
3310 unsigned long node_end_pfn
,
3311 unsigned long *zone_start_pfn
,
3312 unsigned long *zone_end_pfn
)
3314 /* Only adjust if ZONE_MOVABLE is on this node */
3315 if (zone_movable_pfn
[nid
]) {
3316 /* Size ZONE_MOVABLE */
3317 if (zone_type
== ZONE_MOVABLE
) {
3318 *zone_start_pfn
= zone_movable_pfn
[nid
];
3319 *zone_end_pfn
= min(node_end_pfn
,
3320 arch_zone_highest_possible_pfn
[movable_zone
]);
3322 /* Adjust for ZONE_MOVABLE starting within this range */
3323 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3324 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3325 *zone_end_pfn
= zone_movable_pfn
[nid
];
3327 /* Check if this whole range is within ZONE_MOVABLE */
3328 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3329 *zone_start_pfn
= *zone_end_pfn
;
3334 * Return the number of pages a zone spans in a node, including holes
3335 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3337 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3338 unsigned long zone_type
,
3339 unsigned long *ignored
)
3341 unsigned long node_start_pfn
, node_end_pfn
;
3342 unsigned long zone_start_pfn
, zone_end_pfn
;
3344 /* Get the start and end of the node and zone */
3345 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3346 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3347 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3348 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3349 node_start_pfn
, node_end_pfn
,
3350 &zone_start_pfn
, &zone_end_pfn
);
3352 /* Check that this node has pages within the zone's required range */
3353 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3356 /* Move the zone boundaries inside the node if necessary */
3357 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3358 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3360 /* Return the spanned pages */
3361 return zone_end_pfn
- zone_start_pfn
;
3365 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3366 * then all holes in the requested range will be accounted for.
3368 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3369 unsigned long range_start_pfn
,
3370 unsigned long range_end_pfn
)
3373 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3374 unsigned long start_pfn
;
3376 /* Find the end_pfn of the first active range of pfns in the node */
3377 i
= first_active_region_index_in_nid(nid
);
3381 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3383 /* Account for ranges before physical memory on this node */
3384 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3385 hole_pages
= prev_end_pfn
- range_start_pfn
;
3387 /* Find all holes for the zone within the node */
3388 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3390 /* No need to continue if prev_end_pfn is outside the zone */
3391 if (prev_end_pfn
>= range_end_pfn
)
3394 /* Make sure the end of the zone is not within the hole */
3395 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3396 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3398 /* Update the hole size cound and move on */
3399 if (start_pfn
> range_start_pfn
) {
3400 BUG_ON(prev_end_pfn
> start_pfn
);
3401 hole_pages
+= start_pfn
- prev_end_pfn
;
3403 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3406 /* Account for ranges past physical memory on this node */
3407 if (range_end_pfn
> prev_end_pfn
)
3408 hole_pages
+= range_end_pfn
-
3409 max(range_start_pfn
, prev_end_pfn
);
3415 * absent_pages_in_range - Return number of page frames in holes within a range
3416 * @start_pfn: The start PFN to start searching for holes
3417 * @end_pfn: The end PFN to stop searching for holes
3419 * It returns the number of pages frames in memory holes within a range.
3421 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3422 unsigned long end_pfn
)
3424 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3427 /* Return the number of page frames in holes in a zone on a node */
3428 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3429 unsigned long zone_type
,
3430 unsigned long *ignored
)
3432 unsigned long node_start_pfn
, node_end_pfn
;
3433 unsigned long zone_start_pfn
, zone_end_pfn
;
3435 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3436 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3438 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3441 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3442 node_start_pfn
, node_end_pfn
,
3443 &zone_start_pfn
, &zone_end_pfn
);
3444 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3448 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3449 unsigned long zone_type
,
3450 unsigned long *zones_size
)
3452 return zones_size
[zone_type
];
3455 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3456 unsigned long zone_type
,
3457 unsigned long *zholes_size
)
3462 return zholes_size
[zone_type
];
3467 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3468 unsigned long *zones_size
, unsigned long *zholes_size
)
3470 unsigned long realtotalpages
, totalpages
= 0;
3473 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3474 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3476 pgdat
->node_spanned_pages
= totalpages
;
3478 realtotalpages
= totalpages
;
3479 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3481 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3483 pgdat
->node_present_pages
= realtotalpages
;
3484 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3488 #ifndef CONFIG_SPARSEMEM
3490 * Calculate the size of the zone->blockflags rounded to an unsigned long
3491 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3492 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3493 * round what is now in bits to nearest long in bits, then return it in
3496 static unsigned long __init
usemap_size(unsigned long zonesize
)
3498 unsigned long usemapsize
;
3500 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3501 usemapsize
= usemapsize
>> pageblock_order
;
3502 usemapsize
*= NR_PAGEBLOCK_BITS
;
3503 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3505 return usemapsize
/ 8;
3508 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3509 struct zone
*zone
, unsigned long zonesize
)
3511 unsigned long usemapsize
= usemap_size(zonesize
);
3512 zone
->pageblock_flags
= NULL
;
3514 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3517 static void inline setup_usemap(struct pglist_data
*pgdat
,
3518 struct zone
*zone
, unsigned long zonesize
) {}
3519 #endif /* CONFIG_SPARSEMEM */
3521 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3523 /* Return a sensible default order for the pageblock size. */
3524 static inline int pageblock_default_order(void)
3526 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3527 return HUGETLB_PAGE_ORDER
;
3532 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3533 static inline void __init
set_pageblock_order(unsigned int order
)
3535 /* Check that pageblock_nr_pages has not already been setup */
3536 if (pageblock_order
)
3540 * Assume the largest contiguous order of interest is a huge page.
3541 * This value may be variable depending on boot parameters on IA64
3543 pageblock_order
= order
;
3545 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3548 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3549 * and pageblock_default_order() are unused as pageblock_order is set
3550 * at compile-time. See include/linux/pageblock-flags.h for the values of
3551 * pageblock_order based on the kernel config
3553 static inline int pageblock_default_order(unsigned int order
)
3557 #define set_pageblock_order(x) do {} while (0)
3559 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3562 * Set up the zone data structures:
3563 * - mark all pages reserved
3564 * - mark all memory queues empty
3565 * - clear the memory bitmaps
3567 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3568 unsigned long *zones_size
, unsigned long *zholes_size
)
3571 int nid
= pgdat
->node_id
;
3572 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3575 pgdat_resize_init(pgdat
);
3576 pgdat
->nr_zones
= 0;
3577 init_waitqueue_head(&pgdat
->kswapd_wait
);
3578 pgdat
->kswapd_max_order
= 0;
3579 pgdat_page_cgroup_init(pgdat
);
3581 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3582 struct zone
*zone
= pgdat
->node_zones
+ j
;
3583 unsigned long size
, realsize
, memmap_pages
;
3586 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3587 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3591 * Adjust realsize so that it accounts for how much memory
3592 * is used by this zone for memmap. This affects the watermark
3593 * and per-cpu initialisations
3596 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3597 if (realsize
>= memmap_pages
) {
3598 realsize
-= memmap_pages
;
3601 " %s zone: %lu pages used for memmap\n",
3602 zone_names
[j
], memmap_pages
);
3605 " %s zone: %lu pages exceeds realsize %lu\n",
3606 zone_names
[j
], memmap_pages
, realsize
);
3608 /* Account for reserved pages */
3609 if (j
== 0 && realsize
> dma_reserve
) {
3610 realsize
-= dma_reserve
;
3611 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3612 zone_names
[0], dma_reserve
);
3615 if (!is_highmem_idx(j
))
3616 nr_kernel_pages
+= realsize
;
3617 nr_all_pages
+= realsize
;
3619 zone
->spanned_pages
= size
;
3620 zone
->present_pages
= realsize
;
3623 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3625 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3627 zone
->name
= zone_names
[j
];
3628 spin_lock_init(&zone
->lock
);
3629 spin_lock_init(&zone
->lru_lock
);
3630 zone_seqlock_init(zone
);
3631 zone
->zone_pgdat
= pgdat
;
3633 zone
->prev_priority
= DEF_PRIORITY
;
3635 zone_pcp_init(zone
);
3637 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3638 zone
->lru
[l
].nr_scan
= 0;
3640 zone
->reclaim_stat
.recent_rotated
[0] = 0;
3641 zone
->reclaim_stat
.recent_rotated
[1] = 0;
3642 zone
->reclaim_stat
.recent_scanned
[0] = 0;
3643 zone
->reclaim_stat
.recent_scanned
[1] = 0;
3644 zap_zone_vm_stats(zone
);
3649 set_pageblock_order(pageblock_default_order());
3650 setup_usemap(pgdat
, zone
, size
);
3651 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3652 size
, MEMMAP_EARLY
);
3654 memmap_init(size
, nid
, j
, zone_start_pfn
);
3655 zone_start_pfn
+= size
;
3659 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3661 /* Skip empty nodes */
3662 if (!pgdat
->node_spanned_pages
)
3665 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3666 /* ia64 gets its own node_mem_map, before this, without bootmem */
3667 if (!pgdat
->node_mem_map
) {
3668 unsigned long size
, start
, end
;
3672 * The zone's endpoints aren't required to be MAX_ORDER
3673 * aligned but the node_mem_map endpoints must be in order
3674 * for the buddy allocator to function correctly.
3676 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3677 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3678 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3679 size
= (end
- start
) * sizeof(struct page
);
3680 map
= alloc_remap(pgdat
->node_id
, size
);
3682 map
= alloc_bootmem_node(pgdat
, size
);
3683 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3685 #ifndef CONFIG_NEED_MULTIPLE_NODES
3687 * With no DISCONTIG, the global mem_map is just set as node 0's
3689 if (pgdat
== NODE_DATA(0)) {
3690 mem_map
= NODE_DATA(0)->node_mem_map
;
3691 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3692 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3693 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3694 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3697 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3700 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3701 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3703 pg_data_t
*pgdat
= NODE_DATA(nid
);
3705 pgdat
->node_id
= nid
;
3706 pgdat
->node_start_pfn
= node_start_pfn
;
3707 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3709 alloc_node_mem_map(pgdat
);
3710 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3711 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3712 nid
, (unsigned long)pgdat
,
3713 (unsigned long)pgdat
->node_mem_map
);
3716 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3719 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3721 #if MAX_NUMNODES > 1
3723 * Figure out the number of possible node ids.
3725 static void __init
setup_nr_node_ids(void)
3728 unsigned int highest
= 0;
3730 for_each_node_mask(node
, node_possible_map
)
3732 nr_node_ids
= highest
+ 1;
3735 static inline void setup_nr_node_ids(void)
3741 * add_active_range - Register a range of PFNs backed by physical memory
3742 * @nid: The node ID the range resides on
3743 * @start_pfn: The start PFN of the available physical memory
3744 * @end_pfn: The end PFN of the available physical memory
3746 * These ranges are stored in an early_node_map[] and later used by
3747 * free_area_init_nodes() to calculate zone sizes and holes. If the
3748 * range spans a memory hole, it is up to the architecture to ensure
3749 * the memory is not freed by the bootmem allocator. If possible
3750 * the range being registered will be merged with existing ranges.
3752 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3753 unsigned long end_pfn
)
3757 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3758 "Entering add_active_range(%d, %#lx, %#lx) "
3759 "%d entries of %d used\n",
3760 nid
, start_pfn
, end_pfn
,
3761 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3763 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3765 /* Merge with existing active regions if possible */
3766 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3767 if (early_node_map
[i
].nid
!= nid
)
3770 /* Skip if an existing region covers this new one */
3771 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3772 end_pfn
<= early_node_map
[i
].end_pfn
)
3775 /* Merge forward if suitable */
3776 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3777 end_pfn
> early_node_map
[i
].end_pfn
) {
3778 early_node_map
[i
].end_pfn
= end_pfn
;
3782 /* Merge backward if suitable */
3783 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3784 end_pfn
>= early_node_map
[i
].start_pfn
) {
3785 early_node_map
[i
].start_pfn
= start_pfn
;
3790 /* Check that early_node_map is large enough */
3791 if (i
>= MAX_ACTIVE_REGIONS
) {
3792 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3793 MAX_ACTIVE_REGIONS
);
3797 early_node_map
[i
].nid
= nid
;
3798 early_node_map
[i
].start_pfn
= start_pfn
;
3799 early_node_map
[i
].end_pfn
= end_pfn
;
3800 nr_nodemap_entries
= i
+ 1;
3804 * remove_active_range - Shrink an existing registered range of PFNs
3805 * @nid: The node id the range is on that should be shrunk
3806 * @start_pfn: The new PFN of the range
3807 * @end_pfn: The new PFN of the range
3809 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3810 * The map is kept near the end physical page range that has already been
3811 * registered. This function allows an arch to shrink an existing registered
3814 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3815 unsigned long end_pfn
)
3820 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3821 nid
, start_pfn
, end_pfn
);
3823 /* Find the old active region end and shrink */
3824 for_each_active_range_index_in_nid(i
, nid
) {
3825 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3826 early_node_map
[i
].end_pfn
<= end_pfn
) {
3828 early_node_map
[i
].start_pfn
= 0;
3829 early_node_map
[i
].end_pfn
= 0;
3833 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3834 early_node_map
[i
].end_pfn
> start_pfn
) {
3835 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3836 early_node_map
[i
].end_pfn
= start_pfn
;
3837 if (temp_end_pfn
> end_pfn
)
3838 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3841 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3842 early_node_map
[i
].end_pfn
> end_pfn
&&
3843 early_node_map
[i
].start_pfn
< end_pfn
) {
3844 early_node_map
[i
].start_pfn
= end_pfn
;
3852 /* remove the blank ones */
3853 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3854 if (early_node_map
[i
].nid
!= nid
)
3856 if (early_node_map
[i
].end_pfn
)
3858 /* we found it, get rid of it */
3859 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3860 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3861 sizeof(early_node_map
[j
]));
3862 j
= nr_nodemap_entries
- 1;
3863 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3864 nr_nodemap_entries
--;
3869 * remove_all_active_ranges - Remove all currently registered regions
3871 * During discovery, it may be found that a table like SRAT is invalid
3872 * and an alternative discovery method must be used. This function removes
3873 * all currently registered regions.
3875 void __init
remove_all_active_ranges(void)
3877 memset(early_node_map
, 0, sizeof(early_node_map
));
3878 nr_nodemap_entries
= 0;
3881 /* Compare two active node_active_regions */
3882 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3884 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3885 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3887 /* Done this way to avoid overflows */
3888 if (arange
->start_pfn
> brange
->start_pfn
)
3890 if (arange
->start_pfn
< brange
->start_pfn
)
3896 /* sort the node_map by start_pfn */
3897 static void __init
sort_node_map(void)
3899 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3900 sizeof(struct node_active_region
),
3901 cmp_node_active_region
, NULL
);
3904 /* Find the lowest pfn for a node */
3905 static unsigned long __init
find_min_pfn_for_node(int nid
)
3908 unsigned long min_pfn
= ULONG_MAX
;
3910 /* Assuming a sorted map, the first range found has the starting pfn */
3911 for_each_active_range_index_in_nid(i
, nid
)
3912 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3914 if (min_pfn
== ULONG_MAX
) {
3916 "Could not find start_pfn for node %d\n", nid
);
3924 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3926 * It returns the minimum PFN based on information provided via
3927 * add_active_range().
3929 unsigned long __init
find_min_pfn_with_active_regions(void)
3931 return find_min_pfn_for_node(MAX_NUMNODES
);
3935 * early_calculate_totalpages()
3936 * Sum pages in active regions for movable zone.
3937 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3939 static unsigned long __init
early_calculate_totalpages(void)
3942 unsigned long totalpages
= 0;
3944 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3945 unsigned long pages
= early_node_map
[i
].end_pfn
-
3946 early_node_map
[i
].start_pfn
;
3947 totalpages
+= pages
;
3949 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3955 * Find the PFN the Movable zone begins in each node. Kernel memory
3956 * is spread evenly between nodes as long as the nodes have enough
3957 * memory. When they don't, some nodes will have more kernelcore than
3960 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3963 unsigned long usable_startpfn
;
3964 unsigned long kernelcore_node
, kernelcore_remaining
;
3965 unsigned long totalpages
= early_calculate_totalpages();
3966 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3969 * If movablecore was specified, calculate what size of
3970 * kernelcore that corresponds so that memory usable for
3971 * any allocation type is evenly spread. If both kernelcore
3972 * and movablecore are specified, then the value of kernelcore
3973 * will be used for required_kernelcore if it's greater than
3974 * what movablecore would have allowed.
3976 if (required_movablecore
) {
3977 unsigned long corepages
;
3980 * Round-up so that ZONE_MOVABLE is at least as large as what
3981 * was requested by the user
3983 required_movablecore
=
3984 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3985 corepages
= totalpages
- required_movablecore
;
3987 required_kernelcore
= max(required_kernelcore
, corepages
);
3990 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3991 if (!required_kernelcore
)
3994 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3995 find_usable_zone_for_movable();
3996 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3999 /* Spread kernelcore memory as evenly as possible throughout nodes */
4000 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4001 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4003 * Recalculate kernelcore_node if the division per node
4004 * now exceeds what is necessary to satisfy the requested
4005 * amount of memory for the kernel
4007 if (required_kernelcore
< kernelcore_node
)
4008 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4011 * As the map is walked, we track how much memory is usable
4012 * by the kernel using kernelcore_remaining. When it is
4013 * 0, the rest of the node is usable by ZONE_MOVABLE
4015 kernelcore_remaining
= kernelcore_node
;
4017 /* Go through each range of PFNs within this node */
4018 for_each_active_range_index_in_nid(i
, nid
) {
4019 unsigned long start_pfn
, end_pfn
;
4020 unsigned long size_pages
;
4022 start_pfn
= max(early_node_map
[i
].start_pfn
,
4023 zone_movable_pfn
[nid
]);
4024 end_pfn
= early_node_map
[i
].end_pfn
;
4025 if (start_pfn
>= end_pfn
)
4028 /* Account for what is only usable for kernelcore */
4029 if (start_pfn
< usable_startpfn
) {
4030 unsigned long kernel_pages
;
4031 kernel_pages
= min(end_pfn
, usable_startpfn
)
4034 kernelcore_remaining
-= min(kernel_pages
,
4035 kernelcore_remaining
);
4036 required_kernelcore
-= min(kernel_pages
,
4037 required_kernelcore
);
4039 /* Continue if range is now fully accounted */
4040 if (end_pfn
<= usable_startpfn
) {
4043 * Push zone_movable_pfn to the end so
4044 * that if we have to rebalance
4045 * kernelcore across nodes, we will
4046 * not double account here
4048 zone_movable_pfn
[nid
] = end_pfn
;
4051 start_pfn
= usable_startpfn
;
4055 * The usable PFN range for ZONE_MOVABLE is from
4056 * start_pfn->end_pfn. Calculate size_pages as the
4057 * number of pages used as kernelcore
4059 size_pages
= end_pfn
- start_pfn
;
4060 if (size_pages
> kernelcore_remaining
)
4061 size_pages
= kernelcore_remaining
;
4062 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4065 * Some kernelcore has been met, update counts and
4066 * break if the kernelcore for this node has been
4069 required_kernelcore
-= min(required_kernelcore
,
4071 kernelcore_remaining
-= size_pages
;
4072 if (!kernelcore_remaining
)
4078 * If there is still required_kernelcore, we do another pass with one
4079 * less node in the count. This will push zone_movable_pfn[nid] further
4080 * along on the nodes that still have memory until kernelcore is
4084 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4087 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4088 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4089 zone_movable_pfn
[nid
] =
4090 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4093 /* Any regular memory on that node ? */
4094 static void check_for_regular_memory(pg_data_t
*pgdat
)
4096 #ifdef CONFIG_HIGHMEM
4097 enum zone_type zone_type
;
4099 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4100 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4101 if (zone
->present_pages
)
4102 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4108 * free_area_init_nodes - Initialise all pg_data_t and zone data
4109 * @max_zone_pfn: an array of max PFNs for each zone
4111 * This will call free_area_init_node() for each active node in the system.
4112 * Using the page ranges provided by add_active_range(), the size of each
4113 * zone in each node and their holes is calculated. If the maximum PFN
4114 * between two adjacent zones match, it is assumed that the zone is empty.
4115 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4116 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4117 * starts where the previous one ended. For example, ZONE_DMA32 starts
4118 * at arch_max_dma_pfn.
4120 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4125 /* Sort early_node_map as initialisation assumes it is sorted */
4128 /* Record where the zone boundaries are */
4129 memset(arch_zone_lowest_possible_pfn
, 0,
4130 sizeof(arch_zone_lowest_possible_pfn
));
4131 memset(arch_zone_highest_possible_pfn
, 0,
4132 sizeof(arch_zone_highest_possible_pfn
));
4133 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4134 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4135 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4136 if (i
== ZONE_MOVABLE
)
4138 arch_zone_lowest_possible_pfn
[i
] =
4139 arch_zone_highest_possible_pfn
[i
-1];
4140 arch_zone_highest_possible_pfn
[i
] =
4141 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4143 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4144 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4146 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4147 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4148 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4150 /* Print out the zone ranges */
4151 printk("Zone PFN ranges:\n");
4152 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4153 if (i
== ZONE_MOVABLE
)
4155 printk(" %-8s %0#10lx -> %0#10lx\n",
4157 arch_zone_lowest_possible_pfn
[i
],
4158 arch_zone_highest_possible_pfn
[i
]);
4161 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4162 printk("Movable zone start PFN for each node\n");
4163 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4164 if (zone_movable_pfn
[i
])
4165 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4168 /* Print out the early_node_map[] */
4169 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4170 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4171 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4172 early_node_map
[i
].start_pfn
,
4173 early_node_map
[i
].end_pfn
);
4175 /* Initialise every node */
4176 mminit_verify_pageflags_layout();
4177 setup_nr_node_ids();
4178 for_each_online_node(nid
) {
4179 pg_data_t
*pgdat
= NODE_DATA(nid
);
4180 free_area_init_node(nid
, NULL
,
4181 find_min_pfn_for_node(nid
), NULL
);
4183 /* Any memory on that node */
4184 if (pgdat
->node_present_pages
)
4185 node_set_state(nid
, N_HIGH_MEMORY
);
4186 check_for_regular_memory(pgdat
);
4190 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4192 unsigned long long coremem
;
4196 coremem
= memparse(p
, &p
);
4197 *core
= coremem
>> PAGE_SHIFT
;
4199 /* Paranoid check that UL is enough for the coremem value */
4200 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4206 * kernelcore=size sets the amount of memory for use for allocations that
4207 * cannot be reclaimed or migrated.
4209 static int __init
cmdline_parse_kernelcore(char *p
)
4211 return cmdline_parse_core(p
, &required_kernelcore
);
4215 * movablecore=size sets the amount of memory for use for allocations that
4216 * can be reclaimed or migrated.
4218 static int __init
cmdline_parse_movablecore(char *p
)
4220 return cmdline_parse_core(p
, &required_movablecore
);
4223 early_param("kernelcore", cmdline_parse_kernelcore
);
4224 early_param("movablecore", cmdline_parse_movablecore
);
4226 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4229 * set_dma_reserve - set the specified number of pages reserved in the first zone
4230 * @new_dma_reserve: The number of pages to mark reserved
4232 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4233 * In the DMA zone, a significant percentage may be consumed by kernel image
4234 * and other unfreeable allocations which can skew the watermarks badly. This
4235 * function may optionally be used to account for unfreeable pages in the
4236 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4237 * smaller per-cpu batchsize.
4239 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4241 dma_reserve
= new_dma_reserve
;
4244 #ifndef CONFIG_NEED_MULTIPLE_NODES
4245 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4246 EXPORT_SYMBOL(contig_page_data
);
4249 void __init
free_area_init(unsigned long *zones_size
)
4251 free_area_init_node(0, zones_size
,
4252 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4255 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4256 unsigned long action
, void *hcpu
)
4258 int cpu
= (unsigned long)hcpu
;
4260 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4264 * Spill the event counters of the dead processor
4265 * into the current processors event counters.
4266 * This artificially elevates the count of the current
4269 vm_events_fold_cpu(cpu
);
4272 * Zero the differential counters of the dead processor
4273 * so that the vm statistics are consistent.
4275 * This is only okay since the processor is dead and cannot
4276 * race with what we are doing.
4278 refresh_cpu_vm_stats(cpu
);
4283 void __init
page_alloc_init(void)
4285 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4289 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4290 * or min_free_kbytes changes.
4292 static void calculate_totalreserve_pages(void)
4294 struct pglist_data
*pgdat
;
4295 unsigned long reserve_pages
= 0;
4296 enum zone_type i
, j
;
4298 for_each_online_pgdat(pgdat
) {
4299 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4300 struct zone
*zone
= pgdat
->node_zones
+ i
;
4301 unsigned long max
= 0;
4303 /* Find valid and maximum lowmem_reserve in the zone */
4304 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4305 if (zone
->lowmem_reserve
[j
] > max
)
4306 max
= zone
->lowmem_reserve
[j
];
4309 /* we treat pages_high as reserved pages. */
4310 max
+= zone
->pages_high
;
4312 if (max
> zone
->present_pages
)
4313 max
= zone
->present_pages
;
4314 reserve_pages
+= max
;
4317 totalreserve_pages
= reserve_pages
;
4321 * setup_per_zone_lowmem_reserve - called whenever
4322 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4323 * has a correct pages reserved value, so an adequate number of
4324 * pages are left in the zone after a successful __alloc_pages().
4326 static void setup_per_zone_lowmem_reserve(void)
4328 struct pglist_data
*pgdat
;
4329 enum zone_type j
, idx
;
4331 for_each_online_pgdat(pgdat
) {
4332 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4333 struct zone
*zone
= pgdat
->node_zones
+ j
;
4334 unsigned long present_pages
= zone
->present_pages
;
4336 zone
->lowmem_reserve
[j
] = 0;
4340 struct zone
*lower_zone
;
4344 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4345 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4347 lower_zone
= pgdat
->node_zones
+ idx
;
4348 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4349 sysctl_lowmem_reserve_ratio
[idx
];
4350 present_pages
+= lower_zone
->present_pages
;
4355 /* update totalreserve_pages */
4356 calculate_totalreserve_pages();
4360 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4362 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4363 * with respect to min_free_kbytes.
4365 void setup_per_zone_pages_min(void)
4367 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4368 unsigned long lowmem_pages
= 0;
4370 unsigned long flags
;
4372 /* Calculate total number of !ZONE_HIGHMEM pages */
4373 for_each_zone(zone
) {
4374 if (!is_highmem(zone
))
4375 lowmem_pages
+= zone
->present_pages
;
4378 for_each_zone(zone
) {
4381 spin_lock_irqsave(&zone
->lock
, flags
);
4382 tmp
= (u64
)pages_min
* zone
->present_pages
;
4383 do_div(tmp
, lowmem_pages
);
4384 if (is_highmem(zone
)) {
4386 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4387 * need highmem pages, so cap pages_min to a small
4390 * The (pages_high-pages_low) and (pages_low-pages_min)
4391 * deltas controls asynch page reclaim, and so should
4392 * not be capped for highmem.
4396 min_pages
= zone
->present_pages
/ 1024;
4397 if (min_pages
< SWAP_CLUSTER_MAX
)
4398 min_pages
= SWAP_CLUSTER_MAX
;
4399 if (min_pages
> 128)
4401 zone
->pages_min
= min_pages
;
4404 * If it's a lowmem zone, reserve a number of pages
4405 * proportionate to the zone's size.
4407 zone
->pages_min
= tmp
;
4410 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4411 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4412 setup_zone_migrate_reserve(zone
);
4413 spin_unlock_irqrestore(&zone
->lock
, flags
);
4416 /* update totalreserve_pages */
4417 calculate_totalreserve_pages();
4421 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4423 * The inactive anon list should be small enough that the VM never has to
4424 * do too much work, but large enough that each inactive page has a chance
4425 * to be referenced again before it is swapped out.
4427 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4428 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4429 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4430 * the anonymous pages are kept on the inactive list.
4433 * memory ratio inactive anon
4434 * -------------------------------------
4443 static void setup_per_zone_inactive_ratio(void)
4447 for_each_zone(zone
) {
4448 unsigned int gb
, ratio
;
4450 /* Zone size in gigabytes */
4451 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4452 ratio
= int_sqrt(10 * gb
);
4456 zone
->inactive_ratio
= ratio
;
4461 * Initialise min_free_kbytes.
4463 * For small machines we want it small (128k min). For large machines
4464 * we want it large (64MB max). But it is not linear, because network
4465 * bandwidth does not increase linearly with machine size. We use
4467 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4468 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4484 static int __init
init_per_zone_pages_min(void)
4486 unsigned long lowmem_kbytes
;
4488 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4490 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4491 if (min_free_kbytes
< 128)
4492 min_free_kbytes
= 128;
4493 if (min_free_kbytes
> 65536)
4494 min_free_kbytes
= 65536;
4495 setup_per_zone_pages_min();
4496 setup_per_zone_lowmem_reserve();
4497 setup_per_zone_inactive_ratio();
4500 module_init(init_per_zone_pages_min
)
4503 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4504 * that we can call two helper functions whenever min_free_kbytes
4507 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4508 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4510 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4512 setup_per_zone_pages_min();
4517 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4518 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4523 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4528 zone
->min_unmapped_pages
= (zone
->present_pages
*
4529 sysctl_min_unmapped_ratio
) / 100;
4533 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4534 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4539 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4544 zone
->min_slab_pages
= (zone
->present_pages
*
4545 sysctl_min_slab_ratio
) / 100;
4551 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4552 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4553 * whenever sysctl_lowmem_reserve_ratio changes.
4555 * The reserve ratio obviously has absolutely no relation with the
4556 * pages_min watermarks. The lowmem reserve ratio can only make sense
4557 * if in function of the boot time zone sizes.
4559 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4560 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4562 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4563 setup_per_zone_lowmem_reserve();
4568 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4569 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4570 * can have before it gets flushed back to buddy allocator.
4573 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4574 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4580 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4581 if (!write
|| (ret
== -EINVAL
))
4583 for_each_zone(zone
) {
4584 for_each_online_cpu(cpu
) {
4586 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4587 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4593 int hashdist
= HASHDIST_DEFAULT
;
4596 static int __init
set_hashdist(char *str
)
4600 hashdist
= simple_strtoul(str
, &str
, 0);
4603 __setup("hashdist=", set_hashdist
);
4607 * allocate a large system hash table from bootmem
4608 * - it is assumed that the hash table must contain an exact power-of-2
4609 * quantity of entries
4610 * - limit is the number of hash buckets, not the total allocation size
4612 void *__init
alloc_large_system_hash(const char *tablename
,
4613 unsigned long bucketsize
,
4614 unsigned long numentries
,
4617 unsigned int *_hash_shift
,
4618 unsigned int *_hash_mask
,
4619 unsigned long limit
)
4621 unsigned long long max
= limit
;
4622 unsigned long log2qty
, size
;
4625 /* allow the kernel cmdline to have a say */
4627 /* round applicable memory size up to nearest megabyte */
4628 numentries
= nr_kernel_pages
;
4629 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4630 numentries
>>= 20 - PAGE_SHIFT
;
4631 numentries
<<= 20 - PAGE_SHIFT
;
4633 /* limit to 1 bucket per 2^scale bytes of low memory */
4634 if (scale
> PAGE_SHIFT
)
4635 numentries
>>= (scale
- PAGE_SHIFT
);
4637 numentries
<<= (PAGE_SHIFT
- scale
);
4639 /* Make sure we've got at least a 0-order allocation.. */
4640 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4641 numentries
= PAGE_SIZE
/ bucketsize
;
4643 numentries
= roundup_pow_of_two(numentries
);
4645 /* limit allocation size to 1/16 total memory by default */
4647 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4648 do_div(max
, bucketsize
);
4651 if (numentries
> max
)
4654 log2qty
= ilog2(numentries
);
4657 size
= bucketsize
<< log2qty
;
4658 if (flags
& HASH_EARLY
)
4659 table
= alloc_bootmem_nopanic(size
);
4661 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4663 unsigned long order
= get_order(size
);
4665 if (order
< MAX_ORDER
)
4666 table
= (void *)__get_free_pages(GFP_ATOMIC
,
4669 * If bucketsize is not a power-of-two, we may free
4670 * some pages at the end of hash table.
4673 unsigned long alloc_end
= (unsigned long)table
+
4674 (PAGE_SIZE
<< order
);
4675 unsigned long used
= (unsigned long)table
+
4677 split_page(virt_to_page(table
), order
);
4678 while (used
< alloc_end
) {
4684 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4687 panic("Failed to allocate %s hash table\n", tablename
);
4689 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4692 ilog2(size
) - PAGE_SHIFT
,
4696 *_hash_shift
= log2qty
;
4698 *_hash_mask
= (1 << log2qty
) - 1;
4701 * If hashdist is set, the table allocation is done with __vmalloc()
4702 * which invokes the kmemleak_alloc() callback. This function may also
4703 * be called before the slab and kmemleak are initialised when
4704 * kmemleak simply buffers the request to be executed later
4705 * (GFP_ATOMIC flag ignored in this case).
4708 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
4713 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4714 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4717 #ifdef CONFIG_SPARSEMEM
4718 return __pfn_to_section(pfn
)->pageblock_flags
;
4720 return zone
->pageblock_flags
;
4721 #endif /* CONFIG_SPARSEMEM */
4724 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4726 #ifdef CONFIG_SPARSEMEM
4727 pfn
&= (PAGES_PER_SECTION
-1);
4728 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4730 pfn
= pfn
- zone
->zone_start_pfn
;
4731 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4732 #endif /* CONFIG_SPARSEMEM */
4736 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4737 * @page: The page within the block of interest
4738 * @start_bitidx: The first bit of interest to retrieve
4739 * @end_bitidx: The last bit of interest
4740 * returns pageblock_bits flags
4742 unsigned long get_pageblock_flags_group(struct page
*page
,
4743 int start_bitidx
, int end_bitidx
)
4746 unsigned long *bitmap
;
4747 unsigned long pfn
, bitidx
;
4748 unsigned long flags
= 0;
4749 unsigned long value
= 1;
4751 zone
= page_zone(page
);
4752 pfn
= page_to_pfn(page
);
4753 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4754 bitidx
= pfn_to_bitidx(zone
, pfn
);
4756 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4757 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4764 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4765 * @page: The page within the block of interest
4766 * @start_bitidx: The first bit of interest
4767 * @end_bitidx: The last bit of interest
4768 * @flags: The flags to set
4770 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4771 int start_bitidx
, int end_bitidx
)
4774 unsigned long *bitmap
;
4775 unsigned long pfn
, bitidx
;
4776 unsigned long value
= 1;
4778 zone
= page_zone(page
);
4779 pfn
= page_to_pfn(page
);
4780 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4781 bitidx
= pfn_to_bitidx(zone
, pfn
);
4782 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4783 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4785 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4787 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4789 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4793 * This is designed as sub function...plz see page_isolation.c also.
4794 * set/clear page block's type to be ISOLATE.
4795 * page allocater never alloc memory from ISOLATE block.
4798 int set_migratetype_isolate(struct page
*page
)
4801 unsigned long flags
;
4804 zone
= page_zone(page
);
4805 spin_lock_irqsave(&zone
->lock
, flags
);
4807 * In future, more migrate types will be able to be isolation target.
4809 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4811 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4812 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4815 spin_unlock_irqrestore(&zone
->lock
, flags
);
4821 void unset_migratetype_isolate(struct page
*page
)
4824 unsigned long flags
;
4825 zone
= page_zone(page
);
4826 spin_lock_irqsave(&zone
->lock
, flags
);
4827 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4829 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4830 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4832 spin_unlock_irqrestore(&zone
->lock
, flags
);
4835 #ifdef CONFIG_MEMORY_HOTREMOVE
4837 * All pages in the range must be isolated before calling this.
4840 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4846 unsigned long flags
;
4847 /* find the first valid pfn */
4848 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4853 zone
= page_zone(pfn_to_page(pfn
));
4854 spin_lock_irqsave(&zone
->lock
, flags
);
4856 while (pfn
< end_pfn
) {
4857 if (!pfn_valid(pfn
)) {
4861 page
= pfn_to_page(pfn
);
4862 BUG_ON(page_count(page
));
4863 BUG_ON(!PageBuddy(page
));
4864 order
= page_order(page
);
4865 #ifdef CONFIG_DEBUG_VM
4866 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4867 pfn
, 1 << order
, end_pfn
);
4869 list_del(&page
->lru
);
4870 rmv_page_order(page
);
4871 zone
->free_area
[order
].nr_free
--;
4872 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4874 for (i
= 0; i
< (1 << order
); i
++)
4875 SetPageReserved((page
+i
));
4876 pfn
+= (1 << order
);
4878 spin_unlock_irqrestore(&zone
->lock
, flags
);