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
)
171 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
172 PB_migrate
, PB_migrate_end
);
175 #ifdef CONFIG_DEBUG_VM
176 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
180 unsigned long pfn
= page_to_pfn(page
);
183 seq
= zone_span_seqbegin(zone
);
184 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
186 else if (pfn
< zone
->zone_start_pfn
)
188 } while (zone_span_seqretry(zone
, seq
));
193 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
195 if (!pfn_valid_within(page_to_pfn(page
)))
197 if (zone
!= page_zone(page
))
203 * Temporary debugging check for pages not lying within a given zone.
205 static int bad_range(struct zone
*zone
, struct page
*page
)
207 if (page_outside_zone_boundaries(zone
, page
))
209 if (!page_is_consistent(zone
, page
))
215 static inline int bad_range(struct zone
*zone
, struct page
*page
)
221 static void bad_page(struct page
*page
)
223 static unsigned long resume
;
224 static unsigned long nr_shown
;
225 static unsigned long nr_unshown
;
228 * Allow a burst of 60 reports, then keep quiet for that minute;
229 * or allow a steady drip of one report per second.
231 if (nr_shown
== 60) {
232 if (time_before(jiffies
, resume
)) {
238 "BUG: Bad page state: %lu messages suppressed\n",
245 resume
= jiffies
+ 60 * HZ
;
247 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
248 current
->comm
, page_to_pfn(page
));
250 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
251 page
, (void *)page
->flags
, page_count(page
),
252 page_mapcount(page
), page
->mapping
, page
->index
);
256 /* Leave bad fields for debug, except PageBuddy could make trouble */
257 __ClearPageBuddy(page
);
258 add_taint(TAINT_BAD_PAGE
);
262 * Higher-order pages are called "compound pages". They are structured thusly:
264 * The first PAGE_SIZE page is called the "head page".
266 * The remaining PAGE_SIZE pages are called "tail pages".
268 * All pages have PG_compound set. All pages have their ->private pointing at
269 * the head page (even the head page has this).
271 * The first tail page's ->lru.next holds the address of the compound page's
272 * put_page() function. Its ->lru.prev holds the order of allocation.
273 * This usage means that zero-order pages may not be compound.
276 static void free_compound_page(struct page
*page
)
278 __free_pages_ok(page
, compound_order(page
));
281 void prep_compound_page(struct page
*page
, unsigned long order
)
284 int nr_pages
= 1 << order
;
286 set_compound_page_dtor(page
, free_compound_page
);
287 set_compound_order(page
, order
);
289 for (i
= 1; i
< nr_pages
; i
++) {
290 struct page
*p
= page
+ i
;
293 p
->first_page
= page
;
297 #ifdef CONFIG_HUGETLBFS
298 void prep_compound_gigantic_page(struct page
*page
, unsigned long order
)
301 int nr_pages
= 1 << order
;
302 struct page
*p
= page
+ 1;
304 set_compound_page_dtor(page
, free_compound_page
);
305 set_compound_order(page
, order
);
307 for (i
= 1; i
< nr_pages
; i
++, p
= mem_map_next(p
, page
, i
)) {
309 p
->first_page
= page
;
314 static int destroy_compound_page(struct page
*page
, unsigned long order
)
317 int nr_pages
= 1 << order
;
320 if (unlikely(compound_order(page
) != order
) ||
321 unlikely(!PageHead(page
))) {
326 __ClearPageHead(page
);
328 for (i
= 1; i
< nr_pages
; i
++) {
329 struct page
*p
= page
+ i
;
331 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
341 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
346 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
347 * and __GFP_HIGHMEM from hard or soft interrupt context.
349 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
350 for (i
= 0; i
< (1 << order
); i
++)
351 clear_highpage(page
+ i
);
354 static inline void set_page_order(struct page
*page
, int order
)
356 set_page_private(page
, order
);
357 __SetPageBuddy(page
);
360 static inline void rmv_page_order(struct page
*page
)
362 __ClearPageBuddy(page
);
363 set_page_private(page
, 0);
367 * Locate the struct page for both the matching buddy in our
368 * pair (buddy1) and the combined O(n+1) page they form (page).
370 * 1) Any buddy B1 will have an order O twin B2 which satisfies
371 * the following equation:
373 * For example, if the starting buddy (buddy2) is #8 its order
375 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
377 * 2) Any buddy B will have an order O+1 parent P which
378 * satisfies the following equation:
381 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
383 static inline struct page
*
384 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
386 unsigned long buddy_idx
= page_idx
^ (1 << order
);
388 return page
+ (buddy_idx
- page_idx
);
391 static inline unsigned long
392 __find_combined_index(unsigned long page_idx
, unsigned int order
)
394 return (page_idx
& ~(1 << order
));
398 * This function checks whether a page is free && is the buddy
399 * we can do coalesce a page and its buddy if
400 * (a) the buddy is not in a hole &&
401 * (b) the buddy is in the buddy system &&
402 * (c) a page and its buddy have the same order &&
403 * (d) a page and its buddy are in the same zone.
405 * For recording whether a page is in the buddy system, we use PG_buddy.
406 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
408 * For recording page's order, we use page_private(page).
410 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
413 if (!pfn_valid_within(page_to_pfn(buddy
)))
416 if (page_zone_id(page
) != page_zone_id(buddy
))
419 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
420 BUG_ON(page_count(buddy
) != 0);
427 * Freeing function for a buddy system allocator.
429 * The concept of a buddy system is to maintain direct-mapped table
430 * (containing bit values) for memory blocks of various "orders".
431 * The bottom level table contains the map for the smallest allocatable
432 * units of memory (here, pages), and each level above it describes
433 * pairs of units from the levels below, hence, "buddies".
434 * At a high level, all that happens here is marking the table entry
435 * at the bottom level available, and propagating the changes upward
436 * as necessary, plus some accounting needed to play nicely with other
437 * parts of the VM system.
438 * At each level, we keep a list of pages, which are heads of continuous
439 * free pages of length of (1 << order) and marked with PG_buddy. Page's
440 * order is recorded in page_private(page) field.
441 * So when we are allocating or freeing one, we can derive the state of the
442 * other. That is, if we allocate a small block, and both were
443 * free, the remainder of the region must be split into blocks.
444 * If a block is freed, and its buddy is also free, then this
445 * triggers coalescing into a block of larger size.
450 static inline void __free_one_page(struct page
*page
,
451 struct zone
*zone
, unsigned int order
)
453 unsigned long page_idx
;
454 int order_size
= 1 << order
;
455 int migratetype
= get_pageblock_migratetype(page
);
457 if (unlikely(PageCompound(page
)))
458 if (unlikely(destroy_compound_page(page
, order
)))
461 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
463 VM_BUG_ON(page_idx
& (order_size
- 1));
464 VM_BUG_ON(bad_range(zone
, page
));
466 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
467 while (order
< MAX_ORDER
-1) {
468 unsigned long combined_idx
;
471 buddy
= __page_find_buddy(page
, page_idx
, order
);
472 if (!page_is_buddy(page
, buddy
, order
))
475 /* Our buddy is free, merge with it and move up one order. */
476 list_del(&buddy
->lru
);
477 zone
->free_area
[order
].nr_free
--;
478 rmv_page_order(buddy
);
479 combined_idx
= __find_combined_index(page_idx
, order
);
480 page
= page
+ (combined_idx
- page_idx
);
481 page_idx
= combined_idx
;
484 set_page_order(page
, order
);
486 &zone
->free_area
[order
].free_list
[migratetype
]);
487 zone
->free_area
[order
].nr_free
++;
490 static inline int free_pages_check(struct page
*page
)
492 free_page_mlock(page
);
493 if (unlikely(page_mapcount(page
) |
494 (page
->mapping
!= NULL
) |
495 (page_count(page
) != 0) |
496 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
500 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
501 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
506 * Frees a list of pages.
507 * Assumes all pages on list are in same zone, and of same order.
508 * count is the number of pages to free.
510 * If the zone was previously in an "all pages pinned" state then look to
511 * see if this freeing clears that state.
513 * And clear the zone's pages_scanned counter, to hold off the "all pages are
514 * pinned" detection logic.
516 static void free_pages_bulk(struct zone
*zone
, int count
,
517 struct list_head
*list
, int order
)
519 spin_lock(&zone
->lock
);
520 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
521 zone
->pages_scanned
= 0;
525 VM_BUG_ON(list_empty(list
));
526 page
= list_entry(list
->prev
, struct page
, lru
);
527 /* have to delete it as __free_one_page list manipulates */
528 list_del(&page
->lru
);
529 __free_one_page(page
, zone
, order
);
531 spin_unlock(&zone
->lock
);
534 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
536 spin_lock(&zone
->lock
);
537 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
538 zone
->pages_scanned
= 0;
539 __free_one_page(page
, zone
, order
);
540 spin_unlock(&zone
->lock
);
543 static void __free_pages_ok(struct page
*page
, unsigned int order
)
549 for (i
= 0 ; i
< (1 << order
) ; ++i
)
550 bad
+= free_pages_check(page
+ i
);
554 if (!PageHighMem(page
)) {
555 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
556 debug_check_no_obj_freed(page_address(page
),
559 arch_free_page(page
, order
);
560 kernel_map_pages(page
, 1 << order
, 0);
562 local_irq_save(flags
);
563 __count_vm_events(PGFREE
, 1 << order
);
564 free_one_page(page_zone(page
), page
, order
);
565 local_irq_restore(flags
);
569 * permit the bootmem allocator to evade page validation on high-order frees
571 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
574 __ClearPageReserved(page
);
575 set_page_count(page
, 0);
576 set_page_refcounted(page
);
582 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
583 struct page
*p
= &page
[loop
];
585 if (loop
+ 1 < BITS_PER_LONG
)
587 __ClearPageReserved(p
);
588 set_page_count(p
, 0);
591 set_page_refcounted(page
);
592 __free_pages(page
, order
);
598 * The order of subdivision here is critical for the IO subsystem.
599 * Please do not alter this order without good reasons and regression
600 * testing. Specifically, as large blocks of memory are subdivided,
601 * the order in which smaller blocks are delivered depends on the order
602 * they're subdivided in this function. This is the primary factor
603 * influencing the order in which pages are delivered to the IO
604 * subsystem according to empirical testing, and this is also justified
605 * by considering the behavior of a buddy system containing a single
606 * large block of memory acted on by a series of small allocations.
607 * This behavior is a critical factor in sglist merging's success.
611 static inline void expand(struct zone
*zone
, struct page
*page
,
612 int low
, int high
, struct free_area
*area
,
615 unsigned long size
= 1 << high
;
621 VM_BUG_ON(bad_range(zone
, &page
[size
]));
622 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
624 set_page_order(&page
[size
], high
);
629 * This page is about to be returned from the page allocator
631 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
633 if (unlikely(page_mapcount(page
) |
634 (page
->mapping
!= NULL
) |
635 (page_count(page
) != 0) |
636 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
641 set_page_private(page
, 0);
642 set_page_refcounted(page
);
644 arch_alloc_page(page
, order
);
645 kernel_map_pages(page
, 1 << order
, 1);
647 if (gfp_flags
& __GFP_ZERO
)
648 prep_zero_page(page
, order
, gfp_flags
);
650 if (order
&& (gfp_flags
& __GFP_COMP
))
651 prep_compound_page(page
, order
);
657 * Go through the free lists for the given migratetype and remove
658 * the smallest available page from the freelists
660 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
663 unsigned int current_order
;
664 struct free_area
* area
;
667 /* Find a page of the appropriate size in the preferred list */
668 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
669 area
= &(zone
->free_area
[current_order
]);
670 if (list_empty(&area
->free_list
[migratetype
]))
673 page
= list_entry(area
->free_list
[migratetype
].next
,
675 list_del(&page
->lru
);
676 rmv_page_order(page
);
678 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
679 expand(zone
, page
, order
, current_order
, area
, migratetype
);
688 * This array describes the order lists are fallen back to when
689 * the free lists for the desirable migrate type are depleted
691 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
692 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
693 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
694 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
695 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
699 * Move the free pages in a range to the free lists of the requested type.
700 * Note that start_page and end_pages are not aligned on a pageblock
701 * boundary. If alignment is required, use move_freepages_block()
703 static int move_freepages(struct zone
*zone
,
704 struct page
*start_page
, struct page
*end_page
,
711 #ifndef CONFIG_HOLES_IN_ZONE
713 * page_zone is not safe to call in this context when
714 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
715 * anyway as we check zone boundaries in move_freepages_block().
716 * Remove at a later date when no bug reports exist related to
717 * grouping pages by mobility
719 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
722 for (page
= start_page
; page
<= end_page
;) {
723 /* Make sure we are not inadvertently changing nodes */
724 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
726 if (!pfn_valid_within(page_to_pfn(page
))) {
731 if (!PageBuddy(page
)) {
736 order
= page_order(page
);
737 list_del(&page
->lru
);
739 &zone
->free_area
[order
].free_list
[migratetype
]);
741 pages_moved
+= 1 << order
;
747 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
750 unsigned long start_pfn
, end_pfn
;
751 struct page
*start_page
, *end_page
;
753 start_pfn
= page_to_pfn(page
);
754 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
755 start_page
= pfn_to_page(start_pfn
);
756 end_page
= start_page
+ pageblock_nr_pages
- 1;
757 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
759 /* Do not cross zone boundaries */
760 if (start_pfn
< zone
->zone_start_pfn
)
762 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
765 return move_freepages(zone
, start_page
, end_page
, migratetype
);
768 /* Remove an element from the buddy allocator from the fallback list */
769 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
770 int start_migratetype
)
772 struct free_area
* area
;
777 /* Find the largest possible block of pages in the other list */
778 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
780 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
781 migratetype
= fallbacks
[start_migratetype
][i
];
783 /* MIGRATE_RESERVE handled later if necessary */
784 if (migratetype
== MIGRATE_RESERVE
)
787 area
= &(zone
->free_area
[current_order
]);
788 if (list_empty(&area
->free_list
[migratetype
]))
791 page
= list_entry(area
->free_list
[migratetype
].next
,
796 * If breaking a large block of pages, move all free
797 * pages to the preferred allocation list. If falling
798 * back for a reclaimable kernel allocation, be more
799 * agressive about taking ownership of free pages
801 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
802 start_migratetype
== MIGRATE_RECLAIMABLE
) {
804 pages
= move_freepages_block(zone
, page
,
807 /* Claim the whole block if over half of it is free */
808 if (pages
>= (1 << (pageblock_order
-1)))
809 set_pageblock_migratetype(page
,
812 migratetype
= start_migratetype
;
815 /* Remove the page from the freelists */
816 list_del(&page
->lru
);
817 rmv_page_order(page
);
818 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
821 if (current_order
== pageblock_order
)
822 set_pageblock_migratetype(page
,
825 expand(zone
, page
, order
, current_order
, area
, migratetype
);
830 /* Use MIGRATE_RESERVE rather than fail an allocation */
831 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
835 * Do the hard work of removing an element from the buddy allocator.
836 * Call me with the zone->lock already held.
838 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
843 page
= __rmqueue_smallest(zone
, order
, migratetype
);
846 page
= __rmqueue_fallback(zone
, order
, migratetype
);
852 * Obtain a specified number of elements from the buddy allocator, all under
853 * a single hold of the lock, for efficiency. Add them to the supplied list.
854 * Returns the number of new pages which were placed at *list.
856 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
857 unsigned long count
, struct list_head
*list
,
862 spin_lock(&zone
->lock
);
863 for (i
= 0; i
< count
; ++i
) {
864 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
865 if (unlikely(page
== NULL
))
869 * Split buddy pages returned by expand() are received here
870 * in physical page order. The page is added to the callers and
871 * list and the list head then moves forward. From the callers
872 * perspective, the linked list is ordered by page number in
873 * some conditions. This is useful for IO devices that can
874 * merge IO requests if the physical pages are ordered
877 list_add(&page
->lru
, list
);
878 set_page_private(page
, migratetype
);
881 spin_unlock(&zone
->lock
);
887 * Called from the vmstat counter updater to drain pagesets of this
888 * currently executing processor on remote nodes after they have
891 * Note that this function must be called with the thread pinned to
892 * a single processor.
894 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
899 local_irq_save(flags
);
900 if (pcp
->count
>= pcp
->batch
)
901 to_drain
= pcp
->batch
;
903 to_drain
= pcp
->count
;
904 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
905 pcp
->count
-= to_drain
;
906 local_irq_restore(flags
);
911 * Drain pages of the indicated processor.
913 * The processor must either be the current processor and the
914 * thread pinned to the current processor or a processor that
917 static void drain_pages(unsigned int cpu
)
922 for_each_populated_zone(zone
) {
923 struct per_cpu_pageset
*pset
;
924 struct per_cpu_pages
*pcp
;
926 pset
= zone_pcp(zone
, cpu
);
929 local_irq_save(flags
);
930 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
932 local_irq_restore(flags
);
937 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
939 void drain_local_pages(void *arg
)
941 drain_pages(smp_processor_id());
945 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
947 void drain_all_pages(void)
949 on_each_cpu(drain_local_pages
, NULL
, 1);
952 #ifdef CONFIG_HIBERNATION
954 void mark_free_pages(struct zone
*zone
)
956 unsigned long pfn
, max_zone_pfn
;
959 struct list_head
*curr
;
961 if (!zone
->spanned_pages
)
964 spin_lock_irqsave(&zone
->lock
, flags
);
966 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
967 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
968 if (pfn_valid(pfn
)) {
969 struct page
*page
= pfn_to_page(pfn
);
971 if (!swsusp_page_is_forbidden(page
))
972 swsusp_unset_page_free(page
);
975 for_each_migratetype_order(order
, t
) {
976 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
979 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
980 for (i
= 0; i
< (1UL << order
); i
++)
981 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
984 spin_unlock_irqrestore(&zone
->lock
, flags
);
986 #endif /* CONFIG_PM */
989 * Free a 0-order page
991 static void free_hot_cold_page(struct page
*page
, int cold
)
993 struct zone
*zone
= page_zone(page
);
994 struct per_cpu_pages
*pcp
;
998 page
->mapping
= NULL
;
999 if (free_pages_check(page
))
1002 if (!PageHighMem(page
)) {
1003 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1004 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1006 arch_free_page(page
, 0);
1007 kernel_map_pages(page
, 1, 0);
1009 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1010 local_irq_save(flags
);
1011 __count_vm_event(PGFREE
);
1013 list_add_tail(&page
->lru
, &pcp
->list
);
1015 list_add(&page
->lru
, &pcp
->list
);
1016 set_page_private(page
, get_pageblock_migratetype(page
));
1018 if (pcp
->count
>= pcp
->high
) {
1019 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1020 pcp
->count
-= pcp
->batch
;
1022 local_irq_restore(flags
);
1026 void free_hot_page(struct page
*page
)
1028 free_hot_cold_page(page
, 0);
1031 void free_cold_page(struct page
*page
)
1033 free_hot_cold_page(page
, 1);
1037 * split_page takes a non-compound higher-order page, and splits it into
1038 * n (1<<order) sub-pages: page[0..n]
1039 * Each sub-page must be freed individually.
1041 * Note: this is probably too low level an operation for use in drivers.
1042 * Please consult with lkml before using this in your driver.
1044 void split_page(struct page
*page
, unsigned int order
)
1048 VM_BUG_ON(PageCompound(page
));
1049 VM_BUG_ON(!page_count(page
));
1050 for (i
= 1; i
< (1 << order
); i
++)
1051 set_page_refcounted(page
+ i
);
1055 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1056 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1059 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1060 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1062 unsigned long flags
;
1064 int cold
= !!(gfp_flags
& __GFP_COLD
);
1066 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1070 if (likely(order
== 0)) {
1071 struct per_cpu_pages
*pcp
;
1073 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1074 local_irq_save(flags
);
1076 pcp
->count
= rmqueue_bulk(zone
, 0,
1077 pcp
->batch
, &pcp
->list
, migratetype
);
1078 if (unlikely(!pcp
->count
))
1082 /* Find a page of the appropriate migrate type */
1084 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1085 if (page_private(page
) == migratetype
)
1088 list_for_each_entry(page
, &pcp
->list
, lru
)
1089 if (page_private(page
) == migratetype
)
1093 /* Allocate more to the pcp list if necessary */
1094 if (unlikely(&page
->lru
== &pcp
->list
)) {
1095 pcp
->count
+= rmqueue_bulk(zone
, 0,
1096 pcp
->batch
, &pcp
->list
, migratetype
);
1097 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1100 list_del(&page
->lru
);
1103 spin_lock_irqsave(&zone
->lock
, flags
);
1104 page
= __rmqueue(zone
, order
, migratetype
);
1105 spin_unlock(&zone
->lock
);
1110 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1111 zone_statistics(preferred_zone
, zone
);
1112 local_irq_restore(flags
);
1115 VM_BUG_ON(bad_range(zone
, page
));
1116 if (prep_new_page(page
, order
, gfp_flags
))
1121 local_irq_restore(flags
);
1126 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1127 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1128 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1129 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1130 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1131 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1132 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1134 #ifdef CONFIG_FAIL_PAGE_ALLOC
1136 static struct fail_page_alloc_attr
{
1137 struct fault_attr attr
;
1139 u32 ignore_gfp_highmem
;
1140 u32 ignore_gfp_wait
;
1143 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1145 struct dentry
*ignore_gfp_highmem_file
;
1146 struct dentry
*ignore_gfp_wait_file
;
1147 struct dentry
*min_order_file
;
1149 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1151 } fail_page_alloc
= {
1152 .attr
= FAULT_ATTR_INITIALIZER
,
1153 .ignore_gfp_wait
= 1,
1154 .ignore_gfp_highmem
= 1,
1158 static int __init
setup_fail_page_alloc(char *str
)
1160 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1162 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1164 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1166 if (order
< fail_page_alloc
.min_order
)
1168 if (gfp_mask
& __GFP_NOFAIL
)
1170 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1172 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1175 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1178 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1180 static int __init
fail_page_alloc_debugfs(void)
1182 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1186 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1190 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1192 fail_page_alloc
.ignore_gfp_wait_file
=
1193 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1194 &fail_page_alloc
.ignore_gfp_wait
);
1196 fail_page_alloc
.ignore_gfp_highmem_file
=
1197 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1198 &fail_page_alloc
.ignore_gfp_highmem
);
1199 fail_page_alloc
.min_order_file
=
1200 debugfs_create_u32("min-order", mode
, dir
,
1201 &fail_page_alloc
.min_order
);
1203 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1204 !fail_page_alloc
.ignore_gfp_highmem_file
||
1205 !fail_page_alloc
.min_order_file
) {
1207 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1208 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1209 debugfs_remove(fail_page_alloc
.min_order_file
);
1210 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1216 late_initcall(fail_page_alloc_debugfs
);
1218 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1220 #else /* CONFIG_FAIL_PAGE_ALLOC */
1222 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1227 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1230 * Return 1 if free pages are above 'mark'. This takes into account the order
1231 * of the allocation.
1233 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1234 int classzone_idx
, int alloc_flags
)
1236 /* free_pages my go negative - that's OK */
1238 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1241 if (alloc_flags
& ALLOC_HIGH
)
1243 if (alloc_flags
& ALLOC_HARDER
)
1246 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1248 for (o
= 0; o
< order
; o
++) {
1249 /* At the next order, this order's pages become unavailable */
1250 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1252 /* Require fewer higher order pages to be free */
1255 if (free_pages
<= min
)
1263 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1264 * skip over zones that are not allowed by the cpuset, or that have
1265 * been recently (in last second) found to be nearly full. See further
1266 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1267 * that have to skip over a lot of full or unallowed zones.
1269 * If the zonelist cache is present in the passed in zonelist, then
1270 * returns a pointer to the allowed node mask (either the current
1271 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1273 * If the zonelist cache is not available for this zonelist, does
1274 * nothing and returns NULL.
1276 * If the fullzones BITMAP in the zonelist cache is stale (more than
1277 * a second since last zap'd) then we zap it out (clear its bits.)
1279 * We hold off even calling zlc_setup, until after we've checked the
1280 * first zone in the zonelist, on the theory that most allocations will
1281 * be satisfied from that first zone, so best to examine that zone as
1282 * quickly as we can.
1284 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1286 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1287 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1289 zlc
= zonelist
->zlcache_ptr
;
1293 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1294 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1295 zlc
->last_full_zap
= jiffies
;
1298 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1299 &cpuset_current_mems_allowed
:
1300 &node_states
[N_HIGH_MEMORY
];
1301 return allowednodes
;
1305 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1306 * if it is worth looking at further for free memory:
1307 * 1) Check that the zone isn't thought to be full (doesn't have its
1308 * bit set in the zonelist_cache fullzones BITMAP).
1309 * 2) Check that the zones node (obtained from the zonelist_cache
1310 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1311 * Return true (non-zero) if zone is worth looking at further, or
1312 * else return false (zero) if it is not.
1314 * This check -ignores- the distinction between various watermarks,
1315 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1316 * found to be full for any variation of these watermarks, it will
1317 * be considered full for up to one second by all requests, unless
1318 * we are so low on memory on all allowed nodes that we are forced
1319 * into the second scan of the zonelist.
1321 * In the second scan we ignore this zonelist cache and exactly
1322 * apply the watermarks to all zones, even it is slower to do so.
1323 * We are low on memory in the second scan, and should leave no stone
1324 * unturned looking for a free page.
1326 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1327 nodemask_t
*allowednodes
)
1329 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1330 int i
; /* index of *z in zonelist zones */
1331 int n
; /* node that zone *z is on */
1333 zlc
= zonelist
->zlcache_ptr
;
1337 i
= z
- zonelist
->_zonerefs
;
1340 /* This zone is worth trying if it is allowed but not full */
1341 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1345 * Given 'z' scanning a zonelist, set the corresponding bit in
1346 * zlc->fullzones, so that subsequent attempts to allocate a page
1347 * from that zone don't waste time re-examining it.
1349 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1351 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1352 int i
; /* index of *z in zonelist zones */
1354 zlc
= zonelist
->zlcache_ptr
;
1358 i
= z
- zonelist
->_zonerefs
;
1360 set_bit(i
, zlc
->fullzones
);
1363 #else /* CONFIG_NUMA */
1365 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1370 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1371 nodemask_t
*allowednodes
)
1376 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1379 #endif /* CONFIG_NUMA */
1382 * get_page_from_freelist goes through the zonelist trying to allocate
1385 static struct page
*
1386 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1387 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1390 struct page
*page
= NULL
;
1392 struct zone
*zone
, *preferred_zone
;
1393 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1394 int zlc_active
= 0; /* set if using zonelist_cache */
1395 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1397 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1399 if (!preferred_zone
)
1402 classzone_idx
= zone_idx(preferred_zone
);
1406 * Scan zonelist, looking for a zone with enough free.
1407 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1409 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1410 high_zoneidx
, nodemask
) {
1411 if (NUMA_BUILD
&& zlc_active
&&
1412 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1414 if ((alloc_flags
& ALLOC_CPUSET
) &&
1415 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1418 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1420 if (alloc_flags
& ALLOC_WMARK_MIN
)
1421 mark
= zone
->pages_min
;
1422 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1423 mark
= zone
->pages_low
;
1425 mark
= zone
->pages_high
;
1426 if (!zone_watermark_ok(zone
, order
, mark
,
1427 classzone_idx
, alloc_flags
)) {
1428 if (!zone_reclaim_mode
||
1429 !zone_reclaim(zone
, gfp_mask
, order
))
1430 goto this_zone_full
;
1434 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1439 zlc_mark_zone_full(zonelist
, z
);
1441 if (NUMA_BUILD
&& !did_zlc_setup
) {
1442 /* we do zlc_setup after the first zone is tried */
1443 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1449 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1450 /* Disable zlc cache for second zonelist scan */
1458 * This is the 'heart' of the zoned buddy allocator.
1461 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
1462 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1464 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1465 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1469 struct reclaim_state reclaim_state
;
1470 struct task_struct
*p
= current
;
1473 unsigned long did_some_progress
;
1474 unsigned long pages_reclaimed
= 0;
1476 lockdep_trace_alloc(gfp_mask
);
1478 might_sleep_if(wait
);
1480 if (should_fail_alloc_page(gfp_mask
, order
))
1484 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1486 if (unlikely(!z
->zone
)) {
1488 * Happens if we have an empty zonelist as a result of
1489 * GFP_THISNODE being used on a memoryless node
1494 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1495 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1500 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1501 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1502 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1503 * using a larger set of nodes after it has established that the
1504 * allowed per node queues are empty and that nodes are
1507 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1510 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1511 wakeup_kswapd(zone
, order
);
1514 * OK, we're below the kswapd watermark and have kicked background
1515 * reclaim. Now things get more complex, so set up alloc_flags according
1516 * to how we want to proceed.
1518 * The caller may dip into page reserves a bit more if the caller
1519 * cannot run direct reclaim, or if the caller has realtime scheduling
1520 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1521 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1523 alloc_flags
= ALLOC_WMARK_MIN
;
1524 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1525 alloc_flags
|= ALLOC_HARDER
;
1526 if (gfp_mask
& __GFP_HIGH
)
1527 alloc_flags
|= ALLOC_HIGH
;
1529 alloc_flags
|= ALLOC_CPUSET
;
1532 * Go through the zonelist again. Let __GFP_HIGH and allocations
1533 * coming from realtime tasks go deeper into reserves.
1535 * This is the last chance, in general, before the goto nopage.
1536 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1537 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1539 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1540 high_zoneidx
, alloc_flags
);
1544 /* This allocation should allow future memory freeing. */
1547 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1548 && !in_interrupt()) {
1549 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1551 /* go through the zonelist yet again, ignoring mins */
1552 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1553 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1556 if (gfp_mask
& __GFP_NOFAIL
) {
1557 congestion_wait(WRITE
, HZ
/50);
1564 /* Atomic allocations - we can't balance anything */
1570 /* We now go into synchronous reclaim */
1571 cpuset_memory_pressure_bump();
1573 p
->flags
|= PF_MEMALLOC
;
1575 lockdep_set_current_reclaim_state(gfp_mask
);
1576 reclaim_state
.reclaimed_slab
= 0;
1577 p
->reclaim_state
= &reclaim_state
;
1579 did_some_progress
= try_to_free_pages(zonelist
, order
,
1580 gfp_mask
, nodemask
);
1582 p
->reclaim_state
= NULL
;
1583 lockdep_clear_current_reclaim_state();
1584 p
->flags
&= ~PF_MEMALLOC
;
1591 if (likely(did_some_progress
)) {
1592 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1593 zonelist
, high_zoneidx
, alloc_flags
);
1596 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1597 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1598 schedule_timeout_uninterruptible(1);
1603 * Go through the zonelist yet one more time, keep
1604 * very high watermark here, this is only to catch
1605 * a parallel oom killing, we must fail if we're still
1606 * under heavy pressure.
1608 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1609 order
, zonelist
, high_zoneidx
,
1610 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1612 clear_zonelist_oom(zonelist
, gfp_mask
);
1616 /* The OOM killer will not help higher order allocs so fail */
1617 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1618 clear_zonelist_oom(zonelist
, gfp_mask
);
1622 out_of_memory(zonelist
, gfp_mask
, order
);
1623 clear_zonelist_oom(zonelist
, gfp_mask
);
1628 * Don't let big-order allocations loop unless the caller explicitly
1629 * requests that. Wait for some write requests to complete then retry.
1631 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1632 * means __GFP_NOFAIL, but that may not be true in other
1635 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1636 * specified, then we retry until we no longer reclaim any pages
1637 * (above), or we've reclaimed an order of pages at least as
1638 * large as the allocation's order. In both cases, if the
1639 * allocation still fails, we stop retrying.
1641 pages_reclaimed
+= did_some_progress
;
1643 if (!(gfp_mask
& __GFP_NORETRY
)) {
1644 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1647 if (gfp_mask
& __GFP_REPEAT
&&
1648 pages_reclaimed
< (1 << order
))
1651 if (gfp_mask
& __GFP_NOFAIL
)
1655 congestion_wait(WRITE
, HZ
/50);
1660 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1661 printk(KERN_WARNING
"%s: page allocation failure."
1662 " order:%d, mode:0x%x\n",
1663 p
->comm
, order
, gfp_mask
);
1670 EXPORT_SYMBOL(__alloc_pages_nodemask
);
1673 * Common helper functions.
1675 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1678 page
= alloc_pages(gfp_mask
, order
);
1681 return (unsigned long) page_address(page
);
1684 EXPORT_SYMBOL(__get_free_pages
);
1686 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1691 * get_zeroed_page() returns a 32-bit address, which cannot represent
1694 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1696 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1698 return (unsigned long) page_address(page
);
1702 EXPORT_SYMBOL(get_zeroed_page
);
1704 void __pagevec_free(struct pagevec
*pvec
)
1706 int i
= pagevec_count(pvec
);
1709 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1712 void __free_pages(struct page
*page
, unsigned int order
)
1714 if (put_page_testzero(page
)) {
1716 free_hot_page(page
);
1718 __free_pages_ok(page
, order
);
1722 EXPORT_SYMBOL(__free_pages
);
1724 void free_pages(unsigned long addr
, unsigned int order
)
1727 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1728 __free_pages(virt_to_page((void *)addr
), order
);
1732 EXPORT_SYMBOL(free_pages
);
1735 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1736 * @size: the number of bytes to allocate
1737 * @gfp_mask: GFP flags for the allocation
1739 * This function is similar to alloc_pages(), except that it allocates the
1740 * minimum number of pages to satisfy the request. alloc_pages() can only
1741 * allocate memory in power-of-two pages.
1743 * This function is also limited by MAX_ORDER.
1745 * Memory allocated by this function must be released by free_pages_exact().
1747 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1749 unsigned int order
= get_order(size
);
1752 addr
= __get_free_pages(gfp_mask
, order
);
1754 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1755 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1757 split_page(virt_to_page(addr
), order
);
1758 while (used
< alloc_end
) {
1764 return (void *)addr
;
1766 EXPORT_SYMBOL(alloc_pages_exact
);
1769 * free_pages_exact - release memory allocated via alloc_pages_exact()
1770 * @virt: the value returned by alloc_pages_exact.
1771 * @size: size of allocation, same value as passed to alloc_pages_exact().
1773 * Release the memory allocated by a previous call to alloc_pages_exact.
1775 void free_pages_exact(void *virt
, size_t size
)
1777 unsigned long addr
= (unsigned long)virt
;
1778 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1780 while (addr
< end
) {
1785 EXPORT_SYMBOL(free_pages_exact
);
1787 static unsigned int nr_free_zone_pages(int offset
)
1792 /* Just pick one node, since fallback list is circular */
1793 unsigned int sum
= 0;
1795 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1797 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1798 unsigned long size
= zone
->present_pages
;
1799 unsigned long high
= zone
->pages_high
;
1808 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1810 unsigned int nr_free_buffer_pages(void)
1812 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1814 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1817 * Amount of free RAM allocatable within all zones
1819 unsigned int nr_free_pagecache_pages(void)
1821 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1824 static inline void show_node(struct zone
*zone
)
1827 printk("Node %d ", zone_to_nid(zone
));
1830 void si_meminfo(struct sysinfo
*val
)
1832 val
->totalram
= totalram_pages
;
1834 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1835 val
->bufferram
= nr_blockdev_pages();
1836 val
->totalhigh
= totalhigh_pages
;
1837 val
->freehigh
= nr_free_highpages();
1838 val
->mem_unit
= PAGE_SIZE
;
1841 EXPORT_SYMBOL(si_meminfo
);
1844 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1846 pg_data_t
*pgdat
= NODE_DATA(nid
);
1848 val
->totalram
= pgdat
->node_present_pages
;
1849 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1850 #ifdef CONFIG_HIGHMEM
1851 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1852 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1858 val
->mem_unit
= PAGE_SIZE
;
1862 #define K(x) ((x) << (PAGE_SHIFT-10))
1865 * Show free area list (used inside shift_scroll-lock stuff)
1866 * We also calculate the percentage fragmentation. We do this by counting the
1867 * memory on each free list with the exception of the first item on the list.
1869 void show_free_areas(void)
1874 for_each_populated_zone(zone
) {
1876 printk("%s per-cpu:\n", zone
->name
);
1878 for_each_online_cpu(cpu
) {
1879 struct per_cpu_pageset
*pageset
;
1881 pageset
= zone_pcp(zone
, cpu
);
1883 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1884 cpu
, pageset
->pcp
.high
,
1885 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1889 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1890 " inactive_file:%lu"
1891 //TODO: check/adjust line lengths
1892 #ifdef CONFIG_UNEVICTABLE_LRU
1895 " dirty:%lu writeback:%lu unstable:%lu\n"
1896 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1897 global_page_state(NR_ACTIVE_ANON
),
1898 global_page_state(NR_ACTIVE_FILE
),
1899 global_page_state(NR_INACTIVE_ANON
),
1900 global_page_state(NR_INACTIVE_FILE
),
1901 #ifdef CONFIG_UNEVICTABLE_LRU
1902 global_page_state(NR_UNEVICTABLE
),
1904 global_page_state(NR_FILE_DIRTY
),
1905 global_page_state(NR_WRITEBACK
),
1906 global_page_state(NR_UNSTABLE_NFS
),
1907 global_page_state(NR_FREE_PAGES
),
1908 global_page_state(NR_SLAB_RECLAIMABLE
) +
1909 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1910 global_page_state(NR_FILE_MAPPED
),
1911 global_page_state(NR_PAGETABLE
),
1912 global_page_state(NR_BOUNCE
));
1914 for_each_populated_zone(zone
) {
1923 " active_anon:%lukB"
1924 " inactive_anon:%lukB"
1925 " active_file:%lukB"
1926 " inactive_file:%lukB"
1927 #ifdef CONFIG_UNEVICTABLE_LRU
1928 " unevictable:%lukB"
1931 " pages_scanned:%lu"
1932 " all_unreclaimable? %s"
1935 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1938 K(zone
->pages_high
),
1939 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
1940 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
1941 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
1942 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
1943 #ifdef CONFIG_UNEVICTABLE_LRU
1944 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
1946 K(zone
->present_pages
),
1947 zone
->pages_scanned
,
1948 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1950 printk("lowmem_reserve[]:");
1951 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1952 printk(" %lu", zone
->lowmem_reserve
[i
]);
1956 for_each_populated_zone(zone
) {
1957 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1960 printk("%s: ", zone
->name
);
1962 spin_lock_irqsave(&zone
->lock
, flags
);
1963 for (order
= 0; order
< MAX_ORDER
; order
++) {
1964 nr
[order
] = zone
->free_area
[order
].nr_free
;
1965 total
+= nr
[order
] << order
;
1967 spin_unlock_irqrestore(&zone
->lock
, flags
);
1968 for (order
= 0; order
< MAX_ORDER
; order
++)
1969 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1970 printk("= %lukB\n", K(total
));
1973 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
1975 show_swap_cache_info();
1978 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
1980 zoneref
->zone
= zone
;
1981 zoneref
->zone_idx
= zone_idx(zone
);
1985 * Builds allocation fallback zone lists.
1987 * Add all populated zones of a node to the zonelist.
1989 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1990 int nr_zones
, enum zone_type zone_type
)
1994 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1999 zone
= pgdat
->node_zones
+ zone_type
;
2000 if (populated_zone(zone
)) {
2001 zoneref_set_zone(zone
,
2002 &zonelist
->_zonerefs
[nr_zones
++]);
2003 check_highest_zone(zone_type
);
2006 } while (zone_type
);
2013 * 0 = automatic detection of better ordering.
2014 * 1 = order by ([node] distance, -zonetype)
2015 * 2 = order by (-zonetype, [node] distance)
2017 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2018 * the same zonelist. So only NUMA can configure this param.
2020 #define ZONELIST_ORDER_DEFAULT 0
2021 #define ZONELIST_ORDER_NODE 1
2022 #define ZONELIST_ORDER_ZONE 2
2024 /* zonelist order in the kernel.
2025 * set_zonelist_order() will set this to NODE or ZONE.
2027 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2028 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2032 /* The value user specified ....changed by config */
2033 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2034 /* string for sysctl */
2035 #define NUMA_ZONELIST_ORDER_LEN 16
2036 char numa_zonelist_order
[16] = "default";
2039 * interface for configure zonelist ordering.
2040 * command line option "numa_zonelist_order"
2041 * = "[dD]efault - default, automatic configuration.
2042 * = "[nN]ode - order by node locality, then by zone within node
2043 * = "[zZ]one - order by zone, then by locality within zone
2046 static int __parse_numa_zonelist_order(char *s
)
2048 if (*s
== 'd' || *s
== 'D') {
2049 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2050 } else if (*s
== 'n' || *s
== 'N') {
2051 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2052 } else if (*s
== 'z' || *s
== 'Z') {
2053 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2056 "Ignoring invalid numa_zonelist_order value: "
2063 static __init
int setup_numa_zonelist_order(char *s
)
2066 return __parse_numa_zonelist_order(s
);
2069 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2072 * sysctl handler for numa_zonelist_order
2074 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2075 struct file
*file
, void __user
*buffer
, size_t *length
,
2078 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2082 strncpy(saved_string
, (char*)table
->data
,
2083 NUMA_ZONELIST_ORDER_LEN
);
2084 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2088 int oldval
= user_zonelist_order
;
2089 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2091 * bogus value. restore saved string
2093 strncpy((char*)table
->data
, saved_string
,
2094 NUMA_ZONELIST_ORDER_LEN
);
2095 user_zonelist_order
= oldval
;
2096 } else if (oldval
!= user_zonelist_order
)
2097 build_all_zonelists();
2103 #define MAX_NODE_LOAD (num_online_nodes())
2104 static int node_load
[MAX_NUMNODES
];
2107 * find_next_best_node - find the next node that should appear in a given node's fallback list
2108 * @node: node whose fallback list we're appending
2109 * @used_node_mask: nodemask_t of already used nodes
2111 * We use a number of factors to determine which is the next node that should
2112 * appear on a given node's fallback list. The node should not have appeared
2113 * already in @node's fallback list, and it should be the next closest node
2114 * according to the distance array (which contains arbitrary distance values
2115 * from each node to each node in the system), and should also prefer nodes
2116 * with no CPUs, since presumably they'll have very little allocation pressure
2117 * on them otherwise.
2118 * It returns -1 if no node is found.
2120 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2123 int min_val
= INT_MAX
;
2125 const struct cpumask
*tmp
= cpumask_of_node(0);
2127 /* Use the local node if we haven't already */
2128 if (!node_isset(node
, *used_node_mask
)) {
2129 node_set(node
, *used_node_mask
);
2133 for_each_node_state(n
, N_HIGH_MEMORY
) {
2135 /* Don't want a node to appear more than once */
2136 if (node_isset(n
, *used_node_mask
))
2139 /* Use the distance array to find the distance */
2140 val
= node_distance(node
, n
);
2142 /* Penalize nodes under us ("prefer the next node") */
2145 /* Give preference to headless and unused nodes */
2146 tmp
= cpumask_of_node(n
);
2147 if (!cpumask_empty(tmp
))
2148 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2150 /* Slight preference for less loaded node */
2151 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2152 val
+= node_load
[n
];
2154 if (val
< min_val
) {
2161 node_set(best_node
, *used_node_mask
);
2168 * Build zonelists ordered by node and zones within node.
2169 * This results in maximum locality--normal zone overflows into local
2170 * DMA zone, if any--but risks exhausting DMA zone.
2172 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2175 struct zonelist
*zonelist
;
2177 zonelist
= &pgdat
->node_zonelists
[0];
2178 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2180 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2182 zonelist
->_zonerefs
[j
].zone
= NULL
;
2183 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2187 * Build gfp_thisnode zonelists
2189 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2192 struct zonelist
*zonelist
;
2194 zonelist
= &pgdat
->node_zonelists
[1];
2195 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2196 zonelist
->_zonerefs
[j
].zone
= NULL
;
2197 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2201 * Build zonelists ordered by zone and nodes within zones.
2202 * This results in conserving DMA zone[s] until all Normal memory is
2203 * exhausted, but results in overflowing to remote node while memory
2204 * may still exist in local DMA zone.
2206 static int node_order
[MAX_NUMNODES
];
2208 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2211 int zone_type
; /* needs to be signed */
2213 struct zonelist
*zonelist
;
2215 zonelist
= &pgdat
->node_zonelists
[0];
2217 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2218 for (j
= 0; j
< nr_nodes
; j
++) {
2219 node
= node_order
[j
];
2220 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2221 if (populated_zone(z
)) {
2223 &zonelist
->_zonerefs
[pos
++]);
2224 check_highest_zone(zone_type
);
2228 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2229 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2232 static int default_zonelist_order(void)
2235 unsigned long low_kmem_size
,total_size
;
2239 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2240 * If they are really small and used heavily, the system can fall
2241 * into OOM very easily.
2242 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2244 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2247 for_each_online_node(nid
) {
2248 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2249 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2250 if (populated_zone(z
)) {
2251 if (zone_type
< ZONE_NORMAL
)
2252 low_kmem_size
+= z
->present_pages
;
2253 total_size
+= z
->present_pages
;
2257 if (!low_kmem_size
|| /* there are no DMA area. */
2258 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2259 return ZONELIST_ORDER_NODE
;
2261 * look into each node's config.
2262 * If there is a node whose DMA/DMA32 memory is very big area on
2263 * local memory, NODE_ORDER may be suitable.
2265 average_size
= total_size
/
2266 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2267 for_each_online_node(nid
) {
2270 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2271 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2272 if (populated_zone(z
)) {
2273 if (zone_type
< ZONE_NORMAL
)
2274 low_kmem_size
+= z
->present_pages
;
2275 total_size
+= z
->present_pages
;
2278 if (low_kmem_size
&&
2279 total_size
> average_size
&& /* ignore small node */
2280 low_kmem_size
> total_size
* 70/100)
2281 return ZONELIST_ORDER_NODE
;
2283 return ZONELIST_ORDER_ZONE
;
2286 static void set_zonelist_order(void)
2288 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2289 current_zonelist_order
= default_zonelist_order();
2291 current_zonelist_order
= user_zonelist_order
;
2294 static void build_zonelists(pg_data_t
*pgdat
)
2298 nodemask_t used_mask
;
2299 int local_node
, prev_node
;
2300 struct zonelist
*zonelist
;
2301 int order
= current_zonelist_order
;
2303 /* initialize zonelists */
2304 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2305 zonelist
= pgdat
->node_zonelists
+ i
;
2306 zonelist
->_zonerefs
[0].zone
= NULL
;
2307 zonelist
->_zonerefs
[0].zone_idx
= 0;
2310 /* NUMA-aware ordering of nodes */
2311 local_node
= pgdat
->node_id
;
2312 load
= num_online_nodes();
2313 prev_node
= local_node
;
2314 nodes_clear(used_mask
);
2316 memset(node_load
, 0, sizeof(node_load
));
2317 memset(node_order
, 0, sizeof(node_order
));
2320 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2321 int distance
= node_distance(local_node
, node
);
2324 * If another node is sufficiently far away then it is better
2325 * to reclaim pages in a zone before going off node.
2327 if (distance
> RECLAIM_DISTANCE
)
2328 zone_reclaim_mode
= 1;
2331 * We don't want to pressure a particular node.
2332 * So adding penalty to the first node in same
2333 * distance group to make it round-robin.
2335 if (distance
!= node_distance(local_node
, prev_node
))
2336 node_load
[node
] = load
;
2340 if (order
== ZONELIST_ORDER_NODE
)
2341 build_zonelists_in_node_order(pgdat
, node
);
2343 node_order
[j
++] = node
; /* remember order */
2346 if (order
== ZONELIST_ORDER_ZONE
) {
2347 /* calculate node order -- i.e., DMA last! */
2348 build_zonelists_in_zone_order(pgdat
, j
);
2351 build_thisnode_zonelists(pgdat
);
2354 /* Construct the zonelist performance cache - see further mmzone.h */
2355 static void build_zonelist_cache(pg_data_t
*pgdat
)
2357 struct zonelist
*zonelist
;
2358 struct zonelist_cache
*zlc
;
2361 zonelist
= &pgdat
->node_zonelists
[0];
2362 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2363 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2364 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2365 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2369 #else /* CONFIG_NUMA */
2371 static void set_zonelist_order(void)
2373 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2376 static void build_zonelists(pg_data_t
*pgdat
)
2378 int node
, local_node
;
2380 struct zonelist
*zonelist
;
2382 local_node
= pgdat
->node_id
;
2384 zonelist
= &pgdat
->node_zonelists
[0];
2385 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2388 * Now we build the zonelist so that it contains the zones
2389 * of all the other nodes.
2390 * We don't want to pressure a particular node, so when
2391 * building the zones for node N, we make sure that the
2392 * zones coming right after the local ones are those from
2393 * node N+1 (modulo N)
2395 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2396 if (!node_online(node
))
2398 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2401 for (node
= 0; node
< local_node
; node
++) {
2402 if (!node_online(node
))
2404 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2408 zonelist
->_zonerefs
[j
].zone
= NULL
;
2409 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2412 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2413 static void build_zonelist_cache(pg_data_t
*pgdat
)
2415 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2418 #endif /* CONFIG_NUMA */
2420 /* return values int ....just for stop_machine() */
2421 static int __build_all_zonelists(void *dummy
)
2425 for_each_online_node(nid
) {
2426 pg_data_t
*pgdat
= NODE_DATA(nid
);
2428 build_zonelists(pgdat
);
2429 build_zonelist_cache(pgdat
);
2434 void build_all_zonelists(void)
2436 set_zonelist_order();
2438 if (system_state
== SYSTEM_BOOTING
) {
2439 __build_all_zonelists(NULL
);
2440 mminit_verify_zonelist();
2441 cpuset_init_current_mems_allowed();
2443 /* we have to stop all cpus to guarantee there is no user
2445 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2446 /* cpuset refresh routine should be here */
2448 vm_total_pages
= nr_free_pagecache_pages();
2450 * Disable grouping by mobility if the number of pages in the
2451 * system is too low to allow the mechanism to work. It would be
2452 * more accurate, but expensive to check per-zone. This check is
2453 * made on memory-hotadd so a system can start with mobility
2454 * disabled and enable it later
2456 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2457 page_group_by_mobility_disabled
= 1;
2459 page_group_by_mobility_disabled
= 0;
2461 printk("Built %i zonelists in %s order, mobility grouping %s. "
2462 "Total pages: %ld\n",
2464 zonelist_order_name
[current_zonelist_order
],
2465 page_group_by_mobility_disabled
? "off" : "on",
2468 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2473 * Helper functions to size the waitqueue hash table.
2474 * Essentially these want to choose hash table sizes sufficiently
2475 * large so that collisions trying to wait on pages are rare.
2476 * But in fact, the number of active page waitqueues on typical
2477 * systems is ridiculously low, less than 200. So this is even
2478 * conservative, even though it seems large.
2480 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2481 * waitqueues, i.e. the size of the waitq table given the number of pages.
2483 #define PAGES_PER_WAITQUEUE 256
2485 #ifndef CONFIG_MEMORY_HOTPLUG
2486 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2488 unsigned long size
= 1;
2490 pages
/= PAGES_PER_WAITQUEUE
;
2492 while (size
< pages
)
2496 * Once we have dozens or even hundreds of threads sleeping
2497 * on IO we've got bigger problems than wait queue collision.
2498 * Limit the size of the wait table to a reasonable size.
2500 size
= min(size
, 4096UL);
2502 return max(size
, 4UL);
2506 * A zone's size might be changed by hot-add, so it is not possible to determine
2507 * a suitable size for its wait_table. So we use the maximum size now.
2509 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2511 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2512 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2513 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2515 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2516 * or more by the traditional way. (See above). It equals:
2518 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2519 * ia64(16K page size) : = ( 8G + 4M)byte.
2520 * powerpc (64K page size) : = (32G +16M)byte.
2522 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2529 * This is an integer logarithm so that shifts can be used later
2530 * to extract the more random high bits from the multiplicative
2531 * hash function before the remainder is taken.
2533 static inline unsigned long wait_table_bits(unsigned long size
)
2538 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2541 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2542 * of blocks reserved is based on zone->pages_min. The memory within the
2543 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2544 * higher will lead to a bigger reserve which will get freed as contiguous
2545 * blocks as reclaim kicks in
2547 static void setup_zone_migrate_reserve(struct zone
*zone
)
2549 unsigned long start_pfn
, pfn
, end_pfn
;
2551 unsigned long reserve
, block_migratetype
;
2553 /* Get the start pfn, end pfn and the number of blocks to reserve */
2554 start_pfn
= zone
->zone_start_pfn
;
2555 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2556 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2559 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2560 if (!pfn_valid(pfn
))
2562 page
= pfn_to_page(pfn
);
2564 /* Watch out for overlapping nodes */
2565 if (page_to_nid(page
) != zone_to_nid(zone
))
2568 /* Blocks with reserved pages will never free, skip them. */
2569 if (PageReserved(page
))
2572 block_migratetype
= get_pageblock_migratetype(page
);
2574 /* If this block is reserved, account for it */
2575 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2580 /* Suitable for reserving if this block is movable */
2581 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2582 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2583 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2589 * If the reserve is met and this is a previous reserved block,
2592 if (block_migratetype
== MIGRATE_RESERVE
) {
2593 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2594 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2600 * Initially all pages are reserved - free ones are freed
2601 * up by free_all_bootmem() once the early boot process is
2602 * done. Non-atomic initialization, single-pass.
2604 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2605 unsigned long start_pfn
, enum memmap_context context
)
2608 unsigned long end_pfn
= start_pfn
+ size
;
2612 if (highest_memmap_pfn
< end_pfn
- 1)
2613 highest_memmap_pfn
= end_pfn
- 1;
2615 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2616 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2618 * There can be holes in boot-time mem_map[]s
2619 * handed to this function. They do not
2620 * exist on hotplugged memory.
2622 if (context
== MEMMAP_EARLY
) {
2623 if (!early_pfn_valid(pfn
))
2625 if (!early_pfn_in_nid(pfn
, nid
))
2628 page
= pfn_to_page(pfn
);
2629 set_page_links(page
, zone
, nid
, pfn
);
2630 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2631 init_page_count(page
);
2632 reset_page_mapcount(page
);
2633 SetPageReserved(page
);
2635 * Mark the block movable so that blocks are reserved for
2636 * movable at startup. This will force kernel allocations
2637 * to reserve their blocks rather than leaking throughout
2638 * the address space during boot when many long-lived
2639 * kernel allocations are made. Later some blocks near
2640 * the start are marked MIGRATE_RESERVE by
2641 * setup_zone_migrate_reserve()
2643 * bitmap is created for zone's valid pfn range. but memmap
2644 * can be created for invalid pages (for alignment)
2645 * check here not to call set_pageblock_migratetype() against
2648 if ((z
->zone_start_pfn
<= pfn
)
2649 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2650 && !(pfn
& (pageblock_nr_pages
- 1)))
2651 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2653 INIT_LIST_HEAD(&page
->lru
);
2654 #ifdef WANT_PAGE_VIRTUAL
2655 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2656 if (!is_highmem_idx(zone
))
2657 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2662 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2665 for_each_migratetype_order(order
, t
) {
2666 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2667 zone
->free_area
[order
].nr_free
= 0;
2671 #ifndef __HAVE_ARCH_MEMMAP_INIT
2672 #define memmap_init(size, nid, zone, start_pfn) \
2673 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2676 static int zone_batchsize(struct zone
*zone
)
2682 * The per-cpu-pages pools are set to around 1000th of the
2683 * size of the zone. But no more than 1/2 of a meg.
2685 * OK, so we don't know how big the cache is. So guess.
2687 batch
= zone
->present_pages
/ 1024;
2688 if (batch
* PAGE_SIZE
> 512 * 1024)
2689 batch
= (512 * 1024) / PAGE_SIZE
;
2690 batch
/= 4; /* We effectively *= 4 below */
2695 * Clamp the batch to a 2^n - 1 value. Having a power
2696 * of 2 value was found to be more likely to have
2697 * suboptimal cache aliasing properties in some cases.
2699 * For example if 2 tasks are alternately allocating
2700 * batches of pages, one task can end up with a lot
2701 * of pages of one half of the possible page colors
2702 * and the other with pages of the other colors.
2704 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
2709 /* The deferral and batching of frees should be suppressed under NOMMU
2712 * The problem is that NOMMU needs to be able to allocate large chunks
2713 * of contiguous memory as there's no hardware page translation to
2714 * assemble apparent contiguous memory from discontiguous pages.
2716 * Queueing large contiguous runs of pages for batching, however,
2717 * causes the pages to actually be freed in smaller chunks. As there
2718 * can be a significant delay between the individual batches being
2719 * recycled, this leads to the once large chunks of space being
2720 * fragmented and becoming unavailable for high-order allocations.
2726 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2728 struct per_cpu_pages
*pcp
;
2730 memset(p
, 0, sizeof(*p
));
2734 pcp
->high
= 6 * batch
;
2735 pcp
->batch
= max(1UL, 1 * batch
);
2736 INIT_LIST_HEAD(&pcp
->list
);
2740 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2741 * to the value high for the pageset p.
2744 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2747 struct per_cpu_pages
*pcp
;
2751 pcp
->batch
= max(1UL, high
/4);
2752 if ((high
/4) > (PAGE_SHIFT
* 8))
2753 pcp
->batch
= PAGE_SHIFT
* 8;
2759 * Boot pageset table. One per cpu which is going to be used for all
2760 * zones and all nodes. The parameters will be set in such a way
2761 * that an item put on a list will immediately be handed over to
2762 * the buddy list. This is safe since pageset manipulation is done
2763 * with interrupts disabled.
2765 * Some NUMA counter updates may also be caught by the boot pagesets.
2767 * The boot_pagesets must be kept even after bootup is complete for
2768 * unused processors and/or zones. They do play a role for bootstrapping
2769 * hotplugged processors.
2771 * zoneinfo_show() and maybe other functions do
2772 * not check if the processor is online before following the pageset pointer.
2773 * Other parts of the kernel may not check if the zone is available.
2775 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2778 * Dynamically allocate memory for the
2779 * per cpu pageset array in struct zone.
2781 static int __cpuinit
process_zones(int cpu
)
2783 struct zone
*zone
, *dzone
;
2784 int node
= cpu_to_node(cpu
);
2786 node_set_state(node
, N_CPU
); /* this node has a cpu */
2788 for_each_populated_zone(zone
) {
2789 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2791 if (!zone_pcp(zone
, cpu
))
2794 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2796 if (percpu_pagelist_fraction
)
2797 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2798 (zone
->present_pages
/ percpu_pagelist_fraction
));
2803 for_each_zone(dzone
) {
2804 if (!populated_zone(dzone
))
2808 kfree(zone_pcp(dzone
, cpu
));
2809 zone_pcp(dzone
, cpu
) = NULL
;
2814 static inline void free_zone_pagesets(int cpu
)
2818 for_each_zone(zone
) {
2819 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2821 /* Free per_cpu_pageset if it is slab allocated */
2822 if (pset
!= &boot_pageset
[cpu
])
2824 zone_pcp(zone
, cpu
) = NULL
;
2828 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2829 unsigned long action
,
2832 int cpu
= (long)hcpu
;
2833 int ret
= NOTIFY_OK
;
2836 case CPU_UP_PREPARE
:
2837 case CPU_UP_PREPARE_FROZEN
:
2838 if (process_zones(cpu
))
2841 case CPU_UP_CANCELED
:
2842 case CPU_UP_CANCELED_FROZEN
:
2844 case CPU_DEAD_FROZEN
:
2845 free_zone_pagesets(cpu
);
2853 static struct notifier_block __cpuinitdata pageset_notifier
=
2854 { &pageset_cpuup_callback
, NULL
, 0 };
2856 void __init
setup_per_cpu_pageset(void)
2860 /* Initialize per_cpu_pageset for cpu 0.
2861 * A cpuup callback will do this for every cpu
2862 * as it comes online
2864 err
= process_zones(smp_processor_id());
2866 register_cpu_notifier(&pageset_notifier
);
2871 static noinline __init_refok
2872 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2875 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2879 * The per-page waitqueue mechanism uses hashed waitqueues
2882 zone
->wait_table_hash_nr_entries
=
2883 wait_table_hash_nr_entries(zone_size_pages
);
2884 zone
->wait_table_bits
=
2885 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2886 alloc_size
= zone
->wait_table_hash_nr_entries
2887 * sizeof(wait_queue_head_t
);
2889 if (!slab_is_available()) {
2890 zone
->wait_table
= (wait_queue_head_t
*)
2891 alloc_bootmem_node(pgdat
, alloc_size
);
2894 * This case means that a zone whose size was 0 gets new memory
2895 * via memory hot-add.
2896 * But it may be the case that a new node was hot-added. In
2897 * this case vmalloc() will not be able to use this new node's
2898 * memory - this wait_table must be initialized to use this new
2899 * node itself as well.
2900 * To use this new node's memory, further consideration will be
2903 zone
->wait_table
= vmalloc(alloc_size
);
2905 if (!zone
->wait_table
)
2908 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2909 init_waitqueue_head(zone
->wait_table
+ i
);
2914 static __meminit
void zone_pcp_init(struct zone
*zone
)
2917 unsigned long batch
= zone_batchsize(zone
);
2919 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2921 /* Early boot. Slab allocator not functional yet */
2922 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2923 setup_pageset(&boot_pageset
[cpu
],0);
2925 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2928 if (zone
->present_pages
)
2929 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2930 zone
->name
, zone
->present_pages
, batch
);
2933 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2934 unsigned long zone_start_pfn
,
2936 enum memmap_context context
)
2938 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2940 ret
= zone_wait_table_init(zone
, size
);
2943 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2945 zone
->zone_start_pfn
= zone_start_pfn
;
2947 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
2948 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2950 (unsigned long)zone_idx(zone
),
2951 zone_start_pfn
, (zone_start_pfn
+ size
));
2953 zone_init_free_lists(zone
);
2958 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2960 * Basic iterator support. Return the first range of PFNs for a node
2961 * Note: nid == MAX_NUMNODES returns first region regardless of node
2963 static int __meminit
first_active_region_index_in_nid(int nid
)
2967 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2968 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2975 * Basic iterator support. Return the next active range of PFNs for a node
2976 * Note: nid == MAX_NUMNODES returns next region regardless of node
2978 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2980 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2981 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2987 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2989 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2990 * Architectures may implement their own version but if add_active_range()
2991 * was used and there are no special requirements, this is a convenient
2994 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
2998 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2999 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3000 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3002 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3003 return early_node_map
[i
].nid
;
3005 /* This is a memory hole */
3008 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3010 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3014 nid
= __early_pfn_to_nid(pfn
);
3017 /* just returns 0 */
3021 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3022 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3026 nid
= __early_pfn_to_nid(pfn
);
3027 if (nid
>= 0 && nid
!= node
)
3033 /* Basic iterator support to walk early_node_map[] */
3034 #define for_each_active_range_index_in_nid(i, nid) \
3035 for (i = first_active_region_index_in_nid(nid); i != -1; \
3036 i = next_active_region_index_in_nid(i, nid))
3039 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3040 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3041 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3043 * If an architecture guarantees that all ranges registered with
3044 * add_active_ranges() contain no holes and may be freed, this
3045 * this function may be used instead of calling free_bootmem() manually.
3047 void __init
free_bootmem_with_active_regions(int nid
,
3048 unsigned long max_low_pfn
)
3052 for_each_active_range_index_in_nid(i
, nid
) {
3053 unsigned long size_pages
= 0;
3054 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3056 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3059 if (end_pfn
> max_low_pfn
)
3060 end_pfn
= max_low_pfn
;
3062 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3063 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3064 PFN_PHYS(early_node_map
[i
].start_pfn
),
3065 size_pages
<< PAGE_SHIFT
);
3069 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3074 for_each_active_range_index_in_nid(i
, nid
) {
3075 ret
= work_fn(early_node_map
[i
].start_pfn
,
3076 early_node_map
[i
].end_pfn
, data
);
3082 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3083 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3085 * If an architecture guarantees that all ranges registered with
3086 * add_active_ranges() contain no holes and may be freed, this
3087 * function may be used instead of calling memory_present() manually.
3089 void __init
sparse_memory_present_with_active_regions(int nid
)
3093 for_each_active_range_index_in_nid(i
, nid
)
3094 memory_present(early_node_map
[i
].nid
,
3095 early_node_map
[i
].start_pfn
,
3096 early_node_map
[i
].end_pfn
);
3100 * get_pfn_range_for_nid - Return the start and end page frames for a node
3101 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3102 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3103 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3105 * It returns the start and end page frame of a node based on information
3106 * provided by an arch calling add_active_range(). If called for a node
3107 * with no available memory, a warning is printed and the start and end
3110 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3111 unsigned long *start_pfn
, unsigned long *end_pfn
)
3117 for_each_active_range_index_in_nid(i
, nid
) {
3118 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3119 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3122 if (*start_pfn
== -1UL)
3127 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3128 * assumption is made that zones within a node are ordered in monotonic
3129 * increasing memory addresses so that the "highest" populated zone is used
3131 static void __init
find_usable_zone_for_movable(void)
3134 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3135 if (zone_index
== ZONE_MOVABLE
)
3138 if (arch_zone_highest_possible_pfn
[zone_index
] >
3139 arch_zone_lowest_possible_pfn
[zone_index
])
3143 VM_BUG_ON(zone_index
== -1);
3144 movable_zone
= zone_index
;
3148 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3149 * because it is sized independant of architecture. Unlike the other zones,
3150 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3151 * in each node depending on the size of each node and how evenly kernelcore
3152 * is distributed. This helper function adjusts the zone ranges
3153 * provided by the architecture for a given node by using the end of the
3154 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3155 * zones within a node are in order of monotonic increases memory addresses
3157 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3158 unsigned long zone_type
,
3159 unsigned long node_start_pfn
,
3160 unsigned long node_end_pfn
,
3161 unsigned long *zone_start_pfn
,
3162 unsigned long *zone_end_pfn
)
3164 /* Only adjust if ZONE_MOVABLE is on this node */
3165 if (zone_movable_pfn
[nid
]) {
3166 /* Size ZONE_MOVABLE */
3167 if (zone_type
== ZONE_MOVABLE
) {
3168 *zone_start_pfn
= zone_movable_pfn
[nid
];
3169 *zone_end_pfn
= min(node_end_pfn
,
3170 arch_zone_highest_possible_pfn
[movable_zone
]);
3172 /* Adjust for ZONE_MOVABLE starting within this range */
3173 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3174 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3175 *zone_end_pfn
= zone_movable_pfn
[nid
];
3177 /* Check if this whole range is within ZONE_MOVABLE */
3178 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3179 *zone_start_pfn
= *zone_end_pfn
;
3184 * Return the number of pages a zone spans in a node, including holes
3185 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3187 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3188 unsigned long zone_type
,
3189 unsigned long *ignored
)
3191 unsigned long node_start_pfn
, node_end_pfn
;
3192 unsigned long zone_start_pfn
, zone_end_pfn
;
3194 /* Get the start and end of the node and zone */
3195 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3196 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3197 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3198 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3199 node_start_pfn
, node_end_pfn
,
3200 &zone_start_pfn
, &zone_end_pfn
);
3202 /* Check that this node has pages within the zone's required range */
3203 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3206 /* Move the zone boundaries inside the node if necessary */
3207 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3208 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3210 /* Return the spanned pages */
3211 return zone_end_pfn
- zone_start_pfn
;
3215 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3216 * then all holes in the requested range will be accounted for.
3218 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3219 unsigned long range_start_pfn
,
3220 unsigned long range_end_pfn
)
3223 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3224 unsigned long start_pfn
;
3226 /* Find the end_pfn of the first active range of pfns in the node */
3227 i
= first_active_region_index_in_nid(nid
);
3231 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3233 /* Account for ranges before physical memory on this node */
3234 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3235 hole_pages
= prev_end_pfn
- range_start_pfn
;
3237 /* Find all holes for the zone within the node */
3238 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3240 /* No need to continue if prev_end_pfn is outside the zone */
3241 if (prev_end_pfn
>= range_end_pfn
)
3244 /* Make sure the end of the zone is not within the hole */
3245 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3246 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3248 /* Update the hole size cound and move on */
3249 if (start_pfn
> range_start_pfn
) {
3250 BUG_ON(prev_end_pfn
> start_pfn
);
3251 hole_pages
+= start_pfn
- prev_end_pfn
;
3253 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3256 /* Account for ranges past physical memory on this node */
3257 if (range_end_pfn
> prev_end_pfn
)
3258 hole_pages
+= range_end_pfn
-
3259 max(range_start_pfn
, prev_end_pfn
);
3265 * absent_pages_in_range - Return number of page frames in holes within a range
3266 * @start_pfn: The start PFN to start searching for holes
3267 * @end_pfn: The end PFN to stop searching for holes
3269 * It returns the number of pages frames in memory holes within a range.
3271 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3272 unsigned long end_pfn
)
3274 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3277 /* Return the number of page frames in holes in a zone on a node */
3278 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3279 unsigned long zone_type
,
3280 unsigned long *ignored
)
3282 unsigned long node_start_pfn
, node_end_pfn
;
3283 unsigned long zone_start_pfn
, zone_end_pfn
;
3285 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3286 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3288 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3291 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3292 node_start_pfn
, node_end_pfn
,
3293 &zone_start_pfn
, &zone_end_pfn
);
3294 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3298 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3299 unsigned long zone_type
,
3300 unsigned long *zones_size
)
3302 return zones_size
[zone_type
];
3305 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3306 unsigned long zone_type
,
3307 unsigned long *zholes_size
)
3312 return zholes_size
[zone_type
];
3317 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3318 unsigned long *zones_size
, unsigned long *zholes_size
)
3320 unsigned long realtotalpages
, totalpages
= 0;
3323 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3324 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3326 pgdat
->node_spanned_pages
= totalpages
;
3328 realtotalpages
= totalpages
;
3329 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3331 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3333 pgdat
->node_present_pages
= realtotalpages
;
3334 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3338 #ifndef CONFIG_SPARSEMEM
3340 * Calculate the size of the zone->blockflags rounded to an unsigned long
3341 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3342 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3343 * round what is now in bits to nearest long in bits, then return it in
3346 static unsigned long __init
usemap_size(unsigned long zonesize
)
3348 unsigned long usemapsize
;
3350 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3351 usemapsize
= usemapsize
>> pageblock_order
;
3352 usemapsize
*= NR_PAGEBLOCK_BITS
;
3353 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3355 return usemapsize
/ 8;
3358 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3359 struct zone
*zone
, unsigned long zonesize
)
3361 unsigned long usemapsize
= usemap_size(zonesize
);
3362 zone
->pageblock_flags
= NULL
;
3364 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3367 static void inline setup_usemap(struct pglist_data
*pgdat
,
3368 struct zone
*zone
, unsigned long zonesize
) {}
3369 #endif /* CONFIG_SPARSEMEM */
3371 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3373 /* Return a sensible default order for the pageblock size. */
3374 static inline int pageblock_default_order(void)
3376 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3377 return HUGETLB_PAGE_ORDER
;
3382 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3383 static inline void __init
set_pageblock_order(unsigned int order
)
3385 /* Check that pageblock_nr_pages has not already been setup */
3386 if (pageblock_order
)
3390 * Assume the largest contiguous order of interest is a huge page.
3391 * This value may be variable depending on boot parameters on IA64
3393 pageblock_order
= order
;
3395 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3398 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3399 * and pageblock_default_order() are unused as pageblock_order is set
3400 * at compile-time. See include/linux/pageblock-flags.h for the values of
3401 * pageblock_order based on the kernel config
3403 static inline int pageblock_default_order(unsigned int order
)
3407 #define set_pageblock_order(x) do {} while (0)
3409 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3412 * Set up the zone data structures:
3413 * - mark all pages reserved
3414 * - mark all memory queues empty
3415 * - clear the memory bitmaps
3417 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3418 unsigned long *zones_size
, unsigned long *zholes_size
)
3421 int nid
= pgdat
->node_id
;
3422 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3425 pgdat_resize_init(pgdat
);
3426 pgdat
->nr_zones
= 0;
3427 init_waitqueue_head(&pgdat
->kswapd_wait
);
3428 pgdat
->kswapd_max_order
= 0;
3429 pgdat_page_cgroup_init(pgdat
);
3431 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3432 struct zone
*zone
= pgdat
->node_zones
+ j
;
3433 unsigned long size
, realsize
, memmap_pages
;
3436 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3437 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3441 * Adjust realsize so that it accounts for how much memory
3442 * is used by this zone for memmap. This affects the watermark
3443 * and per-cpu initialisations
3446 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3447 if (realsize
>= memmap_pages
) {
3448 realsize
-= memmap_pages
;
3451 " %s zone: %lu pages used for memmap\n",
3452 zone_names
[j
], memmap_pages
);
3455 " %s zone: %lu pages exceeds realsize %lu\n",
3456 zone_names
[j
], memmap_pages
, realsize
);
3458 /* Account for reserved pages */
3459 if (j
== 0 && realsize
> dma_reserve
) {
3460 realsize
-= dma_reserve
;
3461 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3462 zone_names
[0], dma_reserve
);
3465 if (!is_highmem_idx(j
))
3466 nr_kernel_pages
+= realsize
;
3467 nr_all_pages
+= realsize
;
3469 zone
->spanned_pages
= size
;
3470 zone
->present_pages
= realsize
;
3473 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3475 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3477 zone
->name
= zone_names
[j
];
3478 spin_lock_init(&zone
->lock
);
3479 spin_lock_init(&zone
->lru_lock
);
3480 zone_seqlock_init(zone
);
3481 zone
->zone_pgdat
= pgdat
;
3483 zone
->prev_priority
= DEF_PRIORITY
;
3485 zone_pcp_init(zone
);
3487 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3488 zone
->lru
[l
].nr_scan
= 0;
3490 zone
->reclaim_stat
.recent_rotated
[0] = 0;
3491 zone
->reclaim_stat
.recent_rotated
[1] = 0;
3492 zone
->reclaim_stat
.recent_scanned
[0] = 0;
3493 zone
->reclaim_stat
.recent_scanned
[1] = 0;
3494 zap_zone_vm_stats(zone
);
3499 set_pageblock_order(pageblock_default_order());
3500 setup_usemap(pgdat
, zone
, size
);
3501 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3502 size
, MEMMAP_EARLY
);
3504 memmap_init(size
, nid
, j
, zone_start_pfn
);
3505 zone_start_pfn
+= size
;
3509 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3511 /* Skip empty nodes */
3512 if (!pgdat
->node_spanned_pages
)
3515 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3516 /* ia64 gets its own node_mem_map, before this, without bootmem */
3517 if (!pgdat
->node_mem_map
) {
3518 unsigned long size
, start
, end
;
3522 * The zone's endpoints aren't required to be MAX_ORDER
3523 * aligned but the node_mem_map endpoints must be in order
3524 * for the buddy allocator to function correctly.
3526 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3527 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3528 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3529 size
= (end
- start
) * sizeof(struct page
);
3530 map
= alloc_remap(pgdat
->node_id
, size
);
3532 map
= alloc_bootmem_node(pgdat
, size
);
3533 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3535 #ifndef CONFIG_NEED_MULTIPLE_NODES
3537 * With no DISCONTIG, the global mem_map is just set as node 0's
3539 if (pgdat
== NODE_DATA(0)) {
3540 mem_map
= NODE_DATA(0)->node_mem_map
;
3541 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3542 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3543 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3544 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3547 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3550 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3551 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3553 pg_data_t
*pgdat
= NODE_DATA(nid
);
3555 pgdat
->node_id
= nid
;
3556 pgdat
->node_start_pfn
= node_start_pfn
;
3557 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3559 alloc_node_mem_map(pgdat
);
3560 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3561 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3562 nid
, (unsigned long)pgdat
,
3563 (unsigned long)pgdat
->node_mem_map
);
3566 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3569 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3571 #if MAX_NUMNODES > 1
3573 * Figure out the number of possible node ids.
3575 static void __init
setup_nr_node_ids(void)
3578 unsigned int highest
= 0;
3580 for_each_node_mask(node
, node_possible_map
)
3582 nr_node_ids
= highest
+ 1;
3585 static inline void setup_nr_node_ids(void)
3591 * add_active_range - Register a range of PFNs backed by physical memory
3592 * @nid: The node ID the range resides on
3593 * @start_pfn: The start PFN of the available physical memory
3594 * @end_pfn: The end PFN of the available physical memory
3596 * These ranges are stored in an early_node_map[] and later used by
3597 * free_area_init_nodes() to calculate zone sizes and holes. If the
3598 * range spans a memory hole, it is up to the architecture to ensure
3599 * the memory is not freed by the bootmem allocator. If possible
3600 * the range being registered will be merged with existing ranges.
3602 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3603 unsigned long end_pfn
)
3607 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3608 "Entering add_active_range(%d, %#lx, %#lx) "
3609 "%d entries of %d used\n",
3610 nid
, start_pfn
, end_pfn
,
3611 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3613 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3615 /* Merge with existing active regions if possible */
3616 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3617 if (early_node_map
[i
].nid
!= nid
)
3620 /* Skip if an existing region covers this new one */
3621 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3622 end_pfn
<= early_node_map
[i
].end_pfn
)
3625 /* Merge forward if suitable */
3626 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3627 end_pfn
> early_node_map
[i
].end_pfn
) {
3628 early_node_map
[i
].end_pfn
= end_pfn
;
3632 /* Merge backward if suitable */
3633 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3634 end_pfn
>= early_node_map
[i
].start_pfn
) {
3635 early_node_map
[i
].start_pfn
= start_pfn
;
3640 /* Check that early_node_map is large enough */
3641 if (i
>= MAX_ACTIVE_REGIONS
) {
3642 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3643 MAX_ACTIVE_REGIONS
);
3647 early_node_map
[i
].nid
= nid
;
3648 early_node_map
[i
].start_pfn
= start_pfn
;
3649 early_node_map
[i
].end_pfn
= end_pfn
;
3650 nr_nodemap_entries
= i
+ 1;
3654 * remove_active_range - Shrink an existing registered range of PFNs
3655 * @nid: The node id the range is on that should be shrunk
3656 * @start_pfn: The new PFN of the range
3657 * @end_pfn: The new PFN of the range
3659 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3660 * The map is kept near the end physical page range that has already been
3661 * registered. This function allows an arch to shrink an existing registered
3664 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3665 unsigned long end_pfn
)
3670 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3671 nid
, start_pfn
, end_pfn
);
3673 /* Find the old active region end and shrink */
3674 for_each_active_range_index_in_nid(i
, nid
) {
3675 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3676 early_node_map
[i
].end_pfn
<= end_pfn
) {
3678 early_node_map
[i
].start_pfn
= 0;
3679 early_node_map
[i
].end_pfn
= 0;
3683 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3684 early_node_map
[i
].end_pfn
> start_pfn
) {
3685 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3686 early_node_map
[i
].end_pfn
= start_pfn
;
3687 if (temp_end_pfn
> end_pfn
)
3688 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3691 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3692 early_node_map
[i
].end_pfn
> end_pfn
&&
3693 early_node_map
[i
].start_pfn
< end_pfn
) {
3694 early_node_map
[i
].start_pfn
= end_pfn
;
3702 /* remove the blank ones */
3703 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3704 if (early_node_map
[i
].nid
!= nid
)
3706 if (early_node_map
[i
].end_pfn
)
3708 /* we found it, get rid of it */
3709 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3710 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3711 sizeof(early_node_map
[j
]));
3712 j
= nr_nodemap_entries
- 1;
3713 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3714 nr_nodemap_entries
--;
3719 * remove_all_active_ranges - Remove all currently registered regions
3721 * During discovery, it may be found that a table like SRAT is invalid
3722 * and an alternative discovery method must be used. This function removes
3723 * all currently registered regions.
3725 void __init
remove_all_active_ranges(void)
3727 memset(early_node_map
, 0, sizeof(early_node_map
));
3728 nr_nodemap_entries
= 0;
3731 /* Compare two active node_active_regions */
3732 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3734 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3735 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3737 /* Done this way to avoid overflows */
3738 if (arange
->start_pfn
> brange
->start_pfn
)
3740 if (arange
->start_pfn
< brange
->start_pfn
)
3746 /* sort the node_map by start_pfn */
3747 static void __init
sort_node_map(void)
3749 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3750 sizeof(struct node_active_region
),
3751 cmp_node_active_region
, NULL
);
3754 /* Find the lowest pfn for a node */
3755 static unsigned long __init
find_min_pfn_for_node(int nid
)
3758 unsigned long min_pfn
= ULONG_MAX
;
3760 /* Assuming a sorted map, the first range found has the starting pfn */
3761 for_each_active_range_index_in_nid(i
, nid
)
3762 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3764 if (min_pfn
== ULONG_MAX
) {
3766 "Could not find start_pfn for node %d\n", nid
);
3774 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3776 * It returns the minimum PFN based on information provided via
3777 * add_active_range().
3779 unsigned long __init
find_min_pfn_with_active_regions(void)
3781 return find_min_pfn_for_node(MAX_NUMNODES
);
3785 * early_calculate_totalpages()
3786 * Sum pages in active regions for movable zone.
3787 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3789 static unsigned long __init
early_calculate_totalpages(void)
3792 unsigned long totalpages
= 0;
3794 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3795 unsigned long pages
= early_node_map
[i
].end_pfn
-
3796 early_node_map
[i
].start_pfn
;
3797 totalpages
+= pages
;
3799 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3805 * Find the PFN the Movable zone begins in each node. Kernel memory
3806 * is spread evenly between nodes as long as the nodes have enough
3807 * memory. When they don't, some nodes will have more kernelcore than
3810 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3813 unsigned long usable_startpfn
;
3814 unsigned long kernelcore_node
, kernelcore_remaining
;
3815 unsigned long totalpages
= early_calculate_totalpages();
3816 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3819 * If movablecore was specified, calculate what size of
3820 * kernelcore that corresponds so that memory usable for
3821 * any allocation type is evenly spread. If both kernelcore
3822 * and movablecore are specified, then the value of kernelcore
3823 * will be used for required_kernelcore if it's greater than
3824 * what movablecore would have allowed.
3826 if (required_movablecore
) {
3827 unsigned long corepages
;
3830 * Round-up so that ZONE_MOVABLE is at least as large as what
3831 * was requested by the user
3833 required_movablecore
=
3834 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3835 corepages
= totalpages
- required_movablecore
;
3837 required_kernelcore
= max(required_kernelcore
, corepages
);
3840 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3841 if (!required_kernelcore
)
3844 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3845 find_usable_zone_for_movable();
3846 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3849 /* Spread kernelcore memory as evenly as possible throughout nodes */
3850 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3851 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3853 * Recalculate kernelcore_node if the division per node
3854 * now exceeds what is necessary to satisfy the requested
3855 * amount of memory for the kernel
3857 if (required_kernelcore
< kernelcore_node
)
3858 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3861 * As the map is walked, we track how much memory is usable
3862 * by the kernel using kernelcore_remaining. When it is
3863 * 0, the rest of the node is usable by ZONE_MOVABLE
3865 kernelcore_remaining
= kernelcore_node
;
3867 /* Go through each range of PFNs within this node */
3868 for_each_active_range_index_in_nid(i
, nid
) {
3869 unsigned long start_pfn
, end_pfn
;
3870 unsigned long size_pages
;
3872 start_pfn
= max(early_node_map
[i
].start_pfn
,
3873 zone_movable_pfn
[nid
]);
3874 end_pfn
= early_node_map
[i
].end_pfn
;
3875 if (start_pfn
>= end_pfn
)
3878 /* Account for what is only usable for kernelcore */
3879 if (start_pfn
< usable_startpfn
) {
3880 unsigned long kernel_pages
;
3881 kernel_pages
= min(end_pfn
, usable_startpfn
)
3884 kernelcore_remaining
-= min(kernel_pages
,
3885 kernelcore_remaining
);
3886 required_kernelcore
-= min(kernel_pages
,
3887 required_kernelcore
);
3889 /* Continue if range is now fully accounted */
3890 if (end_pfn
<= usable_startpfn
) {
3893 * Push zone_movable_pfn to the end so
3894 * that if we have to rebalance
3895 * kernelcore across nodes, we will
3896 * not double account here
3898 zone_movable_pfn
[nid
] = end_pfn
;
3901 start_pfn
= usable_startpfn
;
3905 * The usable PFN range for ZONE_MOVABLE is from
3906 * start_pfn->end_pfn. Calculate size_pages as the
3907 * number of pages used as kernelcore
3909 size_pages
= end_pfn
- start_pfn
;
3910 if (size_pages
> kernelcore_remaining
)
3911 size_pages
= kernelcore_remaining
;
3912 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3915 * Some kernelcore has been met, update counts and
3916 * break if the kernelcore for this node has been
3919 required_kernelcore
-= min(required_kernelcore
,
3921 kernelcore_remaining
-= size_pages
;
3922 if (!kernelcore_remaining
)
3928 * If there is still required_kernelcore, we do another pass with one
3929 * less node in the count. This will push zone_movable_pfn[nid] further
3930 * along on the nodes that still have memory until kernelcore is
3934 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3937 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3938 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3939 zone_movable_pfn
[nid
] =
3940 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3943 /* Any regular memory on that node ? */
3944 static void check_for_regular_memory(pg_data_t
*pgdat
)
3946 #ifdef CONFIG_HIGHMEM
3947 enum zone_type zone_type
;
3949 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3950 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3951 if (zone
->present_pages
)
3952 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3958 * free_area_init_nodes - Initialise all pg_data_t and zone data
3959 * @max_zone_pfn: an array of max PFNs for each zone
3961 * This will call free_area_init_node() for each active node in the system.
3962 * Using the page ranges provided by add_active_range(), the size of each
3963 * zone in each node and their holes is calculated. If the maximum PFN
3964 * between two adjacent zones match, it is assumed that the zone is empty.
3965 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3966 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3967 * starts where the previous one ended. For example, ZONE_DMA32 starts
3968 * at arch_max_dma_pfn.
3970 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3975 /* Sort early_node_map as initialisation assumes it is sorted */
3978 /* Record where the zone boundaries are */
3979 memset(arch_zone_lowest_possible_pfn
, 0,
3980 sizeof(arch_zone_lowest_possible_pfn
));
3981 memset(arch_zone_highest_possible_pfn
, 0,
3982 sizeof(arch_zone_highest_possible_pfn
));
3983 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3984 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3985 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3986 if (i
== ZONE_MOVABLE
)
3988 arch_zone_lowest_possible_pfn
[i
] =
3989 arch_zone_highest_possible_pfn
[i
-1];
3990 arch_zone_highest_possible_pfn
[i
] =
3991 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3993 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3994 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3996 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3997 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3998 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4000 /* Print out the zone ranges */
4001 printk("Zone PFN ranges:\n");
4002 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4003 if (i
== ZONE_MOVABLE
)
4005 printk(" %-8s %0#10lx -> %0#10lx\n",
4007 arch_zone_lowest_possible_pfn
[i
],
4008 arch_zone_highest_possible_pfn
[i
]);
4011 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4012 printk("Movable zone start PFN for each node\n");
4013 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4014 if (zone_movable_pfn
[i
])
4015 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4018 /* Print out the early_node_map[] */
4019 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4020 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4021 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4022 early_node_map
[i
].start_pfn
,
4023 early_node_map
[i
].end_pfn
);
4025 /* Initialise every node */
4026 mminit_verify_pageflags_layout();
4027 setup_nr_node_ids();
4028 for_each_online_node(nid
) {
4029 pg_data_t
*pgdat
= NODE_DATA(nid
);
4030 free_area_init_node(nid
, NULL
,
4031 find_min_pfn_for_node(nid
), NULL
);
4033 /* Any memory on that node */
4034 if (pgdat
->node_present_pages
)
4035 node_set_state(nid
, N_HIGH_MEMORY
);
4036 check_for_regular_memory(pgdat
);
4040 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4042 unsigned long long coremem
;
4046 coremem
= memparse(p
, &p
);
4047 *core
= coremem
>> PAGE_SHIFT
;
4049 /* Paranoid check that UL is enough for the coremem value */
4050 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4056 * kernelcore=size sets the amount of memory for use for allocations that
4057 * cannot be reclaimed or migrated.
4059 static int __init
cmdline_parse_kernelcore(char *p
)
4061 return cmdline_parse_core(p
, &required_kernelcore
);
4065 * movablecore=size sets the amount of memory for use for allocations that
4066 * can be reclaimed or migrated.
4068 static int __init
cmdline_parse_movablecore(char *p
)
4070 return cmdline_parse_core(p
, &required_movablecore
);
4073 early_param("kernelcore", cmdline_parse_kernelcore
);
4074 early_param("movablecore", cmdline_parse_movablecore
);
4076 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4079 * set_dma_reserve - set the specified number of pages reserved in the first zone
4080 * @new_dma_reserve: The number of pages to mark reserved
4082 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4083 * In the DMA zone, a significant percentage may be consumed by kernel image
4084 * and other unfreeable allocations which can skew the watermarks badly. This
4085 * function may optionally be used to account for unfreeable pages in the
4086 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4087 * smaller per-cpu batchsize.
4089 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4091 dma_reserve
= new_dma_reserve
;
4094 #ifndef CONFIG_NEED_MULTIPLE_NODES
4095 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4096 EXPORT_SYMBOL(contig_page_data
);
4099 void __init
free_area_init(unsigned long *zones_size
)
4101 free_area_init_node(0, zones_size
,
4102 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4105 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4106 unsigned long action
, void *hcpu
)
4108 int cpu
= (unsigned long)hcpu
;
4110 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4114 * Spill the event counters of the dead processor
4115 * into the current processors event counters.
4116 * This artificially elevates the count of the current
4119 vm_events_fold_cpu(cpu
);
4122 * Zero the differential counters of the dead processor
4123 * so that the vm statistics are consistent.
4125 * This is only okay since the processor is dead and cannot
4126 * race with what we are doing.
4128 refresh_cpu_vm_stats(cpu
);
4133 void __init
page_alloc_init(void)
4135 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4139 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4140 * or min_free_kbytes changes.
4142 static void calculate_totalreserve_pages(void)
4144 struct pglist_data
*pgdat
;
4145 unsigned long reserve_pages
= 0;
4146 enum zone_type i
, j
;
4148 for_each_online_pgdat(pgdat
) {
4149 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4150 struct zone
*zone
= pgdat
->node_zones
+ i
;
4151 unsigned long max
= 0;
4153 /* Find valid and maximum lowmem_reserve in the zone */
4154 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4155 if (zone
->lowmem_reserve
[j
] > max
)
4156 max
= zone
->lowmem_reserve
[j
];
4159 /* we treat pages_high as reserved pages. */
4160 max
+= zone
->pages_high
;
4162 if (max
> zone
->present_pages
)
4163 max
= zone
->present_pages
;
4164 reserve_pages
+= max
;
4167 totalreserve_pages
= reserve_pages
;
4171 * setup_per_zone_lowmem_reserve - called whenever
4172 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4173 * has a correct pages reserved value, so an adequate number of
4174 * pages are left in the zone after a successful __alloc_pages().
4176 static void setup_per_zone_lowmem_reserve(void)
4178 struct pglist_data
*pgdat
;
4179 enum zone_type j
, idx
;
4181 for_each_online_pgdat(pgdat
) {
4182 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4183 struct zone
*zone
= pgdat
->node_zones
+ j
;
4184 unsigned long present_pages
= zone
->present_pages
;
4186 zone
->lowmem_reserve
[j
] = 0;
4190 struct zone
*lower_zone
;
4194 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4195 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4197 lower_zone
= pgdat
->node_zones
+ idx
;
4198 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4199 sysctl_lowmem_reserve_ratio
[idx
];
4200 present_pages
+= lower_zone
->present_pages
;
4205 /* update totalreserve_pages */
4206 calculate_totalreserve_pages();
4210 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4212 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4213 * with respect to min_free_kbytes.
4215 void setup_per_zone_pages_min(void)
4217 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4218 unsigned long lowmem_pages
= 0;
4220 unsigned long flags
;
4222 /* Calculate total number of !ZONE_HIGHMEM pages */
4223 for_each_zone(zone
) {
4224 if (!is_highmem(zone
))
4225 lowmem_pages
+= zone
->present_pages
;
4228 for_each_zone(zone
) {
4231 spin_lock_irqsave(&zone
->lock
, flags
);
4232 tmp
= (u64
)pages_min
* zone
->present_pages
;
4233 do_div(tmp
, lowmem_pages
);
4234 if (is_highmem(zone
)) {
4236 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4237 * need highmem pages, so cap pages_min to a small
4240 * The (pages_high-pages_low) and (pages_low-pages_min)
4241 * deltas controls asynch page reclaim, and so should
4242 * not be capped for highmem.
4246 min_pages
= zone
->present_pages
/ 1024;
4247 if (min_pages
< SWAP_CLUSTER_MAX
)
4248 min_pages
= SWAP_CLUSTER_MAX
;
4249 if (min_pages
> 128)
4251 zone
->pages_min
= min_pages
;
4254 * If it's a lowmem zone, reserve a number of pages
4255 * proportionate to the zone's size.
4257 zone
->pages_min
= tmp
;
4260 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4261 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4262 setup_zone_migrate_reserve(zone
);
4263 spin_unlock_irqrestore(&zone
->lock
, flags
);
4266 /* update totalreserve_pages */
4267 calculate_totalreserve_pages();
4271 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4273 * The inactive anon list should be small enough that the VM never has to
4274 * do too much work, but large enough that each inactive page has a chance
4275 * to be referenced again before it is swapped out.
4277 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4278 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4279 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4280 * the anonymous pages are kept on the inactive list.
4283 * memory ratio inactive anon
4284 * -------------------------------------
4293 static void setup_per_zone_inactive_ratio(void)
4297 for_each_zone(zone
) {
4298 unsigned int gb
, ratio
;
4300 /* Zone size in gigabytes */
4301 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4302 ratio
= int_sqrt(10 * gb
);
4306 zone
->inactive_ratio
= ratio
;
4311 * Initialise min_free_kbytes.
4313 * For small machines we want it small (128k min). For large machines
4314 * we want it large (64MB max). But it is not linear, because network
4315 * bandwidth does not increase linearly with machine size. We use
4317 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4318 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4334 static int __init
init_per_zone_pages_min(void)
4336 unsigned long lowmem_kbytes
;
4338 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4340 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4341 if (min_free_kbytes
< 128)
4342 min_free_kbytes
= 128;
4343 if (min_free_kbytes
> 65536)
4344 min_free_kbytes
= 65536;
4345 setup_per_zone_pages_min();
4346 setup_per_zone_lowmem_reserve();
4347 setup_per_zone_inactive_ratio();
4350 module_init(init_per_zone_pages_min
)
4353 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4354 * that we can call two helper functions whenever min_free_kbytes
4357 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4358 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4360 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4362 setup_per_zone_pages_min();
4367 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4368 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4373 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4378 zone
->min_unmapped_pages
= (zone
->present_pages
*
4379 sysctl_min_unmapped_ratio
) / 100;
4383 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4384 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4389 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4394 zone
->min_slab_pages
= (zone
->present_pages
*
4395 sysctl_min_slab_ratio
) / 100;
4401 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4402 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4403 * whenever sysctl_lowmem_reserve_ratio changes.
4405 * The reserve ratio obviously has absolutely no relation with the
4406 * pages_min watermarks. The lowmem reserve ratio can only make sense
4407 * if in function of the boot time zone sizes.
4409 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4410 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4412 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4413 setup_per_zone_lowmem_reserve();
4418 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4419 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4420 * can have before it gets flushed back to buddy allocator.
4423 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4424 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4430 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4431 if (!write
|| (ret
== -EINVAL
))
4433 for_each_zone(zone
) {
4434 for_each_online_cpu(cpu
) {
4436 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4437 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4443 int hashdist
= HASHDIST_DEFAULT
;
4446 static int __init
set_hashdist(char *str
)
4450 hashdist
= simple_strtoul(str
, &str
, 0);
4453 __setup("hashdist=", set_hashdist
);
4457 * allocate a large system hash table from bootmem
4458 * - it is assumed that the hash table must contain an exact power-of-2
4459 * quantity of entries
4460 * - limit is the number of hash buckets, not the total allocation size
4462 void *__init
alloc_large_system_hash(const char *tablename
,
4463 unsigned long bucketsize
,
4464 unsigned long numentries
,
4467 unsigned int *_hash_shift
,
4468 unsigned int *_hash_mask
,
4469 unsigned long limit
)
4471 unsigned long long max
= limit
;
4472 unsigned long log2qty
, size
;
4475 /* allow the kernel cmdline to have a say */
4477 /* round applicable memory size up to nearest megabyte */
4478 numentries
= nr_kernel_pages
;
4479 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4480 numentries
>>= 20 - PAGE_SHIFT
;
4481 numentries
<<= 20 - PAGE_SHIFT
;
4483 /* limit to 1 bucket per 2^scale bytes of low memory */
4484 if (scale
> PAGE_SHIFT
)
4485 numentries
>>= (scale
- PAGE_SHIFT
);
4487 numentries
<<= (PAGE_SHIFT
- scale
);
4489 /* Make sure we've got at least a 0-order allocation.. */
4490 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4491 numentries
= PAGE_SIZE
/ bucketsize
;
4493 numentries
= roundup_pow_of_two(numentries
);
4495 /* limit allocation size to 1/16 total memory by default */
4497 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4498 do_div(max
, bucketsize
);
4501 if (numentries
> max
)
4504 log2qty
= ilog2(numentries
);
4507 size
= bucketsize
<< log2qty
;
4508 if (flags
& HASH_EARLY
)
4509 table
= alloc_bootmem_nopanic(size
);
4511 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4513 unsigned long order
= get_order(size
);
4515 if (order
< MAX_ORDER
)
4516 table
= (void *)__get_free_pages(GFP_ATOMIC
,
4519 * If bucketsize is not a power-of-two, we may free
4520 * some pages at the end of hash table.
4523 unsigned long alloc_end
= (unsigned long)table
+
4524 (PAGE_SIZE
<< order
);
4525 unsigned long used
= (unsigned long)table
+
4527 split_page(virt_to_page(table
), order
);
4528 while (used
< alloc_end
) {
4534 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4537 panic("Failed to allocate %s hash table\n", tablename
);
4539 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4542 ilog2(size
) - PAGE_SHIFT
,
4546 *_hash_shift
= log2qty
;
4548 *_hash_mask
= (1 << log2qty
) - 1;
4551 * If hashdist is set, the table allocation is done with __vmalloc()
4552 * which invokes the kmemleak_alloc() callback. This function may also
4553 * be called before the slab and kmemleak are initialised when
4554 * kmemleak simply buffers the request to be executed later
4555 * (GFP_ATOMIC flag ignored in this case).
4558 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
4563 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4564 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4567 #ifdef CONFIG_SPARSEMEM
4568 return __pfn_to_section(pfn
)->pageblock_flags
;
4570 return zone
->pageblock_flags
;
4571 #endif /* CONFIG_SPARSEMEM */
4574 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4576 #ifdef CONFIG_SPARSEMEM
4577 pfn
&= (PAGES_PER_SECTION
-1);
4578 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4580 pfn
= pfn
- zone
->zone_start_pfn
;
4581 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4582 #endif /* CONFIG_SPARSEMEM */
4586 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4587 * @page: The page within the block of interest
4588 * @start_bitidx: The first bit of interest to retrieve
4589 * @end_bitidx: The last bit of interest
4590 * returns pageblock_bits flags
4592 unsigned long get_pageblock_flags_group(struct page
*page
,
4593 int start_bitidx
, int end_bitidx
)
4596 unsigned long *bitmap
;
4597 unsigned long pfn
, bitidx
;
4598 unsigned long flags
= 0;
4599 unsigned long value
= 1;
4601 zone
= page_zone(page
);
4602 pfn
= page_to_pfn(page
);
4603 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4604 bitidx
= pfn_to_bitidx(zone
, pfn
);
4606 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4607 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4614 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4615 * @page: The page within the block of interest
4616 * @start_bitidx: The first bit of interest
4617 * @end_bitidx: The last bit of interest
4618 * @flags: The flags to set
4620 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4621 int start_bitidx
, int end_bitidx
)
4624 unsigned long *bitmap
;
4625 unsigned long pfn
, bitidx
;
4626 unsigned long value
= 1;
4628 zone
= page_zone(page
);
4629 pfn
= page_to_pfn(page
);
4630 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4631 bitidx
= pfn_to_bitidx(zone
, pfn
);
4632 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4633 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4635 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4637 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4639 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4643 * This is designed as sub function...plz see page_isolation.c also.
4644 * set/clear page block's type to be ISOLATE.
4645 * page allocater never alloc memory from ISOLATE block.
4648 int set_migratetype_isolate(struct page
*page
)
4651 unsigned long flags
;
4654 zone
= page_zone(page
);
4655 spin_lock_irqsave(&zone
->lock
, flags
);
4657 * In future, more migrate types will be able to be isolation target.
4659 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4661 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4662 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4665 spin_unlock_irqrestore(&zone
->lock
, flags
);
4671 void unset_migratetype_isolate(struct page
*page
)
4674 unsigned long flags
;
4675 zone
= page_zone(page
);
4676 spin_lock_irqsave(&zone
->lock
, flags
);
4677 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4679 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4680 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4682 spin_unlock_irqrestore(&zone
->lock
, flags
);
4685 #ifdef CONFIG_MEMORY_HOTREMOVE
4687 * All pages in the range must be isolated before calling this.
4690 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4696 unsigned long flags
;
4697 /* find the first valid pfn */
4698 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4703 zone
= page_zone(pfn_to_page(pfn
));
4704 spin_lock_irqsave(&zone
->lock
, flags
);
4706 while (pfn
< end_pfn
) {
4707 if (!pfn_valid(pfn
)) {
4711 page
= pfn_to_page(pfn
);
4712 BUG_ON(page_count(page
));
4713 BUG_ON(!PageBuddy(page
));
4714 order
= page_order(page
);
4715 #ifdef CONFIG_DEBUG_VM
4716 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4717 pfn
, 1 << order
, end_pfn
);
4719 list_del(&page
->lru
);
4720 rmv_page_order(page
);
4721 zone
->free_area
[order
].nr_free
--;
4722 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4724 for (i
= 0; i
< (1 << order
); i
++)
4725 SetPageReserved((page
+i
));
4726 pfn
+= (1 << order
);
4728 spin_unlock_irqrestore(&zone
->lock
, flags
);