2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
40 #include <linux/sort.h>
41 #include <linux/pfn.h>
42 #include <linux/backing-dev.h>
43 #include <linux/fault-inject.h>
45 #include <asm/tlbflush.h>
46 #include <asm/div64.h>
50 * Array of node states.
52 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
53 [N_POSSIBLE
] = NODE_MASK_ALL
,
54 [N_ONLINE
] = { { [0] = 1UL } },
56 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
58 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
60 [N_CPU
] = { { [0] = 1UL } },
63 EXPORT_SYMBOL(node_states
);
65 unsigned long totalram_pages __read_mostly
;
66 unsigned long totalreserve_pages __read_mostly
;
68 int percpu_pagelist_fraction
;
70 static void __free_pages_ok(struct page
*page
, unsigned int order
);
73 * results with 256, 32 in the lowmem_reserve sysctl:
74 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
75 * 1G machine -> (16M dma, 784M normal, 224M high)
76 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
77 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
78 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
80 * TBD: should special case ZONE_DMA32 machines here - in those we normally
81 * don't need any ZONE_NORMAL reservation
83 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
84 #ifdef CONFIG_ZONE_DMA
87 #ifdef CONFIG_ZONE_DMA32
96 EXPORT_SYMBOL(totalram_pages
);
98 static char * const zone_names
[MAX_NR_ZONES
] = {
99 #ifdef CONFIG_ZONE_DMA
102 #ifdef CONFIG_ZONE_DMA32
106 #ifdef CONFIG_HIGHMEM
112 int min_free_kbytes
= 1024;
114 unsigned long __meminitdata nr_kernel_pages
;
115 unsigned long __meminitdata nr_all_pages
;
116 static unsigned long __meminitdata dma_reserve
;
118 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
120 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
121 * ranges of memory (RAM) that may be registered with add_active_range().
122 * Ranges passed to add_active_range() will be merged if possible
123 * so the number of times add_active_range() can be called is
124 * related to the number of nodes and the number of holes
126 #ifdef CONFIG_MAX_ACTIVE_REGIONS
127 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
128 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
130 #if MAX_NUMNODES >= 32
131 /* If there can be many nodes, allow up to 50 holes per node */
132 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
134 /* By default, allow up to 256 distinct regions */
135 #define MAX_ACTIVE_REGIONS 256
139 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
140 static int __meminitdata nr_nodemap_entries
;
141 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
142 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
143 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
144 static unsigned long __meminitdata node_boundary_start_pfn
[MAX_NUMNODES
];
145 static unsigned long __meminitdata node_boundary_end_pfn
[MAX_NUMNODES
];
146 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
147 unsigned long __initdata required_kernelcore
;
148 unsigned long __initdata required_movablecore
;
149 unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
151 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
153 EXPORT_SYMBOL(movable_zone
);
154 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
157 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
158 EXPORT_SYMBOL(nr_node_ids
);
161 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
162 int page_group_by_mobility_disabled __read_mostly
;
164 static inline int get_pageblock_migratetype(struct page
*page
)
166 if (unlikely(page_group_by_mobility_disabled
))
167 return MIGRATE_UNMOVABLE
;
169 return get_pageblock_flags_group(page
, PB_migrate
, PB_migrate_end
);
172 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
174 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
175 PB_migrate
, PB_migrate_end
);
178 static inline int allocflags_to_migratetype(gfp_t gfp_flags
, int order
)
180 WARN_ON((gfp_flags
& GFP_MOVABLE_MASK
) == GFP_MOVABLE_MASK
);
182 if (unlikely(page_group_by_mobility_disabled
))
183 return MIGRATE_UNMOVABLE
;
185 /* Cluster high-order atomic allocations together */
186 if (unlikely(order
> 0) &&
187 (!(gfp_flags
& __GFP_WAIT
) || in_interrupt()))
188 return MIGRATE_HIGHATOMIC
;
190 /* Cluster based on mobility */
191 return (((gfp_flags
& __GFP_MOVABLE
) != 0) << 1) |
192 ((gfp_flags
& __GFP_RECLAIMABLE
) != 0);
196 static inline int get_pageblock_migratetype(struct page
*page
)
198 return MIGRATE_UNMOVABLE
;
201 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
205 static inline int allocflags_to_migratetype(gfp_t gfp_flags
, int order
)
207 return MIGRATE_UNMOVABLE
;
209 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
211 #ifdef CONFIG_DEBUG_VM
212 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
216 unsigned long pfn
= page_to_pfn(page
);
219 seq
= zone_span_seqbegin(zone
);
220 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
222 else if (pfn
< zone
->zone_start_pfn
)
224 } while (zone_span_seqretry(zone
, seq
));
229 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
231 if (!pfn_valid_within(page_to_pfn(page
)))
233 if (zone
!= page_zone(page
))
239 * Temporary debugging check for pages not lying within a given zone.
241 static int bad_range(struct zone
*zone
, struct page
*page
)
243 if (page_outside_zone_boundaries(zone
, page
))
245 if (!page_is_consistent(zone
, page
))
251 static inline int bad_range(struct zone
*zone
, struct page
*page
)
257 static void bad_page(struct page
*page
)
259 printk(KERN_EMERG
"Bad page state in process '%s'\n"
260 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
261 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
262 KERN_EMERG
"Backtrace:\n",
263 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
264 (unsigned long)page
->flags
, page
->mapping
,
265 page_mapcount(page
), page_count(page
));
267 page
->flags
&= ~(1 << PG_lru
|
277 set_page_count(page
, 0);
278 reset_page_mapcount(page
);
279 page
->mapping
= NULL
;
280 add_taint(TAINT_BAD_PAGE
);
284 * Higher-order pages are called "compound pages". They are structured thusly:
286 * The first PAGE_SIZE page is called the "head page".
288 * The remaining PAGE_SIZE pages are called "tail pages".
290 * All pages have PG_compound set. All pages have their ->private pointing at
291 * the head page (even the head page has this).
293 * The first tail page's ->lru.next holds the address of the compound page's
294 * put_page() function. Its ->lru.prev holds the order of allocation.
295 * This usage means that zero-order pages may not be compound.
298 static void free_compound_page(struct page
*page
)
300 __free_pages_ok(page
, compound_order(page
));
303 static void prep_compound_page(struct page
*page
, unsigned long order
)
306 int nr_pages
= 1 << order
;
308 set_compound_page_dtor(page
, free_compound_page
);
309 set_compound_order(page
, order
);
311 for (i
= 1; i
< nr_pages
; i
++) {
312 struct page
*p
= page
+ i
;
315 p
->first_page
= page
;
319 static void destroy_compound_page(struct page
*page
, unsigned long order
)
322 int nr_pages
= 1 << order
;
324 if (unlikely(compound_order(page
) != order
))
327 if (unlikely(!PageHead(page
)))
329 __ClearPageHead(page
);
330 for (i
= 1; i
< nr_pages
; i
++) {
331 struct page
*p
= page
+ i
;
333 if (unlikely(!PageTail(p
) |
334 (p
->first_page
!= page
)))
340 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
344 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
346 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
347 * and __GFP_HIGHMEM from hard or soft interrupt context.
349 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
350 for (i
= 0; i
< (1 << order
); i
++)
351 clear_highpage(page
+ i
);
355 * function for dealing with page's order in buddy system.
356 * zone->lock is already acquired when we use these.
357 * So, we don't need atomic page->flags operations here.
359 static inline unsigned long page_order(struct page
*page
)
361 return page_private(page
);
364 static inline void set_page_order(struct page
*page
, int order
)
366 set_page_private(page
, order
);
367 __SetPageBuddy(page
);
370 static inline void rmv_page_order(struct page
*page
)
372 __ClearPageBuddy(page
);
373 set_page_private(page
, 0);
377 * Locate the struct page for both the matching buddy in our
378 * pair (buddy1) and the combined O(n+1) page they form (page).
380 * 1) Any buddy B1 will have an order O twin B2 which satisfies
381 * the following equation:
383 * For example, if the starting buddy (buddy2) is #8 its order
385 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
387 * 2) Any buddy B will have an order O+1 parent P which
388 * satisfies the following equation:
391 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
393 static inline struct page
*
394 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
396 unsigned long buddy_idx
= page_idx
^ (1 << order
);
398 return page
+ (buddy_idx
- page_idx
);
401 static inline unsigned long
402 __find_combined_index(unsigned long page_idx
, unsigned int order
)
404 return (page_idx
& ~(1 << order
));
408 * This function checks whether a page is free && is the buddy
409 * we can do coalesce a page and its buddy if
410 * (a) the buddy is not in a hole &&
411 * (b) the buddy is in the buddy system &&
412 * (c) a page and its buddy have the same order &&
413 * (d) a page and its buddy are in the same zone.
415 * For recording whether a page is in the buddy system, we use PG_buddy.
416 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
418 * For recording page's order, we use page_private(page).
420 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
423 if (!pfn_valid_within(page_to_pfn(buddy
)))
426 if (page_zone_id(page
) != page_zone_id(buddy
))
429 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
430 BUG_ON(page_count(buddy
) != 0);
437 * Freeing function for a buddy system allocator.
439 * The concept of a buddy system is to maintain direct-mapped table
440 * (containing bit values) for memory blocks of various "orders".
441 * The bottom level table contains the map for the smallest allocatable
442 * units of memory (here, pages), and each level above it describes
443 * pairs of units from the levels below, hence, "buddies".
444 * At a high level, all that happens here is marking the table entry
445 * at the bottom level available, and propagating the changes upward
446 * as necessary, plus some accounting needed to play nicely with other
447 * parts of the VM system.
448 * At each level, we keep a list of pages, which are heads of continuous
449 * free pages of length of (1 << order) and marked with PG_buddy. Page's
450 * order is recorded in page_private(page) field.
451 * So when we are allocating or freeing one, we can derive the state of the
452 * other. That is, if we allocate a small block, and both were
453 * free, the remainder of the region must be split into blocks.
454 * If a block is freed, and its buddy is also free, then this
455 * triggers coalescing into a block of larger size.
460 static inline void __free_one_page(struct page
*page
,
461 struct zone
*zone
, unsigned int order
)
463 unsigned long page_idx
;
464 int order_size
= 1 << order
;
465 int migratetype
= get_pageblock_migratetype(page
);
467 if (unlikely(PageCompound(page
)))
468 destroy_compound_page(page
, order
);
470 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
472 VM_BUG_ON(page_idx
& (order_size
- 1));
473 VM_BUG_ON(bad_range(zone
, page
));
475 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
476 while (order
< MAX_ORDER
-1) {
477 unsigned long combined_idx
;
480 buddy
= __page_find_buddy(page
, page_idx
, order
);
481 if (!page_is_buddy(page
, buddy
, order
))
482 break; /* Move the buddy up one level. */
484 list_del(&buddy
->lru
);
485 zone
->free_area
[order
].nr_free
--;
486 rmv_page_order(buddy
);
487 combined_idx
= __find_combined_index(page_idx
, order
);
488 page
= page
+ (combined_idx
- page_idx
);
489 page_idx
= combined_idx
;
492 set_page_order(page
, order
);
494 &zone
->free_area
[order
].free_list
[migratetype
]);
495 zone
->free_area
[order
].nr_free
++;
498 static inline int free_pages_check(struct page
*page
)
500 if (unlikely(page_mapcount(page
) |
501 (page
->mapping
!= NULL
) |
502 (page_count(page
) != 0) |
515 __ClearPageDirty(page
);
517 * For now, we report if PG_reserved was found set, but do not
518 * clear it, and do not free the page. But we shall soon need
519 * to do more, for when the ZERO_PAGE count wraps negative.
521 return PageReserved(page
);
525 * Frees a list of pages.
526 * Assumes all pages on list are in same zone, and of same order.
527 * count is the number of pages to free.
529 * If the zone was previously in an "all pages pinned" state then look to
530 * see if this freeing clears that state.
532 * And clear the zone's pages_scanned counter, to hold off the "all pages are
533 * pinned" detection logic.
535 static void free_pages_bulk(struct zone
*zone
, int count
,
536 struct list_head
*list
, int order
)
538 spin_lock(&zone
->lock
);
539 zone
->all_unreclaimable
= 0;
540 zone
->pages_scanned
= 0;
544 VM_BUG_ON(list_empty(list
));
545 page
= list_entry(list
->prev
, struct page
, lru
);
546 /* have to delete it as __free_one_page list manipulates */
547 list_del(&page
->lru
);
548 __free_one_page(page
, zone
, order
);
550 spin_unlock(&zone
->lock
);
553 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
555 spin_lock(&zone
->lock
);
556 zone
->all_unreclaimable
= 0;
557 zone
->pages_scanned
= 0;
558 __free_one_page(page
, zone
, order
);
559 spin_unlock(&zone
->lock
);
562 static void __free_pages_ok(struct page
*page
, unsigned int order
)
568 for (i
= 0 ; i
< (1 << order
) ; ++i
)
569 reserved
+= free_pages_check(page
+ i
);
573 if (!PageHighMem(page
))
574 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
575 arch_free_page(page
, order
);
576 kernel_map_pages(page
, 1 << order
, 0);
578 local_irq_save(flags
);
579 __count_vm_events(PGFREE
, 1 << order
);
580 free_one_page(page_zone(page
), page
, order
);
581 local_irq_restore(flags
);
585 * permit the bootmem allocator to evade page validation on high-order frees
587 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
590 __ClearPageReserved(page
);
591 set_page_count(page
, 0);
592 set_page_refcounted(page
);
598 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
599 struct page
*p
= &page
[loop
];
601 if (loop
+ 1 < BITS_PER_LONG
)
603 __ClearPageReserved(p
);
604 set_page_count(p
, 0);
607 set_page_refcounted(page
);
608 __free_pages(page
, order
);
614 * The order of subdivision here is critical for the IO subsystem.
615 * Please do not alter this order without good reasons and regression
616 * testing. Specifically, as large blocks of memory are subdivided,
617 * the order in which smaller blocks are delivered depends on the order
618 * they're subdivided in this function. This is the primary factor
619 * influencing the order in which pages are delivered to the IO
620 * subsystem according to empirical testing, and this is also justified
621 * by considering the behavior of a buddy system containing a single
622 * large block of memory acted on by a series of small allocations.
623 * This behavior is a critical factor in sglist merging's success.
627 static inline void expand(struct zone
*zone
, struct page
*page
,
628 int low
, int high
, struct free_area
*area
,
631 unsigned long size
= 1 << high
;
637 VM_BUG_ON(bad_range(zone
, &page
[size
]));
638 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
640 set_page_order(&page
[size
], high
);
645 * This page is about to be returned from the page allocator
647 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
649 if (unlikely(page_mapcount(page
) |
650 (page
->mapping
!= NULL
) |
651 (page_count(page
) != 0) |
666 * For now, we report if PG_reserved was found set, but do not
667 * clear it, and do not allocate the page: as a safety net.
669 if (PageReserved(page
))
672 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_readahead
|
673 1 << PG_referenced
| 1 << PG_arch_1
|
674 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
);
675 set_page_private(page
, 0);
676 set_page_refcounted(page
);
678 arch_alloc_page(page
, order
);
679 kernel_map_pages(page
, 1 << order
, 1);
681 if (gfp_flags
& __GFP_ZERO
)
682 prep_zero_page(page
, order
, gfp_flags
);
684 if (order
&& (gfp_flags
& __GFP_COMP
))
685 prep_compound_page(page
, order
);
690 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
692 * This array describes the order lists are fallen back to when
693 * the free lists for the desirable migrate type are depleted
695 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
696 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_HIGHATOMIC
},
697 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_HIGHATOMIC
},
698 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
,MIGRATE_HIGHATOMIC
},
699 [MIGRATE_HIGHATOMIC
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
,MIGRATE_MOVABLE
},
703 * Move the free pages in a range to the free lists of the requested type.
704 * Note that start_page and end_pages are not aligned in a MAX_ORDER_NR_PAGES
705 * boundary. If alignment is required, use move_freepages_block()
707 int move_freepages(struct zone
*zone
,
708 struct page
*start_page
, struct page
*end_page
,
713 int blocks_moved
= 0;
715 #ifndef CONFIG_HOLES_IN_ZONE
717 * page_zone is not safe to call in this context when
718 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
719 * anyway as we check zone boundaries in move_freepages_block().
720 * Remove at a later date when no bug reports exist related to
721 * CONFIG_PAGE_GROUP_BY_MOBILITY
723 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
726 for (page
= start_page
; page
<= end_page
;) {
727 if (!pfn_valid_within(page_to_pfn(page
))) {
732 if (!PageBuddy(page
)) {
737 order
= page_order(page
);
738 list_del(&page
->lru
);
740 &zone
->free_area
[order
].free_list
[migratetype
]);
748 int move_freepages_block(struct zone
*zone
, struct page
*page
, int migratetype
)
750 unsigned long start_pfn
, end_pfn
;
751 struct page
*start_page
, *end_page
;
753 start_pfn
= page_to_pfn(page
);
754 start_pfn
= start_pfn
& ~(MAX_ORDER_NR_PAGES
-1);
755 start_page
= pfn_to_page(start_pfn
);
756 end_page
= start_page
+ MAX_ORDER_NR_PAGES
- 1;
757 end_pfn
= start_pfn
+ MAX_ORDER_NR_PAGES
- 1;
759 /* Do not cross zone boundaries */
760 if (start_pfn
< zone
->zone_start_pfn
)
762 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
765 return move_freepages(zone
, start_page
, end_page
, migratetype
);
768 /* Return the page with the lowest PFN in the list */
769 static struct page
*min_page(struct list_head
*list
)
771 unsigned long min_pfn
= -1UL;
772 struct page
*min_page
= NULL
, *page
;;
774 list_for_each_entry(page
, list
, lru
) {
775 unsigned long pfn
= page_to_pfn(page
);
785 /* Remove an element from the buddy allocator from the fallback list */
786 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
787 int start_migratetype
)
789 struct free_area
* area
;
793 int nonatomic_fallback_atomic
= 0;
796 /* Find the largest possible block of pages in the other list */
797 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
799 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
800 migratetype
= fallbacks
[start_migratetype
][i
];
803 * Make it hard to fallback to blocks used for
804 * high-order atomic allocations
806 if (migratetype
== MIGRATE_HIGHATOMIC
&&
807 start_migratetype
!= MIGRATE_UNMOVABLE
&&
808 !nonatomic_fallback_atomic
)
811 area
= &(zone
->free_area
[current_order
]);
812 if (list_empty(&area
->free_list
[migratetype
]))
815 /* Bias kernel allocations towards low pfns */
816 page
= list_entry(area
->free_list
[migratetype
].next
,
818 if (unlikely(start_migratetype
!= MIGRATE_MOVABLE
))
819 page
= min_page(&area
->free_list
[migratetype
]);
823 * If breaking a large block of pages, move all free
824 * pages to the preferred allocation list. If falling
825 * back for a reclaimable kernel allocation, be more
826 * agressive about taking ownership of free pages
828 if (unlikely(current_order
>= MAX_ORDER
/ 2) ||
829 start_migratetype
== MIGRATE_RECLAIMABLE
) {
831 pages
= move_freepages_block(zone
, page
,
834 /* Claim the whole block if over half of it is free */
835 if ((pages
<< current_order
) >= (1 << (MAX_ORDER
-2)) &&
836 migratetype
!= MIGRATE_HIGHATOMIC
)
837 set_pageblock_migratetype(page
,
840 migratetype
= start_migratetype
;
843 /* Remove the page from the freelists */
844 list_del(&page
->lru
);
845 rmv_page_order(page
);
846 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
849 if (current_order
== MAX_ORDER
- 1)
850 set_pageblock_migratetype(page
,
853 expand(zone
, page
, order
, current_order
, area
, migratetype
);
858 /* Allow fallback to high-order atomic blocks if memory is that low */
859 if (!nonatomic_fallback_atomic
) {
860 nonatomic_fallback_atomic
= 1;
867 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
868 int start_migratetype
)
872 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
875 * Do the hard work of removing an element from the buddy allocator.
876 * Call me with the zone->lock already held.
878 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
881 struct free_area
* area
;
882 unsigned int current_order
;
885 /* Find a page of the appropriate size in the preferred list */
886 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
887 area
= &(zone
->free_area
[current_order
]);
888 if (list_empty(&area
->free_list
[migratetype
]))
891 page
= list_entry(area
->free_list
[migratetype
].next
,
893 list_del(&page
->lru
);
894 rmv_page_order(page
);
896 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
897 expand(zone
, page
, order
, current_order
, area
, migratetype
);
901 page
= __rmqueue_fallback(zone
, order
, migratetype
);
909 * Obtain a specified number of elements from the buddy allocator, all under
910 * a single hold of the lock, for efficiency. Add them to the supplied list.
911 * Returns the number of new pages which were placed at *list.
913 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
914 unsigned long count
, struct list_head
*list
,
919 spin_lock(&zone
->lock
);
920 for (i
= 0; i
< count
; ++i
) {
921 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
922 if (unlikely(page
== NULL
))
924 list_add(&page
->lru
, list
);
925 set_page_private(page
, migratetype
);
927 spin_unlock(&zone
->lock
);
933 * Called from the vmstat counter updater to drain pagesets of this
934 * currently executing processor on remote nodes after they have
937 * Note that this function must be called with the thread pinned to
938 * a single processor.
940 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
945 local_irq_save(flags
);
946 if (pcp
->count
>= pcp
->batch
)
947 to_drain
= pcp
->batch
;
949 to_drain
= pcp
->count
;
950 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
951 pcp
->count
-= to_drain
;
952 local_irq_restore(flags
);
956 static void __drain_pages(unsigned int cpu
)
962 for_each_zone(zone
) {
963 struct per_cpu_pageset
*pset
;
965 if (!populated_zone(zone
))
968 pset
= zone_pcp(zone
, cpu
);
969 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
970 struct per_cpu_pages
*pcp
;
973 local_irq_save(flags
);
974 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
976 local_irq_restore(flags
);
981 #ifdef CONFIG_HIBERNATION
983 void mark_free_pages(struct zone
*zone
)
985 unsigned long pfn
, max_zone_pfn
;
988 struct list_head
*curr
;
990 if (!zone
->spanned_pages
)
993 spin_lock_irqsave(&zone
->lock
, flags
);
995 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
996 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
997 if (pfn_valid(pfn
)) {
998 struct page
*page
= pfn_to_page(pfn
);
1000 if (!swsusp_page_is_forbidden(page
))
1001 swsusp_unset_page_free(page
);
1004 for_each_migratetype_order(order
, t
) {
1005 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1008 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1009 for (i
= 0; i
< (1UL << order
); i
++)
1010 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1013 spin_unlock_irqrestore(&zone
->lock
, flags
);
1015 #endif /* CONFIG_PM */
1017 #if defined(CONFIG_HIBERNATION) || defined(CONFIG_PAGE_GROUP_BY_MOBILITY)
1019 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1021 void drain_local_pages(void)
1023 unsigned long flags
;
1025 local_irq_save(flags
);
1026 __drain_pages(smp_processor_id());
1027 local_irq_restore(flags
);
1030 void smp_drain_local_pages(void *arg
)
1032 drain_local_pages();
1036 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1038 void drain_all_local_pages(void)
1040 unsigned long flags
;
1042 local_irq_save(flags
);
1043 __drain_pages(smp_processor_id());
1044 local_irq_restore(flags
);
1046 smp_call_function(smp_drain_local_pages
, NULL
, 0, 1);
1049 void drain_all_local_pages(void) {}
1050 #endif /* CONFIG_HIBERNATION || CONFIG_PAGE_GROUP_BY_MOBILITY */
1053 * Free a 0-order page
1055 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
1057 struct zone
*zone
= page_zone(page
);
1058 struct per_cpu_pages
*pcp
;
1059 unsigned long flags
;
1062 page
->mapping
= NULL
;
1063 if (free_pages_check(page
))
1066 if (!PageHighMem(page
))
1067 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1068 arch_free_page(page
, 0);
1069 kernel_map_pages(page
, 1, 0);
1071 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
1072 local_irq_save(flags
);
1073 __count_vm_event(PGFREE
);
1074 list_add(&page
->lru
, &pcp
->list
);
1075 set_page_private(page
, get_pageblock_migratetype(page
));
1077 if (pcp
->count
>= pcp
->high
) {
1078 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1079 pcp
->count
-= pcp
->batch
;
1081 local_irq_restore(flags
);
1085 void fastcall
free_hot_page(struct page
*page
)
1087 free_hot_cold_page(page
, 0);
1090 void fastcall
free_cold_page(struct page
*page
)
1092 free_hot_cold_page(page
, 1);
1096 * split_page takes a non-compound higher-order page, and splits it into
1097 * n (1<<order) sub-pages: page[0..n]
1098 * Each sub-page must be freed individually.
1100 * Note: this is probably too low level an operation for use in drivers.
1101 * Please consult with lkml before using this in your driver.
1103 void split_page(struct page
*page
, unsigned int order
)
1107 VM_BUG_ON(PageCompound(page
));
1108 VM_BUG_ON(!page_count(page
));
1109 for (i
= 1; i
< (1 << order
); i
++)
1110 set_page_refcounted(page
+ i
);
1114 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1115 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1118 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
1119 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1121 unsigned long flags
;
1123 int cold
= !!(gfp_flags
& __GFP_COLD
);
1125 int migratetype
= allocflags_to_migratetype(gfp_flags
, order
);
1129 if (likely(order
== 0)) {
1130 struct per_cpu_pages
*pcp
;
1132 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
1133 local_irq_save(flags
);
1135 pcp
->count
= rmqueue_bulk(zone
, 0,
1136 pcp
->batch
, &pcp
->list
, migratetype
);
1137 if (unlikely(!pcp
->count
))
1141 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
1142 /* Find a page of the appropriate migrate type */
1143 list_for_each_entry(page
, &pcp
->list
, lru
)
1144 if (page_private(page
) == migratetype
)
1147 /* Allocate more to the pcp list if necessary */
1148 if (unlikely(&page
->lru
== &pcp
->list
)) {
1149 pcp
->count
+= rmqueue_bulk(zone
, 0,
1150 pcp
->batch
, &pcp
->list
, migratetype
);
1151 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1154 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1155 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
1157 list_del(&page
->lru
);
1160 spin_lock_irqsave(&zone
->lock
, flags
);
1161 page
= __rmqueue(zone
, order
, migratetype
);
1162 spin_unlock(&zone
->lock
);
1167 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1168 zone_statistics(zonelist
, zone
);
1169 local_irq_restore(flags
);
1172 VM_BUG_ON(bad_range(zone
, page
));
1173 if (prep_new_page(page
, order
, gfp_flags
))
1178 local_irq_restore(flags
);
1183 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1184 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1185 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1186 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1187 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1188 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1189 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1191 #ifdef CONFIG_FAIL_PAGE_ALLOC
1193 static struct fail_page_alloc_attr
{
1194 struct fault_attr attr
;
1196 u32 ignore_gfp_highmem
;
1197 u32 ignore_gfp_wait
;
1200 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1202 struct dentry
*ignore_gfp_highmem_file
;
1203 struct dentry
*ignore_gfp_wait_file
;
1204 struct dentry
*min_order_file
;
1206 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1208 } fail_page_alloc
= {
1209 .attr
= FAULT_ATTR_INITIALIZER
,
1210 .ignore_gfp_wait
= 1,
1211 .ignore_gfp_highmem
= 1,
1215 static int __init
setup_fail_page_alloc(char *str
)
1217 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1219 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1221 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1223 if (order
< fail_page_alloc
.min_order
)
1225 if (gfp_mask
& __GFP_NOFAIL
)
1227 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1229 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1232 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1235 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1237 static int __init
fail_page_alloc_debugfs(void)
1239 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1243 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1247 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1249 fail_page_alloc
.ignore_gfp_wait_file
=
1250 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1251 &fail_page_alloc
.ignore_gfp_wait
);
1253 fail_page_alloc
.ignore_gfp_highmem_file
=
1254 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1255 &fail_page_alloc
.ignore_gfp_highmem
);
1256 fail_page_alloc
.min_order_file
=
1257 debugfs_create_u32("min-order", mode
, dir
,
1258 &fail_page_alloc
.min_order
);
1260 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1261 !fail_page_alloc
.ignore_gfp_highmem_file
||
1262 !fail_page_alloc
.min_order_file
) {
1264 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1265 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1266 debugfs_remove(fail_page_alloc
.min_order_file
);
1267 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1273 late_initcall(fail_page_alloc_debugfs
);
1275 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1277 #else /* CONFIG_FAIL_PAGE_ALLOC */
1279 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1284 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1287 * Return 1 if free pages are above 'mark'. This takes into account the order
1288 * of the allocation.
1290 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1291 int classzone_idx
, int alloc_flags
)
1293 /* free_pages my go negative - that's OK */
1295 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1298 if (alloc_flags
& ALLOC_HIGH
)
1300 if (alloc_flags
& ALLOC_HARDER
)
1303 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1305 for (o
= 0; o
< order
; o
++) {
1306 /* At the next order, this order's pages become unavailable */
1307 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1309 /* Require fewer higher order pages to be free */
1312 if (free_pages
<= min
)
1320 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1321 * skip over zones that are not allowed by the cpuset, or that have
1322 * been recently (in last second) found to be nearly full. See further
1323 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1324 * that have to skip over alot of full or unallowed zones.
1326 * If the zonelist cache is present in the passed in zonelist, then
1327 * returns a pointer to the allowed node mask (either the current
1328 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1330 * If the zonelist cache is not available for this zonelist, does
1331 * nothing and returns NULL.
1333 * If the fullzones BITMAP in the zonelist cache is stale (more than
1334 * a second since last zap'd) then we zap it out (clear its bits.)
1336 * We hold off even calling zlc_setup, until after we've checked the
1337 * first zone in the zonelist, on the theory that most allocations will
1338 * be satisfied from that first zone, so best to examine that zone as
1339 * quickly as we can.
1341 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1343 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1344 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1346 zlc
= zonelist
->zlcache_ptr
;
1350 if (jiffies
- zlc
->last_full_zap
> 1 * HZ
) {
1351 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1352 zlc
->last_full_zap
= jiffies
;
1355 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1356 &cpuset_current_mems_allowed
:
1357 &node_states
[N_HIGH_MEMORY
];
1358 return allowednodes
;
1362 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1363 * if it is worth looking at further for free memory:
1364 * 1) Check that the zone isn't thought to be full (doesn't have its
1365 * bit set in the zonelist_cache fullzones BITMAP).
1366 * 2) Check that the zones node (obtained from the zonelist_cache
1367 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1368 * Return true (non-zero) if zone is worth looking at further, or
1369 * else return false (zero) if it is not.
1371 * This check -ignores- the distinction between various watermarks,
1372 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1373 * found to be full for any variation of these watermarks, it will
1374 * be considered full for up to one second by all requests, unless
1375 * we are so low on memory on all allowed nodes that we are forced
1376 * into the second scan of the zonelist.
1378 * In the second scan we ignore this zonelist cache and exactly
1379 * apply the watermarks to all zones, even it is slower to do so.
1380 * We are low on memory in the second scan, and should leave no stone
1381 * unturned looking for a free page.
1383 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1384 nodemask_t
*allowednodes
)
1386 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1387 int i
; /* index of *z in zonelist zones */
1388 int n
; /* node that zone *z is on */
1390 zlc
= zonelist
->zlcache_ptr
;
1394 i
= z
- zonelist
->zones
;
1397 /* This zone is worth trying if it is allowed but not full */
1398 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1402 * Given 'z' scanning a zonelist, set the corresponding bit in
1403 * zlc->fullzones, so that subsequent attempts to allocate a page
1404 * from that zone don't waste time re-examining it.
1406 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1408 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1409 int i
; /* index of *z in zonelist zones */
1411 zlc
= zonelist
->zlcache_ptr
;
1415 i
= z
- zonelist
->zones
;
1417 set_bit(i
, zlc
->fullzones
);
1420 #else /* CONFIG_NUMA */
1422 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1427 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1428 nodemask_t
*allowednodes
)
1433 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1436 #endif /* CONFIG_NUMA */
1439 * get_page_from_freelist goes through the zonelist trying to allocate
1442 static struct page
*
1443 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
1444 struct zonelist
*zonelist
, int alloc_flags
)
1447 struct page
*page
= NULL
;
1448 int classzone_idx
= zone_idx(zonelist
->zones
[0]);
1450 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1451 int zlc_active
= 0; /* set if using zonelist_cache */
1452 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1453 enum zone_type highest_zoneidx
= -1; /* Gets set for policy zonelists */
1457 * Scan zonelist, looking for a zone with enough free.
1458 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1460 z
= zonelist
->zones
;
1464 * In NUMA, this could be a policy zonelist which contains
1465 * zones that may not be allowed by the current gfp_mask.
1466 * Check the zone is allowed by the current flags
1468 if (unlikely(alloc_should_filter_zonelist(zonelist
))) {
1469 if (highest_zoneidx
== -1)
1470 highest_zoneidx
= gfp_zone(gfp_mask
);
1471 if (zone_idx(*z
) > highest_zoneidx
)
1475 if (NUMA_BUILD
&& zlc_active
&&
1476 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1479 if ((alloc_flags
& ALLOC_CPUSET
) &&
1480 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1483 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1485 if (alloc_flags
& ALLOC_WMARK_MIN
)
1486 mark
= zone
->pages_min
;
1487 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1488 mark
= zone
->pages_low
;
1490 mark
= zone
->pages_high
;
1491 if (!zone_watermark_ok(zone
, order
, mark
,
1492 classzone_idx
, alloc_flags
)) {
1493 if (!zone_reclaim_mode
||
1494 !zone_reclaim(zone
, gfp_mask
, order
))
1495 goto this_zone_full
;
1499 page
= buffered_rmqueue(zonelist
, zone
, order
, gfp_mask
);
1504 zlc_mark_zone_full(zonelist
, z
);
1506 if (NUMA_BUILD
&& !did_zlc_setup
) {
1507 /* we do zlc_setup after the first zone is tried */
1508 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1512 } while (*(++z
) != NULL
);
1514 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1515 /* Disable zlc cache for second zonelist scan */
1523 * This is the 'heart' of the zoned buddy allocator.
1525 struct page
* fastcall
1526 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
1527 struct zonelist
*zonelist
)
1529 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1532 struct reclaim_state reclaim_state
;
1533 struct task_struct
*p
= current
;
1536 int did_some_progress
;
1538 might_sleep_if(wait
);
1540 if (should_fail_alloc_page(gfp_mask
, order
))
1544 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
1546 if (unlikely(*z
== NULL
)) {
1548 * Happens if we have an empty zonelist as a result of
1549 * GFP_THISNODE being used on a memoryless node
1554 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1555 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1560 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1561 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1562 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1563 * using a larger set of nodes after it has established that the
1564 * allowed per node queues are empty and that nodes are
1567 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1570 for (z
= zonelist
->zones
; *z
; z
++)
1571 wakeup_kswapd(*z
, order
);
1574 * OK, we're below the kswapd watermark and have kicked background
1575 * reclaim. Now things get more complex, so set up alloc_flags according
1576 * to how we want to proceed.
1578 * The caller may dip into page reserves a bit more if the caller
1579 * cannot run direct reclaim, or if the caller has realtime scheduling
1580 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1581 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1583 alloc_flags
= ALLOC_WMARK_MIN
;
1584 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1585 alloc_flags
|= ALLOC_HARDER
;
1586 if (gfp_mask
& __GFP_HIGH
)
1587 alloc_flags
|= ALLOC_HIGH
;
1589 alloc_flags
|= ALLOC_CPUSET
;
1592 * Go through the zonelist again. Let __GFP_HIGH and allocations
1593 * coming from realtime tasks go deeper into reserves.
1595 * This is the last chance, in general, before the goto nopage.
1596 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1597 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1599 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1603 /* This allocation should allow future memory freeing. */
1606 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1607 && !in_interrupt()) {
1608 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1610 /* go through the zonelist yet again, ignoring mins */
1611 page
= get_page_from_freelist(gfp_mask
, order
,
1612 zonelist
, ALLOC_NO_WATERMARKS
);
1615 if (gfp_mask
& __GFP_NOFAIL
) {
1616 congestion_wait(WRITE
, HZ
/50);
1623 /* Atomic allocations - we can't balance anything */
1629 /* We now go into synchronous reclaim */
1630 cpuset_memory_pressure_bump();
1631 p
->flags
|= PF_MEMALLOC
;
1632 reclaim_state
.reclaimed_slab
= 0;
1633 p
->reclaim_state
= &reclaim_state
;
1635 did_some_progress
= try_to_free_pages(zonelist
->zones
, order
, gfp_mask
);
1637 p
->reclaim_state
= NULL
;
1638 p
->flags
&= ~PF_MEMALLOC
;
1643 drain_all_local_pages();
1645 if (likely(did_some_progress
)) {
1646 page
= get_page_from_freelist(gfp_mask
, order
,
1647 zonelist
, alloc_flags
);
1650 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1652 * Go through the zonelist yet one more time, keep
1653 * very high watermark here, this is only to catch
1654 * a parallel oom killing, we must fail if we're still
1655 * under heavy pressure.
1657 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1658 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1662 /* The OOM killer will not help higher order allocs so fail */
1663 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1666 out_of_memory(zonelist
, gfp_mask
, order
);
1671 * Don't let big-order allocations loop unless the caller explicitly
1672 * requests that. Wait for some write requests to complete then retry.
1674 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1675 * <= 3, but that may not be true in other implementations.
1678 if (!(gfp_mask
& __GFP_NORETRY
)) {
1679 if ((order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
1680 (gfp_mask
& __GFP_REPEAT
))
1682 if (gfp_mask
& __GFP_NOFAIL
)
1686 congestion_wait(WRITE
, HZ
/50);
1691 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1692 printk(KERN_WARNING
"%s: page allocation failure."
1693 " order:%d, mode:0x%x\n",
1694 p
->comm
, order
, gfp_mask
);
1702 EXPORT_SYMBOL(__alloc_pages
);
1705 * Common helper functions.
1707 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1710 page
= alloc_pages(gfp_mask
, order
);
1713 return (unsigned long) page_address(page
);
1716 EXPORT_SYMBOL(__get_free_pages
);
1718 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1723 * get_zeroed_page() returns a 32-bit address, which cannot represent
1726 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1728 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1730 return (unsigned long) page_address(page
);
1734 EXPORT_SYMBOL(get_zeroed_page
);
1736 void __pagevec_free(struct pagevec
*pvec
)
1738 int i
= pagevec_count(pvec
);
1741 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1744 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1746 if (put_page_testzero(page
)) {
1748 free_hot_page(page
);
1750 __free_pages_ok(page
, order
);
1754 EXPORT_SYMBOL(__free_pages
);
1756 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1759 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1760 __free_pages(virt_to_page((void *)addr
), order
);
1764 EXPORT_SYMBOL(free_pages
);
1766 static unsigned int nr_free_zone_pages(int offset
)
1768 /* Just pick one node, since fallback list is circular */
1769 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1770 unsigned int sum
= 0;
1772 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1773 struct zone
**zonep
= zonelist
->zones
;
1776 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1777 unsigned long size
= zone
->present_pages
;
1778 unsigned long high
= zone
->pages_high
;
1787 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1789 unsigned int nr_free_buffer_pages(void)
1791 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1793 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1796 * Amount of free RAM allocatable within all zones
1798 unsigned int nr_free_pagecache_pages(void)
1800 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1803 static inline void show_node(struct zone
*zone
)
1806 printk("Node %d ", zone_to_nid(zone
));
1809 void si_meminfo(struct sysinfo
*val
)
1811 val
->totalram
= totalram_pages
;
1813 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1814 val
->bufferram
= nr_blockdev_pages();
1815 val
->totalhigh
= totalhigh_pages
;
1816 val
->freehigh
= nr_free_highpages();
1817 val
->mem_unit
= PAGE_SIZE
;
1820 EXPORT_SYMBOL(si_meminfo
);
1823 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1825 pg_data_t
*pgdat
= NODE_DATA(nid
);
1827 val
->totalram
= pgdat
->node_present_pages
;
1828 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1829 #ifdef CONFIG_HIGHMEM
1830 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1831 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1837 val
->mem_unit
= PAGE_SIZE
;
1841 #define K(x) ((x) << (PAGE_SHIFT-10))
1844 * Show free area list (used inside shift_scroll-lock stuff)
1845 * We also calculate the percentage fragmentation. We do this by counting the
1846 * memory on each free list with the exception of the first item on the list.
1848 void show_free_areas(void)
1853 for_each_zone(zone
) {
1854 if (!populated_zone(zone
))
1858 printk("%s per-cpu:\n", zone
->name
);
1860 for_each_online_cpu(cpu
) {
1861 struct per_cpu_pageset
*pageset
;
1863 pageset
= zone_pcp(zone
, cpu
);
1865 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1866 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1867 cpu
, pageset
->pcp
[0].high
,
1868 pageset
->pcp
[0].batch
, pageset
->pcp
[0].count
,
1869 pageset
->pcp
[1].high
, pageset
->pcp
[1].batch
,
1870 pageset
->pcp
[1].count
);
1874 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1875 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1876 global_page_state(NR_ACTIVE
),
1877 global_page_state(NR_INACTIVE
),
1878 global_page_state(NR_FILE_DIRTY
),
1879 global_page_state(NR_WRITEBACK
),
1880 global_page_state(NR_UNSTABLE_NFS
),
1881 global_page_state(NR_FREE_PAGES
),
1882 global_page_state(NR_SLAB_RECLAIMABLE
) +
1883 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1884 global_page_state(NR_FILE_MAPPED
),
1885 global_page_state(NR_PAGETABLE
),
1886 global_page_state(NR_BOUNCE
));
1888 for_each_zone(zone
) {
1891 if (!populated_zone(zone
))
1903 " pages_scanned:%lu"
1904 " all_unreclaimable? %s"
1907 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1910 K(zone
->pages_high
),
1911 K(zone_page_state(zone
, NR_ACTIVE
)),
1912 K(zone_page_state(zone
, NR_INACTIVE
)),
1913 K(zone
->present_pages
),
1914 zone
->pages_scanned
,
1915 (zone
->all_unreclaimable
? "yes" : "no")
1917 printk("lowmem_reserve[]:");
1918 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1919 printk(" %lu", zone
->lowmem_reserve
[i
]);
1923 for_each_zone(zone
) {
1924 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1926 if (!populated_zone(zone
))
1930 printk("%s: ", zone
->name
);
1932 spin_lock_irqsave(&zone
->lock
, flags
);
1933 for (order
= 0; order
< MAX_ORDER
; order
++) {
1934 nr
[order
] = zone
->free_area
[order
].nr_free
;
1935 total
+= nr
[order
] << order
;
1937 spin_unlock_irqrestore(&zone
->lock
, flags
);
1938 for (order
= 0; order
< MAX_ORDER
; order
++)
1939 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1940 printk("= %lukB\n", K(total
));
1943 show_swap_cache_info();
1947 * Builds allocation fallback zone lists.
1949 * Add all populated zones of a node to the zonelist.
1951 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1952 int nr_zones
, enum zone_type zone_type
)
1956 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1961 zone
= pgdat
->node_zones
+ zone_type
;
1962 if (populated_zone(zone
)) {
1963 zonelist
->zones
[nr_zones
++] = zone
;
1964 check_highest_zone(zone_type
);
1967 } while (zone_type
);
1974 * 0 = automatic detection of better ordering.
1975 * 1 = order by ([node] distance, -zonetype)
1976 * 2 = order by (-zonetype, [node] distance)
1978 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1979 * the same zonelist. So only NUMA can configure this param.
1981 #define ZONELIST_ORDER_DEFAULT 0
1982 #define ZONELIST_ORDER_NODE 1
1983 #define ZONELIST_ORDER_ZONE 2
1985 /* zonelist order in the kernel.
1986 * set_zonelist_order() will set this to NODE or ZONE.
1988 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1989 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
1993 /* The value user specified ....changed by config */
1994 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1995 /* string for sysctl */
1996 #define NUMA_ZONELIST_ORDER_LEN 16
1997 char numa_zonelist_order
[16] = "default";
2000 * interface for configure zonelist ordering.
2001 * command line option "numa_zonelist_order"
2002 * = "[dD]efault - default, automatic configuration.
2003 * = "[nN]ode - order by node locality, then by zone within node
2004 * = "[zZ]one - order by zone, then by locality within zone
2007 static int __parse_numa_zonelist_order(char *s
)
2009 if (*s
== 'd' || *s
== 'D') {
2010 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2011 } else if (*s
== 'n' || *s
== 'N') {
2012 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2013 } else if (*s
== 'z' || *s
== 'Z') {
2014 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2017 "Ignoring invalid numa_zonelist_order value: "
2024 static __init
int setup_numa_zonelist_order(char *s
)
2027 return __parse_numa_zonelist_order(s
);
2030 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2033 * sysctl handler for numa_zonelist_order
2035 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2036 struct file
*file
, void __user
*buffer
, size_t *length
,
2039 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2043 strncpy(saved_string
, (char*)table
->data
,
2044 NUMA_ZONELIST_ORDER_LEN
);
2045 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2049 int oldval
= user_zonelist_order
;
2050 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2052 * bogus value. restore saved string
2054 strncpy((char*)table
->data
, saved_string
,
2055 NUMA_ZONELIST_ORDER_LEN
);
2056 user_zonelist_order
= oldval
;
2057 } else if (oldval
!= user_zonelist_order
)
2058 build_all_zonelists();
2064 #define MAX_NODE_LOAD (num_online_nodes())
2065 static int node_load
[MAX_NUMNODES
];
2068 * find_next_best_node - find the next node that should appear in a given node's fallback list
2069 * @node: node whose fallback list we're appending
2070 * @used_node_mask: nodemask_t of already used nodes
2072 * We use a number of factors to determine which is the next node that should
2073 * appear on a given node's fallback list. The node should not have appeared
2074 * already in @node's fallback list, and it should be the next closest node
2075 * according to the distance array (which contains arbitrary distance values
2076 * from each node to each node in the system), and should also prefer nodes
2077 * with no CPUs, since presumably they'll have very little allocation pressure
2078 * on them otherwise.
2079 * It returns -1 if no node is found.
2081 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2084 int min_val
= INT_MAX
;
2087 /* Use the local node if we haven't already */
2088 if (!node_isset(node
, *used_node_mask
)) {
2089 node_set(node
, *used_node_mask
);
2093 for_each_node_state(n
, N_HIGH_MEMORY
) {
2096 /* Don't want a node to appear more than once */
2097 if (node_isset(n
, *used_node_mask
))
2100 /* Use the distance array to find the distance */
2101 val
= node_distance(node
, n
);
2103 /* Penalize nodes under us ("prefer the next node") */
2106 /* Give preference to headless and unused nodes */
2107 tmp
= node_to_cpumask(n
);
2108 if (!cpus_empty(tmp
))
2109 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2111 /* Slight preference for less loaded node */
2112 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2113 val
+= node_load
[n
];
2115 if (val
< min_val
) {
2122 node_set(best_node
, *used_node_mask
);
2129 * Build zonelists ordered by node and zones within node.
2130 * This results in maximum locality--normal zone overflows into local
2131 * DMA zone, if any--but risks exhausting DMA zone.
2133 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2137 struct zonelist
*zonelist
;
2139 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2140 zonelist
= pgdat
->node_zonelists
+ i
;
2141 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++)
2143 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2144 zonelist
->zones
[j
] = NULL
;
2149 * Build gfp_thisnode zonelists
2151 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2155 struct zonelist
*zonelist
;
2157 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2158 zonelist
= pgdat
->node_zonelists
+ MAX_NR_ZONES
+ i
;
2159 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2160 zonelist
->zones
[j
] = NULL
;
2165 * Build zonelists ordered by zone and nodes within zones.
2166 * This results in conserving DMA zone[s] until all Normal memory is
2167 * exhausted, but results in overflowing to remote node while memory
2168 * may still exist in local DMA zone.
2170 static int node_order
[MAX_NUMNODES
];
2172 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2176 int zone_type
; /* needs to be signed */
2178 struct zonelist
*zonelist
;
2180 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2181 zonelist
= pgdat
->node_zonelists
+ i
;
2183 for (zone_type
= i
; zone_type
>= 0; zone_type
--) {
2184 for (j
= 0; j
< nr_nodes
; j
++) {
2185 node
= node_order
[j
];
2186 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2187 if (populated_zone(z
)) {
2188 zonelist
->zones
[pos
++] = z
;
2189 check_highest_zone(zone_type
);
2193 zonelist
->zones
[pos
] = NULL
;
2197 static int default_zonelist_order(void)
2200 unsigned long low_kmem_size
,total_size
;
2204 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2205 * If they are really small and used heavily, the system can fall
2206 * into OOM very easily.
2207 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2209 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2212 for_each_online_node(nid
) {
2213 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2214 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2215 if (populated_zone(z
)) {
2216 if (zone_type
< ZONE_NORMAL
)
2217 low_kmem_size
+= z
->present_pages
;
2218 total_size
+= z
->present_pages
;
2222 if (!low_kmem_size
|| /* there are no DMA area. */
2223 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2224 return ZONELIST_ORDER_NODE
;
2226 * look into each node's config.
2227 * If there is a node whose DMA/DMA32 memory is very big area on
2228 * local memory, NODE_ORDER may be suitable.
2230 average_size
= total_size
/
2231 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2232 for_each_online_node(nid
) {
2235 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2236 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2237 if (populated_zone(z
)) {
2238 if (zone_type
< ZONE_NORMAL
)
2239 low_kmem_size
+= z
->present_pages
;
2240 total_size
+= z
->present_pages
;
2243 if (low_kmem_size
&&
2244 total_size
> average_size
&& /* ignore small node */
2245 low_kmem_size
> total_size
* 70/100)
2246 return ZONELIST_ORDER_NODE
;
2248 return ZONELIST_ORDER_ZONE
;
2251 static void set_zonelist_order(void)
2253 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2254 current_zonelist_order
= default_zonelist_order();
2256 current_zonelist_order
= user_zonelist_order
;
2259 static void build_zonelists(pg_data_t
*pgdat
)
2263 nodemask_t used_mask
;
2264 int local_node
, prev_node
;
2265 struct zonelist
*zonelist
;
2266 int order
= current_zonelist_order
;
2268 /* initialize zonelists */
2269 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2270 zonelist
= pgdat
->node_zonelists
+ i
;
2271 zonelist
->zones
[0] = NULL
;
2274 /* NUMA-aware ordering of nodes */
2275 local_node
= pgdat
->node_id
;
2276 load
= num_online_nodes();
2277 prev_node
= local_node
;
2278 nodes_clear(used_mask
);
2280 memset(node_load
, 0, sizeof(node_load
));
2281 memset(node_order
, 0, sizeof(node_order
));
2284 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2285 int distance
= node_distance(local_node
, node
);
2288 * If another node is sufficiently far away then it is better
2289 * to reclaim pages in a zone before going off node.
2291 if (distance
> RECLAIM_DISTANCE
)
2292 zone_reclaim_mode
= 1;
2295 * We don't want to pressure a particular node.
2296 * So adding penalty to the first node in same
2297 * distance group to make it round-robin.
2299 if (distance
!= node_distance(local_node
, prev_node
))
2300 node_load
[node
] = load
;
2304 if (order
== ZONELIST_ORDER_NODE
)
2305 build_zonelists_in_node_order(pgdat
, node
);
2307 node_order
[j
++] = node
; /* remember order */
2310 if (order
== ZONELIST_ORDER_ZONE
) {
2311 /* calculate node order -- i.e., DMA last! */
2312 build_zonelists_in_zone_order(pgdat
, j
);
2315 build_thisnode_zonelists(pgdat
);
2318 /* Construct the zonelist performance cache - see further mmzone.h */
2319 static void build_zonelist_cache(pg_data_t
*pgdat
)
2323 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2324 struct zonelist
*zonelist
;
2325 struct zonelist_cache
*zlc
;
2328 zonelist
= pgdat
->node_zonelists
+ i
;
2329 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2330 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2331 for (z
= zonelist
->zones
; *z
; z
++)
2332 zlc
->z_to_n
[z
- zonelist
->zones
] = zone_to_nid(*z
);
2337 #else /* CONFIG_NUMA */
2339 static void set_zonelist_order(void)
2341 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2344 static void build_zonelists(pg_data_t
*pgdat
)
2346 int node
, local_node
;
2349 local_node
= pgdat
->node_id
;
2350 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2351 struct zonelist
*zonelist
;
2353 zonelist
= pgdat
->node_zonelists
+ i
;
2355 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2357 * Now we build the zonelist so that it contains the zones
2358 * of all the other nodes.
2359 * We don't want to pressure a particular node, so when
2360 * building the zones for node N, we make sure that the
2361 * zones coming right after the local ones are those from
2362 * node N+1 (modulo N)
2364 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2365 if (!node_online(node
))
2367 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2369 for (node
= 0; node
< local_node
; node
++) {
2370 if (!node_online(node
))
2372 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2375 zonelist
->zones
[j
] = NULL
;
2379 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2380 static void build_zonelist_cache(pg_data_t
*pgdat
)
2384 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2385 pgdat
->node_zonelists
[i
].zlcache_ptr
= NULL
;
2388 #endif /* CONFIG_NUMA */
2390 /* return values int ....just for stop_machine_run() */
2391 static int __build_all_zonelists(void *dummy
)
2395 for_each_online_node(nid
) {
2396 pg_data_t
*pgdat
= NODE_DATA(nid
);
2398 build_zonelists(pgdat
);
2399 build_zonelist_cache(pgdat
);
2404 void build_all_zonelists(void)
2406 set_zonelist_order();
2408 if (system_state
== SYSTEM_BOOTING
) {
2409 __build_all_zonelists(NULL
);
2410 cpuset_init_current_mems_allowed();
2412 /* we have to stop all cpus to guaranntee there is no user
2414 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
2415 /* cpuset refresh routine should be here */
2417 vm_total_pages
= nr_free_pagecache_pages();
2419 * Disable grouping by mobility if the number of pages in the
2420 * system is too low to allow the mechanism to work. It would be
2421 * more accurate, but expensive to check per-zone. This check is
2422 * made on memory-hotadd so a system can start with mobility
2423 * disabled and enable it later
2425 if (vm_total_pages
< (MAX_ORDER_NR_PAGES
* MIGRATE_TYPES
))
2426 page_group_by_mobility_disabled
= 1;
2428 page_group_by_mobility_disabled
= 0;
2430 printk("Built %i zonelists in %s order, mobility grouping %s. "
2431 "Total pages: %ld\n",
2433 zonelist_order_name
[current_zonelist_order
],
2434 page_group_by_mobility_disabled
? "off" : "on",
2437 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2442 * Helper functions to size the waitqueue hash table.
2443 * Essentially these want to choose hash table sizes sufficiently
2444 * large so that collisions trying to wait on pages are rare.
2445 * But in fact, the number of active page waitqueues on typical
2446 * systems is ridiculously low, less than 200. So this is even
2447 * conservative, even though it seems large.
2449 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2450 * waitqueues, i.e. the size of the waitq table given the number of pages.
2452 #define PAGES_PER_WAITQUEUE 256
2454 #ifndef CONFIG_MEMORY_HOTPLUG
2455 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2457 unsigned long size
= 1;
2459 pages
/= PAGES_PER_WAITQUEUE
;
2461 while (size
< pages
)
2465 * Once we have dozens or even hundreds of threads sleeping
2466 * on IO we've got bigger problems than wait queue collision.
2467 * Limit the size of the wait table to a reasonable size.
2469 size
= min(size
, 4096UL);
2471 return max(size
, 4UL);
2475 * A zone's size might be changed by hot-add, so it is not possible to determine
2476 * a suitable size for its wait_table. So we use the maximum size now.
2478 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2480 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2481 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2482 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2484 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2485 * or more by the traditional way. (See above). It equals:
2487 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2488 * ia64(16K page size) : = ( 8G + 4M)byte.
2489 * powerpc (64K page size) : = (32G +16M)byte.
2491 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2498 * This is an integer logarithm so that shifts can be used later
2499 * to extract the more random high bits from the multiplicative
2500 * hash function before the remainder is taken.
2502 static inline unsigned long wait_table_bits(unsigned long size
)
2507 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2510 * Initially all pages are reserved - free ones are freed
2511 * up by free_all_bootmem() once the early boot process is
2512 * done. Non-atomic initialization, single-pass.
2514 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2515 unsigned long start_pfn
, enum memmap_context context
)
2518 unsigned long end_pfn
= start_pfn
+ size
;
2521 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2523 * There can be holes in boot-time mem_map[]s
2524 * handed to this function. They do not
2525 * exist on hotplugged memory.
2527 if (context
== MEMMAP_EARLY
) {
2528 if (!early_pfn_valid(pfn
))
2530 if (!early_pfn_in_nid(pfn
, nid
))
2533 page
= pfn_to_page(pfn
);
2534 set_page_links(page
, zone
, nid
, pfn
);
2535 init_page_count(page
);
2536 reset_page_mapcount(page
);
2537 SetPageReserved(page
);
2540 * Mark the block movable so that blocks are reserved for
2541 * movable at startup. This will force kernel allocations
2542 * to reserve their blocks rather than leaking throughout
2543 * the address space during boot when many long-lived
2544 * kernel allocations are made
2546 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2548 INIT_LIST_HEAD(&page
->lru
);
2549 #ifdef WANT_PAGE_VIRTUAL
2550 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2551 if (!is_highmem_idx(zone
))
2552 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2557 static void __meminit
zone_init_free_lists(struct pglist_data
*pgdat
,
2558 struct zone
*zone
, unsigned long size
)
2561 for_each_migratetype_order(order
, t
) {
2562 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2563 zone
->free_area
[order
].nr_free
= 0;
2567 #ifndef __HAVE_ARCH_MEMMAP_INIT
2568 #define memmap_init(size, nid, zone, start_pfn) \
2569 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2572 static int __devinit
zone_batchsize(struct zone
*zone
)
2577 * The per-cpu-pages pools are set to around 1000th of the
2578 * size of the zone. But no more than 1/2 of a meg.
2580 * OK, so we don't know how big the cache is. So guess.
2582 batch
= zone
->present_pages
/ 1024;
2583 if (batch
* PAGE_SIZE
> 512 * 1024)
2584 batch
= (512 * 1024) / PAGE_SIZE
;
2585 batch
/= 4; /* We effectively *= 4 below */
2590 * Clamp the batch to a 2^n - 1 value. Having a power
2591 * of 2 value was found to be more likely to have
2592 * suboptimal cache aliasing properties in some cases.
2594 * For example if 2 tasks are alternately allocating
2595 * batches of pages, one task can end up with a lot
2596 * of pages of one half of the possible page colors
2597 * and the other with pages of the other colors.
2599 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2604 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2606 struct per_cpu_pages
*pcp
;
2608 memset(p
, 0, sizeof(*p
));
2610 pcp
= &p
->pcp
[0]; /* hot */
2612 pcp
->high
= 6 * batch
;
2613 pcp
->batch
= max(1UL, 1 * batch
);
2614 INIT_LIST_HEAD(&pcp
->list
);
2616 pcp
= &p
->pcp
[1]; /* cold*/
2618 pcp
->high
= 2 * batch
;
2619 pcp
->batch
= max(1UL, batch
/2);
2620 INIT_LIST_HEAD(&pcp
->list
);
2624 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2625 * to the value high for the pageset p.
2628 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2631 struct per_cpu_pages
*pcp
;
2633 pcp
= &p
->pcp
[0]; /* hot list */
2635 pcp
->batch
= max(1UL, high
/4);
2636 if ((high
/4) > (PAGE_SHIFT
* 8))
2637 pcp
->batch
= PAGE_SHIFT
* 8;
2643 * Boot pageset table. One per cpu which is going to be used for all
2644 * zones and all nodes. The parameters will be set in such a way
2645 * that an item put on a list will immediately be handed over to
2646 * the buddy list. This is safe since pageset manipulation is done
2647 * with interrupts disabled.
2649 * Some NUMA counter updates may also be caught by the boot pagesets.
2651 * The boot_pagesets must be kept even after bootup is complete for
2652 * unused processors and/or zones. They do play a role for bootstrapping
2653 * hotplugged processors.
2655 * zoneinfo_show() and maybe other functions do
2656 * not check if the processor is online before following the pageset pointer.
2657 * Other parts of the kernel may not check if the zone is available.
2659 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2662 * Dynamically allocate memory for the
2663 * per cpu pageset array in struct zone.
2665 static int __cpuinit
process_zones(int cpu
)
2667 struct zone
*zone
, *dzone
;
2668 int node
= cpu_to_node(cpu
);
2670 node_set_state(node
, N_CPU
); /* this node has a cpu */
2672 for_each_zone(zone
) {
2674 if (!populated_zone(zone
))
2677 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2679 if (!zone_pcp(zone
, cpu
))
2682 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2684 if (percpu_pagelist_fraction
)
2685 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2686 (zone
->present_pages
/ percpu_pagelist_fraction
));
2691 for_each_zone(dzone
) {
2692 if (!populated_zone(dzone
))
2696 kfree(zone_pcp(dzone
, cpu
));
2697 zone_pcp(dzone
, cpu
) = NULL
;
2702 static inline void free_zone_pagesets(int cpu
)
2706 for_each_zone(zone
) {
2707 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2709 /* Free per_cpu_pageset if it is slab allocated */
2710 if (pset
!= &boot_pageset
[cpu
])
2712 zone_pcp(zone
, cpu
) = NULL
;
2716 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2717 unsigned long action
,
2720 int cpu
= (long)hcpu
;
2721 int ret
= NOTIFY_OK
;
2724 case CPU_UP_PREPARE
:
2725 case CPU_UP_PREPARE_FROZEN
:
2726 if (process_zones(cpu
))
2729 case CPU_UP_CANCELED
:
2730 case CPU_UP_CANCELED_FROZEN
:
2732 case CPU_DEAD_FROZEN
:
2733 free_zone_pagesets(cpu
);
2741 static struct notifier_block __cpuinitdata pageset_notifier
=
2742 { &pageset_cpuup_callback
, NULL
, 0 };
2744 void __init
setup_per_cpu_pageset(void)
2748 /* Initialize per_cpu_pageset for cpu 0.
2749 * A cpuup callback will do this for every cpu
2750 * as it comes online
2752 err
= process_zones(smp_processor_id());
2754 register_cpu_notifier(&pageset_notifier
);
2759 static noinline __init_refok
2760 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2763 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2767 * The per-page waitqueue mechanism uses hashed waitqueues
2770 zone
->wait_table_hash_nr_entries
=
2771 wait_table_hash_nr_entries(zone_size_pages
);
2772 zone
->wait_table_bits
=
2773 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2774 alloc_size
= zone
->wait_table_hash_nr_entries
2775 * sizeof(wait_queue_head_t
);
2777 if (system_state
== SYSTEM_BOOTING
) {
2778 zone
->wait_table
= (wait_queue_head_t
*)
2779 alloc_bootmem_node(pgdat
, alloc_size
);
2782 * This case means that a zone whose size was 0 gets new memory
2783 * via memory hot-add.
2784 * But it may be the case that a new node was hot-added. In
2785 * this case vmalloc() will not be able to use this new node's
2786 * memory - this wait_table must be initialized to use this new
2787 * node itself as well.
2788 * To use this new node's memory, further consideration will be
2791 zone
->wait_table
= vmalloc(alloc_size
);
2793 if (!zone
->wait_table
)
2796 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2797 init_waitqueue_head(zone
->wait_table
+ i
);
2802 static __meminit
void zone_pcp_init(struct zone
*zone
)
2805 unsigned long batch
= zone_batchsize(zone
);
2807 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2809 /* Early boot. Slab allocator not functional yet */
2810 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2811 setup_pageset(&boot_pageset
[cpu
],0);
2813 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2816 if (zone
->present_pages
)
2817 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2818 zone
->name
, zone
->present_pages
, batch
);
2821 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2822 unsigned long zone_start_pfn
,
2824 enum memmap_context context
)
2826 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2828 ret
= zone_wait_table_init(zone
, size
);
2831 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2833 zone
->zone_start_pfn
= zone_start_pfn
;
2835 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
2837 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2842 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2844 * Basic iterator support. Return the first range of PFNs for a node
2845 * Note: nid == MAX_NUMNODES returns first region regardless of node
2847 static int __meminit
first_active_region_index_in_nid(int nid
)
2851 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2852 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2859 * Basic iterator support. Return the next active range of PFNs for a node
2860 * Note: nid == MAX_NUMNODES returns next region regardles of node
2862 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2864 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2865 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2871 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2873 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2874 * Architectures may implement their own version but if add_active_range()
2875 * was used and there are no special requirements, this is a convenient
2878 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2882 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2883 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2884 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2886 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2887 return early_node_map
[i
].nid
;
2892 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2894 /* Basic iterator support to walk early_node_map[] */
2895 #define for_each_active_range_index_in_nid(i, nid) \
2896 for (i = first_active_region_index_in_nid(nid); i != -1; \
2897 i = next_active_region_index_in_nid(i, nid))
2900 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2901 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2902 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2904 * If an architecture guarantees that all ranges registered with
2905 * add_active_ranges() contain no holes and may be freed, this
2906 * this function may be used instead of calling free_bootmem() manually.
2908 void __init
free_bootmem_with_active_regions(int nid
,
2909 unsigned long max_low_pfn
)
2913 for_each_active_range_index_in_nid(i
, nid
) {
2914 unsigned long size_pages
= 0;
2915 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2917 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2920 if (end_pfn
> max_low_pfn
)
2921 end_pfn
= max_low_pfn
;
2923 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2924 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2925 PFN_PHYS(early_node_map
[i
].start_pfn
),
2926 size_pages
<< PAGE_SHIFT
);
2931 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2932 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2934 * If an architecture guarantees that all ranges registered with
2935 * add_active_ranges() contain no holes and may be freed, this
2936 * function may be used instead of calling memory_present() manually.
2938 void __init
sparse_memory_present_with_active_regions(int nid
)
2942 for_each_active_range_index_in_nid(i
, nid
)
2943 memory_present(early_node_map
[i
].nid
,
2944 early_node_map
[i
].start_pfn
,
2945 early_node_map
[i
].end_pfn
);
2949 * push_node_boundaries - Push node boundaries to at least the requested boundary
2950 * @nid: The nid of the node to push the boundary for
2951 * @start_pfn: The start pfn of the node
2952 * @end_pfn: The end pfn of the node
2954 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2955 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2956 * be hotplugged even though no physical memory exists. This function allows
2957 * an arch to push out the node boundaries so mem_map is allocated that can
2960 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2961 void __init
push_node_boundaries(unsigned int nid
,
2962 unsigned long start_pfn
, unsigned long end_pfn
)
2964 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2965 nid
, start_pfn
, end_pfn
);
2967 /* Initialise the boundary for this node if necessary */
2968 if (node_boundary_end_pfn
[nid
] == 0)
2969 node_boundary_start_pfn
[nid
] = -1UL;
2971 /* Update the boundaries */
2972 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2973 node_boundary_start_pfn
[nid
] = start_pfn
;
2974 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2975 node_boundary_end_pfn
[nid
] = end_pfn
;
2978 /* If necessary, push the node boundary out for reserve hotadd */
2979 static void __meminit
account_node_boundary(unsigned int nid
,
2980 unsigned long *start_pfn
, unsigned long *end_pfn
)
2982 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2983 nid
, *start_pfn
, *end_pfn
);
2985 /* Return if boundary information has not been provided */
2986 if (node_boundary_end_pfn
[nid
] == 0)
2989 /* Check the boundaries and update if necessary */
2990 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
2991 *start_pfn
= node_boundary_start_pfn
[nid
];
2992 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
2993 *end_pfn
= node_boundary_end_pfn
[nid
];
2996 void __init
push_node_boundaries(unsigned int nid
,
2997 unsigned long start_pfn
, unsigned long end_pfn
) {}
2999 static void __meminit
account_node_boundary(unsigned int nid
,
3000 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3005 * get_pfn_range_for_nid - Return the start and end page frames for a node
3006 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3007 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3008 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3010 * It returns the start and end page frame of a node based on information
3011 * provided by an arch calling add_active_range(). If called for a node
3012 * with no available memory, a warning is printed and the start and end
3015 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3016 unsigned long *start_pfn
, unsigned long *end_pfn
)
3022 for_each_active_range_index_in_nid(i
, nid
) {
3023 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3024 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3027 if (*start_pfn
== -1UL)
3030 /* Push the node boundaries out if requested */
3031 account_node_boundary(nid
, start_pfn
, end_pfn
);
3035 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3036 * assumption is made that zones within a node are ordered in monotonic
3037 * increasing memory addresses so that the "highest" populated zone is used
3039 void __init
find_usable_zone_for_movable(void)
3042 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3043 if (zone_index
== ZONE_MOVABLE
)
3046 if (arch_zone_highest_possible_pfn
[zone_index
] >
3047 arch_zone_lowest_possible_pfn
[zone_index
])
3051 VM_BUG_ON(zone_index
== -1);
3052 movable_zone
= zone_index
;
3056 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3057 * because it is sized independant of architecture. Unlike the other zones,
3058 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3059 * in each node depending on the size of each node and how evenly kernelcore
3060 * is distributed. This helper function adjusts the zone ranges
3061 * provided by the architecture for a given node by using the end of the
3062 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3063 * zones within a node are in order of monotonic increases memory addresses
3065 void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3066 unsigned long zone_type
,
3067 unsigned long node_start_pfn
,
3068 unsigned long node_end_pfn
,
3069 unsigned long *zone_start_pfn
,
3070 unsigned long *zone_end_pfn
)
3072 /* Only adjust if ZONE_MOVABLE is on this node */
3073 if (zone_movable_pfn
[nid
]) {
3074 /* Size ZONE_MOVABLE */
3075 if (zone_type
== ZONE_MOVABLE
) {
3076 *zone_start_pfn
= zone_movable_pfn
[nid
];
3077 *zone_end_pfn
= min(node_end_pfn
,
3078 arch_zone_highest_possible_pfn
[movable_zone
]);
3080 /* Adjust for ZONE_MOVABLE starting within this range */
3081 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3082 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3083 *zone_end_pfn
= zone_movable_pfn
[nid
];
3085 /* Check if this whole range is within ZONE_MOVABLE */
3086 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3087 *zone_start_pfn
= *zone_end_pfn
;
3092 * Return the number of pages a zone spans in a node, including holes
3093 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3095 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3096 unsigned long zone_type
,
3097 unsigned long *ignored
)
3099 unsigned long node_start_pfn
, node_end_pfn
;
3100 unsigned long zone_start_pfn
, zone_end_pfn
;
3102 /* Get the start and end of the node and zone */
3103 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3104 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3105 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3106 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3107 node_start_pfn
, node_end_pfn
,
3108 &zone_start_pfn
, &zone_end_pfn
);
3110 /* Check that this node has pages within the zone's required range */
3111 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3114 /* Move the zone boundaries inside the node if necessary */
3115 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3116 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3118 /* Return the spanned pages */
3119 return zone_end_pfn
- zone_start_pfn
;
3123 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3124 * then all holes in the requested range will be accounted for.
3126 unsigned long __meminit
__absent_pages_in_range(int nid
,
3127 unsigned long range_start_pfn
,
3128 unsigned long range_end_pfn
)
3131 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3132 unsigned long start_pfn
;
3134 /* Find the end_pfn of the first active range of pfns in the node */
3135 i
= first_active_region_index_in_nid(nid
);
3139 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3141 /* Account for ranges before physical memory on this node */
3142 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3143 hole_pages
= prev_end_pfn
- range_start_pfn
;
3145 /* Find all holes for the zone within the node */
3146 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3148 /* No need to continue if prev_end_pfn is outside the zone */
3149 if (prev_end_pfn
>= range_end_pfn
)
3152 /* Make sure the end of the zone is not within the hole */
3153 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3154 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3156 /* Update the hole size cound and move on */
3157 if (start_pfn
> range_start_pfn
) {
3158 BUG_ON(prev_end_pfn
> start_pfn
);
3159 hole_pages
+= start_pfn
- prev_end_pfn
;
3161 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3164 /* Account for ranges past physical memory on this node */
3165 if (range_end_pfn
> prev_end_pfn
)
3166 hole_pages
+= range_end_pfn
-
3167 max(range_start_pfn
, prev_end_pfn
);
3173 * absent_pages_in_range - Return number of page frames in holes within a range
3174 * @start_pfn: The start PFN to start searching for holes
3175 * @end_pfn: The end PFN to stop searching for holes
3177 * It returns the number of pages frames in memory holes within a range.
3179 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3180 unsigned long end_pfn
)
3182 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3185 /* Return the number of page frames in holes in a zone on a node */
3186 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3187 unsigned long zone_type
,
3188 unsigned long *ignored
)
3190 unsigned long node_start_pfn
, node_end_pfn
;
3191 unsigned long zone_start_pfn
, zone_end_pfn
;
3193 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3194 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3196 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3199 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3200 node_start_pfn
, node_end_pfn
,
3201 &zone_start_pfn
, &zone_end_pfn
);
3202 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3206 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3207 unsigned long zone_type
,
3208 unsigned long *zones_size
)
3210 return zones_size
[zone_type
];
3213 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3214 unsigned long zone_type
,
3215 unsigned long *zholes_size
)
3220 return zholes_size
[zone_type
];
3225 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3226 unsigned long *zones_size
, unsigned long *zholes_size
)
3228 unsigned long realtotalpages
, totalpages
= 0;
3231 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3232 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3234 pgdat
->node_spanned_pages
= totalpages
;
3236 realtotalpages
= totalpages
;
3237 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3239 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3241 pgdat
->node_present_pages
= realtotalpages
;
3242 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3246 #ifndef CONFIG_SPARSEMEM
3248 * Calculate the size of the zone->blockflags rounded to an unsigned long
3249 * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up
3250 * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally
3251 * round what is now in bits to nearest long in bits, then return it in
3254 static unsigned long __init
usemap_size(unsigned long zonesize
)
3256 unsigned long usemapsize
;
3258 usemapsize
= roundup(zonesize
, MAX_ORDER_NR_PAGES
);
3259 usemapsize
= usemapsize
>> (MAX_ORDER
-1);
3260 usemapsize
*= NR_PAGEBLOCK_BITS
;
3261 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3263 return usemapsize
/ 8;
3266 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3267 struct zone
*zone
, unsigned long zonesize
)
3269 unsigned long usemapsize
= usemap_size(zonesize
);
3270 zone
->pageblock_flags
= NULL
;
3272 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3273 memset(zone
->pageblock_flags
, 0, usemapsize
);
3277 static void inline setup_usemap(struct pglist_data
*pgdat
,
3278 struct zone
*zone
, unsigned long zonesize
) {}
3279 #endif /* CONFIG_SPARSEMEM */
3282 * Set up the zone data structures:
3283 * - mark all pages reserved
3284 * - mark all memory queues empty
3285 * - clear the memory bitmaps
3287 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
3288 unsigned long *zones_size
, unsigned long *zholes_size
)
3291 int nid
= pgdat
->node_id
;
3292 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3295 pgdat_resize_init(pgdat
);
3296 pgdat
->nr_zones
= 0;
3297 init_waitqueue_head(&pgdat
->kswapd_wait
);
3298 pgdat
->kswapd_max_order
= 0;
3300 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3301 struct zone
*zone
= pgdat
->node_zones
+ j
;
3302 unsigned long size
, realsize
, memmap_pages
;
3304 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3305 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3309 * Adjust realsize so that it accounts for how much memory
3310 * is used by this zone for memmap. This affects the watermark
3311 * and per-cpu initialisations
3313 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3314 if (realsize
>= memmap_pages
) {
3315 realsize
-= memmap_pages
;
3317 " %s zone: %lu pages used for memmap\n",
3318 zone_names
[j
], memmap_pages
);
3321 " %s zone: %lu pages exceeds realsize %lu\n",
3322 zone_names
[j
], memmap_pages
, realsize
);
3324 /* Account for reserved pages */
3325 if (j
== 0 && realsize
> dma_reserve
) {
3326 realsize
-= dma_reserve
;
3327 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3328 zone_names
[0], dma_reserve
);
3331 if (!is_highmem_idx(j
))
3332 nr_kernel_pages
+= realsize
;
3333 nr_all_pages
+= realsize
;
3335 zone
->spanned_pages
= size
;
3336 zone
->present_pages
= realsize
;
3339 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3341 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3343 zone
->name
= zone_names
[j
];
3344 spin_lock_init(&zone
->lock
);
3345 spin_lock_init(&zone
->lru_lock
);
3346 zone_seqlock_init(zone
);
3347 zone
->zone_pgdat
= pgdat
;
3349 zone
->prev_priority
= DEF_PRIORITY
;
3351 zone_pcp_init(zone
);
3352 INIT_LIST_HEAD(&zone
->active_list
);
3353 INIT_LIST_HEAD(&zone
->inactive_list
);
3354 zone
->nr_scan_active
= 0;
3355 zone
->nr_scan_inactive
= 0;
3356 zap_zone_vm_stats(zone
);
3357 atomic_set(&zone
->reclaim_in_progress
, 0);
3361 setup_usemap(pgdat
, zone
, size
);
3362 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3363 size
, MEMMAP_EARLY
);
3365 zone_start_pfn
+= size
;
3369 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3371 /* Skip empty nodes */
3372 if (!pgdat
->node_spanned_pages
)
3375 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3376 /* ia64 gets its own node_mem_map, before this, without bootmem */
3377 if (!pgdat
->node_mem_map
) {
3378 unsigned long size
, start
, end
;
3382 * The zone's endpoints aren't required to be MAX_ORDER
3383 * aligned but the node_mem_map endpoints must be in order
3384 * for the buddy allocator to function correctly.
3386 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3387 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3388 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3389 size
= (end
- start
) * sizeof(struct page
);
3390 map
= alloc_remap(pgdat
->node_id
, size
);
3392 map
= alloc_bootmem_node(pgdat
, size
);
3393 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3395 #ifndef CONFIG_NEED_MULTIPLE_NODES
3397 * With no DISCONTIG, the global mem_map is just set as node 0's
3399 if (pgdat
== NODE_DATA(0)) {
3400 mem_map
= NODE_DATA(0)->node_mem_map
;
3401 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3402 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3403 mem_map
-= pgdat
->node_start_pfn
;
3404 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3407 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3410 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
3411 unsigned long *zones_size
, unsigned long node_start_pfn
,
3412 unsigned long *zholes_size
)
3414 pgdat
->node_id
= nid
;
3415 pgdat
->node_start_pfn
= node_start_pfn
;
3416 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3418 alloc_node_mem_map(pgdat
);
3420 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3423 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3425 #if MAX_NUMNODES > 1
3427 * Figure out the number of possible node ids.
3429 static void __init
setup_nr_node_ids(void)
3432 unsigned int highest
= 0;
3434 for_each_node_mask(node
, node_possible_map
)
3436 nr_node_ids
= highest
+ 1;
3439 static inline void setup_nr_node_ids(void)
3445 * add_active_range - Register a range of PFNs backed by physical memory
3446 * @nid: The node ID the range resides on
3447 * @start_pfn: The start PFN of the available physical memory
3448 * @end_pfn: The end PFN of the available physical memory
3450 * These ranges are stored in an early_node_map[] and later used by
3451 * free_area_init_nodes() to calculate zone sizes and holes. If the
3452 * range spans a memory hole, it is up to the architecture to ensure
3453 * the memory is not freed by the bootmem allocator. If possible
3454 * the range being registered will be merged with existing ranges.
3456 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3457 unsigned long end_pfn
)
3461 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
3462 "%d entries of %d used\n",
3463 nid
, start_pfn
, end_pfn
,
3464 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3466 /* Merge with existing active regions if possible */
3467 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3468 if (early_node_map
[i
].nid
!= nid
)
3471 /* Skip if an existing region covers this new one */
3472 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3473 end_pfn
<= early_node_map
[i
].end_pfn
)
3476 /* Merge forward if suitable */
3477 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3478 end_pfn
> early_node_map
[i
].end_pfn
) {
3479 early_node_map
[i
].end_pfn
= end_pfn
;
3483 /* Merge backward if suitable */
3484 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3485 end_pfn
>= early_node_map
[i
].start_pfn
) {
3486 early_node_map
[i
].start_pfn
= start_pfn
;
3491 /* Check that early_node_map is large enough */
3492 if (i
>= MAX_ACTIVE_REGIONS
) {
3493 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3494 MAX_ACTIVE_REGIONS
);
3498 early_node_map
[i
].nid
= nid
;
3499 early_node_map
[i
].start_pfn
= start_pfn
;
3500 early_node_map
[i
].end_pfn
= end_pfn
;
3501 nr_nodemap_entries
= i
+ 1;
3505 * shrink_active_range - Shrink an existing registered range of PFNs
3506 * @nid: The node id the range is on that should be shrunk
3507 * @old_end_pfn: The old end PFN of the range
3508 * @new_end_pfn: The new PFN of the range
3510 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3511 * The map is kept at the end physical page range that has already been
3512 * registered with add_active_range(). This function allows an arch to shrink
3513 * an existing registered range.
3515 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
3516 unsigned long new_end_pfn
)
3520 /* Find the old active region end and shrink */
3521 for_each_active_range_index_in_nid(i
, nid
)
3522 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
3523 early_node_map
[i
].end_pfn
= new_end_pfn
;
3529 * remove_all_active_ranges - Remove all currently registered regions
3531 * During discovery, it may be found that a table like SRAT is invalid
3532 * and an alternative discovery method must be used. This function removes
3533 * all currently registered regions.
3535 void __init
remove_all_active_ranges(void)
3537 memset(early_node_map
, 0, sizeof(early_node_map
));
3538 nr_nodemap_entries
= 0;
3539 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3540 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3541 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3542 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3545 /* Compare two active node_active_regions */
3546 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3548 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3549 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3551 /* Done this way to avoid overflows */
3552 if (arange
->start_pfn
> brange
->start_pfn
)
3554 if (arange
->start_pfn
< brange
->start_pfn
)
3560 /* sort the node_map by start_pfn */
3561 static void __init
sort_node_map(void)
3563 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3564 sizeof(struct node_active_region
),
3565 cmp_node_active_region
, NULL
);
3568 /* Find the lowest pfn for a node */
3569 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
3572 unsigned long min_pfn
= ULONG_MAX
;
3574 /* Assuming a sorted map, the first range found has the starting pfn */
3575 for_each_active_range_index_in_nid(i
, nid
)
3576 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3578 if (min_pfn
== ULONG_MAX
) {
3580 "Could not find start_pfn for node %lu\n", nid
);
3588 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3590 * It returns the minimum PFN based on information provided via
3591 * add_active_range().
3593 unsigned long __init
find_min_pfn_with_active_regions(void)
3595 return find_min_pfn_for_node(MAX_NUMNODES
);
3599 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3601 * It returns the maximum PFN based on information provided via
3602 * add_active_range().
3604 unsigned long __init
find_max_pfn_with_active_regions(void)
3607 unsigned long max_pfn
= 0;
3609 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3610 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
3616 * early_calculate_totalpages()
3617 * Sum pages in active regions for movable zone.
3618 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3620 unsigned long __init
early_calculate_totalpages(void)
3623 unsigned long totalpages
= 0;
3625 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3626 unsigned long pages
= early_node_map
[i
].end_pfn
-
3627 early_node_map
[i
].start_pfn
;
3628 totalpages
+= pages
;
3630 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3636 * Find the PFN the Movable zone begins in each node. Kernel memory
3637 * is spread evenly between nodes as long as the nodes have enough
3638 * memory. When they don't, some nodes will have more kernelcore than
3641 void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3644 unsigned long usable_startpfn
;
3645 unsigned long kernelcore_node
, kernelcore_remaining
;
3646 unsigned long totalpages
= early_calculate_totalpages();
3647 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3650 * If movablecore was specified, calculate what size of
3651 * kernelcore that corresponds so that memory usable for
3652 * any allocation type is evenly spread. If both kernelcore
3653 * and movablecore are specified, then the value of kernelcore
3654 * will be used for required_kernelcore if it's greater than
3655 * what movablecore would have allowed.
3657 if (required_movablecore
) {
3658 unsigned long corepages
;
3661 * Round-up so that ZONE_MOVABLE is at least as large as what
3662 * was requested by the user
3664 required_movablecore
=
3665 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3666 corepages
= totalpages
- required_movablecore
;
3668 required_kernelcore
= max(required_kernelcore
, corepages
);
3671 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3672 if (!required_kernelcore
)
3675 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3676 find_usable_zone_for_movable();
3677 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3680 /* Spread kernelcore memory as evenly as possible throughout nodes */
3681 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3682 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3684 * Recalculate kernelcore_node if the division per node
3685 * now exceeds what is necessary to satisfy the requested
3686 * amount of memory for the kernel
3688 if (required_kernelcore
< kernelcore_node
)
3689 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3692 * As the map is walked, we track how much memory is usable
3693 * by the kernel using kernelcore_remaining. When it is
3694 * 0, the rest of the node is usable by ZONE_MOVABLE
3696 kernelcore_remaining
= kernelcore_node
;
3698 /* Go through each range of PFNs within this node */
3699 for_each_active_range_index_in_nid(i
, nid
) {
3700 unsigned long start_pfn
, end_pfn
;
3701 unsigned long size_pages
;
3703 start_pfn
= max(early_node_map
[i
].start_pfn
,
3704 zone_movable_pfn
[nid
]);
3705 end_pfn
= early_node_map
[i
].end_pfn
;
3706 if (start_pfn
>= end_pfn
)
3709 /* Account for what is only usable for kernelcore */
3710 if (start_pfn
< usable_startpfn
) {
3711 unsigned long kernel_pages
;
3712 kernel_pages
= min(end_pfn
, usable_startpfn
)
3715 kernelcore_remaining
-= min(kernel_pages
,
3716 kernelcore_remaining
);
3717 required_kernelcore
-= min(kernel_pages
,
3718 required_kernelcore
);
3720 /* Continue if range is now fully accounted */
3721 if (end_pfn
<= usable_startpfn
) {
3724 * Push zone_movable_pfn to the end so
3725 * that if we have to rebalance
3726 * kernelcore across nodes, we will
3727 * not double account here
3729 zone_movable_pfn
[nid
] = end_pfn
;
3732 start_pfn
= usable_startpfn
;
3736 * The usable PFN range for ZONE_MOVABLE is from
3737 * start_pfn->end_pfn. Calculate size_pages as the
3738 * number of pages used as kernelcore
3740 size_pages
= end_pfn
- start_pfn
;
3741 if (size_pages
> kernelcore_remaining
)
3742 size_pages
= kernelcore_remaining
;
3743 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3746 * Some kernelcore has been met, update counts and
3747 * break if the kernelcore for this node has been
3750 required_kernelcore
-= min(required_kernelcore
,
3752 kernelcore_remaining
-= size_pages
;
3753 if (!kernelcore_remaining
)
3759 * If there is still required_kernelcore, we do another pass with one
3760 * less node in the count. This will push zone_movable_pfn[nid] further
3761 * along on the nodes that still have memory until kernelcore is
3765 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3768 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3769 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3770 zone_movable_pfn
[nid
] =
3771 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3774 /* Any regular memory on that node ? */
3775 static void check_for_regular_memory(pg_data_t
*pgdat
)
3777 #ifdef CONFIG_HIGHMEM
3778 enum zone_type zone_type
;
3780 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3781 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3782 if (zone
->present_pages
)
3783 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3789 * free_area_init_nodes - Initialise all pg_data_t and zone data
3790 * @max_zone_pfn: an array of max PFNs for each zone
3792 * This will call free_area_init_node() for each active node in the system.
3793 * Using the page ranges provided by add_active_range(), the size of each
3794 * zone in each node and their holes is calculated. If the maximum PFN
3795 * between two adjacent zones match, it is assumed that the zone is empty.
3796 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3797 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3798 * starts where the previous one ended. For example, ZONE_DMA32 starts
3799 * at arch_max_dma_pfn.
3801 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3806 /* Sort early_node_map as initialisation assumes it is sorted */
3809 /* Record where the zone boundaries are */
3810 memset(arch_zone_lowest_possible_pfn
, 0,
3811 sizeof(arch_zone_lowest_possible_pfn
));
3812 memset(arch_zone_highest_possible_pfn
, 0,
3813 sizeof(arch_zone_highest_possible_pfn
));
3814 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3815 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3816 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3817 if (i
== ZONE_MOVABLE
)
3819 arch_zone_lowest_possible_pfn
[i
] =
3820 arch_zone_highest_possible_pfn
[i
-1];
3821 arch_zone_highest_possible_pfn
[i
] =
3822 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3824 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3825 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3827 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3828 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3829 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3831 /* Print out the zone ranges */
3832 printk("Zone PFN ranges:\n");
3833 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3834 if (i
== ZONE_MOVABLE
)
3836 printk(" %-8s %8lu -> %8lu\n",
3838 arch_zone_lowest_possible_pfn
[i
],
3839 arch_zone_highest_possible_pfn
[i
]);
3842 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3843 printk("Movable zone start PFN for each node\n");
3844 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
3845 if (zone_movable_pfn
[i
])
3846 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
3849 /* Print out the early_node_map[] */
3850 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
3851 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3852 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
3853 early_node_map
[i
].start_pfn
,
3854 early_node_map
[i
].end_pfn
);
3856 /* Initialise every node */
3857 setup_nr_node_ids();
3858 for_each_online_node(nid
) {
3859 pg_data_t
*pgdat
= NODE_DATA(nid
);
3860 free_area_init_node(nid
, pgdat
, NULL
,
3861 find_min_pfn_for_node(nid
), NULL
);
3863 /* Any memory on that node */
3864 if (pgdat
->node_present_pages
)
3865 node_set_state(nid
, N_HIGH_MEMORY
);
3866 check_for_regular_memory(pgdat
);
3870 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
3872 unsigned long long coremem
;
3876 coremem
= memparse(p
, &p
);
3877 *core
= coremem
>> PAGE_SHIFT
;
3879 /* Paranoid check that UL is enough for the coremem value */
3880 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
3886 * kernelcore=size sets the amount of memory for use for allocations that
3887 * cannot be reclaimed or migrated.
3889 static int __init
cmdline_parse_kernelcore(char *p
)
3891 return cmdline_parse_core(p
, &required_kernelcore
);
3895 * movablecore=size sets the amount of memory for use for allocations that
3896 * can be reclaimed or migrated.
3898 static int __init
cmdline_parse_movablecore(char *p
)
3900 return cmdline_parse_core(p
, &required_movablecore
);
3903 early_param("kernelcore", cmdline_parse_kernelcore
);
3904 early_param("movablecore", cmdline_parse_movablecore
);
3906 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3909 * set_dma_reserve - set the specified number of pages reserved in the first zone
3910 * @new_dma_reserve: The number of pages to mark reserved
3912 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3913 * In the DMA zone, a significant percentage may be consumed by kernel image
3914 * and other unfreeable allocations which can skew the watermarks badly. This
3915 * function may optionally be used to account for unfreeable pages in the
3916 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3917 * smaller per-cpu batchsize.
3919 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
3921 dma_reserve
= new_dma_reserve
;
3924 #ifndef CONFIG_NEED_MULTIPLE_NODES
3925 static bootmem_data_t contig_bootmem_data
;
3926 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
3928 EXPORT_SYMBOL(contig_page_data
);
3931 void __init
free_area_init(unsigned long *zones_size
)
3933 free_area_init_node(0, NODE_DATA(0), zones_size
,
3934 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
3937 static int page_alloc_cpu_notify(struct notifier_block
*self
,
3938 unsigned long action
, void *hcpu
)
3940 int cpu
= (unsigned long)hcpu
;
3942 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3943 local_irq_disable();
3945 vm_events_fold_cpu(cpu
);
3947 refresh_cpu_vm_stats(cpu
);
3952 void __init
page_alloc_init(void)
3954 hotcpu_notifier(page_alloc_cpu_notify
, 0);
3958 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3959 * or min_free_kbytes changes.
3961 static void calculate_totalreserve_pages(void)
3963 struct pglist_data
*pgdat
;
3964 unsigned long reserve_pages
= 0;
3965 enum zone_type i
, j
;
3967 for_each_online_pgdat(pgdat
) {
3968 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3969 struct zone
*zone
= pgdat
->node_zones
+ i
;
3970 unsigned long max
= 0;
3972 /* Find valid and maximum lowmem_reserve in the zone */
3973 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
3974 if (zone
->lowmem_reserve
[j
] > max
)
3975 max
= zone
->lowmem_reserve
[j
];
3978 /* we treat pages_high as reserved pages. */
3979 max
+= zone
->pages_high
;
3981 if (max
> zone
->present_pages
)
3982 max
= zone
->present_pages
;
3983 reserve_pages
+= max
;
3986 totalreserve_pages
= reserve_pages
;
3990 * setup_per_zone_lowmem_reserve - called whenever
3991 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3992 * has a correct pages reserved value, so an adequate number of
3993 * pages are left in the zone after a successful __alloc_pages().
3995 static void setup_per_zone_lowmem_reserve(void)
3997 struct pglist_data
*pgdat
;
3998 enum zone_type j
, idx
;
4000 for_each_online_pgdat(pgdat
) {
4001 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4002 struct zone
*zone
= pgdat
->node_zones
+ j
;
4003 unsigned long present_pages
= zone
->present_pages
;
4005 zone
->lowmem_reserve
[j
] = 0;
4009 struct zone
*lower_zone
;
4013 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4014 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4016 lower_zone
= pgdat
->node_zones
+ idx
;
4017 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4018 sysctl_lowmem_reserve_ratio
[idx
];
4019 present_pages
+= lower_zone
->present_pages
;
4024 /* update totalreserve_pages */
4025 calculate_totalreserve_pages();
4029 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4031 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4032 * with respect to min_free_kbytes.
4034 void setup_per_zone_pages_min(void)
4036 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4037 unsigned long lowmem_pages
= 0;
4039 unsigned long flags
;
4041 /* Calculate total number of !ZONE_HIGHMEM pages */
4042 for_each_zone(zone
) {
4043 if (!is_highmem(zone
))
4044 lowmem_pages
+= zone
->present_pages
;
4047 for_each_zone(zone
) {
4050 spin_lock_irqsave(&zone
->lru_lock
, flags
);
4051 tmp
= (u64
)pages_min
* zone
->present_pages
;
4052 do_div(tmp
, lowmem_pages
);
4053 if (is_highmem(zone
)) {
4055 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4056 * need highmem pages, so cap pages_min to a small
4059 * The (pages_high-pages_low) and (pages_low-pages_min)
4060 * deltas controls asynch page reclaim, and so should
4061 * not be capped for highmem.
4065 min_pages
= zone
->present_pages
/ 1024;
4066 if (min_pages
< SWAP_CLUSTER_MAX
)
4067 min_pages
= SWAP_CLUSTER_MAX
;
4068 if (min_pages
> 128)
4070 zone
->pages_min
= min_pages
;
4073 * If it's a lowmem zone, reserve a number of pages
4074 * proportionate to the zone's size.
4076 zone
->pages_min
= tmp
;
4079 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4080 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4081 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4084 /* update totalreserve_pages */
4085 calculate_totalreserve_pages();
4089 * Initialise min_free_kbytes.
4091 * For small machines we want it small (128k min). For large machines
4092 * we want it large (64MB max). But it is not linear, because network
4093 * bandwidth does not increase linearly with machine size. We use
4095 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4096 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4112 static int __init
init_per_zone_pages_min(void)
4114 unsigned long lowmem_kbytes
;
4116 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4118 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4119 if (min_free_kbytes
< 128)
4120 min_free_kbytes
= 128;
4121 if (min_free_kbytes
> 65536)
4122 min_free_kbytes
= 65536;
4123 setup_per_zone_pages_min();
4124 setup_per_zone_lowmem_reserve();
4127 module_init(init_per_zone_pages_min
)
4130 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4131 * that we can call two helper functions whenever min_free_kbytes
4134 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4135 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4137 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4139 setup_per_zone_pages_min();
4144 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4145 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4150 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4155 zone
->min_unmapped_pages
= (zone
->present_pages
*
4156 sysctl_min_unmapped_ratio
) / 100;
4160 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4161 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4166 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4171 zone
->min_slab_pages
= (zone
->present_pages
*
4172 sysctl_min_slab_ratio
) / 100;
4178 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4179 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4180 * whenever sysctl_lowmem_reserve_ratio changes.
4182 * The reserve ratio obviously has absolutely no relation with the
4183 * pages_min watermarks. The lowmem reserve ratio can only make sense
4184 * if in function of the boot time zone sizes.
4186 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4187 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4189 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4190 setup_per_zone_lowmem_reserve();
4195 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4196 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4197 * can have before it gets flushed back to buddy allocator.
4200 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4201 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4207 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4208 if (!write
|| (ret
== -EINVAL
))
4210 for_each_zone(zone
) {
4211 for_each_online_cpu(cpu
) {
4213 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4214 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4220 int hashdist
= HASHDIST_DEFAULT
;
4223 static int __init
set_hashdist(char *str
)
4227 hashdist
= simple_strtoul(str
, &str
, 0);
4230 __setup("hashdist=", set_hashdist
);
4234 * allocate a large system hash table from bootmem
4235 * - it is assumed that the hash table must contain an exact power-of-2
4236 * quantity of entries
4237 * - limit is the number of hash buckets, not the total allocation size
4239 void *__init
alloc_large_system_hash(const char *tablename
,
4240 unsigned long bucketsize
,
4241 unsigned long numentries
,
4244 unsigned int *_hash_shift
,
4245 unsigned int *_hash_mask
,
4246 unsigned long limit
)
4248 unsigned long long max
= limit
;
4249 unsigned long log2qty
, size
;
4252 /* allow the kernel cmdline to have a say */
4254 /* round applicable memory size up to nearest megabyte */
4255 numentries
= nr_kernel_pages
;
4256 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4257 numentries
>>= 20 - PAGE_SHIFT
;
4258 numentries
<<= 20 - PAGE_SHIFT
;
4260 /* limit to 1 bucket per 2^scale bytes of low memory */
4261 if (scale
> PAGE_SHIFT
)
4262 numentries
>>= (scale
- PAGE_SHIFT
);
4264 numentries
<<= (PAGE_SHIFT
- scale
);
4266 /* Make sure we've got at least a 0-order allocation.. */
4267 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4268 numentries
= PAGE_SIZE
/ bucketsize
;
4270 numentries
= roundup_pow_of_two(numentries
);
4272 /* limit allocation size to 1/16 total memory by default */
4274 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4275 do_div(max
, bucketsize
);
4278 if (numentries
> max
)
4281 log2qty
= ilog2(numentries
);
4284 size
= bucketsize
<< log2qty
;
4285 if (flags
& HASH_EARLY
)
4286 table
= alloc_bootmem(size
);
4288 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4290 unsigned long order
;
4291 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
4293 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4295 * If bucketsize is not a power-of-two, we may free
4296 * some pages at the end of hash table.
4299 unsigned long alloc_end
= (unsigned long)table
+
4300 (PAGE_SIZE
<< order
);
4301 unsigned long used
= (unsigned long)table
+
4303 split_page(virt_to_page(table
), order
);
4304 while (used
< alloc_end
) {
4310 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4313 panic("Failed to allocate %s hash table\n", tablename
);
4315 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4318 ilog2(size
) - PAGE_SHIFT
,
4322 *_hash_shift
= log2qty
;
4324 *_hash_mask
= (1 << log2qty
) - 1;
4329 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4330 struct page
*pfn_to_page(unsigned long pfn
)
4332 return __pfn_to_page(pfn
);
4334 unsigned long page_to_pfn(struct page
*page
)
4336 return __page_to_pfn(page
);
4338 EXPORT_SYMBOL(pfn_to_page
);
4339 EXPORT_SYMBOL(page_to_pfn
);
4340 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4342 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4343 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4346 #ifdef CONFIG_SPARSEMEM
4347 return __pfn_to_section(pfn
)->pageblock_flags
;
4349 return zone
->pageblock_flags
;
4350 #endif /* CONFIG_SPARSEMEM */
4353 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4355 #ifdef CONFIG_SPARSEMEM
4356 pfn
&= (PAGES_PER_SECTION
-1);
4357 return (pfn
>> (MAX_ORDER
-1)) * NR_PAGEBLOCK_BITS
;
4359 pfn
= pfn
- zone
->zone_start_pfn
;
4360 return (pfn
>> (MAX_ORDER
-1)) * NR_PAGEBLOCK_BITS
;
4361 #endif /* CONFIG_SPARSEMEM */
4365 * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages
4366 * @page: The page within the block of interest
4367 * @start_bitidx: The first bit of interest to retrieve
4368 * @end_bitidx: The last bit of interest
4369 * returns pageblock_bits flags
4371 unsigned long get_pageblock_flags_group(struct page
*page
,
4372 int start_bitidx
, int end_bitidx
)
4375 unsigned long *bitmap
;
4376 unsigned long pfn
, bitidx
;
4377 unsigned long flags
= 0;
4378 unsigned long value
= 1;
4380 zone
= page_zone(page
);
4381 pfn
= page_to_pfn(page
);
4382 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4383 bitidx
= pfn_to_bitidx(zone
, pfn
);
4385 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4386 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4393 * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages
4394 * @page: The page within the block of interest
4395 * @start_bitidx: The first bit of interest
4396 * @end_bitidx: The last bit of interest
4397 * @flags: The flags to set
4399 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4400 int start_bitidx
, int end_bitidx
)
4403 unsigned long *bitmap
;
4404 unsigned long pfn
, bitidx
;
4405 unsigned long value
= 1;
4407 zone
= page_zone(page
);
4408 pfn
= page_to_pfn(page
);
4409 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4410 bitidx
= pfn_to_bitidx(zone
, pfn
);
4412 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4414 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4416 __clear_bit(bitidx
+ start_bitidx
, bitmap
);