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 static inline int get_pageblock_migratetype(struct page
*page
)
164 return get_pageblock_flags_group(page
, PB_migrate
, PB_migrate_end
);
167 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
169 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
170 PB_migrate
, PB_migrate_end
);
173 static inline int gfpflags_to_migratetype(gfp_t gfp_flags
)
175 WARN_ON((gfp_flags
& GFP_MOVABLE_MASK
) == GFP_MOVABLE_MASK
);
177 return (((gfp_flags
& __GFP_MOVABLE
) != 0) << 1) |
178 ((gfp_flags
& __GFP_RECLAIMABLE
) != 0);
182 static inline int get_pageblock_migratetype(struct page
*page
)
184 return MIGRATE_UNMOVABLE
;
187 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
191 static inline int gfpflags_to_migratetype(gfp_t gfp_flags
)
193 return MIGRATE_UNMOVABLE
;
195 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
197 #ifdef CONFIG_DEBUG_VM
198 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
202 unsigned long pfn
= page_to_pfn(page
);
205 seq
= zone_span_seqbegin(zone
);
206 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
208 else if (pfn
< zone
->zone_start_pfn
)
210 } while (zone_span_seqretry(zone
, seq
));
215 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
217 if (!pfn_valid_within(page_to_pfn(page
)))
219 if (zone
!= page_zone(page
))
225 * Temporary debugging check for pages not lying within a given zone.
227 static int bad_range(struct zone
*zone
, struct page
*page
)
229 if (page_outside_zone_boundaries(zone
, page
))
231 if (!page_is_consistent(zone
, page
))
237 static inline int bad_range(struct zone
*zone
, struct page
*page
)
243 static void bad_page(struct page
*page
)
245 printk(KERN_EMERG
"Bad page state in process '%s'\n"
246 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
247 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
248 KERN_EMERG
"Backtrace:\n",
249 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
250 (unsigned long)page
->flags
, page
->mapping
,
251 page_mapcount(page
), page_count(page
));
253 page
->flags
&= ~(1 << PG_lru
|
263 set_page_count(page
, 0);
264 reset_page_mapcount(page
);
265 page
->mapping
= NULL
;
266 add_taint(TAINT_BAD_PAGE
);
270 * Higher-order pages are called "compound pages". They are structured thusly:
272 * The first PAGE_SIZE page is called the "head page".
274 * The remaining PAGE_SIZE pages are called "tail pages".
276 * All pages have PG_compound set. All pages have their ->private pointing at
277 * the head page (even the head page has this).
279 * The first tail page's ->lru.next holds the address of the compound page's
280 * put_page() function. Its ->lru.prev holds the order of allocation.
281 * This usage means that zero-order pages may not be compound.
284 static void free_compound_page(struct page
*page
)
286 __free_pages_ok(page
, compound_order(page
));
289 static void prep_compound_page(struct page
*page
, unsigned long order
)
292 int nr_pages
= 1 << order
;
294 set_compound_page_dtor(page
, free_compound_page
);
295 set_compound_order(page
, order
);
297 for (i
= 1; i
< nr_pages
; i
++) {
298 struct page
*p
= page
+ i
;
301 p
->first_page
= page
;
305 static void destroy_compound_page(struct page
*page
, unsigned long order
)
308 int nr_pages
= 1 << order
;
310 if (unlikely(compound_order(page
) != order
))
313 if (unlikely(!PageHead(page
)))
315 __ClearPageHead(page
);
316 for (i
= 1; i
< nr_pages
; i
++) {
317 struct page
*p
= page
+ i
;
319 if (unlikely(!PageTail(p
) |
320 (p
->first_page
!= page
)))
326 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
330 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
332 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
333 * and __GFP_HIGHMEM from hard or soft interrupt context.
335 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
336 for (i
= 0; i
< (1 << order
); i
++)
337 clear_highpage(page
+ i
);
341 * function for dealing with page's order in buddy system.
342 * zone->lock is already acquired when we use these.
343 * So, we don't need atomic page->flags operations here.
345 static inline unsigned long page_order(struct page
*page
)
347 return page_private(page
);
350 static inline void set_page_order(struct page
*page
, int order
)
352 set_page_private(page
, order
);
353 __SetPageBuddy(page
);
356 static inline void rmv_page_order(struct page
*page
)
358 __ClearPageBuddy(page
);
359 set_page_private(page
, 0);
363 * Locate the struct page for both the matching buddy in our
364 * pair (buddy1) and the combined O(n+1) page they form (page).
366 * 1) Any buddy B1 will have an order O twin B2 which satisfies
367 * the following equation:
369 * For example, if the starting buddy (buddy2) is #8 its order
371 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
373 * 2) Any buddy B will have an order O+1 parent P which
374 * satisfies the following equation:
377 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
379 static inline struct page
*
380 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
382 unsigned long buddy_idx
= page_idx
^ (1 << order
);
384 return page
+ (buddy_idx
- page_idx
);
387 static inline unsigned long
388 __find_combined_index(unsigned long page_idx
, unsigned int order
)
390 return (page_idx
& ~(1 << order
));
394 * This function checks whether a page is free && is the buddy
395 * we can do coalesce a page and its buddy if
396 * (a) the buddy is not in a hole &&
397 * (b) the buddy is in the buddy system &&
398 * (c) a page and its buddy have the same order &&
399 * (d) a page and its buddy are in the same zone.
401 * For recording whether a page is in the buddy system, we use PG_buddy.
402 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
404 * For recording page's order, we use page_private(page).
406 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
409 if (!pfn_valid_within(page_to_pfn(buddy
)))
412 if (page_zone_id(page
) != page_zone_id(buddy
))
415 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
416 BUG_ON(page_count(buddy
) != 0);
423 * Freeing function for a buddy system allocator.
425 * The concept of a buddy system is to maintain direct-mapped table
426 * (containing bit values) for memory blocks of various "orders".
427 * The bottom level table contains the map for the smallest allocatable
428 * units of memory (here, pages), and each level above it describes
429 * pairs of units from the levels below, hence, "buddies".
430 * At a high level, all that happens here is marking the table entry
431 * at the bottom level available, and propagating the changes upward
432 * as necessary, plus some accounting needed to play nicely with other
433 * parts of the VM system.
434 * At each level, we keep a list of pages, which are heads of continuous
435 * free pages of length of (1 << order) and marked with PG_buddy. Page's
436 * order is recorded in page_private(page) field.
437 * So when we are allocating or freeing one, we can derive the state of the
438 * other. That is, if we allocate a small block, and both were
439 * free, the remainder of the region must be split into blocks.
440 * If a block is freed, and its buddy is also free, then this
441 * triggers coalescing into a block of larger size.
446 static inline void __free_one_page(struct page
*page
,
447 struct zone
*zone
, unsigned int order
)
449 unsigned long page_idx
;
450 int order_size
= 1 << order
;
451 int migratetype
= get_pageblock_migratetype(page
);
453 if (unlikely(PageCompound(page
)))
454 destroy_compound_page(page
, order
);
456 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
458 VM_BUG_ON(page_idx
& (order_size
- 1));
459 VM_BUG_ON(bad_range(zone
, page
));
461 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
462 while (order
< MAX_ORDER
-1) {
463 unsigned long combined_idx
;
466 buddy
= __page_find_buddy(page
, page_idx
, order
);
467 if (!page_is_buddy(page
, buddy
, order
))
468 break; /* Move the buddy up one level. */
470 list_del(&buddy
->lru
);
471 zone
->free_area
[order
].nr_free
--;
472 rmv_page_order(buddy
);
473 combined_idx
= __find_combined_index(page_idx
, order
);
474 page
= page
+ (combined_idx
- page_idx
);
475 page_idx
= combined_idx
;
478 set_page_order(page
, order
);
480 &zone
->free_area
[order
].free_list
[migratetype
]);
481 zone
->free_area
[order
].nr_free
++;
484 static inline int free_pages_check(struct page
*page
)
486 if (unlikely(page_mapcount(page
) |
487 (page
->mapping
!= NULL
) |
488 (page_count(page
) != 0) |
501 __ClearPageDirty(page
);
503 * For now, we report if PG_reserved was found set, but do not
504 * clear it, and do not free the page. But we shall soon need
505 * to do more, for when the ZERO_PAGE count wraps negative.
507 return PageReserved(page
);
511 * Frees a list of pages.
512 * Assumes all pages on list are in same zone, and of same order.
513 * count is the number of pages to free.
515 * If the zone was previously in an "all pages pinned" state then look to
516 * see if this freeing clears that state.
518 * And clear the zone's pages_scanned counter, to hold off the "all pages are
519 * pinned" detection logic.
521 static void free_pages_bulk(struct zone
*zone
, int count
,
522 struct list_head
*list
, int order
)
524 spin_lock(&zone
->lock
);
525 zone
->all_unreclaimable
= 0;
526 zone
->pages_scanned
= 0;
530 VM_BUG_ON(list_empty(list
));
531 page
= list_entry(list
->prev
, struct page
, lru
);
532 /* have to delete it as __free_one_page list manipulates */
533 list_del(&page
->lru
);
534 __free_one_page(page
, zone
, order
);
536 spin_unlock(&zone
->lock
);
539 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
541 spin_lock(&zone
->lock
);
542 zone
->all_unreclaimable
= 0;
543 zone
->pages_scanned
= 0;
544 __free_one_page(page
, zone
, order
);
545 spin_unlock(&zone
->lock
);
548 static void __free_pages_ok(struct page
*page
, unsigned int order
)
554 for (i
= 0 ; i
< (1 << order
) ; ++i
)
555 reserved
+= free_pages_check(page
+ i
);
559 if (!PageHighMem(page
))
560 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
561 arch_free_page(page
, order
);
562 kernel_map_pages(page
, 1 << order
, 0);
564 local_irq_save(flags
);
565 __count_vm_events(PGFREE
, 1 << order
);
566 free_one_page(page_zone(page
), page
, order
);
567 local_irq_restore(flags
);
571 * permit the bootmem allocator to evade page validation on high-order frees
573 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
576 __ClearPageReserved(page
);
577 set_page_count(page
, 0);
578 set_page_refcounted(page
);
584 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
585 struct page
*p
= &page
[loop
];
587 if (loop
+ 1 < BITS_PER_LONG
)
589 __ClearPageReserved(p
);
590 set_page_count(p
, 0);
593 set_page_refcounted(page
);
594 __free_pages(page
, order
);
600 * The order of subdivision here is critical for the IO subsystem.
601 * Please do not alter this order without good reasons and regression
602 * testing. Specifically, as large blocks of memory are subdivided,
603 * the order in which smaller blocks are delivered depends on the order
604 * they're subdivided in this function. This is the primary factor
605 * influencing the order in which pages are delivered to the IO
606 * subsystem according to empirical testing, and this is also justified
607 * by considering the behavior of a buddy system containing a single
608 * large block of memory acted on by a series of small allocations.
609 * This behavior is a critical factor in sglist merging's success.
613 static inline void expand(struct zone
*zone
, struct page
*page
,
614 int low
, int high
, struct free_area
*area
,
617 unsigned long size
= 1 << high
;
623 VM_BUG_ON(bad_range(zone
, &page
[size
]));
624 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
626 set_page_order(&page
[size
], high
);
631 * This page is about to be returned from the page allocator
633 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
635 if (unlikely(page_mapcount(page
) |
636 (page
->mapping
!= NULL
) |
637 (page_count(page
) != 0) |
652 * For now, we report if PG_reserved was found set, but do not
653 * clear it, and do not allocate the page: as a safety net.
655 if (PageReserved(page
))
658 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_readahead
|
659 1 << PG_referenced
| 1 << PG_arch_1
|
660 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
);
661 set_page_private(page
, 0);
662 set_page_refcounted(page
);
664 arch_alloc_page(page
, order
);
665 kernel_map_pages(page
, 1 << order
, 1);
667 if (gfp_flags
& __GFP_ZERO
)
668 prep_zero_page(page
, order
, gfp_flags
);
670 if (order
&& (gfp_flags
& __GFP_COMP
))
671 prep_compound_page(page
, order
);
676 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
678 * This array describes the order lists are fallen back to when
679 * the free lists for the desirable migrate type are depleted
681 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
682 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
},
683 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
},
684 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
},
688 * Move the free pages in a range to the free lists of the requested type.
689 * Note that start_page and end_pages are not aligned in a MAX_ORDER_NR_PAGES
690 * boundary. If alignment is required, use move_freepages_block()
692 int move_freepages(struct zone
*zone
,
693 struct page
*start_page
, struct page
*end_page
,
698 int blocks_moved
= 0;
700 #ifndef CONFIG_HOLES_IN_ZONE
702 * page_zone is not safe to call in this context when
703 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
704 * anyway as we check zone boundaries in move_freepages_block().
705 * Remove at a later date when no bug reports exist related to
706 * CONFIG_PAGE_GROUP_BY_MOBILITY
708 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
711 for (page
= start_page
; page
<= end_page
;) {
712 if (!pfn_valid_within(page_to_pfn(page
))) {
717 if (!PageBuddy(page
)) {
722 order
= page_order(page
);
723 list_del(&page
->lru
);
725 &zone
->free_area
[order
].free_list
[migratetype
]);
733 int move_freepages_block(struct zone
*zone
, struct page
*page
, int migratetype
)
735 unsigned long start_pfn
, end_pfn
;
736 struct page
*start_page
, *end_page
;
738 start_pfn
= page_to_pfn(page
);
739 start_pfn
= start_pfn
& ~(MAX_ORDER_NR_PAGES
-1);
740 start_page
= pfn_to_page(start_pfn
);
741 end_page
= start_page
+ MAX_ORDER_NR_PAGES
- 1;
742 end_pfn
= start_pfn
+ MAX_ORDER_NR_PAGES
- 1;
744 /* Do not cross zone boundaries */
745 if (start_pfn
< zone
->zone_start_pfn
)
747 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
750 return move_freepages(zone
, start_page
, end_page
, migratetype
);
753 /* Remove an element from the buddy allocator from the fallback list */
754 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
755 int start_migratetype
)
757 struct free_area
* area
;
762 /* Find the largest possible block of pages in the other list */
763 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
765 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
766 migratetype
= fallbacks
[start_migratetype
][i
];
768 area
= &(zone
->free_area
[current_order
]);
769 if (list_empty(&area
->free_list
[migratetype
]))
772 page
= list_entry(area
->free_list
[migratetype
].next
,
777 * If breaking a large block of pages, move all free
778 * pages to the preferred allocation list
780 if (unlikely(current_order
>= MAX_ORDER
/ 2)) {
781 migratetype
= start_migratetype
;
782 move_freepages_block(zone
, page
, migratetype
);
785 /* Remove the page from the freelists */
786 list_del(&page
->lru
);
787 rmv_page_order(page
);
788 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
791 if (current_order
== MAX_ORDER
- 1)
792 set_pageblock_migratetype(page
,
795 expand(zone
, page
, order
, current_order
, area
, migratetype
);
803 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
804 int start_migratetype
)
808 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
811 * Do the hard work of removing an element from the buddy allocator.
812 * Call me with the zone->lock already held.
814 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
817 struct free_area
* area
;
818 unsigned int current_order
;
821 /* Find a page of the appropriate size in the preferred list */
822 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
823 area
= &(zone
->free_area
[current_order
]);
824 if (list_empty(&area
->free_list
[migratetype
]))
827 page
= list_entry(area
->free_list
[migratetype
].next
,
829 list_del(&page
->lru
);
830 rmv_page_order(page
);
832 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
833 expand(zone
, page
, order
, current_order
, area
, migratetype
);
837 page
= __rmqueue_fallback(zone
, order
, migratetype
);
845 * Obtain a specified number of elements from the buddy allocator, all under
846 * a single hold of the lock, for efficiency. Add them to the supplied list.
847 * Returns the number of new pages which were placed at *list.
849 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
850 unsigned long count
, struct list_head
*list
,
855 spin_lock(&zone
->lock
);
856 for (i
= 0; i
< count
; ++i
) {
857 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
858 if (unlikely(page
== NULL
))
860 list_add(&page
->lru
, list
);
861 set_page_private(page
, migratetype
);
863 spin_unlock(&zone
->lock
);
869 * Called from the vmstat counter updater to drain pagesets of this
870 * currently executing processor on remote nodes after they have
873 * Note that this function must be called with the thread pinned to
874 * a single processor.
876 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
881 local_irq_save(flags
);
882 if (pcp
->count
>= pcp
->batch
)
883 to_drain
= pcp
->batch
;
885 to_drain
= pcp
->count
;
886 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
887 pcp
->count
-= to_drain
;
888 local_irq_restore(flags
);
892 static void __drain_pages(unsigned int cpu
)
898 for_each_zone(zone
) {
899 struct per_cpu_pageset
*pset
;
901 if (!populated_zone(zone
))
904 pset
= zone_pcp(zone
, cpu
);
905 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
906 struct per_cpu_pages
*pcp
;
909 local_irq_save(flags
);
910 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
912 local_irq_restore(flags
);
917 #ifdef CONFIG_HIBERNATION
919 void mark_free_pages(struct zone
*zone
)
921 unsigned long pfn
, max_zone_pfn
;
924 struct list_head
*curr
;
926 if (!zone
->spanned_pages
)
929 spin_lock_irqsave(&zone
->lock
, flags
);
931 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
932 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
933 if (pfn_valid(pfn
)) {
934 struct page
*page
= pfn_to_page(pfn
);
936 if (!swsusp_page_is_forbidden(page
))
937 swsusp_unset_page_free(page
);
940 for_each_migratetype_order(order
, t
) {
941 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
944 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
945 for (i
= 0; i
< (1UL << order
); i
++)
946 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
949 spin_unlock_irqrestore(&zone
->lock
, flags
);
951 #endif /* CONFIG_PM */
953 #if defined(CONFIG_HIBERNATION) || defined(CONFIG_PAGE_GROUP_BY_MOBILITY)
955 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
957 void drain_local_pages(void)
961 local_irq_save(flags
);
962 __drain_pages(smp_processor_id());
963 local_irq_restore(flags
);
966 void smp_drain_local_pages(void *arg
)
972 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
974 void drain_all_local_pages(void)
978 local_irq_save(flags
);
979 __drain_pages(smp_processor_id());
980 local_irq_restore(flags
);
982 smp_call_function(smp_drain_local_pages
, NULL
, 0, 1);
985 void drain_all_local_pages(void) {}
986 #endif /* CONFIG_HIBERNATION || CONFIG_PAGE_GROUP_BY_MOBILITY */
989 * Free a 0-order page
991 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
993 struct zone
*zone
= page_zone(page
);
994 struct per_cpu_pages
*pcp
;
998 page
->mapping
= NULL
;
999 if (free_pages_check(page
))
1002 if (!PageHighMem(page
))
1003 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1004 arch_free_page(page
, 0);
1005 kernel_map_pages(page
, 1, 0);
1007 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
1008 local_irq_save(flags
);
1009 __count_vm_event(PGFREE
);
1010 list_add(&page
->lru
, &pcp
->list
);
1011 set_page_private(page
, get_pageblock_migratetype(page
));
1013 if (pcp
->count
>= pcp
->high
) {
1014 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1015 pcp
->count
-= pcp
->batch
;
1017 local_irq_restore(flags
);
1021 void fastcall
free_hot_page(struct page
*page
)
1023 free_hot_cold_page(page
, 0);
1026 void fastcall
free_cold_page(struct page
*page
)
1028 free_hot_cold_page(page
, 1);
1032 * split_page takes a non-compound higher-order page, and splits it into
1033 * n (1<<order) sub-pages: page[0..n]
1034 * Each sub-page must be freed individually.
1036 * Note: this is probably too low level an operation for use in drivers.
1037 * Please consult with lkml before using this in your driver.
1039 void split_page(struct page
*page
, unsigned int order
)
1043 VM_BUG_ON(PageCompound(page
));
1044 VM_BUG_ON(!page_count(page
));
1045 for (i
= 1; i
< (1 << order
); i
++)
1046 set_page_refcounted(page
+ i
);
1050 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1051 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1054 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
1055 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1057 unsigned long flags
;
1059 int cold
= !!(gfp_flags
& __GFP_COLD
);
1061 int migratetype
= gfpflags_to_migratetype(gfp_flags
);
1065 if (likely(order
== 0)) {
1066 struct per_cpu_pages
*pcp
;
1068 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
1069 local_irq_save(flags
);
1071 pcp
->count
= rmqueue_bulk(zone
, 0,
1072 pcp
->batch
, &pcp
->list
, migratetype
);
1073 if (unlikely(!pcp
->count
))
1077 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
1078 /* Find a page of the appropriate migrate type */
1079 list_for_each_entry(page
, &pcp
->list
, lru
)
1080 if (page_private(page
) == migratetype
)
1083 /* Allocate more to the pcp list if necessary */
1084 if (unlikely(&page
->lru
== &pcp
->list
)) {
1085 pcp
->count
+= rmqueue_bulk(zone
, 0,
1086 pcp
->batch
, &pcp
->list
, migratetype
);
1087 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1090 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1091 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
1093 list_del(&page
->lru
);
1096 spin_lock_irqsave(&zone
->lock
, flags
);
1097 page
= __rmqueue(zone
, order
, migratetype
);
1098 spin_unlock(&zone
->lock
);
1103 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1104 zone_statistics(zonelist
, zone
);
1105 local_irq_restore(flags
);
1108 VM_BUG_ON(bad_range(zone
, page
));
1109 if (prep_new_page(page
, order
, gfp_flags
))
1114 local_irq_restore(flags
);
1119 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1120 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1121 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1122 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1123 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1124 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1125 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1127 #ifdef CONFIG_FAIL_PAGE_ALLOC
1129 static struct fail_page_alloc_attr
{
1130 struct fault_attr attr
;
1132 u32 ignore_gfp_highmem
;
1133 u32 ignore_gfp_wait
;
1136 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1138 struct dentry
*ignore_gfp_highmem_file
;
1139 struct dentry
*ignore_gfp_wait_file
;
1140 struct dentry
*min_order_file
;
1142 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1144 } fail_page_alloc
= {
1145 .attr
= FAULT_ATTR_INITIALIZER
,
1146 .ignore_gfp_wait
= 1,
1147 .ignore_gfp_highmem
= 1,
1151 static int __init
setup_fail_page_alloc(char *str
)
1153 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1155 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1157 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1159 if (order
< fail_page_alloc
.min_order
)
1161 if (gfp_mask
& __GFP_NOFAIL
)
1163 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1165 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1168 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1171 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1173 static int __init
fail_page_alloc_debugfs(void)
1175 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1179 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1183 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1185 fail_page_alloc
.ignore_gfp_wait_file
=
1186 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1187 &fail_page_alloc
.ignore_gfp_wait
);
1189 fail_page_alloc
.ignore_gfp_highmem_file
=
1190 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1191 &fail_page_alloc
.ignore_gfp_highmem
);
1192 fail_page_alloc
.min_order_file
=
1193 debugfs_create_u32("min-order", mode
, dir
,
1194 &fail_page_alloc
.min_order
);
1196 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1197 !fail_page_alloc
.ignore_gfp_highmem_file
||
1198 !fail_page_alloc
.min_order_file
) {
1200 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1201 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1202 debugfs_remove(fail_page_alloc
.min_order_file
);
1203 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1209 late_initcall(fail_page_alloc_debugfs
);
1211 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1213 #else /* CONFIG_FAIL_PAGE_ALLOC */
1215 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1220 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1223 * Return 1 if free pages are above 'mark'. This takes into account the order
1224 * of the allocation.
1226 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1227 int classzone_idx
, int alloc_flags
)
1229 /* free_pages my go negative - that's OK */
1231 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1234 if (alloc_flags
& ALLOC_HIGH
)
1236 if (alloc_flags
& ALLOC_HARDER
)
1239 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1241 for (o
= 0; o
< order
; o
++) {
1242 /* At the next order, this order's pages become unavailable */
1243 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1245 /* Require fewer higher order pages to be free */
1248 if (free_pages
<= min
)
1256 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1257 * skip over zones that are not allowed by the cpuset, or that have
1258 * been recently (in last second) found to be nearly full. See further
1259 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1260 * that have to skip over alot of full or unallowed zones.
1262 * If the zonelist cache is present in the passed in zonelist, then
1263 * returns a pointer to the allowed node mask (either the current
1264 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1266 * If the zonelist cache is not available for this zonelist, does
1267 * nothing and returns NULL.
1269 * If the fullzones BITMAP in the zonelist cache is stale (more than
1270 * a second since last zap'd) then we zap it out (clear its bits.)
1272 * We hold off even calling zlc_setup, until after we've checked the
1273 * first zone in the zonelist, on the theory that most allocations will
1274 * be satisfied from that first zone, so best to examine that zone as
1275 * quickly as we can.
1277 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1279 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1280 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1282 zlc
= zonelist
->zlcache_ptr
;
1286 if (jiffies
- zlc
->last_full_zap
> 1 * HZ
) {
1287 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1288 zlc
->last_full_zap
= jiffies
;
1291 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1292 &cpuset_current_mems_allowed
:
1293 &node_states
[N_HIGH_MEMORY
];
1294 return allowednodes
;
1298 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1299 * if it is worth looking at further for free memory:
1300 * 1) Check that the zone isn't thought to be full (doesn't have its
1301 * bit set in the zonelist_cache fullzones BITMAP).
1302 * 2) Check that the zones node (obtained from the zonelist_cache
1303 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1304 * Return true (non-zero) if zone is worth looking at further, or
1305 * else return false (zero) if it is not.
1307 * This check -ignores- the distinction between various watermarks,
1308 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1309 * found to be full for any variation of these watermarks, it will
1310 * be considered full for up to one second by all requests, unless
1311 * we are so low on memory on all allowed nodes that we are forced
1312 * into the second scan of the zonelist.
1314 * In the second scan we ignore this zonelist cache and exactly
1315 * apply the watermarks to all zones, even it is slower to do so.
1316 * We are low on memory in the second scan, and should leave no stone
1317 * unturned looking for a free page.
1319 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1320 nodemask_t
*allowednodes
)
1322 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1323 int i
; /* index of *z in zonelist zones */
1324 int n
; /* node that zone *z is on */
1326 zlc
= zonelist
->zlcache_ptr
;
1330 i
= z
- zonelist
->zones
;
1333 /* This zone is worth trying if it is allowed but not full */
1334 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1338 * Given 'z' scanning a zonelist, set the corresponding bit in
1339 * zlc->fullzones, so that subsequent attempts to allocate a page
1340 * from that zone don't waste time re-examining it.
1342 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1344 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1345 int i
; /* index of *z in zonelist zones */
1347 zlc
= zonelist
->zlcache_ptr
;
1351 i
= z
- zonelist
->zones
;
1353 set_bit(i
, zlc
->fullzones
);
1356 #else /* CONFIG_NUMA */
1358 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1363 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1364 nodemask_t
*allowednodes
)
1369 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1372 #endif /* CONFIG_NUMA */
1375 * get_page_from_freelist goes through the zonelist trying to allocate
1378 static struct page
*
1379 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
1380 struct zonelist
*zonelist
, int alloc_flags
)
1383 struct page
*page
= NULL
;
1384 int classzone_idx
= zone_idx(zonelist
->zones
[0]);
1386 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1387 int zlc_active
= 0; /* set if using zonelist_cache */
1388 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1389 enum zone_type highest_zoneidx
= -1; /* Gets set for policy zonelists */
1393 * Scan zonelist, looking for a zone with enough free.
1394 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1396 z
= zonelist
->zones
;
1400 * In NUMA, this could be a policy zonelist which contains
1401 * zones that may not be allowed by the current gfp_mask.
1402 * Check the zone is allowed by the current flags
1404 if (unlikely(alloc_should_filter_zonelist(zonelist
))) {
1405 if (highest_zoneidx
== -1)
1406 highest_zoneidx
= gfp_zone(gfp_mask
);
1407 if (zone_idx(*z
) > highest_zoneidx
)
1411 if (NUMA_BUILD
&& zlc_active
&&
1412 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1415 if ((alloc_flags
& ALLOC_CPUSET
) &&
1416 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1419 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1421 if (alloc_flags
& ALLOC_WMARK_MIN
)
1422 mark
= zone
->pages_min
;
1423 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1424 mark
= zone
->pages_low
;
1426 mark
= zone
->pages_high
;
1427 if (!zone_watermark_ok(zone
, order
, mark
,
1428 classzone_idx
, alloc_flags
)) {
1429 if (!zone_reclaim_mode
||
1430 !zone_reclaim(zone
, gfp_mask
, order
))
1431 goto this_zone_full
;
1435 page
= buffered_rmqueue(zonelist
, zone
, order
, gfp_mask
);
1440 zlc_mark_zone_full(zonelist
, z
);
1442 if (NUMA_BUILD
&& !did_zlc_setup
) {
1443 /* we do zlc_setup after the first zone is tried */
1444 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1448 } while (*(++z
) != NULL
);
1450 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1451 /* Disable zlc cache for second zonelist scan */
1459 * This is the 'heart' of the zoned buddy allocator.
1461 struct page
* fastcall
1462 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
1463 struct zonelist
*zonelist
)
1465 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1468 struct reclaim_state reclaim_state
;
1469 struct task_struct
*p
= current
;
1472 int did_some_progress
;
1474 might_sleep_if(wait
);
1476 if (should_fail_alloc_page(gfp_mask
, order
))
1480 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
1482 if (unlikely(*z
== NULL
)) {
1484 * Happens if we have an empty zonelist as a result of
1485 * GFP_THISNODE being used on a memoryless node
1490 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1491 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1496 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1497 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1498 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1499 * using a larger set of nodes after it has established that the
1500 * allowed per node queues are empty and that nodes are
1503 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1506 for (z
= zonelist
->zones
; *z
; z
++)
1507 wakeup_kswapd(*z
, order
);
1510 * OK, we're below the kswapd watermark and have kicked background
1511 * reclaim. Now things get more complex, so set up alloc_flags according
1512 * to how we want to proceed.
1514 * The caller may dip into page reserves a bit more if the caller
1515 * cannot run direct reclaim, or if the caller has realtime scheduling
1516 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1517 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1519 alloc_flags
= ALLOC_WMARK_MIN
;
1520 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1521 alloc_flags
|= ALLOC_HARDER
;
1522 if (gfp_mask
& __GFP_HIGH
)
1523 alloc_flags
|= ALLOC_HIGH
;
1525 alloc_flags
|= ALLOC_CPUSET
;
1528 * Go through the zonelist again. Let __GFP_HIGH and allocations
1529 * coming from realtime tasks go deeper into reserves.
1531 * This is the last chance, in general, before the goto nopage.
1532 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1533 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1535 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1539 /* This allocation should allow future memory freeing. */
1542 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1543 && !in_interrupt()) {
1544 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1546 /* go through the zonelist yet again, ignoring mins */
1547 page
= get_page_from_freelist(gfp_mask
, order
,
1548 zonelist
, ALLOC_NO_WATERMARKS
);
1551 if (gfp_mask
& __GFP_NOFAIL
) {
1552 congestion_wait(WRITE
, HZ
/50);
1559 /* Atomic allocations - we can't balance anything */
1565 /* We now go into synchronous reclaim */
1566 cpuset_memory_pressure_bump();
1567 p
->flags
|= PF_MEMALLOC
;
1568 reclaim_state
.reclaimed_slab
= 0;
1569 p
->reclaim_state
= &reclaim_state
;
1571 did_some_progress
= try_to_free_pages(zonelist
->zones
, order
, gfp_mask
);
1573 p
->reclaim_state
= NULL
;
1574 p
->flags
&= ~PF_MEMALLOC
;
1579 drain_all_local_pages();
1581 if (likely(did_some_progress
)) {
1582 page
= get_page_from_freelist(gfp_mask
, order
,
1583 zonelist
, alloc_flags
);
1586 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1588 * Go through the zonelist yet one more time, keep
1589 * very high watermark here, this is only to catch
1590 * a parallel oom killing, we must fail if we're still
1591 * under heavy pressure.
1593 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1594 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1598 /* The OOM killer will not help higher order allocs so fail */
1599 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1602 out_of_memory(zonelist
, gfp_mask
, order
);
1607 * Don't let big-order allocations loop unless the caller explicitly
1608 * requests that. Wait for some write requests to complete then retry.
1610 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1611 * <= 3, but that may not be true in other implementations.
1614 if (!(gfp_mask
& __GFP_NORETRY
)) {
1615 if ((order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
1616 (gfp_mask
& __GFP_REPEAT
))
1618 if (gfp_mask
& __GFP_NOFAIL
)
1622 congestion_wait(WRITE
, HZ
/50);
1627 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1628 printk(KERN_WARNING
"%s: page allocation failure."
1629 " order:%d, mode:0x%x\n",
1630 p
->comm
, order
, gfp_mask
);
1638 EXPORT_SYMBOL(__alloc_pages
);
1641 * Common helper functions.
1643 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1646 page
= alloc_pages(gfp_mask
, order
);
1649 return (unsigned long) page_address(page
);
1652 EXPORT_SYMBOL(__get_free_pages
);
1654 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1659 * get_zeroed_page() returns a 32-bit address, which cannot represent
1662 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1664 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1666 return (unsigned long) page_address(page
);
1670 EXPORT_SYMBOL(get_zeroed_page
);
1672 void __pagevec_free(struct pagevec
*pvec
)
1674 int i
= pagevec_count(pvec
);
1677 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1680 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1682 if (put_page_testzero(page
)) {
1684 free_hot_page(page
);
1686 __free_pages_ok(page
, order
);
1690 EXPORT_SYMBOL(__free_pages
);
1692 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1695 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1696 __free_pages(virt_to_page((void *)addr
), order
);
1700 EXPORT_SYMBOL(free_pages
);
1702 static unsigned int nr_free_zone_pages(int offset
)
1704 /* Just pick one node, since fallback list is circular */
1705 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1706 unsigned int sum
= 0;
1708 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1709 struct zone
**zonep
= zonelist
->zones
;
1712 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1713 unsigned long size
= zone
->present_pages
;
1714 unsigned long high
= zone
->pages_high
;
1723 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1725 unsigned int nr_free_buffer_pages(void)
1727 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1729 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1732 * Amount of free RAM allocatable within all zones
1734 unsigned int nr_free_pagecache_pages(void)
1736 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1739 static inline void show_node(struct zone
*zone
)
1742 printk("Node %d ", zone_to_nid(zone
));
1745 void si_meminfo(struct sysinfo
*val
)
1747 val
->totalram
= totalram_pages
;
1749 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1750 val
->bufferram
= nr_blockdev_pages();
1751 val
->totalhigh
= totalhigh_pages
;
1752 val
->freehigh
= nr_free_highpages();
1753 val
->mem_unit
= PAGE_SIZE
;
1756 EXPORT_SYMBOL(si_meminfo
);
1759 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1761 pg_data_t
*pgdat
= NODE_DATA(nid
);
1763 val
->totalram
= pgdat
->node_present_pages
;
1764 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1765 #ifdef CONFIG_HIGHMEM
1766 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1767 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1773 val
->mem_unit
= PAGE_SIZE
;
1777 #define K(x) ((x) << (PAGE_SHIFT-10))
1780 * Show free area list (used inside shift_scroll-lock stuff)
1781 * We also calculate the percentage fragmentation. We do this by counting the
1782 * memory on each free list with the exception of the first item on the list.
1784 void show_free_areas(void)
1789 for_each_zone(zone
) {
1790 if (!populated_zone(zone
))
1794 printk("%s per-cpu:\n", zone
->name
);
1796 for_each_online_cpu(cpu
) {
1797 struct per_cpu_pageset
*pageset
;
1799 pageset
= zone_pcp(zone
, cpu
);
1801 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1802 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1803 cpu
, pageset
->pcp
[0].high
,
1804 pageset
->pcp
[0].batch
, pageset
->pcp
[0].count
,
1805 pageset
->pcp
[1].high
, pageset
->pcp
[1].batch
,
1806 pageset
->pcp
[1].count
);
1810 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1811 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1812 global_page_state(NR_ACTIVE
),
1813 global_page_state(NR_INACTIVE
),
1814 global_page_state(NR_FILE_DIRTY
),
1815 global_page_state(NR_WRITEBACK
),
1816 global_page_state(NR_UNSTABLE_NFS
),
1817 global_page_state(NR_FREE_PAGES
),
1818 global_page_state(NR_SLAB_RECLAIMABLE
) +
1819 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1820 global_page_state(NR_FILE_MAPPED
),
1821 global_page_state(NR_PAGETABLE
),
1822 global_page_state(NR_BOUNCE
));
1824 for_each_zone(zone
) {
1827 if (!populated_zone(zone
))
1839 " pages_scanned:%lu"
1840 " all_unreclaimable? %s"
1843 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1846 K(zone
->pages_high
),
1847 K(zone_page_state(zone
, NR_ACTIVE
)),
1848 K(zone_page_state(zone
, NR_INACTIVE
)),
1849 K(zone
->present_pages
),
1850 zone
->pages_scanned
,
1851 (zone
->all_unreclaimable
? "yes" : "no")
1853 printk("lowmem_reserve[]:");
1854 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1855 printk(" %lu", zone
->lowmem_reserve
[i
]);
1859 for_each_zone(zone
) {
1860 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1862 if (!populated_zone(zone
))
1866 printk("%s: ", zone
->name
);
1868 spin_lock_irqsave(&zone
->lock
, flags
);
1869 for (order
= 0; order
< MAX_ORDER
; order
++) {
1870 nr
[order
] = zone
->free_area
[order
].nr_free
;
1871 total
+= nr
[order
] << order
;
1873 spin_unlock_irqrestore(&zone
->lock
, flags
);
1874 for (order
= 0; order
< MAX_ORDER
; order
++)
1875 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1876 printk("= %lukB\n", K(total
));
1879 show_swap_cache_info();
1883 * Builds allocation fallback zone lists.
1885 * Add all populated zones of a node to the zonelist.
1887 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1888 int nr_zones
, enum zone_type zone_type
)
1892 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1897 zone
= pgdat
->node_zones
+ zone_type
;
1898 if (populated_zone(zone
)) {
1899 zonelist
->zones
[nr_zones
++] = zone
;
1900 check_highest_zone(zone_type
);
1903 } while (zone_type
);
1910 * 0 = automatic detection of better ordering.
1911 * 1 = order by ([node] distance, -zonetype)
1912 * 2 = order by (-zonetype, [node] distance)
1914 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1915 * the same zonelist. So only NUMA can configure this param.
1917 #define ZONELIST_ORDER_DEFAULT 0
1918 #define ZONELIST_ORDER_NODE 1
1919 #define ZONELIST_ORDER_ZONE 2
1921 /* zonelist order in the kernel.
1922 * set_zonelist_order() will set this to NODE or ZONE.
1924 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1925 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
1929 /* The value user specified ....changed by config */
1930 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1931 /* string for sysctl */
1932 #define NUMA_ZONELIST_ORDER_LEN 16
1933 char numa_zonelist_order
[16] = "default";
1936 * interface for configure zonelist ordering.
1937 * command line option "numa_zonelist_order"
1938 * = "[dD]efault - default, automatic configuration.
1939 * = "[nN]ode - order by node locality, then by zone within node
1940 * = "[zZ]one - order by zone, then by locality within zone
1943 static int __parse_numa_zonelist_order(char *s
)
1945 if (*s
== 'd' || *s
== 'D') {
1946 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1947 } else if (*s
== 'n' || *s
== 'N') {
1948 user_zonelist_order
= ZONELIST_ORDER_NODE
;
1949 } else if (*s
== 'z' || *s
== 'Z') {
1950 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
1953 "Ignoring invalid numa_zonelist_order value: "
1960 static __init
int setup_numa_zonelist_order(char *s
)
1963 return __parse_numa_zonelist_order(s
);
1966 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
1969 * sysctl handler for numa_zonelist_order
1971 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
1972 struct file
*file
, void __user
*buffer
, size_t *length
,
1975 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
1979 strncpy(saved_string
, (char*)table
->data
,
1980 NUMA_ZONELIST_ORDER_LEN
);
1981 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
1985 int oldval
= user_zonelist_order
;
1986 if (__parse_numa_zonelist_order((char*)table
->data
)) {
1988 * bogus value. restore saved string
1990 strncpy((char*)table
->data
, saved_string
,
1991 NUMA_ZONELIST_ORDER_LEN
);
1992 user_zonelist_order
= oldval
;
1993 } else if (oldval
!= user_zonelist_order
)
1994 build_all_zonelists();
2000 #define MAX_NODE_LOAD (num_online_nodes())
2001 static int node_load
[MAX_NUMNODES
];
2004 * find_next_best_node - find the next node that should appear in a given node's fallback list
2005 * @node: node whose fallback list we're appending
2006 * @used_node_mask: nodemask_t of already used nodes
2008 * We use a number of factors to determine which is the next node that should
2009 * appear on a given node's fallback list. The node should not have appeared
2010 * already in @node's fallback list, and it should be the next closest node
2011 * according to the distance array (which contains arbitrary distance values
2012 * from each node to each node in the system), and should also prefer nodes
2013 * with no CPUs, since presumably they'll have very little allocation pressure
2014 * on them otherwise.
2015 * It returns -1 if no node is found.
2017 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2020 int min_val
= INT_MAX
;
2023 /* Use the local node if we haven't already */
2024 if (!node_isset(node
, *used_node_mask
)) {
2025 node_set(node
, *used_node_mask
);
2029 for_each_node_state(n
, N_HIGH_MEMORY
) {
2032 /* Don't want a node to appear more than once */
2033 if (node_isset(n
, *used_node_mask
))
2036 /* Use the distance array to find the distance */
2037 val
= node_distance(node
, n
);
2039 /* Penalize nodes under us ("prefer the next node") */
2042 /* Give preference to headless and unused nodes */
2043 tmp
= node_to_cpumask(n
);
2044 if (!cpus_empty(tmp
))
2045 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2047 /* Slight preference for less loaded node */
2048 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2049 val
+= node_load
[n
];
2051 if (val
< min_val
) {
2058 node_set(best_node
, *used_node_mask
);
2065 * Build zonelists ordered by node and zones within node.
2066 * This results in maximum locality--normal zone overflows into local
2067 * DMA zone, if any--but risks exhausting DMA zone.
2069 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2073 struct zonelist
*zonelist
;
2075 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2076 zonelist
= pgdat
->node_zonelists
+ i
;
2077 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++)
2079 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2080 zonelist
->zones
[j
] = NULL
;
2085 * Build gfp_thisnode zonelists
2087 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2091 struct zonelist
*zonelist
;
2093 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2094 zonelist
= pgdat
->node_zonelists
+ MAX_NR_ZONES
+ i
;
2095 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2096 zonelist
->zones
[j
] = NULL
;
2101 * Build zonelists ordered by zone and nodes within zones.
2102 * This results in conserving DMA zone[s] until all Normal memory is
2103 * exhausted, but results in overflowing to remote node while memory
2104 * may still exist in local DMA zone.
2106 static int node_order
[MAX_NUMNODES
];
2108 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2112 int zone_type
; /* needs to be signed */
2114 struct zonelist
*zonelist
;
2116 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2117 zonelist
= pgdat
->node_zonelists
+ i
;
2119 for (zone_type
= i
; zone_type
>= 0; zone_type
--) {
2120 for (j
= 0; j
< nr_nodes
; j
++) {
2121 node
= node_order
[j
];
2122 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2123 if (populated_zone(z
)) {
2124 zonelist
->zones
[pos
++] = z
;
2125 check_highest_zone(zone_type
);
2129 zonelist
->zones
[pos
] = NULL
;
2133 static int default_zonelist_order(void)
2136 unsigned long low_kmem_size
,total_size
;
2140 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2141 * If they are really small and used heavily, the system can fall
2142 * into OOM very easily.
2143 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2145 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2148 for_each_online_node(nid
) {
2149 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2150 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2151 if (populated_zone(z
)) {
2152 if (zone_type
< ZONE_NORMAL
)
2153 low_kmem_size
+= z
->present_pages
;
2154 total_size
+= z
->present_pages
;
2158 if (!low_kmem_size
|| /* there are no DMA area. */
2159 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2160 return ZONELIST_ORDER_NODE
;
2162 * look into each node's config.
2163 * If there is a node whose DMA/DMA32 memory is very big area on
2164 * local memory, NODE_ORDER may be suitable.
2166 average_size
= total_size
/
2167 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2168 for_each_online_node(nid
) {
2171 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2172 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2173 if (populated_zone(z
)) {
2174 if (zone_type
< ZONE_NORMAL
)
2175 low_kmem_size
+= z
->present_pages
;
2176 total_size
+= z
->present_pages
;
2179 if (low_kmem_size
&&
2180 total_size
> average_size
&& /* ignore small node */
2181 low_kmem_size
> total_size
* 70/100)
2182 return ZONELIST_ORDER_NODE
;
2184 return ZONELIST_ORDER_ZONE
;
2187 static void set_zonelist_order(void)
2189 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2190 current_zonelist_order
= default_zonelist_order();
2192 current_zonelist_order
= user_zonelist_order
;
2195 static void build_zonelists(pg_data_t
*pgdat
)
2199 nodemask_t used_mask
;
2200 int local_node
, prev_node
;
2201 struct zonelist
*zonelist
;
2202 int order
= current_zonelist_order
;
2204 /* initialize zonelists */
2205 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2206 zonelist
= pgdat
->node_zonelists
+ i
;
2207 zonelist
->zones
[0] = NULL
;
2210 /* NUMA-aware ordering of nodes */
2211 local_node
= pgdat
->node_id
;
2212 load
= num_online_nodes();
2213 prev_node
= local_node
;
2214 nodes_clear(used_mask
);
2216 memset(node_load
, 0, sizeof(node_load
));
2217 memset(node_order
, 0, sizeof(node_order
));
2220 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2221 int distance
= node_distance(local_node
, node
);
2224 * If another node is sufficiently far away then it is better
2225 * to reclaim pages in a zone before going off node.
2227 if (distance
> RECLAIM_DISTANCE
)
2228 zone_reclaim_mode
= 1;
2231 * We don't want to pressure a particular node.
2232 * So adding penalty to the first node in same
2233 * distance group to make it round-robin.
2235 if (distance
!= node_distance(local_node
, prev_node
))
2236 node_load
[node
] = load
;
2240 if (order
== ZONELIST_ORDER_NODE
)
2241 build_zonelists_in_node_order(pgdat
, node
);
2243 node_order
[j
++] = node
; /* remember order */
2246 if (order
== ZONELIST_ORDER_ZONE
) {
2247 /* calculate node order -- i.e., DMA last! */
2248 build_zonelists_in_zone_order(pgdat
, j
);
2251 build_thisnode_zonelists(pgdat
);
2254 /* Construct the zonelist performance cache - see further mmzone.h */
2255 static void build_zonelist_cache(pg_data_t
*pgdat
)
2259 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2260 struct zonelist
*zonelist
;
2261 struct zonelist_cache
*zlc
;
2264 zonelist
= pgdat
->node_zonelists
+ i
;
2265 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2266 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2267 for (z
= zonelist
->zones
; *z
; z
++)
2268 zlc
->z_to_n
[z
- zonelist
->zones
] = zone_to_nid(*z
);
2273 #else /* CONFIG_NUMA */
2275 static void set_zonelist_order(void)
2277 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2280 static void build_zonelists(pg_data_t
*pgdat
)
2282 int node
, local_node
;
2285 local_node
= pgdat
->node_id
;
2286 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2287 struct zonelist
*zonelist
;
2289 zonelist
= pgdat
->node_zonelists
+ i
;
2291 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2293 * Now we build the zonelist so that it contains the zones
2294 * of all the other nodes.
2295 * We don't want to pressure a particular node, so when
2296 * building the zones for node N, we make sure that the
2297 * zones coming right after the local ones are those from
2298 * node N+1 (modulo N)
2300 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2301 if (!node_online(node
))
2303 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2305 for (node
= 0; node
< local_node
; node
++) {
2306 if (!node_online(node
))
2308 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2311 zonelist
->zones
[j
] = NULL
;
2315 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2316 static void build_zonelist_cache(pg_data_t
*pgdat
)
2320 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2321 pgdat
->node_zonelists
[i
].zlcache_ptr
= NULL
;
2324 #endif /* CONFIG_NUMA */
2326 /* return values int ....just for stop_machine_run() */
2327 static int __build_all_zonelists(void *dummy
)
2331 for_each_online_node(nid
) {
2332 pg_data_t
*pgdat
= NODE_DATA(nid
);
2334 build_zonelists(pgdat
);
2335 build_zonelist_cache(pgdat
);
2340 void build_all_zonelists(void)
2342 set_zonelist_order();
2344 if (system_state
== SYSTEM_BOOTING
) {
2345 __build_all_zonelists(NULL
);
2346 cpuset_init_current_mems_allowed();
2348 /* we have to stop all cpus to guaranntee there is no user
2350 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
2351 /* cpuset refresh routine should be here */
2353 vm_total_pages
= nr_free_pagecache_pages();
2354 printk("Built %i zonelists in %s order. Total pages: %ld\n",
2356 zonelist_order_name
[current_zonelist_order
],
2359 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2364 * Helper functions to size the waitqueue hash table.
2365 * Essentially these want to choose hash table sizes sufficiently
2366 * large so that collisions trying to wait on pages are rare.
2367 * But in fact, the number of active page waitqueues on typical
2368 * systems is ridiculously low, less than 200. So this is even
2369 * conservative, even though it seems large.
2371 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2372 * waitqueues, i.e. the size of the waitq table given the number of pages.
2374 #define PAGES_PER_WAITQUEUE 256
2376 #ifndef CONFIG_MEMORY_HOTPLUG
2377 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2379 unsigned long size
= 1;
2381 pages
/= PAGES_PER_WAITQUEUE
;
2383 while (size
< pages
)
2387 * Once we have dozens or even hundreds of threads sleeping
2388 * on IO we've got bigger problems than wait queue collision.
2389 * Limit the size of the wait table to a reasonable size.
2391 size
= min(size
, 4096UL);
2393 return max(size
, 4UL);
2397 * A zone's size might be changed by hot-add, so it is not possible to determine
2398 * a suitable size for its wait_table. So we use the maximum size now.
2400 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2402 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2403 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2404 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2406 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2407 * or more by the traditional way. (See above). It equals:
2409 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2410 * ia64(16K page size) : = ( 8G + 4M)byte.
2411 * powerpc (64K page size) : = (32G +16M)byte.
2413 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2420 * This is an integer logarithm so that shifts can be used later
2421 * to extract the more random high bits from the multiplicative
2422 * hash function before the remainder is taken.
2424 static inline unsigned long wait_table_bits(unsigned long size
)
2429 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2432 * Initially all pages are reserved - free ones are freed
2433 * up by free_all_bootmem() once the early boot process is
2434 * done. Non-atomic initialization, single-pass.
2436 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2437 unsigned long start_pfn
, enum memmap_context context
)
2440 unsigned long end_pfn
= start_pfn
+ size
;
2443 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2445 * There can be holes in boot-time mem_map[]s
2446 * handed to this function. They do not
2447 * exist on hotplugged memory.
2449 if (context
== MEMMAP_EARLY
) {
2450 if (!early_pfn_valid(pfn
))
2452 if (!early_pfn_in_nid(pfn
, nid
))
2455 page
= pfn_to_page(pfn
);
2456 set_page_links(page
, zone
, nid
, pfn
);
2457 init_page_count(page
);
2458 reset_page_mapcount(page
);
2459 SetPageReserved(page
);
2462 * Mark the block movable so that blocks are reserved for
2463 * movable at startup. This will force kernel allocations
2464 * to reserve their blocks rather than leaking throughout
2465 * the address space during boot when many long-lived
2466 * kernel allocations are made
2468 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2470 INIT_LIST_HEAD(&page
->lru
);
2471 #ifdef WANT_PAGE_VIRTUAL
2472 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2473 if (!is_highmem_idx(zone
))
2474 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2479 static void __meminit
zone_init_free_lists(struct pglist_data
*pgdat
,
2480 struct zone
*zone
, unsigned long size
)
2483 for_each_migratetype_order(order
, t
) {
2484 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2485 zone
->free_area
[order
].nr_free
= 0;
2489 #ifndef __HAVE_ARCH_MEMMAP_INIT
2490 #define memmap_init(size, nid, zone, start_pfn) \
2491 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2494 static int __devinit
zone_batchsize(struct zone
*zone
)
2499 * The per-cpu-pages pools are set to around 1000th of the
2500 * size of the zone. But no more than 1/2 of a meg.
2502 * OK, so we don't know how big the cache is. So guess.
2504 batch
= zone
->present_pages
/ 1024;
2505 if (batch
* PAGE_SIZE
> 512 * 1024)
2506 batch
= (512 * 1024) / PAGE_SIZE
;
2507 batch
/= 4; /* We effectively *= 4 below */
2512 * Clamp the batch to a 2^n - 1 value. Having a power
2513 * of 2 value was found to be more likely to have
2514 * suboptimal cache aliasing properties in some cases.
2516 * For example if 2 tasks are alternately allocating
2517 * batches of pages, one task can end up with a lot
2518 * of pages of one half of the possible page colors
2519 * and the other with pages of the other colors.
2521 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2526 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2528 struct per_cpu_pages
*pcp
;
2530 memset(p
, 0, sizeof(*p
));
2532 pcp
= &p
->pcp
[0]; /* hot */
2534 pcp
->high
= 6 * batch
;
2535 pcp
->batch
= max(1UL, 1 * batch
);
2536 INIT_LIST_HEAD(&pcp
->list
);
2538 pcp
= &p
->pcp
[1]; /* cold*/
2540 pcp
->high
= 2 * batch
;
2541 pcp
->batch
= max(1UL, batch
/2);
2542 INIT_LIST_HEAD(&pcp
->list
);
2546 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2547 * to the value high for the pageset p.
2550 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2553 struct per_cpu_pages
*pcp
;
2555 pcp
= &p
->pcp
[0]; /* hot list */
2557 pcp
->batch
= max(1UL, high
/4);
2558 if ((high
/4) > (PAGE_SHIFT
* 8))
2559 pcp
->batch
= PAGE_SHIFT
* 8;
2565 * Boot pageset table. One per cpu which is going to be used for all
2566 * zones and all nodes. The parameters will be set in such a way
2567 * that an item put on a list will immediately be handed over to
2568 * the buddy list. This is safe since pageset manipulation is done
2569 * with interrupts disabled.
2571 * Some NUMA counter updates may also be caught by the boot pagesets.
2573 * The boot_pagesets must be kept even after bootup is complete for
2574 * unused processors and/or zones. They do play a role for bootstrapping
2575 * hotplugged processors.
2577 * zoneinfo_show() and maybe other functions do
2578 * not check if the processor is online before following the pageset pointer.
2579 * Other parts of the kernel may not check if the zone is available.
2581 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2584 * Dynamically allocate memory for the
2585 * per cpu pageset array in struct zone.
2587 static int __cpuinit
process_zones(int cpu
)
2589 struct zone
*zone
, *dzone
;
2590 int node
= cpu_to_node(cpu
);
2592 node_set_state(node
, N_CPU
); /* this node has a cpu */
2594 for_each_zone(zone
) {
2596 if (!populated_zone(zone
))
2599 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2601 if (!zone_pcp(zone
, cpu
))
2604 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2606 if (percpu_pagelist_fraction
)
2607 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2608 (zone
->present_pages
/ percpu_pagelist_fraction
));
2613 for_each_zone(dzone
) {
2614 if (!populated_zone(dzone
))
2618 kfree(zone_pcp(dzone
, cpu
));
2619 zone_pcp(dzone
, cpu
) = NULL
;
2624 static inline void free_zone_pagesets(int cpu
)
2628 for_each_zone(zone
) {
2629 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2631 /* Free per_cpu_pageset if it is slab allocated */
2632 if (pset
!= &boot_pageset
[cpu
])
2634 zone_pcp(zone
, cpu
) = NULL
;
2638 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2639 unsigned long action
,
2642 int cpu
= (long)hcpu
;
2643 int ret
= NOTIFY_OK
;
2646 case CPU_UP_PREPARE
:
2647 case CPU_UP_PREPARE_FROZEN
:
2648 if (process_zones(cpu
))
2651 case CPU_UP_CANCELED
:
2652 case CPU_UP_CANCELED_FROZEN
:
2654 case CPU_DEAD_FROZEN
:
2655 free_zone_pagesets(cpu
);
2663 static struct notifier_block __cpuinitdata pageset_notifier
=
2664 { &pageset_cpuup_callback
, NULL
, 0 };
2666 void __init
setup_per_cpu_pageset(void)
2670 /* Initialize per_cpu_pageset for cpu 0.
2671 * A cpuup callback will do this for every cpu
2672 * as it comes online
2674 err
= process_zones(smp_processor_id());
2676 register_cpu_notifier(&pageset_notifier
);
2681 static noinline __init_refok
2682 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2685 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2689 * The per-page waitqueue mechanism uses hashed waitqueues
2692 zone
->wait_table_hash_nr_entries
=
2693 wait_table_hash_nr_entries(zone_size_pages
);
2694 zone
->wait_table_bits
=
2695 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2696 alloc_size
= zone
->wait_table_hash_nr_entries
2697 * sizeof(wait_queue_head_t
);
2699 if (system_state
== SYSTEM_BOOTING
) {
2700 zone
->wait_table
= (wait_queue_head_t
*)
2701 alloc_bootmem_node(pgdat
, alloc_size
);
2704 * This case means that a zone whose size was 0 gets new memory
2705 * via memory hot-add.
2706 * But it may be the case that a new node was hot-added. In
2707 * this case vmalloc() will not be able to use this new node's
2708 * memory - this wait_table must be initialized to use this new
2709 * node itself as well.
2710 * To use this new node's memory, further consideration will be
2713 zone
->wait_table
= vmalloc(alloc_size
);
2715 if (!zone
->wait_table
)
2718 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2719 init_waitqueue_head(zone
->wait_table
+ i
);
2724 static __meminit
void zone_pcp_init(struct zone
*zone
)
2727 unsigned long batch
= zone_batchsize(zone
);
2729 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2731 /* Early boot. Slab allocator not functional yet */
2732 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2733 setup_pageset(&boot_pageset
[cpu
],0);
2735 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2738 if (zone
->present_pages
)
2739 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2740 zone
->name
, zone
->present_pages
, batch
);
2743 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2744 unsigned long zone_start_pfn
,
2746 enum memmap_context context
)
2748 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2750 ret
= zone_wait_table_init(zone
, size
);
2753 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2755 zone
->zone_start_pfn
= zone_start_pfn
;
2757 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
2759 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2764 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2766 * Basic iterator support. Return the first range of PFNs for a node
2767 * Note: nid == MAX_NUMNODES returns first region regardless of node
2769 static int __meminit
first_active_region_index_in_nid(int nid
)
2773 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2774 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2781 * Basic iterator support. Return the next active range of PFNs for a node
2782 * Note: nid == MAX_NUMNODES returns next region regardles of node
2784 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2786 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2787 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2793 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2795 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2796 * Architectures may implement their own version but if add_active_range()
2797 * was used and there are no special requirements, this is a convenient
2800 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2804 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2805 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2806 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2808 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2809 return early_node_map
[i
].nid
;
2814 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2816 /* Basic iterator support to walk early_node_map[] */
2817 #define for_each_active_range_index_in_nid(i, nid) \
2818 for (i = first_active_region_index_in_nid(nid); i != -1; \
2819 i = next_active_region_index_in_nid(i, nid))
2822 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2823 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2824 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2826 * If an architecture guarantees that all ranges registered with
2827 * add_active_ranges() contain no holes and may be freed, this
2828 * this function may be used instead of calling free_bootmem() manually.
2830 void __init
free_bootmem_with_active_regions(int nid
,
2831 unsigned long max_low_pfn
)
2835 for_each_active_range_index_in_nid(i
, nid
) {
2836 unsigned long size_pages
= 0;
2837 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2839 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2842 if (end_pfn
> max_low_pfn
)
2843 end_pfn
= max_low_pfn
;
2845 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2846 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2847 PFN_PHYS(early_node_map
[i
].start_pfn
),
2848 size_pages
<< PAGE_SHIFT
);
2853 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2854 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2856 * If an architecture guarantees that all ranges registered with
2857 * add_active_ranges() contain no holes and may be freed, this
2858 * function may be used instead of calling memory_present() manually.
2860 void __init
sparse_memory_present_with_active_regions(int nid
)
2864 for_each_active_range_index_in_nid(i
, nid
)
2865 memory_present(early_node_map
[i
].nid
,
2866 early_node_map
[i
].start_pfn
,
2867 early_node_map
[i
].end_pfn
);
2871 * push_node_boundaries - Push node boundaries to at least the requested boundary
2872 * @nid: The nid of the node to push the boundary for
2873 * @start_pfn: The start pfn of the node
2874 * @end_pfn: The end pfn of the node
2876 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2877 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2878 * be hotplugged even though no physical memory exists. This function allows
2879 * an arch to push out the node boundaries so mem_map is allocated that can
2882 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2883 void __init
push_node_boundaries(unsigned int nid
,
2884 unsigned long start_pfn
, unsigned long end_pfn
)
2886 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2887 nid
, start_pfn
, end_pfn
);
2889 /* Initialise the boundary for this node if necessary */
2890 if (node_boundary_end_pfn
[nid
] == 0)
2891 node_boundary_start_pfn
[nid
] = -1UL;
2893 /* Update the boundaries */
2894 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2895 node_boundary_start_pfn
[nid
] = start_pfn
;
2896 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2897 node_boundary_end_pfn
[nid
] = end_pfn
;
2900 /* If necessary, push the node boundary out for reserve hotadd */
2901 static void __meminit
account_node_boundary(unsigned int nid
,
2902 unsigned long *start_pfn
, unsigned long *end_pfn
)
2904 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2905 nid
, *start_pfn
, *end_pfn
);
2907 /* Return if boundary information has not been provided */
2908 if (node_boundary_end_pfn
[nid
] == 0)
2911 /* Check the boundaries and update if necessary */
2912 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
2913 *start_pfn
= node_boundary_start_pfn
[nid
];
2914 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
2915 *end_pfn
= node_boundary_end_pfn
[nid
];
2918 void __init
push_node_boundaries(unsigned int nid
,
2919 unsigned long start_pfn
, unsigned long end_pfn
) {}
2921 static void __meminit
account_node_boundary(unsigned int nid
,
2922 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
2927 * get_pfn_range_for_nid - Return the start and end page frames for a node
2928 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2929 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2930 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2932 * It returns the start and end page frame of a node based on information
2933 * provided by an arch calling add_active_range(). If called for a node
2934 * with no available memory, a warning is printed and the start and end
2937 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
2938 unsigned long *start_pfn
, unsigned long *end_pfn
)
2944 for_each_active_range_index_in_nid(i
, nid
) {
2945 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
2946 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
2949 if (*start_pfn
== -1UL)
2952 /* Push the node boundaries out if requested */
2953 account_node_boundary(nid
, start_pfn
, end_pfn
);
2957 * This finds a zone that can be used for ZONE_MOVABLE pages. The
2958 * assumption is made that zones within a node are ordered in monotonic
2959 * increasing memory addresses so that the "highest" populated zone is used
2961 void __init
find_usable_zone_for_movable(void)
2964 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
2965 if (zone_index
== ZONE_MOVABLE
)
2968 if (arch_zone_highest_possible_pfn
[zone_index
] >
2969 arch_zone_lowest_possible_pfn
[zone_index
])
2973 VM_BUG_ON(zone_index
== -1);
2974 movable_zone
= zone_index
;
2978 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
2979 * because it is sized independant of architecture. Unlike the other zones,
2980 * the starting point for ZONE_MOVABLE is not fixed. It may be different
2981 * in each node depending on the size of each node and how evenly kernelcore
2982 * is distributed. This helper function adjusts the zone ranges
2983 * provided by the architecture for a given node by using the end of the
2984 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
2985 * zones within a node are in order of monotonic increases memory addresses
2987 void __meminit
adjust_zone_range_for_zone_movable(int nid
,
2988 unsigned long zone_type
,
2989 unsigned long node_start_pfn
,
2990 unsigned long node_end_pfn
,
2991 unsigned long *zone_start_pfn
,
2992 unsigned long *zone_end_pfn
)
2994 /* Only adjust if ZONE_MOVABLE is on this node */
2995 if (zone_movable_pfn
[nid
]) {
2996 /* Size ZONE_MOVABLE */
2997 if (zone_type
== ZONE_MOVABLE
) {
2998 *zone_start_pfn
= zone_movable_pfn
[nid
];
2999 *zone_end_pfn
= min(node_end_pfn
,
3000 arch_zone_highest_possible_pfn
[movable_zone
]);
3002 /* Adjust for ZONE_MOVABLE starting within this range */
3003 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3004 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3005 *zone_end_pfn
= zone_movable_pfn
[nid
];
3007 /* Check if this whole range is within ZONE_MOVABLE */
3008 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3009 *zone_start_pfn
= *zone_end_pfn
;
3014 * Return the number of pages a zone spans in a node, including holes
3015 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3017 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3018 unsigned long zone_type
,
3019 unsigned long *ignored
)
3021 unsigned long node_start_pfn
, node_end_pfn
;
3022 unsigned long zone_start_pfn
, zone_end_pfn
;
3024 /* Get the start and end of the node and zone */
3025 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3026 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3027 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3028 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3029 node_start_pfn
, node_end_pfn
,
3030 &zone_start_pfn
, &zone_end_pfn
);
3032 /* Check that this node has pages within the zone's required range */
3033 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3036 /* Move the zone boundaries inside the node if necessary */
3037 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3038 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3040 /* Return the spanned pages */
3041 return zone_end_pfn
- zone_start_pfn
;
3045 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3046 * then all holes in the requested range will be accounted for.
3048 unsigned long __meminit
__absent_pages_in_range(int nid
,
3049 unsigned long range_start_pfn
,
3050 unsigned long range_end_pfn
)
3053 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3054 unsigned long start_pfn
;
3056 /* Find the end_pfn of the first active range of pfns in the node */
3057 i
= first_active_region_index_in_nid(nid
);
3061 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3063 /* Account for ranges before physical memory on this node */
3064 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3065 hole_pages
= prev_end_pfn
- range_start_pfn
;
3067 /* Find all holes for the zone within the node */
3068 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3070 /* No need to continue if prev_end_pfn is outside the zone */
3071 if (prev_end_pfn
>= range_end_pfn
)
3074 /* Make sure the end of the zone is not within the hole */
3075 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3076 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3078 /* Update the hole size cound and move on */
3079 if (start_pfn
> range_start_pfn
) {
3080 BUG_ON(prev_end_pfn
> start_pfn
);
3081 hole_pages
+= start_pfn
- prev_end_pfn
;
3083 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3086 /* Account for ranges past physical memory on this node */
3087 if (range_end_pfn
> prev_end_pfn
)
3088 hole_pages
+= range_end_pfn
-
3089 max(range_start_pfn
, prev_end_pfn
);
3095 * absent_pages_in_range - Return number of page frames in holes within a range
3096 * @start_pfn: The start PFN to start searching for holes
3097 * @end_pfn: The end PFN to stop searching for holes
3099 * It returns the number of pages frames in memory holes within a range.
3101 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3102 unsigned long end_pfn
)
3104 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3107 /* Return the number of page frames in holes in a zone on a node */
3108 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3109 unsigned long zone_type
,
3110 unsigned long *ignored
)
3112 unsigned long node_start_pfn
, node_end_pfn
;
3113 unsigned long zone_start_pfn
, zone_end_pfn
;
3115 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3116 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3118 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3121 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3122 node_start_pfn
, node_end_pfn
,
3123 &zone_start_pfn
, &zone_end_pfn
);
3124 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3128 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3129 unsigned long zone_type
,
3130 unsigned long *zones_size
)
3132 return zones_size
[zone_type
];
3135 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3136 unsigned long zone_type
,
3137 unsigned long *zholes_size
)
3142 return zholes_size
[zone_type
];
3147 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3148 unsigned long *zones_size
, unsigned long *zholes_size
)
3150 unsigned long realtotalpages
, totalpages
= 0;
3153 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3154 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3156 pgdat
->node_spanned_pages
= totalpages
;
3158 realtotalpages
= totalpages
;
3159 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3161 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3163 pgdat
->node_present_pages
= realtotalpages
;
3164 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3168 #ifndef CONFIG_SPARSEMEM
3170 * Calculate the size of the zone->blockflags rounded to an unsigned long
3171 * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up
3172 * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally
3173 * round what is now in bits to nearest long in bits, then return it in
3176 static unsigned long __init
usemap_size(unsigned long zonesize
)
3178 unsigned long usemapsize
;
3180 usemapsize
= roundup(zonesize
, MAX_ORDER_NR_PAGES
);
3181 usemapsize
= usemapsize
>> (MAX_ORDER
-1);
3182 usemapsize
*= NR_PAGEBLOCK_BITS
;
3183 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3185 return usemapsize
/ 8;
3188 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3189 struct zone
*zone
, unsigned long zonesize
)
3191 unsigned long usemapsize
= usemap_size(zonesize
);
3192 zone
->pageblock_flags
= NULL
;
3194 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3195 memset(zone
->pageblock_flags
, 0, usemapsize
);
3199 static void inline setup_usemap(struct pglist_data
*pgdat
,
3200 struct zone
*zone
, unsigned long zonesize
) {}
3201 #endif /* CONFIG_SPARSEMEM */
3204 * Set up the zone data structures:
3205 * - mark all pages reserved
3206 * - mark all memory queues empty
3207 * - clear the memory bitmaps
3209 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
3210 unsigned long *zones_size
, unsigned long *zholes_size
)
3213 int nid
= pgdat
->node_id
;
3214 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3217 pgdat_resize_init(pgdat
);
3218 pgdat
->nr_zones
= 0;
3219 init_waitqueue_head(&pgdat
->kswapd_wait
);
3220 pgdat
->kswapd_max_order
= 0;
3222 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3223 struct zone
*zone
= pgdat
->node_zones
+ j
;
3224 unsigned long size
, realsize
, memmap_pages
;
3226 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3227 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3231 * Adjust realsize so that it accounts for how much memory
3232 * is used by this zone for memmap. This affects the watermark
3233 * and per-cpu initialisations
3235 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3236 if (realsize
>= memmap_pages
) {
3237 realsize
-= memmap_pages
;
3239 " %s zone: %lu pages used for memmap\n",
3240 zone_names
[j
], memmap_pages
);
3243 " %s zone: %lu pages exceeds realsize %lu\n",
3244 zone_names
[j
], memmap_pages
, realsize
);
3246 /* Account for reserved pages */
3247 if (j
== 0 && realsize
> dma_reserve
) {
3248 realsize
-= dma_reserve
;
3249 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3250 zone_names
[0], dma_reserve
);
3253 if (!is_highmem_idx(j
))
3254 nr_kernel_pages
+= realsize
;
3255 nr_all_pages
+= realsize
;
3257 zone
->spanned_pages
= size
;
3258 zone
->present_pages
= realsize
;
3261 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3263 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3265 zone
->name
= zone_names
[j
];
3266 spin_lock_init(&zone
->lock
);
3267 spin_lock_init(&zone
->lru_lock
);
3268 zone_seqlock_init(zone
);
3269 zone
->zone_pgdat
= pgdat
;
3271 zone
->prev_priority
= DEF_PRIORITY
;
3273 zone_pcp_init(zone
);
3274 INIT_LIST_HEAD(&zone
->active_list
);
3275 INIT_LIST_HEAD(&zone
->inactive_list
);
3276 zone
->nr_scan_active
= 0;
3277 zone
->nr_scan_inactive
= 0;
3278 zap_zone_vm_stats(zone
);
3279 atomic_set(&zone
->reclaim_in_progress
, 0);
3283 setup_usemap(pgdat
, zone
, size
);
3284 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3285 size
, MEMMAP_EARLY
);
3287 zone_start_pfn
+= size
;
3291 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3293 /* Skip empty nodes */
3294 if (!pgdat
->node_spanned_pages
)
3297 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3298 /* ia64 gets its own node_mem_map, before this, without bootmem */
3299 if (!pgdat
->node_mem_map
) {
3300 unsigned long size
, start
, end
;
3304 * The zone's endpoints aren't required to be MAX_ORDER
3305 * aligned but the node_mem_map endpoints must be in order
3306 * for the buddy allocator to function correctly.
3308 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3309 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3310 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3311 size
= (end
- start
) * sizeof(struct page
);
3312 map
= alloc_remap(pgdat
->node_id
, size
);
3314 map
= alloc_bootmem_node(pgdat
, size
);
3315 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3317 #ifndef CONFIG_NEED_MULTIPLE_NODES
3319 * With no DISCONTIG, the global mem_map is just set as node 0's
3321 if (pgdat
== NODE_DATA(0)) {
3322 mem_map
= NODE_DATA(0)->node_mem_map
;
3323 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3324 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3325 mem_map
-= pgdat
->node_start_pfn
;
3326 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3329 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3332 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
3333 unsigned long *zones_size
, unsigned long node_start_pfn
,
3334 unsigned long *zholes_size
)
3336 pgdat
->node_id
= nid
;
3337 pgdat
->node_start_pfn
= node_start_pfn
;
3338 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3340 alloc_node_mem_map(pgdat
);
3342 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3345 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3347 #if MAX_NUMNODES > 1
3349 * Figure out the number of possible node ids.
3351 static void __init
setup_nr_node_ids(void)
3354 unsigned int highest
= 0;
3356 for_each_node_mask(node
, node_possible_map
)
3358 nr_node_ids
= highest
+ 1;
3361 static inline void setup_nr_node_ids(void)
3367 * add_active_range - Register a range of PFNs backed by physical memory
3368 * @nid: The node ID the range resides on
3369 * @start_pfn: The start PFN of the available physical memory
3370 * @end_pfn: The end PFN of the available physical memory
3372 * These ranges are stored in an early_node_map[] and later used by
3373 * free_area_init_nodes() to calculate zone sizes and holes. If the
3374 * range spans a memory hole, it is up to the architecture to ensure
3375 * the memory is not freed by the bootmem allocator. If possible
3376 * the range being registered will be merged with existing ranges.
3378 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3379 unsigned long end_pfn
)
3383 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
3384 "%d entries of %d used\n",
3385 nid
, start_pfn
, end_pfn
,
3386 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3388 /* Merge with existing active regions if possible */
3389 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3390 if (early_node_map
[i
].nid
!= nid
)
3393 /* Skip if an existing region covers this new one */
3394 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3395 end_pfn
<= early_node_map
[i
].end_pfn
)
3398 /* Merge forward if suitable */
3399 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3400 end_pfn
> early_node_map
[i
].end_pfn
) {
3401 early_node_map
[i
].end_pfn
= end_pfn
;
3405 /* Merge backward if suitable */
3406 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3407 end_pfn
>= early_node_map
[i
].start_pfn
) {
3408 early_node_map
[i
].start_pfn
= start_pfn
;
3413 /* Check that early_node_map is large enough */
3414 if (i
>= MAX_ACTIVE_REGIONS
) {
3415 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3416 MAX_ACTIVE_REGIONS
);
3420 early_node_map
[i
].nid
= nid
;
3421 early_node_map
[i
].start_pfn
= start_pfn
;
3422 early_node_map
[i
].end_pfn
= end_pfn
;
3423 nr_nodemap_entries
= i
+ 1;
3427 * shrink_active_range - Shrink an existing registered range of PFNs
3428 * @nid: The node id the range is on that should be shrunk
3429 * @old_end_pfn: The old end PFN of the range
3430 * @new_end_pfn: The new PFN of the range
3432 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3433 * The map is kept at the end physical page range that has already been
3434 * registered with add_active_range(). This function allows an arch to shrink
3435 * an existing registered range.
3437 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
3438 unsigned long new_end_pfn
)
3442 /* Find the old active region end and shrink */
3443 for_each_active_range_index_in_nid(i
, nid
)
3444 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
3445 early_node_map
[i
].end_pfn
= new_end_pfn
;
3451 * remove_all_active_ranges - Remove all currently registered regions
3453 * During discovery, it may be found that a table like SRAT is invalid
3454 * and an alternative discovery method must be used. This function removes
3455 * all currently registered regions.
3457 void __init
remove_all_active_ranges(void)
3459 memset(early_node_map
, 0, sizeof(early_node_map
));
3460 nr_nodemap_entries
= 0;
3461 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3462 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3463 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3464 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3467 /* Compare two active node_active_regions */
3468 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3470 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3471 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3473 /* Done this way to avoid overflows */
3474 if (arange
->start_pfn
> brange
->start_pfn
)
3476 if (arange
->start_pfn
< brange
->start_pfn
)
3482 /* sort the node_map by start_pfn */
3483 static void __init
sort_node_map(void)
3485 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3486 sizeof(struct node_active_region
),
3487 cmp_node_active_region
, NULL
);
3490 /* Find the lowest pfn for a node */
3491 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
3494 unsigned long min_pfn
= ULONG_MAX
;
3496 /* Assuming a sorted map, the first range found has the starting pfn */
3497 for_each_active_range_index_in_nid(i
, nid
)
3498 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3500 if (min_pfn
== ULONG_MAX
) {
3502 "Could not find start_pfn for node %lu\n", nid
);
3510 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3512 * It returns the minimum PFN based on information provided via
3513 * add_active_range().
3515 unsigned long __init
find_min_pfn_with_active_regions(void)
3517 return find_min_pfn_for_node(MAX_NUMNODES
);
3521 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3523 * It returns the maximum PFN based on information provided via
3524 * add_active_range().
3526 unsigned long __init
find_max_pfn_with_active_regions(void)
3529 unsigned long max_pfn
= 0;
3531 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3532 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
3538 * early_calculate_totalpages()
3539 * Sum pages in active regions for movable zone.
3540 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3542 unsigned long __init
early_calculate_totalpages(void)
3545 unsigned long totalpages
= 0;
3547 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3548 unsigned long pages
= early_node_map
[i
].end_pfn
-
3549 early_node_map
[i
].start_pfn
;
3550 totalpages
+= pages
;
3552 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3558 * Find the PFN the Movable zone begins in each node. Kernel memory
3559 * is spread evenly between nodes as long as the nodes have enough
3560 * memory. When they don't, some nodes will have more kernelcore than
3563 void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3566 unsigned long usable_startpfn
;
3567 unsigned long kernelcore_node
, kernelcore_remaining
;
3568 unsigned long totalpages
= early_calculate_totalpages();
3569 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3572 * If movablecore was specified, calculate what size of
3573 * kernelcore that corresponds so that memory usable for
3574 * any allocation type is evenly spread. If both kernelcore
3575 * and movablecore are specified, then the value of kernelcore
3576 * will be used for required_kernelcore if it's greater than
3577 * what movablecore would have allowed.
3579 if (required_movablecore
) {
3580 unsigned long corepages
;
3583 * Round-up so that ZONE_MOVABLE is at least as large as what
3584 * was requested by the user
3586 required_movablecore
=
3587 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3588 corepages
= totalpages
- required_movablecore
;
3590 required_kernelcore
= max(required_kernelcore
, corepages
);
3593 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3594 if (!required_kernelcore
)
3597 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3598 find_usable_zone_for_movable();
3599 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3602 /* Spread kernelcore memory as evenly as possible throughout nodes */
3603 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3604 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3606 * Recalculate kernelcore_node if the division per node
3607 * now exceeds what is necessary to satisfy the requested
3608 * amount of memory for the kernel
3610 if (required_kernelcore
< kernelcore_node
)
3611 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3614 * As the map is walked, we track how much memory is usable
3615 * by the kernel using kernelcore_remaining. When it is
3616 * 0, the rest of the node is usable by ZONE_MOVABLE
3618 kernelcore_remaining
= kernelcore_node
;
3620 /* Go through each range of PFNs within this node */
3621 for_each_active_range_index_in_nid(i
, nid
) {
3622 unsigned long start_pfn
, end_pfn
;
3623 unsigned long size_pages
;
3625 start_pfn
= max(early_node_map
[i
].start_pfn
,
3626 zone_movable_pfn
[nid
]);
3627 end_pfn
= early_node_map
[i
].end_pfn
;
3628 if (start_pfn
>= end_pfn
)
3631 /* Account for what is only usable for kernelcore */
3632 if (start_pfn
< usable_startpfn
) {
3633 unsigned long kernel_pages
;
3634 kernel_pages
= min(end_pfn
, usable_startpfn
)
3637 kernelcore_remaining
-= min(kernel_pages
,
3638 kernelcore_remaining
);
3639 required_kernelcore
-= min(kernel_pages
,
3640 required_kernelcore
);
3642 /* Continue if range is now fully accounted */
3643 if (end_pfn
<= usable_startpfn
) {
3646 * Push zone_movable_pfn to the end so
3647 * that if we have to rebalance
3648 * kernelcore across nodes, we will
3649 * not double account here
3651 zone_movable_pfn
[nid
] = end_pfn
;
3654 start_pfn
= usable_startpfn
;
3658 * The usable PFN range for ZONE_MOVABLE is from
3659 * start_pfn->end_pfn. Calculate size_pages as the
3660 * number of pages used as kernelcore
3662 size_pages
= end_pfn
- start_pfn
;
3663 if (size_pages
> kernelcore_remaining
)
3664 size_pages
= kernelcore_remaining
;
3665 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3668 * Some kernelcore has been met, update counts and
3669 * break if the kernelcore for this node has been
3672 required_kernelcore
-= min(required_kernelcore
,
3674 kernelcore_remaining
-= size_pages
;
3675 if (!kernelcore_remaining
)
3681 * If there is still required_kernelcore, we do another pass with one
3682 * less node in the count. This will push zone_movable_pfn[nid] further
3683 * along on the nodes that still have memory until kernelcore is
3687 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3690 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3691 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3692 zone_movable_pfn
[nid
] =
3693 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3696 /* Any regular memory on that node ? */
3697 static void check_for_regular_memory(pg_data_t
*pgdat
)
3699 #ifdef CONFIG_HIGHMEM
3700 enum zone_type zone_type
;
3702 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3703 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3704 if (zone
->present_pages
)
3705 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3711 * free_area_init_nodes - Initialise all pg_data_t and zone data
3712 * @max_zone_pfn: an array of max PFNs for each zone
3714 * This will call free_area_init_node() for each active node in the system.
3715 * Using the page ranges provided by add_active_range(), the size of each
3716 * zone in each node and their holes is calculated. If the maximum PFN
3717 * between two adjacent zones match, it is assumed that the zone is empty.
3718 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3719 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3720 * starts where the previous one ended. For example, ZONE_DMA32 starts
3721 * at arch_max_dma_pfn.
3723 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3728 /* Sort early_node_map as initialisation assumes it is sorted */
3731 /* Record where the zone boundaries are */
3732 memset(arch_zone_lowest_possible_pfn
, 0,
3733 sizeof(arch_zone_lowest_possible_pfn
));
3734 memset(arch_zone_highest_possible_pfn
, 0,
3735 sizeof(arch_zone_highest_possible_pfn
));
3736 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3737 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3738 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3739 if (i
== ZONE_MOVABLE
)
3741 arch_zone_lowest_possible_pfn
[i
] =
3742 arch_zone_highest_possible_pfn
[i
-1];
3743 arch_zone_highest_possible_pfn
[i
] =
3744 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3746 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3747 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3749 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3750 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3751 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3753 /* Print out the zone ranges */
3754 printk("Zone PFN ranges:\n");
3755 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3756 if (i
== ZONE_MOVABLE
)
3758 printk(" %-8s %8lu -> %8lu\n",
3760 arch_zone_lowest_possible_pfn
[i
],
3761 arch_zone_highest_possible_pfn
[i
]);
3764 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3765 printk("Movable zone start PFN for each node\n");
3766 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
3767 if (zone_movable_pfn
[i
])
3768 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
3771 /* Print out the early_node_map[] */
3772 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
3773 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3774 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
3775 early_node_map
[i
].start_pfn
,
3776 early_node_map
[i
].end_pfn
);
3778 /* Initialise every node */
3779 setup_nr_node_ids();
3780 for_each_online_node(nid
) {
3781 pg_data_t
*pgdat
= NODE_DATA(nid
);
3782 free_area_init_node(nid
, pgdat
, NULL
,
3783 find_min_pfn_for_node(nid
), NULL
);
3785 /* Any memory on that node */
3786 if (pgdat
->node_present_pages
)
3787 node_set_state(nid
, N_HIGH_MEMORY
);
3788 check_for_regular_memory(pgdat
);
3792 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
3794 unsigned long long coremem
;
3798 coremem
= memparse(p
, &p
);
3799 *core
= coremem
>> PAGE_SHIFT
;
3801 /* Paranoid check that UL is enough for the coremem value */
3802 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
3808 * kernelcore=size sets the amount of memory for use for allocations that
3809 * cannot be reclaimed or migrated.
3811 static int __init
cmdline_parse_kernelcore(char *p
)
3813 return cmdline_parse_core(p
, &required_kernelcore
);
3817 * movablecore=size sets the amount of memory for use for allocations that
3818 * can be reclaimed or migrated.
3820 static int __init
cmdline_parse_movablecore(char *p
)
3822 return cmdline_parse_core(p
, &required_movablecore
);
3825 early_param("kernelcore", cmdline_parse_kernelcore
);
3826 early_param("movablecore", cmdline_parse_movablecore
);
3828 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3831 * set_dma_reserve - set the specified number of pages reserved in the first zone
3832 * @new_dma_reserve: The number of pages to mark reserved
3834 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3835 * In the DMA zone, a significant percentage may be consumed by kernel image
3836 * and other unfreeable allocations which can skew the watermarks badly. This
3837 * function may optionally be used to account for unfreeable pages in the
3838 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3839 * smaller per-cpu batchsize.
3841 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
3843 dma_reserve
= new_dma_reserve
;
3846 #ifndef CONFIG_NEED_MULTIPLE_NODES
3847 static bootmem_data_t contig_bootmem_data
;
3848 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
3850 EXPORT_SYMBOL(contig_page_data
);
3853 void __init
free_area_init(unsigned long *zones_size
)
3855 free_area_init_node(0, NODE_DATA(0), zones_size
,
3856 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
3859 static int page_alloc_cpu_notify(struct notifier_block
*self
,
3860 unsigned long action
, void *hcpu
)
3862 int cpu
= (unsigned long)hcpu
;
3864 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3865 local_irq_disable();
3867 vm_events_fold_cpu(cpu
);
3869 refresh_cpu_vm_stats(cpu
);
3874 void __init
page_alloc_init(void)
3876 hotcpu_notifier(page_alloc_cpu_notify
, 0);
3880 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3881 * or min_free_kbytes changes.
3883 static void calculate_totalreserve_pages(void)
3885 struct pglist_data
*pgdat
;
3886 unsigned long reserve_pages
= 0;
3887 enum zone_type i
, j
;
3889 for_each_online_pgdat(pgdat
) {
3890 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3891 struct zone
*zone
= pgdat
->node_zones
+ i
;
3892 unsigned long max
= 0;
3894 /* Find valid and maximum lowmem_reserve in the zone */
3895 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
3896 if (zone
->lowmem_reserve
[j
] > max
)
3897 max
= zone
->lowmem_reserve
[j
];
3900 /* we treat pages_high as reserved pages. */
3901 max
+= zone
->pages_high
;
3903 if (max
> zone
->present_pages
)
3904 max
= zone
->present_pages
;
3905 reserve_pages
+= max
;
3908 totalreserve_pages
= reserve_pages
;
3912 * setup_per_zone_lowmem_reserve - called whenever
3913 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3914 * has a correct pages reserved value, so an adequate number of
3915 * pages are left in the zone after a successful __alloc_pages().
3917 static void setup_per_zone_lowmem_reserve(void)
3919 struct pglist_data
*pgdat
;
3920 enum zone_type j
, idx
;
3922 for_each_online_pgdat(pgdat
) {
3923 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3924 struct zone
*zone
= pgdat
->node_zones
+ j
;
3925 unsigned long present_pages
= zone
->present_pages
;
3927 zone
->lowmem_reserve
[j
] = 0;
3931 struct zone
*lower_zone
;
3935 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
3936 sysctl_lowmem_reserve_ratio
[idx
] = 1;
3938 lower_zone
= pgdat
->node_zones
+ idx
;
3939 lower_zone
->lowmem_reserve
[j
] = present_pages
/
3940 sysctl_lowmem_reserve_ratio
[idx
];
3941 present_pages
+= lower_zone
->present_pages
;
3946 /* update totalreserve_pages */
3947 calculate_totalreserve_pages();
3951 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3953 * Ensures that the pages_{min,low,high} values for each zone are set correctly
3954 * with respect to min_free_kbytes.
3956 void setup_per_zone_pages_min(void)
3958 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
3959 unsigned long lowmem_pages
= 0;
3961 unsigned long flags
;
3963 /* Calculate total number of !ZONE_HIGHMEM pages */
3964 for_each_zone(zone
) {
3965 if (!is_highmem(zone
))
3966 lowmem_pages
+= zone
->present_pages
;
3969 for_each_zone(zone
) {
3972 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3973 tmp
= (u64
)pages_min
* zone
->present_pages
;
3974 do_div(tmp
, lowmem_pages
);
3975 if (is_highmem(zone
)) {
3977 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3978 * need highmem pages, so cap pages_min to a small
3981 * The (pages_high-pages_low) and (pages_low-pages_min)
3982 * deltas controls asynch page reclaim, and so should
3983 * not be capped for highmem.
3987 min_pages
= zone
->present_pages
/ 1024;
3988 if (min_pages
< SWAP_CLUSTER_MAX
)
3989 min_pages
= SWAP_CLUSTER_MAX
;
3990 if (min_pages
> 128)
3992 zone
->pages_min
= min_pages
;
3995 * If it's a lowmem zone, reserve a number of pages
3996 * proportionate to the zone's size.
3998 zone
->pages_min
= tmp
;
4001 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4002 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4003 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4006 /* update totalreserve_pages */
4007 calculate_totalreserve_pages();
4011 * Initialise min_free_kbytes.
4013 * For small machines we want it small (128k min). For large machines
4014 * we want it large (64MB max). But it is not linear, because network
4015 * bandwidth does not increase linearly with machine size. We use
4017 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4018 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4034 static int __init
init_per_zone_pages_min(void)
4036 unsigned long lowmem_kbytes
;
4038 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4040 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4041 if (min_free_kbytes
< 128)
4042 min_free_kbytes
= 128;
4043 if (min_free_kbytes
> 65536)
4044 min_free_kbytes
= 65536;
4045 setup_per_zone_pages_min();
4046 setup_per_zone_lowmem_reserve();
4049 module_init(init_per_zone_pages_min
)
4052 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4053 * that we can call two helper functions whenever min_free_kbytes
4056 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4057 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4059 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4061 setup_per_zone_pages_min();
4066 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4067 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4072 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4077 zone
->min_unmapped_pages
= (zone
->present_pages
*
4078 sysctl_min_unmapped_ratio
) / 100;
4082 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4083 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4088 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4093 zone
->min_slab_pages
= (zone
->present_pages
*
4094 sysctl_min_slab_ratio
) / 100;
4100 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4101 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4102 * whenever sysctl_lowmem_reserve_ratio changes.
4104 * The reserve ratio obviously has absolutely no relation with the
4105 * pages_min watermarks. The lowmem reserve ratio can only make sense
4106 * if in function of the boot time zone sizes.
4108 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4109 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4111 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4112 setup_per_zone_lowmem_reserve();
4117 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4118 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4119 * can have before it gets flushed back to buddy allocator.
4122 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4123 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4129 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4130 if (!write
|| (ret
== -EINVAL
))
4132 for_each_zone(zone
) {
4133 for_each_online_cpu(cpu
) {
4135 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4136 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4142 int hashdist
= HASHDIST_DEFAULT
;
4145 static int __init
set_hashdist(char *str
)
4149 hashdist
= simple_strtoul(str
, &str
, 0);
4152 __setup("hashdist=", set_hashdist
);
4156 * allocate a large system hash table from bootmem
4157 * - it is assumed that the hash table must contain an exact power-of-2
4158 * quantity of entries
4159 * - limit is the number of hash buckets, not the total allocation size
4161 void *__init
alloc_large_system_hash(const char *tablename
,
4162 unsigned long bucketsize
,
4163 unsigned long numentries
,
4166 unsigned int *_hash_shift
,
4167 unsigned int *_hash_mask
,
4168 unsigned long limit
)
4170 unsigned long long max
= limit
;
4171 unsigned long log2qty
, size
;
4174 /* allow the kernel cmdline to have a say */
4176 /* round applicable memory size up to nearest megabyte */
4177 numentries
= nr_kernel_pages
;
4178 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4179 numentries
>>= 20 - PAGE_SHIFT
;
4180 numentries
<<= 20 - PAGE_SHIFT
;
4182 /* limit to 1 bucket per 2^scale bytes of low memory */
4183 if (scale
> PAGE_SHIFT
)
4184 numentries
>>= (scale
- PAGE_SHIFT
);
4186 numentries
<<= (PAGE_SHIFT
- scale
);
4188 /* Make sure we've got at least a 0-order allocation.. */
4189 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4190 numentries
= PAGE_SIZE
/ bucketsize
;
4192 numentries
= roundup_pow_of_two(numentries
);
4194 /* limit allocation size to 1/16 total memory by default */
4196 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4197 do_div(max
, bucketsize
);
4200 if (numentries
> max
)
4203 log2qty
= ilog2(numentries
);
4206 size
= bucketsize
<< log2qty
;
4207 if (flags
& HASH_EARLY
)
4208 table
= alloc_bootmem(size
);
4210 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4212 unsigned long order
;
4213 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
4215 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4217 * If bucketsize is not a power-of-two, we may free
4218 * some pages at the end of hash table.
4221 unsigned long alloc_end
= (unsigned long)table
+
4222 (PAGE_SIZE
<< order
);
4223 unsigned long used
= (unsigned long)table
+
4225 split_page(virt_to_page(table
), order
);
4226 while (used
< alloc_end
) {
4232 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4235 panic("Failed to allocate %s hash table\n", tablename
);
4237 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4240 ilog2(size
) - PAGE_SHIFT
,
4244 *_hash_shift
= log2qty
;
4246 *_hash_mask
= (1 << log2qty
) - 1;
4251 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4252 struct page
*pfn_to_page(unsigned long pfn
)
4254 return __pfn_to_page(pfn
);
4256 unsigned long page_to_pfn(struct page
*page
)
4258 return __page_to_pfn(page
);
4260 EXPORT_SYMBOL(pfn_to_page
);
4261 EXPORT_SYMBOL(page_to_pfn
);
4262 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4264 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4265 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4268 #ifdef CONFIG_SPARSEMEM
4269 return __pfn_to_section(pfn
)->pageblock_flags
;
4271 return zone
->pageblock_flags
;
4272 #endif /* CONFIG_SPARSEMEM */
4275 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4277 #ifdef CONFIG_SPARSEMEM
4278 pfn
&= (PAGES_PER_SECTION
-1);
4279 return (pfn
>> (MAX_ORDER
-1)) * NR_PAGEBLOCK_BITS
;
4281 pfn
= pfn
- zone
->zone_start_pfn
;
4282 return (pfn
>> (MAX_ORDER
-1)) * NR_PAGEBLOCK_BITS
;
4283 #endif /* CONFIG_SPARSEMEM */
4287 * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages
4288 * @page: The page within the block of interest
4289 * @start_bitidx: The first bit of interest to retrieve
4290 * @end_bitidx: The last bit of interest
4291 * returns pageblock_bits flags
4293 unsigned long get_pageblock_flags_group(struct page
*page
,
4294 int start_bitidx
, int end_bitidx
)
4297 unsigned long *bitmap
;
4298 unsigned long pfn
, bitidx
;
4299 unsigned long flags
= 0;
4300 unsigned long value
= 1;
4302 zone
= page_zone(page
);
4303 pfn
= page_to_pfn(page
);
4304 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4305 bitidx
= pfn_to_bitidx(zone
, pfn
);
4307 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4308 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4315 * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages
4316 * @page: The page within the block of interest
4317 * @start_bitidx: The first bit of interest
4318 * @end_bitidx: The last bit of interest
4319 * @flags: The flags to set
4321 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4322 int start_bitidx
, int end_bitidx
)
4325 unsigned long *bitmap
;
4326 unsigned long pfn
, bitidx
;
4327 unsigned long value
= 1;
4329 zone
= page_zone(page
);
4330 pfn
= page_to_pfn(page
);
4331 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4332 bitidx
= pfn_to_bitidx(zone
, pfn
);
4334 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4336 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4338 __clear_bit(bitidx
+ start_bitidx
, bitmap
);