2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/oom.h>
32 #include <linux/notifier.h>
33 #include <linux/topology.h>
34 #include <linux/sysctl.h>
35 #include <linux/cpu.h>
36 #include <linux/cpuset.h>
37 #include <linux/memory_hotplug.h>
38 #include <linux/nodemask.h>
39 #include <linux/vmalloc.h>
40 #include <linux/mempolicy.h>
41 #include <linux/stop_machine.h>
42 #include <linux/sort.h>
43 #include <linux/pfn.h>
44 #include <linux/backing-dev.h>
45 #include <linux/fault-inject.h>
46 #include <linux/page-isolation.h>
47 #include <linux/page_cgroup.h>
48 #include <linux/debugobjects.h>
50 #include <asm/tlbflush.h>
51 #include <asm/div64.h>
55 * Array of node states.
57 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
58 [N_POSSIBLE
] = NODE_MASK_ALL
,
59 [N_ONLINE
] = { { [0] = 1UL } },
61 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
63 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
65 [N_CPU
] = { { [0] = 1UL } },
68 EXPORT_SYMBOL(node_states
);
70 unsigned long totalram_pages __read_mostly
;
71 unsigned long totalreserve_pages __read_mostly
;
72 unsigned long highest_memmap_pfn __read_mostly
;
73 int percpu_pagelist_fraction
;
75 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
76 int pageblock_order __read_mostly
;
79 static void __free_pages_ok(struct page
*page
, unsigned int order
);
82 * results with 256, 32 in the lowmem_reserve sysctl:
83 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
84 * 1G machine -> (16M dma, 784M normal, 224M high)
85 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
86 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
87 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
89 * TBD: should special case ZONE_DMA32 machines here - in those we normally
90 * don't need any ZONE_NORMAL reservation
92 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
93 #ifdef CONFIG_ZONE_DMA
96 #ifdef CONFIG_ZONE_DMA32
105 EXPORT_SYMBOL(totalram_pages
);
107 static char * const zone_names
[MAX_NR_ZONES
] = {
108 #ifdef CONFIG_ZONE_DMA
111 #ifdef CONFIG_ZONE_DMA32
115 #ifdef CONFIG_HIGHMEM
121 int min_free_kbytes
= 1024;
123 unsigned long __meminitdata nr_kernel_pages
;
124 unsigned long __meminitdata nr_all_pages
;
125 static unsigned long __meminitdata dma_reserve
;
127 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
129 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
130 * ranges of memory (RAM) that may be registered with add_active_range().
131 * Ranges passed to add_active_range() will be merged if possible
132 * so the number of times add_active_range() can be called is
133 * related to the number of nodes and the number of holes
135 #ifdef CONFIG_MAX_ACTIVE_REGIONS
136 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
137 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
139 #if MAX_NUMNODES >= 32
140 /* If there can be many nodes, allow up to 50 holes per node */
141 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
143 /* By default, allow up to 256 distinct regions */
144 #define MAX_ACTIVE_REGIONS 256
148 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
149 static int __meminitdata nr_nodemap_entries
;
150 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
151 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
152 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
153 static unsigned long __meminitdata node_boundary_start_pfn
[MAX_NUMNODES
];
154 static unsigned long __meminitdata node_boundary_end_pfn
[MAX_NUMNODES
];
155 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
156 static unsigned long __initdata required_kernelcore
;
157 static unsigned long __initdata required_movablecore
;
158 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
160 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
162 EXPORT_SYMBOL(movable_zone
);
163 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
166 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
167 EXPORT_SYMBOL(nr_node_ids
);
170 int page_group_by_mobility_disabled __read_mostly
;
172 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
174 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
175 PB_migrate
, PB_migrate_end
);
178 #ifdef CONFIG_DEBUG_VM
179 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
183 unsigned long pfn
= page_to_pfn(page
);
186 seq
= zone_span_seqbegin(zone
);
187 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
189 else if (pfn
< zone
->zone_start_pfn
)
191 } while (zone_span_seqretry(zone
, seq
));
196 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
198 if (!pfn_valid_within(page_to_pfn(page
)))
200 if (zone
!= page_zone(page
))
206 * Temporary debugging check for pages not lying within a given zone.
208 static int bad_range(struct zone
*zone
, struct page
*page
)
210 if (page_outside_zone_boundaries(zone
, page
))
212 if (!page_is_consistent(zone
, page
))
218 static inline int bad_range(struct zone
*zone
, struct page
*page
)
224 static void bad_page(struct page
*page
)
226 static unsigned long resume
;
227 static unsigned long nr_shown
;
228 static unsigned long nr_unshown
;
231 * Allow a burst of 60 reports, then keep quiet for that minute;
232 * or allow a steady drip of one report per second.
234 if (nr_shown
== 60) {
235 if (time_before(jiffies
, resume
)) {
241 "Bad page state: %lu messages suppressed\n",
248 resume
= jiffies
+ 60 * HZ
;
250 printk(KERN_EMERG
"Bad page state in process %s pfn:%05lx\n",
251 current
->comm
, page_to_pfn(page
));
253 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
254 page
, (void *)page
->flags
, page_count(page
),
255 page_mapcount(page
), page
->mapping
, page
->index
);
256 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n");
260 /* Leave bad fields for debug, except PageBuddy could make trouble */
261 __ClearPageBuddy(page
);
262 add_taint(TAINT_BAD_PAGE
);
266 * Higher-order pages are called "compound pages". They are structured thusly:
268 * The first PAGE_SIZE page is called the "head page".
270 * The remaining PAGE_SIZE pages are called "tail pages".
272 * All pages have PG_compound set. All pages have their ->private pointing at
273 * the head page (even the head page has this).
275 * The first tail page's ->lru.next holds the address of the compound page's
276 * put_page() function. Its ->lru.prev holds the order of allocation.
277 * This usage means that zero-order pages may not be compound.
280 static void free_compound_page(struct page
*page
)
282 __free_pages_ok(page
, compound_order(page
));
285 void prep_compound_page(struct page
*page
, unsigned long order
)
288 int nr_pages
= 1 << order
;
290 set_compound_page_dtor(page
, free_compound_page
);
291 set_compound_order(page
, order
);
293 for (i
= 1; i
< nr_pages
; i
++) {
294 struct page
*p
= page
+ i
;
297 p
->first_page
= page
;
301 #ifdef CONFIG_HUGETLBFS
302 void prep_compound_gigantic_page(struct page
*page
, unsigned long order
)
305 int nr_pages
= 1 << order
;
306 struct page
*p
= page
+ 1;
308 set_compound_page_dtor(page
, free_compound_page
);
309 set_compound_order(page
, order
);
311 for (i
= 1; i
< nr_pages
; i
++, p
= mem_map_next(p
, page
, i
)) {
313 p
->first_page
= page
;
318 static int destroy_compound_page(struct page
*page
, unsigned long order
)
321 int nr_pages
= 1 << order
;
324 if (unlikely(compound_order(page
) != order
) ||
325 unlikely(!PageHead(page
))) {
330 __ClearPageHead(page
);
332 for (i
= 1; i
< nr_pages
; i
++) {
333 struct page
*p
= page
+ i
;
335 if (unlikely(!PageTail(p
) | (p
->first_page
!= page
))) {
345 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
350 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
351 * and __GFP_HIGHMEM from hard or soft interrupt context.
353 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
354 for (i
= 0; i
< (1 << order
); i
++)
355 clear_highpage(page
+ i
);
358 static inline void set_page_order(struct page
*page
, int order
)
360 set_page_private(page
, order
);
361 __SetPageBuddy(page
);
364 static inline void rmv_page_order(struct page
*page
)
366 __ClearPageBuddy(page
);
367 set_page_private(page
, 0);
371 * Locate the struct page for both the matching buddy in our
372 * pair (buddy1) and the combined O(n+1) page they form (page).
374 * 1) Any buddy B1 will have an order O twin B2 which satisfies
375 * the following equation:
377 * For example, if the starting buddy (buddy2) is #8 its order
379 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
381 * 2) Any buddy B will have an order O+1 parent P which
382 * satisfies the following equation:
385 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
387 static inline struct page
*
388 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
390 unsigned long buddy_idx
= page_idx
^ (1 << order
);
392 return page
+ (buddy_idx
- page_idx
);
395 static inline unsigned long
396 __find_combined_index(unsigned long page_idx
, unsigned int order
)
398 return (page_idx
& ~(1 << order
));
402 * This function checks whether a page is free && is the buddy
403 * we can do coalesce a page and its buddy if
404 * (a) the buddy is not in a hole &&
405 * (b) the buddy is in the buddy system &&
406 * (c) a page and its buddy have the same order &&
407 * (d) a page and its buddy are in the same zone.
409 * For recording whether a page is in the buddy system, we use PG_buddy.
410 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
412 * For recording page's order, we use page_private(page).
414 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
417 if (!pfn_valid_within(page_to_pfn(buddy
)))
420 if (page_zone_id(page
) != page_zone_id(buddy
))
423 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
424 BUG_ON(page_count(buddy
) != 0);
431 * Freeing function for a buddy system allocator.
433 * The concept of a buddy system is to maintain direct-mapped table
434 * (containing bit values) for memory blocks of various "orders".
435 * The bottom level table contains the map for the smallest allocatable
436 * units of memory (here, pages), and each level above it describes
437 * pairs of units from the levels below, hence, "buddies".
438 * At a high level, all that happens here is marking the table entry
439 * at the bottom level available, and propagating the changes upward
440 * as necessary, plus some accounting needed to play nicely with other
441 * parts of the VM system.
442 * At each level, we keep a list of pages, which are heads of continuous
443 * free pages of length of (1 << order) and marked with PG_buddy. Page's
444 * order is recorded in page_private(page) field.
445 * So when we are allocating or freeing one, we can derive the state of the
446 * other. That is, if we allocate a small block, and both were
447 * free, the remainder of the region must be split into blocks.
448 * If a block is freed, and its buddy is also free, then this
449 * triggers coalescing into a block of larger size.
454 static inline void __free_one_page(struct page
*page
,
455 struct zone
*zone
, unsigned int order
)
457 unsigned long page_idx
;
458 int order_size
= 1 << order
;
459 int migratetype
= get_pageblock_migratetype(page
);
461 if (unlikely(PageCompound(page
)))
462 if (unlikely(destroy_compound_page(page
, order
)))
465 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
467 VM_BUG_ON(page_idx
& (order_size
- 1));
468 VM_BUG_ON(bad_range(zone
, page
));
470 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
471 while (order
< MAX_ORDER
-1) {
472 unsigned long combined_idx
;
475 buddy
= __page_find_buddy(page
, page_idx
, order
);
476 if (!page_is_buddy(page
, buddy
, order
))
479 /* Our buddy is free, merge with it and move up one order. */
480 list_del(&buddy
->lru
);
481 zone
->free_area
[order
].nr_free
--;
482 rmv_page_order(buddy
);
483 combined_idx
= __find_combined_index(page_idx
, order
);
484 page
= page
+ (combined_idx
- page_idx
);
485 page_idx
= combined_idx
;
488 set_page_order(page
, order
);
490 &zone
->free_area
[order
].free_list
[migratetype
]);
491 zone
->free_area
[order
].nr_free
++;
494 static inline int free_pages_check(struct page
*page
)
496 free_page_mlock(page
);
497 if (unlikely(page_mapcount(page
) |
498 (page
->mapping
!= NULL
) |
499 (page_count(page
) != 0) |
500 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
504 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
505 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
510 * Frees a list of pages.
511 * Assumes all pages on list are in same zone, and of same order.
512 * count is the number of pages to free.
514 * If the zone was previously in an "all pages pinned" state then look to
515 * see if this freeing clears that state.
517 * And clear the zone's pages_scanned counter, to hold off the "all pages are
518 * pinned" detection logic.
520 static void free_pages_bulk(struct zone
*zone
, int count
,
521 struct list_head
*list
, int order
)
523 spin_lock(&zone
->lock
);
524 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
525 zone
->pages_scanned
= 0;
529 VM_BUG_ON(list_empty(list
));
530 page
= list_entry(list
->prev
, struct page
, lru
);
531 /* have to delete it as __free_one_page list manipulates */
532 list_del(&page
->lru
);
533 __free_one_page(page
, zone
, order
);
535 spin_unlock(&zone
->lock
);
538 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
540 spin_lock(&zone
->lock
);
541 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
542 zone
->pages_scanned
= 0;
543 __free_one_page(page
, zone
, order
);
544 spin_unlock(&zone
->lock
);
547 static void __free_pages_ok(struct page
*page
, unsigned int order
)
553 for (i
= 0 ; i
< (1 << order
) ; ++i
)
554 bad
+= free_pages_check(page
+ i
);
558 if (!PageHighMem(page
)) {
559 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
560 debug_check_no_obj_freed(page_address(page
),
563 arch_free_page(page
, order
);
564 kernel_map_pages(page
, 1 << order
, 0);
566 local_irq_save(flags
);
567 __count_vm_events(PGFREE
, 1 << order
);
568 free_one_page(page_zone(page
), page
, order
);
569 local_irq_restore(flags
);
573 * permit the bootmem allocator to evade page validation on high-order frees
575 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
578 __ClearPageReserved(page
);
579 set_page_count(page
, 0);
580 set_page_refcounted(page
);
586 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
587 struct page
*p
= &page
[loop
];
589 if (loop
+ 1 < BITS_PER_LONG
)
591 __ClearPageReserved(p
);
592 set_page_count(p
, 0);
595 set_page_refcounted(page
);
596 __free_pages(page
, order
);
602 * The order of subdivision here is critical for the IO subsystem.
603 * Please do not alter this order without good reasons and regression
604 * testing. Specifically, as large blocks of memory are subdivided,
605 * the order in which smaller blocks are delivered depends on the order
606 * they're subdivided in this function. This is the primary factor
607 * influencing the order in which pages are delivered to the IO
608 * subsystem according to empirical testing, and this is also justified
609 * by considering the behavior of a buddy system containing a single
610 * large block of memory acted on by a series of small allocations.
611 * This behavior is a critical factor in sglist merging's success.
615 static inline void expand(struct zone
*zone
, struct page
*page
,
616 int low
, int high
, struct free_area
*area
,
619 unsigned long size
= 1 << high
;
625 VM_BUG_ON(bad_range(zone
, &page
[size
]));
626 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
628 set_page_order(&page
[size
], high
);
633 * This page is about to be returned from the page allocator
635 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
637 if (unlikely(page_mapcount(page
) |
638 (page
->mapping
!= NULL
) |
639 (page_count(page
) != 0) |
640 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
645 set_page_private(page
, 0);
646 set_page_refcounted(page
);
648 arch_alloc_page(page
, order
);
649 kernel_map_pages(page
, 1 << order
, 1);
651 if (gfp_flags
& __GFP_ZERO
)
652 prep_zero_page(page
, order
, gfp_flags
);
654 if (order
&& (gfp_flags
& __GFP_COMP
))
655 prep_compound_page(page
, order
);
661 * Go through the free lists for the given migratetype and remove
662 * the smallest available page from the freelists
664 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
667 unsigned int current_order
;
668 struct free_area
* area
;
671 /* Find a page of the appropriate size in the preferred list */
672 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
673 area
= &(zone
->free_area
[current_order
]);
674 if (list_empty(&area
->free_list
[migratetype
]))
677 page
= list_entry(area
->free_list
[migratetype
].next
,
679 list_del(&page
->lru
);
680 rmv_page_order(page
);
682 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
683 expand(zone
, page
, order
, current_order
, area
, migratetype
);
692 * This array describes the order lists are fallen back to when
693 * the free lists for the desirable migrate type are depleted
695 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
696 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
697 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
698 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
699 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
703 * Move the free pages in a range to the free lists of the requested type.
704 * Note that start_page and end_pages are not aligned on a pageblock
705 * boundary. If alignment is required, use move_freepages_block()
707 static int move_freepages(struct zone
*zone
,
708 struct page
*start_page
, struct page
*end_page
,
715 #ifndef CONFIG_HOLES_IN_ZONE
717 * page_zone is not safe to call in this context when
718 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
719 * anyway as we check zone boundaries in move_freepages_block().
720 * Remove at a later date when no bug reports exist related to
721 * grouping pages by mobility
723 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
726 for (page
= start_page
; page
<= end_page
;) {
727 /* Make sure we are not inadvertently changing nodes */
728 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
730 if (!pfn_valid_within(page_to_pfn(page
))) {
735 if (!PageBuddy(page
)) {
740 order
= page_order(page
);
741 list_del(&page
->lru
);
743 &zone
->free_area
[order
].free_list
[migratetype
]);
745 pages_moved
+= 1 << order
;
751 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
754 unsigned long start_pfn
, end_pfn
;
755 struct page
*start_page
, *end_page
;
757 start_pfn
= page_to_pfn(page
);
758 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
759 start_page
= pfn_to_page(start_pfn
);
760 end_page
= start_page
+ pageblock_nr_pages
- 1;
761 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
763 /* Do not cross zone boundaries */
764 if (start_pfn
< zone
->zone_start_pfn
)
766 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
769 return move_freepages(zone
, start_page
, end_page
, migratetype
);
772 /* Remove an element from the buddy allocator from the fallback list */
773 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
774 int start_migratetype
)
776 struct free_area
* area
;
781 /* Find the largest possible block of pages in the other list */
782 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
784 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
785 migratetype
= fallbacks
[start_migratetype
][i
];
787 /* MIGRATE_RESERVE handled later if necessary */
788 if (migratetype
== MIGRATE_RESERVE
)
791 area
= &(zone
->free_area
[current_order
]);
792 if (list_empty(&area
->free_list
[migratetype
]))
795 page
= list_entry(area
->free_list
[migratetype
].next
,
800 * If breaking a large block of pages, move all free
801 * pages to the preferred allocation list. If falling
802 * back for a reclaimable kernel allocation, be more
803 * agressive about taking ownership of free pages
805 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
806 start_migratetype
== MIGRATE_RECLAIMABLE
) {
808 pages
= move_freepages_block(zone
, page
,
811 /* Claim the whole block if over half of it is free */
812 if (pages
>= (1 << (pageblock_order
-1)))
813 set_pageblock_migratetype(page
,
816 migratetype
= start_migratetype
;
819 /* Remove the page from the freelists */
820 list_del(&page
->lru
);
821 rmv_page_order(page
);
822 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
825 if (current_order
== pageblock_order
)
826 set_pageblock_migratetype(page
,
829 expand(zone
, page
, order
, current_order
, area
, migratetype
);
834 /* Use MIGRATE_RESERVE rather than fail an allocation */
835 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
839 * Do the hard work of removing an element from the buddy allocator.
840 * Call me with the zone->lock already held.
842 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
847 page
= __rmqueue_smallest(zone
, order
, migratetype
);
850 page
= __rmqueue_fallback(zone
, order
, migratetype
);
856 * Obtain a specified number of elements from the buddy allocator, all under
857 * a single hold of the lock, for efficiency. Add them to the supplied list.
858 * Returns the number of new pages which were placed at *list.
860 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
861 unsigned long count
, struct list_head
*list
,
866 spin_lock(&zone
->lock
);
867 for (i
= 0; i
< count
; ++i
) {
868 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
869 if (unlikely(page
== NULL
))
873 * Split buddy pages returned by expand() are received here
874 * in physical page order. The page is added to the callers and
875 * list and the list head then moves forward. From the callers
876 * perspective, the linked list is ordered by page number in
877 * some conditions. This is useful for IO devices that can
878 * merge IO requests if the physical pages are ordered
881 list_add(&page
->lru
, list
);
882 set_page_private(page
, migratetype
);
885 spin_unlock(&zone
->lock
);
891 * Called from the vmstat counter updater to drain pagesets of this
892 * currently executing processor on remote nodes after they have
895 * Note that this function must be called with the thread pinned to
896 * a single processor.
898 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
903 local_irq_save(flags
);
904 if (pcp
->count
>= pcp
->batch
)
905 to_drain
= pcp
->batch
;
907 to_drain
= pcp
->count
;
908 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
909 pcp
->count
-= to_drain
;
910 local_irq_restore(flags
);
915 * Drain pages of the indicated processor.
917 * The processor must either be the current processor and the
918 * thread pinned to the current processor or a processor that
921 static void drain_pages(unsigned int cpu
)
926 for_each_zone(zone
) {
927 struct per_cpu_pageset
*pset
;
928 struct per_cpu_pages
*pcp
;
930 if (!populated_zone(zone
))
933 pset
= zone_pcp(zone
, cpu
);
936 local_irq_save(flags
);
937 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
939 local_irq_restore(flags
);
944 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
946 void drain_local_pages(void *arg
)
948 drain_pages(smp_processor_id());
952 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
954 void drain_all_pages(void)
956 on_each_cpu(drain_local_pages
, NULL
, 1);
959 #ifdef CONFIG_HIBERNATION
961 void mark_free_pages(struct zone
*zone
)
963 unsigned long pfn
, max_zone_pfn
;
966 struct list_head
*curr
;
968 if (!zone
->spanned_pages
)
971 spin_lock_irqsave(&zone
->lock
, flags
);
973 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
974 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
975 if (pfn_valid(pfn
)) {
976 struct page
*page
= pfn_to_page(pfn
);
978 if (!swsusp_page_is_forbidden(page
))
979 swsusp_unset_page_free(page
);
982 for_each_migratetype_order(order
, t
) {
983 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
986 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
987 for (i
= 0; i
< (1UL << order
); i
++)
988 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
991 spin_unlock_irqrestore(&zone
->lock
, flags
);
993 #endif /* CONFIG_PM */
996 * Free a 0-order page
998 static void free_hot_cold_page(struct page
*page
, int cold
)
1000 struct zone
*zone
= page_zone(page
);
1001 struct per_cpu_pages
*pcp
;
1002 unsigned long flags
;
1005 page
->mapping
= NULL
;
1006 if (free_pages_check(page
))
1009 if (!PageHighMem(page
)) {
1010 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1011 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1013 arch_free_page(page
, 0);
1014 kernel_map_pages(page
, 1, 0);
1016 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1017 local_irq_save(flags
);
1018 __count_vm_event(PGFREE
);
1020 list_add_tail(&page
->lru
, &pcp
->list
);
1022 list_add(&page
->lru
, &pcp
->list
);
1023 set_page_private(page
, get_pageblock_migratetype(page
));
1025 if (pcp
->count
>= pcp
->high
) {
1026 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1027 pcp
->count
-= pcp
->batch
;
1029 local_irq_restore(flags
);
1033 void free_hot_page(struct page
*page
)
1035 free_hot_cold_page(page
, 0);
1038 void free_cold_page(struct page
*page
)
1040 free_hot_cold_page(page
, 1);
1044 * split_page takes a non-compound higher-order page, and splits it into
1045 * n (1<<order) sub-pages: page[0..n]
1046 * Each sub-page must be freed individually.
1048 * Note: this is probably too low level an operation for use in drivers.
1049 * Please consult with lkml before using this in your driver.
1051 void split_page(struct page
*page
, unsigned int order
)
1055 VM_BUG_ON(PageCompound(page
));
1056 VM_BUG_ON(!page_count(page
));
1057 for (i
= 1; i
< (1 << order
); i
++)
1058 set_page_refcounted(page
+ i
);
1062 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1063 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1066 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1067 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1069 unsigned long flags
;
1071 int cold
= !!(gfp_flags
& __GFP_COLD
);
1073 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1077 if (likely(order
== 0)) {
1078 struct per_cpu_pages
*pcp
;
1080 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1081 local_irq_save(flags
);
1083 pcp
->count
= rmqueue_bulk(zone
, 0,
1084 pcp
->batch
, &pcp
->list
, migratetype
);
1085 if (unlikely(!pcp
->count
))
1089 /* Find a page of the appropriate migrate type */
1091 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1092 if (page_private(page
) == migratetype
)
1095 list_for_each_entry(page
, &pcp
->list
, lru
)
1096 if (page_private(page
) == migratetype
)
1100 /* Allocate more to the pcp list if necessary */
1101 if (unlikely(&page
->lru
== &pcp
->list
)) {
1102 pcp
->count
+= rmqueue_bulk(zone
, 0,
1103 pcp
->batch
, &pcp
->list
, migratetype
);
1104 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1107 list_del(&page
->lru
);
1110 spin_lock_irqsave(&zone
->lock
, flags
);
1111 page
= __rmqueue(zone
, order
, migratetype
);
1112 spin_unlock(&zone
->lock
);
1117 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1118 zone_statistics(preferred_zone
, zone
);
1119 local_irq_restore(flags
);
1122 VM_BUG_ON(bad_range(zone
, page
));
1123 if (prep_new_page(page
, order
, gfp_flags
))
1128 local_irq_restore(flags
);
1133 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1134 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1135 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1136 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1137 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1138 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1139 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1141 #ifdef CONFIG_FAIL_PAGE_ALLOC
1143 static struct fail_page_alloc_attr
{
1144 struct fault_attr attr
;
1146 u32 ignore_gfp_highmem
;
1147 u32 ignore_gfp_wait
;
1150 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1152 struct dentry
*ignore_gfp_highmem_file
;
1153 struct dentry
*ignore_gfp_wait_file
;
1154 struct dentry
*min_order_file
;
1156 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1158 } fail_page_alloc
= {
1159 .attr
= FAULT_ATTR_INITIALIZER
,
1160 .ignore_gfp_wait
= 1,
1161 .ignore_gfp_highmem
= 1,
1165 static int __init
setup_fail_page_alloc(char *str
)
1167 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1169 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1171 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1173 if (order
< fail_page_alloc
.min_order
)
1175 if (gfp_mask
& __GFP_NOFAIL
)
1177 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1179 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1182 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1185 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1187 static int __init
fail_page_alloc_debugfs(void)
1189 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1193 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1197 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1199 fail_page_alloc
.ignore_gfp_wait_file
=
1200 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1201 &fail_page_alloc
.ignore_gfp_wait
);
1203 fail_page_alloc
.ignore_gfp_highmem_file
=
1204 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1205 &fail_page_alloc
.ignore_gfp_highmem
);
1206 fail_page_alloc
.min_order_file
=
1207 debugfs_create_u32("min-order", mode
, dir
,
1208 &fail_page_alloc
.min_order
);
1210 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1211 !fail_page_alloc
.ignore_gfp_highmem_file
||
1212 !fail_page_alloc
.min_order_file
) {
1214 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1215 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1216 debugfs_remove(fail_page_alloc
.min_order_file
);
1217 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1223 late_initcall(fail_page_alloc_debugfs
);
1225 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1227 #else /* CONFIG_FAIL_PAGE_ALLOC */
1229 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1234 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1237 * Return 1 if free pages are above 'mark'. This takes into account the order
1238 * of the allocation.
1240 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1241 int classzone_idx
, int alloc_flags
)
1243 /* free_pages my go negative - that's OK */
1245 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1248 if (alloc_flags
& ALLOC_HIGH
)
1250 if (alloc_flags
& ALLOC_HARDER
)
1253 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1255 for (o
= 0; o
< order
; o
++) {
1256 /* At the next order, this order's pages become unavailable */
1257 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1259 /* Require fewer higher order pages to be free */
1262 if (free_pages
<= min
)
1270 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1271 * skip over zones that are not allowed by the cpuset, or that have
1272 * been recently (in last second) found to be nearly full. See further
1273 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1274 * that have to skip over a lot of full or unallowed zones.
1276 * If the zonelist cache is present in the passed in zonelist, then
1277 * returns a pointer to the allowed node mask (either the current
1278 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1280 * If the zonelist cache is not available for this zonelist, does
1281 * nothing and returns NULL.
1283 * If the fullzones BITMAP in the zonelist cache is stale (more than
1284 * a second since last zap'd) then we zap it out (clear its bits.)
1286 * We hold off even calling zlc_setup, until after we've checked the
1287 * first zone in the zonelist, on the theory that most allocations will
1288 * be satisfied from that first zone, so best to examine that zone as
1289 * quickly as we can.
1291 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1293 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1294 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1296 zlc
= zonelist
->zlcache_ptr
;
1300 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1301 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1302 zlc
->last_full_zap
= jiffies
;
1305 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1306 &cpuset_current_mems_allowed
:
1307 &node_states
[N_HIGH_MEMORY
];
1308 return allowednodes
;
1312 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1313 * if it is worth looking at further for free memory:
1314 * 1) Check that the zone isn't thought to be full (doesn't have its
1315 * bit set in the zonelist_cache fullzones BITMAP).
1316 * 2) Check that the zones node (obtained from the zonelist_cache
1317 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1318 * Return true (non-zero) if zone is worth looking at further, or
1319 * else return false (zero) if it is not.
1321 * This check -ignores- the distinction between various watermarks,
1322 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1323 * found to be full for any variation of these watermarks, it will
1324 * be considered full for up to one second by all requests, unless
1325 * we are so low on memory on all allowed nodes that we are forced
1326 * into the second scan of the zonelist.
1328 * In the second scan we ignore this zonelist cache and exactly
1329 * apply the watermarks to all zones, even it is slower to do so.
1330 * We are low on memory in the second scan, and should leave no stone
1331 * unturned looking for a free page.
1333 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1334 nodemask_t
*allowednodes
)
1336 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1337 int i
; /* index of *z in zonelist zones */
1338 int n
; /* node that zone *z is on */
1340 zlc
= zonelist
->zlcache_ptr
;
1344 i
= z
- zonelist
->_zonerefs
;
1347 /* This zone is worth trying if it is allowed but not full */
1348 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1352 * Given 'z' scanning a zonelist, set the corresponding bit in
1353 * zlc->fullzones, so that subsequent attempts to allocate a page
1354 * from that zone don't waste time re-examining it.
1356 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1358 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1359 int i
; /* index of *z in zonelist zones */
1361 zlc
= zonelist
->zlcache_ptr
;
1365 i
= z
- zonelist
->_zonerefs
;
1367 set_bit(i
, zlc
->fullzones
);
1370 #else /* CONFIG_NUMA */
1372 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1377 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1378 nodemask_t
*allowednodes
)
1383 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1386 #endif /* CONFIG_NUMA */
1389 * get_page_from_freelist goes through the zonelist trying to allocate
1392 static struct page
*
1393 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1394 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1397 struct page
*page
= NULL
;
1399 struct zone
*zone
, *preferred_zone
;
1400 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1401 int zlc_active
= 0; /* set if using zonelist_cache */
1402 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1404 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1406 if (!preferred_zone
)
1409 classzone_idx
= zone_idx(preferred_zone
);
1413 * Scan zonelist, looking for a zone with enough free.
1414 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1416 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1417 high_zoneidx
, nodemask
) {
1418 if (NUMA_BUILD
&& zlc_active
&&
1419 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1421 if ((alloc_flags
& ALLOC_CPUSET
) &&
1422 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1425 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1427 if (alloc_flags
& ALLOC_WMARK_MIN
)
1428 mark
= zone
->pages_min
;
1429 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1430 mark
= zone
->pages_low
;
1432 mark
= zone
->pages_high
;
1433 if (!zone_watermark_ok(zone
, order
, mark
,
1434 classzone_idx
, alloc_flags
)) {
1435 if (!zone_reclaim_mode
||
1436 !zone_reclaim(zone
, gfp_mask
, order
))
1437 goto this_zone_full
;
1441 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1446 zlc_mark_zone_full(zonelist
, z
);
1448 if (NUMA_BUILD
&& !did_zlc_setup
) {
1449 /* we do zlc_setup after the first zone is tried */
1450 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1456 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1457 /* Disable zlc cache for second zonelist scan */
1465 * This is the 'heart' of the zoned buddy allocator.
1468 __alloc_pages_internal(gfp_t gfp_mask
, unsigned int order
,
1469 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1471 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1472 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1476 struct reclaim_state reclaim_state
;
1477 struct task_struct
*p
= current
;
1480 unsigned long did_some_progress
;
1481 unsigned long pages_reclaimed
= 0;
1483 might_sleep_if(wait
);
1485 if (should_fail_alloc_page(gfp_mask
, order
))
1489 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1491 if (unlikely(!z
->zone
)) {
1493 * Happens if we have an empty zonelist as a result of
1494 * GFP_THISNODE being used on a memoryless node
1499 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1500 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1505 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1506 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1507 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1508 * using a larger set of nodes after it has established that the
1509 * allowed per node queues are empty and that nodes are
1512 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1515 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1516 wakeup_kswapd(zone
, order
);
1519 * OK, we're below the kswapd watermark and have kicked background
1520 * reclaim. Now things get more complex, so set up alloc_flags according
1521 * to how we want to proceed.
1523 * The caller may dip into page reserves a bit more if the caller
1524 * cannot run direct reclaim, or if the caller has realtime scheduling
1525 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1526 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1528 alloc_flags
= ALLOC_WMARK_MIN
;
1529 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1530 alloc_flags
|= ALLOC_HARDER
;
1531 if (gfp_mask
& __GFP_HIGH
)
1532 alloc_flags
|= ALLOC_HIGH
;
1534 alloc_flags
|= ALLOC_CPUSET
;
1537 * Go through the zonelist again. Let __GFP_HIGH and allocations
1538 * coming from realtime tasks go deeper into reserves.
1540 * This is the last chance, in general, before the goto nopage.
1541 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1542 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1544 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1545 high_zoneidx
, alloc_flags
);
1549 /* This allocation should allow future memory freeing. */
1552 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1553 && !in_interrupt()) {
1554 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1556 /* go through the zonelist yet again, ignoring mins */
1557 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1558 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1561 if (gfp_mask
& __GFP_NOFAIL
) {
1562 congestion_wait(WRITE
, HZ
/50);
1569 /* Atomic allocations - we can't balance anything */
1575 /* We now go into synchronous reclaim */
1576 cpuset_memory_pressure_bump();
1578 * The task's cpuset might have expanded its set of allowable nodes
1580 cpuset_update_task_memory_state();
1581 p
->flags
|= PF_MEMALLOC
;
1582 reclaim_state
.reclaimed_slab
= 0;
1583 p
->reclaim_state
= &reclaim_state
;
1585 did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
);
1587 p
->reclaim_state
= NULL
;
1588 p
->flags
&= ~PF_MEMALLOC
;
1595 if (likely(did_some_progress
)) {
1596 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1597 zonelist
, high_zoneidx
, alloc_flags
);
1600 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1601 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1602 schedule_timeout_uninterruptible(1);
1607 * Go through the zonelist yet one more time, keep
1608 * very high watermark here, this is only to catch
1609 * a parallel oom killing, we must fail if we're still
1610 * under heavy pressure.
1612 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1613 order
, zonelist
, high_zoneidx
,
1614 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1616 clear_zonelist_oom(zonelist
, gfp_mask
);
1620 /* The OOM killer will not help higher order allocs so fail */
1621 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1622 clear_zonelist_oom(zonelist
, gfp_mask
);
1626 out_of_memory(zonelist
, gfp_mask
, order
);
1627 clear_zonelist_oom(zonelist
, gfp_mask
);
1632 * Don't let big-order allocations loop unless the caller explicitly
1633 * requests that. Wait for some write requests to complete then retry.
1635 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1636 * means __GFP_NOFAIL, but that may not be true in other
1639 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1640 * specified, then we retry until we no longer reclaim any pages
1641 * (above), or we've reclaimed an order of pages at least as
1642 * large as the allocation's order. In both cases, if the
1643 * allocation still fails, we stop retrying.
1645 pages_reclaimed
+= did_some_progress
;
1647 if (!(gfp_mask
& __GFP_NORETRY
)) {
1648 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1651 if (gfp_mask
& __GFP_REPEAT
&&
1652 pages_reclaimed
< (1 << order
))
1655 if (gfp_mask
& __GFP_NOFAIL
)
1659 congestion_wait(WRITE
, HZ
/50);
1664 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1665 printk(KERN_WARNING
"%s: page allocation failure."
1666 " order:%d, mode:0x%x\n",
1667 p
->comm
, order
, gfp_mask
);
1674 EXPORT_SYMBOL(__alloc_pages_internal
);
1677 * Common helper functions.
1679 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1682 page
= alloc_pages(gfp_mask
, order
);
1685 return (unsigned long) page_address(page
);
1688 EXPORT_SYMBOL(__get_free_pages
);
1690 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1695 * get_zeroed_page() returns a 32-bit address, which cannot represent
1698 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1700 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1702 return (unsigned long) page_address(page
);
1706 EXPORT_SYMBOL(get_zeroed_page
);
1708 void __pagevec_free(struct pagevec
*pvec
)
1710 int i
= pagevec_count(pvec
);
1713 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1716 void __free_pages(struct page
*page
, unsigned int order
)
1718 if (put_page_testzero(page
)) {
1720 free_hot_page(page
);
1722 __free_pages_ok(page
, order
);
1726 EXPORT_SYMBOL(__free_pages
);
1728 void free_pages(unsigned long addr
, unsigned int order
)
1731 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1732 __free_pages(virt_to_page((void *)addr
), order
);
1736 EXPORT_SYMBOL(free_pages
);
1739 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1740 * @size: the number of bytes to allocate
1741 * @gfp_mask: GFP flags for the allocation
1743 * This function is similar to alloc_pages(), except that it allocates the
1744 * minimum number of pages to satisfy the request. alloc_pages() can only
1745 * allocate memory in power-of-two pages.
1747 * This function is also limited by MAX_ORDER.
1749 * Memory allocated by this function must be released by free_pages_exact().
1751 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1753 unsigned int order
= get_order(size
);
1756 addr
= __get_free_pages(gfp_mask
, order
);
1758 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1759 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1761 split_page(virt_to_page(addr
), order
);
1762 while (used
< alloc_end
) {
1768 return (void *)addr
;
1770 EXPORT_SYMBOL(alloc_pages_exact
);
1773 * free_pages_exact - release memory allocated via alloc_pages_exact()
1774 * @virt: the value returned by alloc_pages_exact.
1775 * @size: size of allocation, same value as passed to alloc_pages_exact().
1777 * Release the memory allocated by a previous call to alloc_pages_exact.
1779 void free_pages_exact(void *virt
, size_t size
)
1781 unsigned long addr
= (unsigned long)virt
;
1782 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1784 while (addr
< end
) {
1789 EXPORT_SYMBOL(free_pages_exact
);
1791 static unsigned int nr_free_zone_pages(int offset
)
1796 /* Just pick one node, since fallback list is circular */
1797 unsigned int sum
= 0;
1799 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1801 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1802 unsigned long size
= zone
->present_pages
;
1803 unsigned long high
= zone
->pages_high
;
1812 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1814 unsigned int nr_free_buffer_pages(void)
1816 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1818 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1821 * Amount of free RAM allocatable within all zones
1823 unsigned int nr_free_pagecache_pages(void)
1825 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1828 static inline void show_node(struct zone
*zone
)
1831 printk("Node %d ", zone_to_nid(zone
));
1834 void si_meminfo(struct sysinfo
*val
)
1836 val
->totalram
= totalram_pages
;
1838 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1839 val
->bufferram
= nr_blockdev_pages();
1840 val
->totalhigh
= totalhigh_pages
;
1841 val
->freehigh
= nr_free_highpages();
1842 val
->mem_unit
= PAGE_SIZE
;
1845 EXPORT_SYMBOL(si_meminfo
);
1848 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1850 pg_data_t
*pgdat
= NODE_DATA(nid
);
1852 val
->totalram
= pgdat
->node_present_pages
;
1853 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1854 #ifdef CONFIG_HIGHMEM
1855 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1856 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1862 val
->mem_unit
= PAGE_SIZE
;
1866 #define K(x) ((x) << (PAGE_SHIFT-10))
1869 * Show free area list (used inside shift_scroll-lock stuff)
1870 * We also calculate the percentage fragmentation. We do this by counting the
1871 * memory on each free list with the exception of the first item on the list.
1873 void show_free_areas(void)
1878 for_each_zone(zone
) {
1879 if (!populated_zone(zone
))
1883 printk("%s per-cpu:\n", zone
->name
);
1885 for_each_online_cpu(cpu
) {
1886 struct per_cpu_pageset
*pageset
;
1888 pageset
= zone_pcp(zone
, cpu
);
1890 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1891 cpu
, pageset
->pcp
.high
,
1892 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1896 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1897 " inactive_file:%lu"
1898 //TODO: check/adjust line lengths
1899 #ifdef CONFIG_UNEVICTABLE_LRU
1902 " dirty:%lu writeback:%lu unstable:%lu\n"
1903 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1904 global_page_state(NR_ACTIVE_ANON
),
1905 global_page_state(NR_ACTIVE_FILE
),
1906 global_page_state(NR_INACTIVE_ANON
),
1907 global_page_state(NR_INACTIVE_FILE
),
1908 #ifdef CONFIG_UNEVICTABLE_LRU
1909 global_page_state(NR_UNEVICTABLE
),
1911 global_page_state(NR_FILE_DIRTY
),
1912 global_page_state(NR_WRITEBACK
),
1913 global_page_state(NR_UNSTABLE_NFS
),
1914 global_page_state(NR_FREE_PAGES
),
1915 global_page_state(NR_SLAB_RECLAIMABLE
) +
1916 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1917 global_page_state(NR_FILE_MAPPED
),
1918 global_page_state(NR_PAGETABLE
),
1919 global_page_state(NR_BOUNCE
));
1921 for_each_zone(zone
) {
1924 if (!populated_zone(zone
))
1933 " active_anon:%lukB"
1934 " inactive_anon:%lukB"
1935 " active_file:%lukB"
1936 " inactive_file:%lukB"
1937 #ifdef CONFIG_UNEVICTABLE_LRU
1938 " unevictable:%lukB"
1941 " pages_scanned:%lu"
1942 " all_unreclaimable? %s"
1945 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1948 K(zone
->pages_high
),
1949 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
1950 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
1951 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
1952 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
1953 #ifdef CONFIG_UNEVICTABLE_LRU
1954 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
1956 K(zone
->present_pages
),
1957 zone
->pages_scanned
,
1958 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1960 printk("lowmem_reserve[]:");
1961 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1962 printk(" %lu", zone
->lowmem_reserve
[i
]);
1966 for_each_zone(zone
) {
1967 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1969 if (!populated_zone(zone
))
1973 printk("%s: ", zone
->name
);
1975 spin_lock_irqsave(&zone
->lock
, flags
);
1976 for (order
= 0; order
< MAX_ORDER
; order
++) {
1977 nr
[order
] = zone
->free_area
[order
].nr_free
;
1978 total
+= nr
[order
] << order
;
1980 spin_unlock_irqrestore(&zone
->lock
, flags
);
1981 for (order
= 0; order
< MAX_ORDER
; order
++)
1982 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1983 printk("= %lukB\n", K(total
));
1986 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
1988 show_swap_cache_info();
1991 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
1993 zoneref
->zone
= zone
;
1994 zoneref
->zone_idx
= zone_idx(zone
);
1998 * Builds allocation fallback zone lists.
2000 * Add all populated zones of a node to the zonelist.
2002 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2003 int nr_zones
, enum zone_type zone_type
)
2007 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2012 zone
= pgdat
->node_zones
+ zone_type
;
2013 if (populated_zone(zone
)) {
2014 zoneref_set_zone(zone
,
2015 &zonelist
->_zonerefs
[nr_zones
++]);
2016 check_highest_zone(zone_type
);
2019 } while (zone_type
);
2026 * 0 = automatic detection of better ordering.
2027 * 1 = order by ([node] distance, -zonetype)
2028 * 2 = order by (-zonetype, [node] distance)
2030 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2031 * the same zonelist. So only NUMA can configure this param.
2033 #define ZONELIST_ORDER_DEFAULT 0
2034 #define ZONELIST_ORDER_NODE 1
2035 #define ZONELIST_ORDER_ZONE 2
2037 /* zonelist order in the kernel.
2038 * set_zonelist_order() will set this to NODE or ZONE.
2040 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2041 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2045 /* The value user specified ....changed by config */
2046 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2047 /* string for sysctl */
2048 #define NUMA_ZONELIST_ORDER_LEN 16
2049 char numa_zonelist_order
[16] = "default";
2052 * interface for configure zonelist ordering.
2053 * command line option "numa_zonelist_order"
2054 * = "[dD]efault - default, automatic configuration.
2055 * = "[nN]ode - order by node locality, then by zone within node
2056 * = "[zZ]one - order by zone, then by locality within zone
2059 static int __parse_numa_zonelist_order(char *s
)
2061 if (*s
== 'd' || *s
== 'D') {
2062 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2063 } else if (*s
== 'n' || *s
== 'N') {
2064 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2065 } else if (*s
== 'z' || *s
== 'Z') {
2066 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2069 "Ignoring invalid numa_zonelist_order value: "
2076 static __init
int setup_numa_zonelist_order(char *s
)
2079 return __parse_numa_zonelist_order(s
);
2082 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2085 * sysctl handler for numa_zonelist_order
2087 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2088 struct file
*file
, void __user
*buffer
, size_t *length
,
2091 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2095 strncpy(saved_string
, (char*)table
->data
,
2096 NUMA_ZONELIST_ORDER_LEN
);
2097 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2101 int oldval
= user_zonelist_order
;
2102 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2104 * bogus value. restore saved string
2106 strncpy((char*)table
->data
, saved_string
,
2107 NUMA_ZONELIST_ORDER_LEN
);
2108 user_zonelist_order
= oldval
;
2109 } else if (oldval
!= user_zonelist_order
)
2110 build_all_zonelists();
2116 #define MAX_NODE_LOAD (num_online_nodes())
2117 static int node_load
[MAX_NUMNODES
];
2120 * find_next_best_node - find the next node that should appear in a given node's fallback list
2121 * @node: node whose fallback list we're appending
2122 * @used_node_mask: nodemask_t of already used nodes
2124 * We use a number of factors to determine which is the next node that should
2125 * appear on a given node's fallback list. The node should not have appeared
2126 * already in @node's fallback list, and it should be the next closest node
2127 * according to the distance array (which contains arbitrary distance values
2128 * from each node to each node in the system), and should also prefer nodes
2129 * with no CPUs, since presumably they'll have very little allocation pressure
2130 * on them otherwise.
2131 * It returns -1 if no node is found.
2133 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2136 int min_val
= INT_MAX
;
2138 node_to_cpumask_ptr(tmp
, 0);
2140 /* Use the local node if we haven't already */
2141 if (!node_isset(node
, *used_node_mask
)) {
2142 node_set(node
, *used_node_mask
);
2146 for_each_node_state(n
, N_HIGH_MEMORY
) {
2148 /* Don't want a node to appear more than once */
2149 if (node_isset(n
, *used_node_mask
))
2152 /* Use the distance array to find the distance */
2153 val
= node_distance(node
, n
);
2155 /* Penalize nodes under us ("prefer the next node") */
2158 /* Give preference to headless and unused nodes */
2159 node_to_cpumask_ptr_next(tmp
, n
);
2160 if (!cpus_empty(*tmp
))
2161 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2163 /* Slight preference for less loaded node */
2164 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2165 val
+= node_load
[n
];
2167 if (val
< min_val
) {
2174 node_set(best_node
, *used_node_mask
);
2181 * Build zonelists ordered by node and zones within node.
2182 * This results in maximum locality--normal zone overflows into local
2183 * DMA zone, if any--but risks exhausting DMA zone.
2185 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2188 struct zonelist
*zonelist
;
2190 zonelist
= &pgdat
->node_zonelists
[0];
2191 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2193 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2195 zonelist
->_zonerefs
[j
].zone
= NULL
;
2196 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2200 * Build gfp_thisnode zonelists
2202 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2205 struct zonelist
*zonelist
;
2207 zonelist
= &pgdat
->node_zonelists
[1];
2208 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2209 zonelist
->_zonerefs
[j
].zone
= NULL
;
2210 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2214 * Build zonelists ordered by zone and nodes within zones.
2215 * This results in conserving DMA zone[s] until all Normal memory is
2216 * exhausted, but results in overflowing to remote node while memory
2217 * may still exist in local DMA zone.
2219 static int node_order
[MAX_NUMNODES
];
2221 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2224 int zone_type
; /* needs to be signed */
2226 struct zonelist
*zonelist
;
2228 zonelist
= &pgdat
->node_zonelists
[0];
2230 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2231 for (j
= 0; j
< nr_nodes
; j
++) {
2232 node
= node_order
[j
];
2233 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2234 if (populated_zone(z
)) {
2236 &zonelist
->_zonerefs
[pos
++]);
2237 check_highest_zone(zone_type
);
2241 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2242 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2245 static int default_zonelist_order(void)
2248 unsigned long low_kmem_size
,total_size
;
2252 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2253 * If they are really small and used heavily, the system can fall
2254 * into OOM very easily.
2255 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2257 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2260 for_each_online_node(nid
) {
2261 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2262 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2263 if (populated_zone(z
)) {
2264 if (zone_type
< ZONE_NORMAL
)
2265 low_kmem_size
+= z
->present_pages
;
2266 total_size
+= z
->present_pages
;
2270 if (!low_kmem_size
|| /* there are no DMA area. */
2271 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2272 return ZONELIST_ORDER_NODE
;
2274 * look into each node's config.
2275 * If there is a node whose DMA/DMA32 memory is very big area on
2276 * local memory, NODE_ORDER may be suitable.
2278 average_size
= total_size
/
2279 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2280 for_each_online_node(nid
) {
2283 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2284 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2285 if (populated_zone(z
)) {
2286 if (zone_type
< ZONE_NORMAL
)
2287 low_kmem_size
+= z
->present_pages
;
2288 total_size
+= z
->present_pages
;
2291 if (low_kmem_size
&&
2292 total_size
> average_size
&& /* ignore small node */
2293 low_kmem_size
> total_size
* 70/100)
2294 return ZONELIST_ORDER_NODE
;
2296 return ZONELIST_ORDER_ZONE
;
2299 static void set_zonelist_order(void)
2301 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2302 current_zonelist_order
= default_zonelist_order();
2304 current_zonelist_order
= user_zonelist_order
;
2307 static void build_zonelists(pg_data_t
*pgdat
)
2311 nodemask_t used_mask
;
2312 int local_node
, prev_node
;
2313 struct zonelist
*zonelist
;
2314 int order
= current_zonelist_order
;
2316 /* initialize zonelists */
2317 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2318 zonelist
= pgdat
->node_zonelists
+ i
;
2319 zonelist
->_zonerefs
[0].zone
= NULL
;
2320 zonelist
->_zonerefs
[0].zone_idx
= 0;
2323 /* NUMA-aware ordering of nodes */
2324 local_node
= pgdat
->node_id
;
2325 load
= num_online_nodes();
2326 prev_node
= local_node
;
2327 nodes_clear(used_mask
);
2329 memset(node_load
, 0, sizeof(node_load
));
2330 memset(node_order
, 0, sizeof(node_order
));
2333 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2334 int distance
= node_distance(local_node
, node
);
2337 * If another node is sufficiently far away then it is better
2338 * to reclaim pages in a zone before going off node.
2340 if (distance
> RECLAIM_DISTANCE
)
2341 zone_reclaim_mode
= 1;
2344 * We don't want to pressure a particular node.
2345 * So adding penalty to the first node in same
2346 * distance group to make it round-robin.
2348 if (distance
!= node_distance(local_node
, prev_node
))
2349 node_load
[node
] = load
;
2353 if (order
== ZONELIST_ORDER_NODE
)
2354 build_zonelists_in_node_order(pgdat
, node
);
2356 node_order
[j
++] = node
; /* remember order */
2359 if (order
== ZONELIST_ORDER_ZONE
) {
2360 /* calculate node order -- i.e., DMA last! */
2361 build_zonelists_in_zone_order(pgdat
, j
);
2364 build_thisnode_zonelists(pgdat
);
2367 /* Construct the zonelist performance cache - see further mmzone.h */
2368 static void build_zonelist_cache(pg_data_t
*pgdat
)
2370 struct zonelist
*zonelist
;
2371 struct zonelist_cache
*zlc
;
2374 zonelist
= &pgdat
->node_zonelists
[0];
2375 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2376 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2377 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2378 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2382 #else /* CONFIG_NUMA */
2384 static void set_zonelist_order(void)
2386 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2389 static void build_zonelists(pg_data_t
*pgdat
)
2391 int node
, local_node
;
2393 struct zonelist
*zonelist
;
2395 local_node
= pgdat
->node_id
;
2397 zonelist
= &pgdat
->node_zonelists
[0];
2398 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2401 * Now we build the zonelist so that it contains the zones
2402 * of all the other nodes.
2403 * We don't want to pressure a particular node, so when
2404 * building the zones for node N, we make sure that the
2405 * zones coming right after the local ones are those from
2406 * node N+1 (modulo N)
2408 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2409 if (!node_online(node
))
2411 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2414 for (node
= 0; node
< local_node
; node
++) {
2415 if (!node_online(node
))
2417 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2421 zonelist
->_zonerefs
[j
].zone
= NULL
;
2422 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2425 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2426 static void build_zonelist_cache(pg_data_t
*pgdat
)
2428 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2431 #endif /* CONFIG_NUMA */
2433 /* return values int ....just for stop_machine() */
2434 static int __build_all_zonelists(void *dummy
)
2438 for_each_online_node(nid
) {
2439 pg_data_t
*pgdat
= NODE_DATA(nid
);
2441 build_zonelists(pgdat
);
2442 build_zonelist_cache(pgdat
);
2447 void build_all_zonelists(void)
2449 set_zonelist_order();
2451 if (system_state
== SYSTEM_BOOTING
) {
2452 __build_all_zonelists(NULL
);
2453 mminit_verify_zonelist();
2454 cpuset_init_current_mems_allowed();
2456 /* we have to stop all cpus to guarantee there is no user
2458 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2459 /* cpuset refresh routine should be here */
2461 vm_total_pages
= nr_free_pagecache_pages();
2463 * Disable grouping by mobility if the number of pages in the
2464 * system is too low to allow the mechanism to work. It would be
2465 * more accurate, but expensive to check per-zone. This check is
2466 * made on memory-hotadd so a system can start with mobility
2467 * disabled and enable it later
2469 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2470 page_group_by_mobility_disabled
= 1;
2472 page_group_by_mobility_disabled
= 0;
2474 printk("Built %i zonelists in %s order, mobility grouping %s. "
2475 "Total pages: %ld\n",
2477 zonelist_order_name
[current_zonelist_order
],
2478 page_group_by_mobility_disabled
? "off" : "on",
2481 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2486 * Helper functions to size the waitqueue hash table.
2487 * Essentially these want to choose hash table sizes sufficiently
2488 * large so that collisions trying to wait on pages are rare.
2489 * But in fact, the number of active page waitqueues on typical
2490 * systems is ridiculously low, less than 200. So this is even
2491 * conservative, even though it seems large.
2493 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2494 * waitqueues, i.e. the size of the waitq table given the number of pages.
2496 #define PAGES_PER_WAITQUEUE 256
2498 #ifndef CONFIG_MEMORY_HOTPLUG
2499 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2501 unsigned long size
= 1;
2503 pages
/= PAGES_PER_WAITQUEUE
;
2505 while (size
< pages
)
2509 * Once we have dozens or even hundreds of threads sleeping
2510 * on IO we've got bigger problems than wait queue collision.
2511 * Limit the size of the wait table to a reasonable size.
2513 size
= min(size
, 4096UL);
2515 return max(size
, 4UL);
2519 * A zone's size might be changed by hot-add, so it is not possible to determine
2520 * a suitable size for its wait_table. So we use the maximum size now.
2522 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2524 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2525 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2526 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2528 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2529 * or more by the traditional way. (See above). It equals:
2531 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2532 * ia64(16K page size) : = ( 8G + 4M)byte.
2533 * powerpc (64K page size) : = (32G +16M)byte.
2535 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2542 * This is an integer logarithm so that shifts can be used later
2543 * to extract the more random high bits from the multiplicative
2544 * hash function before the remainder is taken.
2546 static inline unsigned long wait_table_bits(unsigned long size
)
2551 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2554 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2555 * of blocks reserved is based on zone->pages_min. The memory within the
2556 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2557 * higher will lead to a bigger reserve which will get freed as contiguous
2558 * blocks as reclaim kicks in
2560 static void setup_zone_migrate_reserve(struct zone
*zone
)
2562 unsigned long start_pfn
, pfn
, end_pfn
;
2564 unsigned long reserve
, block_migratetype
;
2566 /* Get the start pfn, end pfn and the number of blocks to reserve */
2567 start_pfn
= zone
->zone_start_pfn
;
2568 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2569 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2572 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2573 if (!pfn_valid(pfn
))
2575 page
= pfn_to_page(pfn
);
2577 /* Watch out for overlapping nodes */
2578 if (page_to_nid(page
) != zone_to_nid(zone
))
2581 /* Blocks with reserved pages will never free, skip them. */
2582 if (PageReserved(page
))
2585 block_migratetype
= get_pageblock_migratetype(page
);
2587 /* If this block is reserved, account for it */
2588 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2593 /* Suitable for reserving if this block is movable */
2594 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2595 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2596 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2602 * If the reserve is met and this is a previous reserved block,
2605 if (block_migratetype
== MIGRATE_RESERVE
) {
2606 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2607 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2613 * Initially all pages are reserved - free ones are freed
2614 * up by free_all_bootmem() once the early boot process is
2615 * done. Non-atomic initialization, single-pass.
2617 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2618 unsigned long start_pfn
, enum memmap_context context
)
2621 unsigned long end_pfn
= start_pfn
+ size
;
2625 if (highest_memmap_pfn
< end_pfn
- 1)
2626 highest_memmap_pfn
= end_pfn
- 1;
2628 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2629 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2631 * There can be holes in boot-time mem_map[]s
2632 * handed to this function. They do not
2633 * exist on hotplugged memory.
2635 if (context
== MEMMAP_EARLY
) {
2636 if (!early_pfn_valid(pfn
))
2638 if (!early_pfn_in_nid(pfn
, nid
))
2641 page
= pfn_to_page(pfn
);
2642 set_page_links(page
, zone
, nid
, pfn
);
2643 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2644 init_page_count(page
);
2645 reset_page_mapcount(page
);
2646 SetPageReserved(page
);
2648 * Mark the block movable so that blocks are reserved for
2649 * movable at startup. This will force kernel allocations
2650 * to reserve their blocks rather than leaking throughout
2651 * the address space during boot when many long-lived
2652 * kernel allocations are made. Later some blocks near
2653 * the start are marked MIGRATE_RESERVE by
2654 * setup_zone_migrate_reserve()
2656 * bitmap is created for zone's valid pfn range. but memmap
2657 * can be created for invalid pages (for alignment)
2658 * check here not to call set_pageblock_migratetype() against
2661 if ((z
->zone_start_pfn
<= pfn
)
2662 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2663 && !(pfn
& (pageblock_nr_pages
- 1)))
2664 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2666 INIT_LIST_HEAD(&page
->lru
);
2667 #ifdef WANT_PAGE_VIRTUAL
2668 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2669 if (!is_highmem_idx(zone
))
2670 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2675 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2678 for_each_migratetype_order(order
, t
) {
2679 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2680 zone
->free_area
[order
].nr_free
= 0;
2684 #ifndef __HAVE_ARCH_MEMMAP_INIT
2685 #define memmap_init(size, nid, zone, start_pfn) \
2686 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2689 static int zone_batchsize(struct zone
*zone
)
2694 * The per-cpu-pages pools are set to around 1000th of the
2695 * size of the zone. But no more than 1/2 of a meg.
2697 * OK, so we don't know how big the cache is. So guess.
2699 batch
= zone
->present_pages
/ 1024;
2700 if (batch
* PAGE_SIZE
> 512 * 1024)
2701 batch
= (512 * 1024) / PAGE_SIZE
;
2702 batch
/= 4; /* We effectively *= 4 below */
2707 * Clamp the batch to a 2^n - 1 value. Having a power
2708 * of 2 value was found to be more likely to have
2709 * suboptimal cache aliasing properties in some cases.
2711 * For example if 2 tasks are alternately allocating
2712 * batches of pages, one task can end up with a lot
2713 * of pages of one half of the possible page colors
2714 * and the other with pages of the other colors.
2716 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2721 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2723 struct per_cpu_pages
*pcp
;
2725 memset(p
, 0, sizeof(*p
));
2729 pcp
->high
= 6 * batch
;
2730 pcp
->batch
= max(1UL, 1 * batch
);
2731 INIT_LIST_HEAD(&pcp
->list
);
2735 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2736 * to the value high for the pageset p.
2739 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2742 struct per_cpu_pages
*pcp
;
2746 pcp
->batch
= max(1UL, high
/4);
2747 if ((high
/4) > (PAGE_SHIFT
* 8))
2748 pcp
->batch
= PAGE_SHIFT
* 8;
2754 * Boot pageset table. One per cpu which is going to be used for all
2755 * zones and all nodes. The parameters will be set in such a way
2756 * that an item put on a list will immediately be handed over to
2757 * the buddy list. This is safe since pageset manipulation is done
2758 * with interrupts disabled.
2760 * Some NUMA counter updates may also be caught by the boot pagesets.
2762 * The boot_pagesets must be kept even after bootup is complete for
2763 * unused processors and/or zones. They do play a role for bootstrapping
2764 * hotplugged processors.
2766 * zoneinfo_show() and maybe other functions do
2767 * not check if the processor is online before following the pageset pointer.
2768 * Other parts of the kernel may not check if the zone is available.
2770 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2773 * Dynamically allocate memory for the
2774 * per cpu pageset array in struct zone.
2776 static int __cpuinit
process_zones(int cpu
)
2778 struct zone
*zone
, *dzone
;
2779 int node
= cpu_to_node(cpu
);
2781 node_set_state(node
, N_CPU
); /* this node has a cpu */
2783 for_each_zone(zone
) {
2785 if (!populated_zone(zone
))
2788 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2790 if (!zone_pcp(zone
, cpu
))
2793 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2795 if (percpu_pagelist_fraction
)
2796 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2797 (zone
->present_pages
/ percpu_pagelist_fraction
));
2802 for_each_zone(dzone
) {
2803 if (!populated_zone(dzone
))
2807 kfree(zone_pcp(dzone
, cpu
));
2808 zone_pcp(dzone
, cpu
) = NULL
;
2813 static inline void free_zone_pagesets(int cpu
)
2817 for_each_zone(zone
) {
2818 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2820 /* Free per_cpu_pageset if it is slab allocated */
2821 if (pset
!= &boot_pageset
[cpu
])
2823 zone_pcp(zone
, cpu
) = NULL
;
2827 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2828 unsigned long action
,
2831 int cpu
= (long)hcpu
;
2832 int ret
= NOTIFY_OK
;
2835 case CPU_UP_PREPARE
:
2836 case CPU_UP_PREPARE_FROZEN
:
2837 if (process_zones(cpu
))
2840 case CPU_UP_CANCELED
:
2841 case CPU_UP_CANCELED_FROZEN
:
2843 case CPU_DEAD_FROZEN
:
2844 free_zone_pagesets(cpu
);
2852 static struct notifier_block __cpuinitdata pageset_notifier
=
2853 { &pageset_cpuup_callback
, NULL
, 0 };
2855 void __init
setup_per_cpu_pageset(void)
2859 /* Initialize per_cpu_pageset for cpu 0.
2860 * A cpuup callback will do this for every cpu
2861 * as it comes online
2863 err
= process_zones(smp_processor_id());
2865 register_cpu_notifier(&pageset_notifier
);
2870 static noinline __init_refok
2871 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2874 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2878 * The per-page waitqueue mechanism uses hashed waitqueues
2881 zone
->wait_table_hash_nr_entries
=
2882 wait_table_hash_nr_entries(zone_size_pages
);
2883 zone
->wait_table_bits
=
2884 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2885 alloc_size
= zone
->wait_table_hash_nr_entries
2886 * sizeof(wait_queue_head_t
);
2888 if (!slab_is_available()) {
2889 zone
->wait_table
= (wait_queue_head_t
*)
2890 alloc_bootmem_node(pgdat
, alloc_size
);
2893 * This case means that a zone whose size was 0 gets new memory
2894 * via memory hot-add.
2895 * But it may be the case that a new node was hot-added. In
2896 * this case vmalloc() will not be able to use this new node's
2897 * memory - this wait_table must be initialized to use this new
2898 * node itself as well.
2899 * To use this new node's memory, further consideration will be
2902 zone
->wait_table
= vmalloc(alloc_size
);
2904 if (!zone
->wait_table
)
2907 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2908 init_waitqueue_head(zone
->wait_table
+ i
);
2913 static __meminit
void zone_pcp_init(struct zone
*zone
)
2916 unsigned long batch
= zone_batchsize(zone
);
2918 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2920 /* Early boot. Slab allocator not functional yet */
2921 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2922 setup_pageset(&boot_pageset
[cpu
],0);
2924 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2927 if (zone
->present_pages
)
2928 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2929 zone
->name
, zone
->present_pages
, batch
);
2932 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2933 unsigned long zone_start_pfn
,
2935 enum memmap_context context
)
2937 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2939 ret
= zone_wait_table_init(zone
, size
);
2942 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2944 zone
->zone_start_pfn
= zone_start_pfn
;
2946 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
2947 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2949 (unsigned long)zone_idx(zone
),
2950 zone_start_pfn
, (zone_start_pfn
+ size
));
2952 zone_init_free_lists(zone
);
2957 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2959 * Basic iterator support. Return the first range of PFNs for a node
2960 * Note: nid == MAX_NUMNODES returns first region regardless of node
2962 static int __meminit
first_active_region_index_in_nid(int nid
)
2966 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2967 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2974 * Basic iterator support. Return the next active range of PFNs for a node
2975 * Note: nid == MAX_NUMNODES returns next region regardless of node
2977 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2979 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2980 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2986 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2988 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2989 * Architectures may implement their own version but if add_active_range()
2990 * was used and there are no special requirements, this is a convenient
2993 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2997 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2998 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2999 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3001 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3002 return early_node_map
[i
].nid
;
3007 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3009 /* Basic iterator support to walk early_node_map[] */
3010 #define for_each_active_range_index_in_nid(i, nid) \
3011 for (i = first_active_region_index_in_nid(nid); i != -1; \
3012 i = next_active_region_index_in_nid(i, nid))
3015 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3016 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3017 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3019 * If an architecture guarantees that all ranges registered with
3020 * add_active_ranges() contain no holes and may be freed, this
3021 * this function may be used instead of calling free_bootmem() manually.
3023 void __init
free_bootmem_with_active_regions(int nid
,
3024 unsigned long max_low_pfn
)
3028 for_each_active_range_index_in_nid(i
, nid
) {
3029 unsigned long size_pages
= 0;
3030 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3032 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3035 if (end_pfn
> max_low_pfn
)
3036 end_pfn
= max_low_pfn
;
3038 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3039 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3040 PFN_PHYS(early_node_map
[i
].start_pfn
),
3041 size_pages
<< PAGE_SHIFT
);
3045 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3050 for_each_active_range_index_in_nid(i
, nid
) {
3051 ret
= work_fn(early_node_map
[i
].start_pfn
,
3052 early_node_map
[i
].end_pfn
, data
);
3058 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3059 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3061 * If an architecture guarantees that all ranges registered with
3062 * add_active_ranges() contain no holes and may be freed, this
3063 * function may be used instead of calling memory_present() manually.
3065 void __init
sparse_memory_present_with_active_regions(int nid
)
3069 for_each_active_range_index_in_nid(i
, nid
)
3070 memory_present(early_node_map
[i
].nid
,
3071 early_node_map
[i
].start_pfn
,
3072 early_node_map
[i
].end_pfn
);
3076 * push_node_boundaries - Push node boundaries to at least the requested boundary
3077 * @nid: The nid of the node to push the boundary for
3078 * @start_pfn: The start pfn of the node
3079 * @end_pfn: The end pfn of the node
3081 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3082 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3083 * be hotplugged even though no physical memory exists. This function allows
3084 * an arch to push out the node boundaries so mem_map is allocated that can
3087 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3088 void __init
push_node_boundaries(unsigned int nid
,
3089 unsigned long start_pfn
, unsigned long end_pfn
)
3091 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3092 "Entering push_node_boundaries(%u, %lu, %lu)\n",
3093 nid
, start_pfn
, end_pfn
);
3095 /* Initialise the boundary for this node if necessary */
3096 if (node_boundary_end_pfn
[nid
] == 0)
3097 node_boundary_start_pfn
[nid
] = -1UL;
3099 /* Update the boundaries */
3100 if (node_boundary_start_pfn
[nid
] > start_pfn
)
3101 node_boundary_start_pfn
[nid
] = start_pfn
;
3102 if (node_boundary_end_pfn
[nid
] < end_pfn
)
3103 node_boundary_end_pfn
[nid
] = end_pfn
;
3106 /* If necessary, push the node boundary out for reserve hotadd */
3107 static void __meminit
account_node_boundary(unsigned int nid
,
3108 unsigned long *start_pfn
, unsigned long *end_pfn
)
3110 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3111 "Entering account_node_boundary(%u, %lu, %lu)\n",
3112 nid
, *start_pfn
, *end_pfn
);
3114 /* Return if boundary information has not been provided */
3115 if (node_boundary_end_pfn
[nid
] == 0)
3118 /* Check the boundaries and update if necessary */
3119 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
3120 *start_pfn
= node_boundary_start_pfn
[nid
];
3121 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
3122 *end_pfn
= node_boundary_end_pfn
[nid
];
3125 void __init
push_node_boundaries(unsigned int nid
,
3126 unsigned long start_pfn
, unsigned long end_pfn
) {}
3128 static void __meminit
account_node_boundary(unsigned int nid
,
3129 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3134 * get_pfn_range_for_nid - Return the start and end page frames for a node
3135 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3136 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3137 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3139 * It returns the start and end page frame of a node based on information
3140 * provided by an arch calling add_active_range(). If called for a node
3141 * with no available memory, a warning is printed and the start and end
3144 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3145 unsigned long *start_pfn
, unsigned long *end_pfn
)
3151 for_each_active_range_index_in_nid(i
, nid
) {
3152 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3153 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3156 if (*start_pfn
== -1UL)
3159 /* Push the node boundaries out if requested */
3160 account_node_boundary(nid
, start_pfn
, end_pfn
);
3164 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3165 * assumption is made that zones within a node are ordered in monotonic
3166 * increasing memory addresses so that the "highest" populated zone is used
3168 static void __init
find_usable_zone_for_movable(void)
3171 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3172 if (zone_index
== ZONE_MOVABLE
)
3175 if (arch_zone_highest_possible_pfn
[zone_index
] >
3176 arch_zone_lowest_possible_pfn
[zone_index
])
3180 VM_BUG_ON(zone_index
== -1);
3181 movable_zone
= zone_index
;
3185 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3186 * because it is sized independant of architecture. Unlike the other zones,
3187 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3188 * in each node depending on the size of each node and how evenly kernelcore
3189 * is distributed. This helper function adjusts the zone ranges
3190 * provided by the architecture for a given node by using the end of the
3191 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3192 * zones within a node are in order of monotonic increases memory addresses
3194 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3195 unsigned long zone_type
,
3196 unsigned long node_start_pfn
,
3197 unsigned long node_end_pfn
,
3198 unsigned long *zone_start_pfn
,
3199 unsigned long *zone_end_pfn
)
3201 /* Only adjust if ZONE_MOVABLE is on this node */
3202 if (zone_movable_pfn
[nid
]) {
3203 /* Size ZONE_MOVABLE */
3204 if (zone_type
== ZONE_MOVABLE
) {
3205 *zone_start_pfn
= zone_movable_pfn
[nid
];
3206 *zone_end_pfn
= min(node_end_pfn
,
3207 arch_zone_highest_possible_pfn
[movable_zone
]);
3209 /* Adjust for ZONE_MOVABLE starting within this range */
3210 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3211 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3212 *zone_end_pfn
= zone_movable_pfn
[nid
];
3214 /* Check if this whole range is within ZONE_MOVABLE */
3215 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3216 *zone_start_pfn
= *zone_end_pfn
;
3221 * Return the number of pages a zone spans in a node, including holes
3222 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3224 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3225 unsigned long zone_type
,
3226 unsigned long *ignored
)
3228 unsigned long node_start_pfn
, node_end_pfn
;
3229 unsigned long zone_start_pfn
, zone_end_pfn
;
3231 /* Get the start and end of the node and zone */
3232 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3233 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3234 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3235 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3236 node_start_pfn
, node_end_pfn
,
3237 &zone_start_pfn
, &zone_end_pfn
);
3239 /* Check that this node has pages within the zone's required range */
3240 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3243 /* Move the zone boundaries inside the node if necessary */
3244 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3245 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3247 /* Return the spanned pages */
3248 return zone_end_pfn
- zone_start_pfn
;
3252 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3253 * then all holes in the requested range will be accounted for.
3255 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3256 unsigned long range_start_pfn
,
3257 unsigned long range_end_pfn
)
3260 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3261 unsigned long start_pfn
;
3263 /* Find the end_pfn of the first active range of pfns in the node */
3264 i
= first_active_region_index_in_nid(nid
);
3268 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3270 /* Account for ranges before physical memory on this node */
3271 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3272 hole_pages
= prev_end_pfn
- range_start_pfn
;
3274 /* Find all holes for the zone within the node */
3275 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3277 /* No need to continue if prev_end_pfn is outside the zone */
3278 if (prev_end_pfn
>= range_end_pfn
)
3281 /* Make sure the end of the zone is not within the hole */
3282 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3283 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3285 /* Update the hole size cound and move on */
3286 if (start_pfn
> range_start_pfn
) {
3287 BUG_ON(prev_end_pfn
> start_pfn
);
3288 hole_pages
+= start_pfn
- prev_end_pfn
;
3290 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3293 /* Account for ranges past physical memory on this node */
3294 if (range_end_pfn
> prev_end_pfn
)
3295 hole_pages
+= range_end_pfn
-
3296 max(range_start_pfn
, prev_end_pfn
);
3302 * absent_pages_in_range - Return number of page frames in holes within a range
3303 * @start_pfn: The start PFN to start searching for holes
3304 * @end_pfn: The end PFN to stop searching for holes
3306 * It returns the number of pages frames in memory holes within a range.
3308 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3309 unsigned long end_pfn
)
3311 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3314 /* Return the number of page frames in holes in a zone on a node */
3315 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3316 unsigned long zone_type
,
3317 unsigned long *ignored
)
3319 unsigned long node_start_pfn
, node_end_pfn
;
3320 unsigned long zone_start_pfn
, zone_end_pfn
;
3322 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3323 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3325 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3328 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3329 node_start_pfn
, node_end_pfn
,
3330 &zone_start_pfn
, &zone_end_pfn
);
3331 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3335 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3336 unsigned long zone_type
,
3337 unsigned long *zones_size
)
3339 return zones_size
[zone_type
];
3342 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3343 unsigned long zone_type
,
3344 unsigned long *zholes_size
)
3349 return zholes_size
[zone_type
];
3354 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3355 unsigned long *zones_size
, unsigned long *zholes_size
)
3357 unsigned long realtotalpages
, totalpages
= 0;
3360 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3361 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3363 pgdat
->node_spanned_pages
= totalpages
;
3365 realtotalpages
= totalpages
;
3366 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3368 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3370 pgdat
->node_present_pages
= realtotalpages
;
3371 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3375 #ifndef CONFIG_SPARSEMEM
3377 * Calculate the size of the zone->blockflags rounded to an unsigned long
3378 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3379 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3380 * round what is now in bits to nearest long in bits, then return it in
3383 static unsigned long __init
usemap_size(unsigned long zonesize
)
3385 unsigned long usemapsize
;
3387 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3388 usemapsize
= usemapsize
>> pageblock_order
;
3389 usemapsize
*= NR_PAGEBLOCK_BITS
;
3390 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3392 return usemapsize
/ 8;
3395 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3396 struct zone
*zone
, unsigned long zonesize
)
3398 unsigned long usemapsize
= usemap_size(zonesize
);
3399 zone
->pageblock_flags
= NULL
;
3401 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3404 static void inline setup_usemap(struct pglist_data
*pgdat
,
3405 struct zone
*zone
, unsigned long zonesize
) {}
3406 #endif /* CONFIG_SPARSEMEM */
3408 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3410 /* Return a sensible default order for the pageblock size. */
3411 static inline int pageblock_default_order(void)
3413 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3414 return HUGETLB_PAGE_ORDER
;
3419 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3420 static inline void __init
set_pageblock_order(unsigned int order
)
3422 /* Check that pageblock_nr_pages has not already been setup */
3423 if (pageblock_order
)
3427 * Assume the largest contiguous order of interest is a huge page.
3428 * This value may be variable depending on boot parameters on IA64
3430 pageblock_order
= order
;
3432 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3435 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3436 * and pageblock_default_order() are unused as pageblock_order is set
3437 * at compile-time. See include/linux/pageblock-flags.h for the values of
3438 * pageblock_order based on the kernel config
3440 static inline int pageblock_default_order(unsigned int order
)
3444 #define set_pageblock_order(x) do {} while (0)
3446 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3449 * Set up the zone data structures:
3450 * - mark all pages reserved
3451 * - mark all memory queues empty
3452 * - clear the memory bitmaps
3454 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3455 unsigned long *zones_size
, unsigned long *zholes_size
)
3458 int nid
= pgdat
->node_id
;
3459 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3462 pgdat_resize_init(pgdat
);
3463 pgdat
->nr_zones
= 0;
3464 init_waitqueue_head(&pgdat
->kswapd_wait
);
3465 pgdat
->kswapd_max_order
= 0;
3466 pgdat_page_cgroup_init(pgdat
);
3468 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3469 struct zone
*zone
= pgdat
->node_zones
+ j
;
3470 unsigned long size
, realsize
, memmap_pages
;
3473 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3474 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3478 * Adjust realsize so that it accounts for how much memory
3479 * is used by this zone for memmap. This affects the watermark
3480 * and per-cpu initialisations
3483 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3484 if (realsize
>= memmap_pages
) {
3485 realsize
-= memmap_pages
;
3488 " %s zone: %lu pages used for memmap\n",
3489 zone_names
[j
], memmap_pages
);
3492 " %s zone: %lu pages exceeds realsize %lu\n",
3493 zone_names
[j
], memmap_pages
, realsize
);
3495 /* Account for reserved pages */
3496 if (j
== 0 && realsize
> dma_reserve
) {
3497 realsize
-= dma_reserve
;
3498 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3499 zone_names
[0], dma_reserve
);
3502 if (!is_highmem_idx(j
))
3503 nr_kernel_pages
+= realsize
;
3504 nr_all_pages
+= realsize
;
3506 zone
->spanned_pages
= size
;
3507 zone
->present_pages
= realsize
;
3510 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3512 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3514 zone
->name
= zone_names
[j
];
3515 spin_lock_init(&zone
->lock
);
3516 spin_lock_init(&zone
->lru_lock
);
3517 zone_seqlock_init(zone
);
3518 zone
->zone_pgdat
= pgdat
;
3520 zone
->prev_priority
= DEF_PRIORITY
;
3522 zone_pcp_init(zone
);
3524 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3525 zone
->lru
[l
].nr_scan
= 0;
3527 zone
->recent_rotated
[0] = 0;
3528 zone
->recent_rotated
[1] = 0;
3529 zone
->recent_scanned
[0] = 0;
3530 zone
->recent_scanned
[1] = 0;
3531 zap_zone_vm_stats(zone
);
3536 set_pageblock_order(pageblock_default_order());
3537 setup_usemap(pgdat
, zone
, size
);
3538 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3539 size
, MEMMAP_EARLY
);
3541 memmap_init(size
, nid
, j
, zone_start_pfn
);
3542 zone_start_pfn
+= size
;
3546 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3548 /* Skip empty nodes */
3549 if (!pgdat
->node_spanned_pages
)
3552 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3553 /* ia64 gets its own node_mem_map, before this, without bootmem */
3554 if (!pgdat
->node_mem_map
) {
3555 unsigned long size
, start
, end
;
3559 * The zone's endpoints aren't required to be MAX_ORDER
3560 * aligned but the node_mem_map endpoints must be in order
3561 * for the buddy allocator to function correctly.
3563 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3564 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3565 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3566 size
= (end
- start
) * sizeof(struct page
);
3567 map
= alloc_remap(pgdat
->node_id
, size
);
3569 map
= alloc_bootmem_node(pgdat
, size
);
3570 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3572 #ifndef CONFIG_NEED_MULTIPLE_NODES
3574 * With no DISCONTIG, the global mem_map is just set as node 0's
3576 if (pgdat
== NODE_DATA(0)) {
3577 mem_map
= NODE_DATA(0)->node_mem_map
;
3578 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3579 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3580 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3581 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3584 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3587 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3588 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3590 pg_data_t
*pgdat
= NODE_DATA(nid
);
3592 pgdat
->node_id
= nid
;
3593 pgdat
->node_start_pfn
= node_start_pfn
;
3594 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3596 alloc_node_mem_map(pgdat
);
3597 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3598 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3599 nid
, (unsigned long)pgdat
,
3600 (unsigned long)pgdat
->node_mem_map
);
3603 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3606 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3608 #if MAX_NUMNODES > 1
3610 * Figure out the number of possible node ids.
3612 static void __init
setup_nr_node_ids(void)
3615 unsigned int highest
= 0;
3617 for_each_node_mask(node
, node_possible_map
)
3619 nr_node_ids
= highest
+ 1;
3622 static inline void setup_nr_node_ids(void)
3628 * add_active_range - Register a range of PFNs backed by physical memory
3629 * @nid: The node ID the range resides on
3630 * @start_pfn: The start PFN of the available physical memory
3631 * @end_pfn: The end PFN of the available physical memory
3633 * These ranges are stored in an early_node_map[] and later used by
3634 * free_area_init_nodes() to calculate zone sizes and holes. If the
3635 * range spans a memory hole, it is up to the architecture to ensure
3636 * the memory is not freed by the bootmem allocator. If possible
3637 * the range being registered will be merged with existing ranges.
3639 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3640 unsigned long end_pfn
)
3644 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3645 "Entering add_active_range(%d, %#lx, %#lx) "
3646 "%d entries of %d used\n",
3647 nid
, start_pfn
, end_pfn
,
3648 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3650 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3652 /* Merge with existing active regions if possible */
3653 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3654 if (early_node_map
[i
].nid
!= nid
)
3657 /* Skip if an existing region covers this new one */
3658 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3659 end_pfn
<= early_node_map
[i
].end_pfn
)
3662 /* Merge forward if suitable */
3663 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3664 end_pfn
> early_node_map
[i
].end_pfn
) {
3665 early_node_map
[i
].end_pfn
= end_pfn
;
3669 /* Merge backward if suitable */
3670 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3671 end_pfn
>= early_node_map
[i
].start_pfn
) {
3672 early_node_map
[i
].start_pfn
= start_pfn
;
3677 /* Check that early_node_map is large enough */
3678 if (i
>= MAX_ACTIVE_REGIONS
) {
3679 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3680 MAX_ACTIVE_REGIONS
);
3684 early_node_map
[i
].nid
= nid
;
3685 early_node_map
[i
].start_pfn
= start_pfn
;
3686 early_node_map
[i
].end_pfn
= end_pfn
;
3687 nr_nodemap_entries
= i
+ 1;
3691 * remove_active_range - Shrink an existing registered range of PFNs
3692 * @nid: The node id the range is on that should be shrunk
3693 * @start_pfn: The new PFN of the range
3694 * @end_pfn: The new PFN of the range
3696 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3697 * The map is kept near the end physical page range that has already been
3698 * registered. This function allows an arch to shrink an existing registered
3701 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3702 unsigned long end_pfn
)
3707 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3708 nid
, start_pfn
, end_pfn
);
3710 /* Find the old active region end and shrink */
3711 for_each_active_range_index_in_nid(i
, nid
) {
3712 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3713 early_node_map
[i
].end_pfn
<= end_pfn
) {
3715 early_node_map
[i
].start_pfn
= 0;
3716 early_node_map
[i
].end_pfn
= 0;
3720 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3721 early_node_map
[i
].end_pfn
> start_pfn
) {
3722 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3723 early_node_map
[i
].end_pfn
= start_pfn
;
3724 if (temp_end_pfn
> end_pfn
)
3725 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3728 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3729 early_node_map
[i
].end_pfn
> end_pfn
&&
3730 early_node_map
[i
].start_pfn
< end_pfn
) {
3731 early_node_map
[i
].start_pfn
= end_pfn
;
3739 /* remove the blank ones */
3740 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3741 if (early_node_map
[i
].nid
!= nid
)
3743 if (early_node_map
[i
].end_pfn
)
3745 /* we found it, get rid of it */
3746 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3747 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3748 sizeof(early_node_map
[j
]));
3749 j
= nr_nodemap_entries
- 1;
3750 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3751 nr_nodemap_entries
--;
3756 * remove_all_active_ranges - Remove all currently registered regions
3758 * During discovery, it may be found that a table like SRAT is invalid
3759 * and an alternative discovery method must be used. This function removes
3760 * all currently registered regions.
3762 void __init
remove_all_active_ranges(void)
3764 memset(early_node_map
, 0, sizeof(early_node_map
));
3765 nr_nodemap_entries
= 0;
3766 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3767 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3768 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3769 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3772 /* Compare two active node_active_regions */
3773 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3775 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3776 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3778 /* Done this way to avoid overflows */
3779 if (arange
->start_pfn
> brange
->start_pfn
)
3781 if (arange
->start_pfn
< brange
->start_pfn
)
3787 /* sort the node_map by start_pfn */
3788 static void __init
sort_node_map(void)
3790 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3791 sizeof(struct node_active_region
),
3792 cmp_node_active_region
, NULL
);
3795 /* Find the lowest pfn for a node */
3796 static unsigned long __init
find_min_pfn_for_node(int nid
)
3799 unsigned long min_pfn
= ULONG_MAX
;
3801 /* Assuming a sorted map, the first range found has the starting pfn */
3802 for_each_active_range_index_in_nid(i
, nid
)
3803 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3805 if (min_pfn
== ULONG_MAX
) {
3807 "Could not find start_pfn for node %d\n", nid
);
3815 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3817 * It returns the minimum PFN based on information provided via
3818 * add_active_range().
3820 unsigned long __init
find_min_pfn_with_active_regions(void)
3822 return find_min_pfn_for_node(MAX_NUMNODES
);
3826 * early_calculate_totalpages()
3827 * Sum pages in active regions for movable zone.
3828 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3830 static unsigned long __init
early_calculate_totalpages(void)
3833 unsigned long totalpages
= 0;
3835 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3836 unsigned long pages
= early_node_map
[i
].end_pfn
-
3837 early_node_map
[i
].start_pfn
;
3838 totalpages
+= pages
;
3840 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3846 * Find the PFN the Movable zone begins in each node. Kernel memory
3847 * is spread evenly between nodes as long as the nodes have enough
3848 * memory. When they don't, some nodes will have more kernelcore than
3851 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3854 unsigned long usable_startpfn
;
3855 unsigned long kernelcore_node
, kernelcore_remaining
;
3856 unsigned long totalpages
= early_calculate_totalpages();
3857 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3860 * If movablecore was specified, calculate what size of
3861 * kernelcore that corresponds so that memory usable for
3862 * any allocation type is evenly spread. If both kernelcore
3863 * and movablecore are specified, then the value of kernelcore
3864 * will be used for required_kernelcore if it's greater than
3865 * what movablecore would have allowed.
3867 if (required_movablecore
) {
3868 unsigned long corepages
;
3871 * Round-up so that ZONE_MOVABLE is at least as large as what
3872 * was requested by the user
3874 required_movablecore
=
3875 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3876 corepages
= totalpages
- required_movablecore
;
3878 required_kernelcore
= max(required_kernelcore
, corepages
);
3881 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3882 if (!required_kernelcore
)
3885 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3886 find_usable_zone_for_movable();
3887 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3890 /* Spread kernelcore memory as evenly as possible throughout nodes */
3891 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3892 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3894 * Recalculate kernelcore_node if the division per node
3895 * now exceeds what is necessary to satisfy the requested
3896 * amount of memory for the kernel
3898 if (required_kernelcore
< kernelcore_node
)
3899 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3902 * As the map is walked, we track how much memory is usable
3903 * by the kernel using kernelcore_remaining. When it is
3904 * 0, the rest of the node is usable by ZONE_MOVABLE
3906 kernelcore_remaining
= kernelcore_node
;
3908 /* Go through each range of PFNs within this node */
3909 for_each_active_range_index_in_nid(i
, nid
) {
3910 unsigned long start_pfn
, end_pfn
;
3911 unsigned long size_pages
;
3913 start_pfn
= max(early_node_map
[i
].start_pfn
,
3914 zone_movable_pfn
[nid
]);
3915 end_pfn
= early_node_map
[i
].end_pfn
;
3916 if (start_pfn
>= end_pfn
)
3919 /* Account for what is only usable for kernelcore */
3920 if (start_pfn
< usable_startpfn
) {
3921 unsigned long kernel_pages
;
3922 kernel_pages
= min(end_pfn
, usable_startpfn
)
3925 kernelcore_remaining
-= min(kernel_pages
,
3926 kernelcore_remaining
);
3927 required_kernelcore
-= min(kernel_pages
,
3928 required_kernelcore
);
3930 /* Continue if range is now fully accounted */
3931 if (end_pfn
<= usable_startpfn
) {
3934 * Push zone_movable_pfn to the end so
3935 * that if we have to rebalance
3936 * kernelcore across nodes, we will
3937 * not double account here
3939 zone_movable_pfn
[nid
] = end_pfn
;
3942 start_pfn
= usable_startpfn
;
3946 * The usable PFN range for ZONE_MOVABLE is from
3947 * start_pfn->end_pfn. Calculate size_pages as the
3948 * number of pages used as kernelcore
3950 size_pages
= end_pfn
- start_pfn
;
3951 if (size_pages
> kernelcore_remaining
)
3952 size_pages
= kernelcore_remaining
;
3953 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3956 * Some kernelcore has been met, update counts and
3957 * break if the kernelcore for this node has been
3960 required_kernelcore
-= min(required_kernelcore
,
3962 kernelcore_remaining
-= size_pages
;
3963 if (!kernelcore_remaining
)
3969 * If there is still required_kernelcore, we do another pass with one
3970 * less node in the count. This will push zone_movable_pfn[nid] further
3971 * along on the nodes that still have memory until kernelcore is
3975 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3978 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3979 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3980 zone_movable_pfn
[nid
] =
3981 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3984 /* Any regular memory on that node ? */
3985 static void check_for_regular_memory(pg_data_t
*pgdat
)
3987 #ifdef CONFIG_HIGHMEM
3988 enum zone_type zone_type
;
3990 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3991 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3992 if (zone
->present_pages
)
3993 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3999 * free_area_init_nodes - Initialise all pg_data_t and zone data
4000 * @max_zone_pfn: an array of max PFNs for each zone
4002 * This will call free_area_init_node() for each active node in the system.
4003 * Using the page ranges provided by add_active_range(), the size of each
4004 * zone in each node and their holes is calculated. If the maximum PFN
4005 * between two adjacent zones match, it is assumed that the zone is empty.
4006 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4007 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4008 * starts where the previous one ended. For example, ZONE_DMA32 starts
4009 * at arch_max_dma_pfn.
4011 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4016 /* Sort early_node_map as initialisation assumes it is sorted */
4019 /* Record where the zone boundaries are */
4020 memset(arch_zone_lowest_possible_pfn
, 0,
4021 sizeof(arch_zone_lowest_possible_pfn
));
4022 memset(arch_zone_highest_possible_pfn
, 0,
4023 sizeof(arch_zone_highest_possible_pfn
));
4024 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4025 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4026 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4027 if (i
== ZONE_MOVABLE
)
4029 arch_zone_lowest_possible_pfn
[i
] =
4030 arch_zone_highest_possible_pfn
[i
-1];
4031 arch_zone_highest_possible_pfn
[i
] =
4032 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4034 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4035 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4037 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4038 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4039 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4041 /* Print out the zone ranges */
4042 printk("Zone PFN ranges:\n");
4043 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4044 if (i
== ZONE_MOVABLE
)
4046 printk(" %-8s %0#10lx -> %0#10lx\n",
4048 arch_zone_lowest_possible_pfn
[i
],
4049 arch_zone_highest_possible_pfn
[i
]);
4052 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4053 printk("Movable zone start PFN for each node\n");
4054 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4055 if (zone_movable_pfn
[i
])
4056 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4059 /* Print out the early_node_map[] */
4060 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4061 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4062 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4063 early_node_map
[i
].start_pfn
,
4064 early_node_map
[i
].end_pfn
);
4066 /* Initialise every node */
4067 mminit_verify_pageflags_layout();
4068 setup_nr_node_ids();
4069 for_each_online_node(nid
) {
4070 pg_data_t
*pgdat
= NODE_DATA(nid
);
4071 free_area_init_node(nid
, NULL
,
4072 find_min_pfn_for_node(nid
), NULL
);
4074 /* Any memory on that node */
4075 if (pgdat
->node_present_pages
)
4076 node_set_state(nid
, N_HIGH_MEMORY
);
4077 check_for_regular_memory(pgdat
);
4081 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4083 unsigned long long coremem
;
4087 coremem
= memparse(p
, &p
);
4088 *core
= coremem
>> PAGE_SHIFT
;
4090 /* Paranoid check that UL is enough for the coremem value */
4091 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4097 * kernelcore=size sets the amount of memory for use for allocations that
4098 * cannot be reclaimed or migrated.
4100 static int __init
cmdline_parse_kernelcore(char *p
)
4102 return cmdline_parse_core(p
, &required_kernelcore
);
4106 * movablecore=size sets the amount of memory for use for allocations that
4107 * can be reclaimed or migrated.
4109 static int __init
cmdline_parse_movablecore(char *p
)
4111 return cmdline_parse_core(p
, &required_movablecore
);
4114 early_param("kernelcore", cmdline_parse_kernelcore
);
4115 early_param("movablecore", cmdline_parse_movablecore
);
4117 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4120 * set_dma_reserve - set the specified number of pages reserved in the first zone
4121 * @new_dma_reserve: The number of pages to mark reserved
4123 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4124 * In the DMA zone, a significant percentage may be consumed by kernel image
4125 * and other unfreeable allocations which can skew the watermarks badly. This
4126 * function may optionally be used to account for unfreeable pages in the
4127 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4128 * smaller per-cpu batchsize.
4130 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4132 dma_reserve
= new_dma_reserve
;
4135 #ifndef CONFIG_NEED_MULTIPLE_NODES
4136 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4137 EXPORT_SYMBOL(contig_page_data
);
4140 void __init
free_area_init(unsigned long *zones_size
)
4142 free_area_init_node(0, zones_size
,
4143 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4146 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4147 unsigned long action
, void *hcpu
)
4149 int cpu
= (unsigned long)hcpu
;
4151 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4155 * Spill the event counters of the dead processor
4156 * into the current processors event counters.
4157 * This artificially elevates the count of the current
4160 vm_events_fold_cpu(cpu
);
4163 * Zero the differential counters of the dead processor
4164 * so that the vm statistics are consistent.
4166 * This is only okay since the processor is dead and cannot
4167 * race with what we are doing.
4169 refresh_cpu_vm_stats(cpu
);
4174 void __init
page_alloc_init(void)
4176 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4180 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4181 * or min_free_kbytes changes.
4183 static void calculate_totalreserve_pages(void)
4185 struct pglist_data
*pgdat
;
4186 unsigned long reserve_pages
= 0;
4187 enum zone_type i
, j
;
4189 for_each_online_pgdat(pgdat
) {
4190 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4191 struct zone
*zone
= pgdat
->node_zones
+ i
;
4192 unsigned long max
= 0;
4194 /* Find valid and maximum lowmem_reserve in the zone */
4195 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4196 if (zone
->lowmem_reserve
[j
] > max
)
4197 max
= zone
->lowmem_reserve
[j
];
4200 /* we treat pages_high as reserved pages. */
4201 max
+= zone
->pages_high
;
4203 if (max
> zone
->present_pages
)
4204 max
= zone
->present_pages
;
4205 reserve_pages
+= max
;
4208 totalreserve_pages
= reserve_pages
;
4212 * setup_per_zone_lowmem_reserve - called whenever
4213 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4214 * has a correct pages reserved value, so an adequate number of
4215 * pages are left in the zone after a successful __alloc_pages().
4217 static void setup_per_zone_lowmem_reserve(void)
4219 struct pglist_data
*pgdat
;
4220 enum zone_type j
, idx
;
4222 for_each_online_pgdat(pgdat
) {
4223 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4224 struct zone
*zone
= pgdat
->node_zones
+ j
;
4225 unsigned long present_pages
= zone
->present_pages
;
4227 zone
->lowmem_reserve
[j
] = 0;
4231 struct zone
*lower_zone
;
4235 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4236 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4238 lower_zone
= pgdat
->node_zones
+ idx
;
4239 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4240 sysctl_lowmem_reserve_ratio
[idx
];
4241 present_pages
+= lower_zone
->present_pages
;
4246 /* update totalreserve_pages */
4247 calculate_totalreserve_pages();
4251 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4253 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4254 * with respect to min_free_kbytes.
4256 void setup_per_zone_pages_min(void)
4258 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4259 unsigned long lowmem_pages
= 0;
4261 unsigned long flags
;
4263 /* Calculate total number of !ZONE_HIGHMEM pages */
4264 for_each_zone(zone
) {
4265 if (!is_highmem(zone
))
4266 lowmem_pages
+= zone
->present_pages
;
4269 for_each_zone(zone
) {
4272 spin_lock_irqsave(&zone
->lock
, flags
);
4273 tmp
= (u64
)pages_min
* zone
->present_pages
;
4274 do_div(tmp
, lowmem_pages
);
4275 if (is_highmem(zone
)) {
4277 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4278 * need highmem pages, so cap pages_min to a small
4281 * The (pages_high-pages_low) and (pages_low-pages_min)
4282 * deltas controls asynch page reclaim, and so should
4283 * not be capped for highmem.
4287 min_pages
= zone
->present_pages
/ 1024;
4288 if (min_pages
< SWAP_CLUSTER_MAX
)
4289 min_pages
= SWAP_CLUSTER_MAX
;
4290 if (min_pages
> 128)
4292 zone
->pages_min
= min_pages
;
4295 * If it's a lowmem zone, reserve a number of pages
4296 * proportionate to the zone's size.
4298 zone
->pages_min
= tmp
;
4301 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4302 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4303 setup_zone_migrate_reserve(zone
);
4304 spin_unlock_irqrestore(&zone
->lock
, flags
);
4307 /* update totalreserve_pages */
4308 calculate_totalreserve_pages();
4312 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4314 * The inactive anon list should be small enough that the VM never has to
4315 * do too much work, but large enough that each inactive page has a chance
4316 * to be referenced again before it is swapped out.
4318 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4319 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4320 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4321 * the anonymous pages are kept on the inactive list.
4324 * memory ratio inactive anon
4325 * -------------------------------------
4334 static void setup_per_zone_inactive_ratio(void)
4338 for_each_zone(zone
) {
4339 unsigned int gb
, ratio
;
4341 /* Zone size in gigabytes */
4342 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4343 ratio
= int_sqrt(10 * gb
);
4347 zone
->inactive_ratio
= ratio
;
4352 * Initialise min_free_kbytes.
4354 * For small machines we want it small (128k min). For large machines
4355 * we want it large (64MB max). But it is not linear, because network
4356 * bandwidth does not increase linearly with machine size. We use
4358 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4359 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4375 static int __init
init_per_zone_pages_min(void)
4377 unsigned long lowmem_kbytes
;
4379 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4381 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4382 if (min_free_kbytes
< 128)
4383 min_free_kbytes
= 128;
4384 if (min_free_kbytes
> 65536)
4385 min_free_kbytes
= 65536;
4386 setup_per_zone_pages_min();
4387 setup_per_zone_lowmem_reserve();
4388 setup_per_zone_inactive_ratio();
4391 module_init(init_per_zone_pages_min
)
4394 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4395 * that we can call two helper functions whenever min_free_kbytes
4398 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4399 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4401 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4403 setup_per_zone_pages_min();
4408 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4409 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4414 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4419 zone
->min_unmapped_pages
= (zone
->present_pages
*
4420 sysctl_min_unmapped_ratio
) / 100;
4424 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4425 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4430 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4435 zone
->min_slab_pages
= (zone
->present_pages
*
4436 sysctl_min_slab_ratio
) / 100;
4442 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4443 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4444 * whenever sysctl_lowmem_reserve_ratio changes.
4446 * The reserve ratio obviously has absolutely no relation with the
4447 * pages_min watermarks. The lowmem reserve ratio can only make sense
4448 * if in function of the boot time zone sizes.
4450 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4451 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4453 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4454 setup_per_zone_lowmem_reserve();
4459 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4460 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4461 * can have before it gets flushed back to buddy allocator.
4464 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4465 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4471 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4472 if (!write
|| (ret
== -EINVAL
))
4474 for_each_zone(zone
) {
4475 for_each_online_cpu(cpu
) {
4477 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4478 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4484 int hashdist
= HASHDIST_DEFAULT
;
4487 static int __init
set_hashdist(char *str
)
4491 hashdist
= simple_strtoul(str
, &str
, 0);
4494 __setup("hashdist=", set_hashdist
);
4498 * allocate a large system hash table from bootmem
4499 * - it is assumed that the hash table must contain an exact power-of-2
4500 * quantity of entries
4501 * - limit is the number of hash buckets, not the total allocation size
4503 void *__init
alloc_large_system_hash(const char *tablename
,
4504 unsigned long bucketsize
,
4505 unsigned long numentries
,
4508 unsigned int *_hash_shift
,
4509 unsigned int *_hash_mask
,
4510 unsigned long limit
)
4512 unsigned long long max
= limit
;
4513 unsigned long log2qty
, size
;
4516 /* allow the kernel cmdline to have a say */
4518 /* round applicable memory size up to nearest megabyte */
4519 numentries
= nr_kernel_pages
;
4520 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4521 numentries
>>= 20 - PAGE_SHIFT
;
4522 numentries
<<= 20 - PAGE_SHIFT
;
4524 /* limit to 1 bucket per 2^scale bytes of low memory */
4525 if (scale
> PAGE_SHIFT
)
4526 numentries
>>= (scale
- PAGE_SHIFT
);
4528 numentries
<<= (PAGE_SHIFT
- scale
);
4530 /* Make sure we've got at least a 0-order allocation.. */
4531 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4532 numentries
= PAGE_SIZE
/ bucketsize
;
4534 numentries
= roundup_pow_of_two(numentries
);
4536 /* limit allocation size to 1/16 total memory by default */
4538 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4539 do_div(max
, bucketsize
);
4542 if (numentries
> max
)
4545 log2qty
= ilog2(numentries
);
4548 size
= bucketsize
<< log2qty
;
4549 if (flags
& HASH_EARLY
)
4550 table
= alloc_bootmem_nopanic(size
);
4552 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4554 unsigned long order
= get_order(size
);
4555 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4557 * If bucketsize is not a power-of-two, we may free
4558 * some pages at the end of hash table.
4561 unsigned long alloc_end
= (unsigned long)table
+
4562 (PAGE_SIZE
<< order
);
4563 unsigned long used
= (unsigned long)table
+
4565 split_page(virt_to_page(table
), order
);
4566 while (used
< alloc_end
) {
4572 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4575 panic("Failed to allocate %s hash table\n", tablename
);
4577 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4580 ilog2(size
) - PAGE_SHIFT
,
4584 *_hash_shift
= log2qty
;
4586 *_hash_mask
= (1 << log2qty
) - 1;
4591 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4592 struct page
*pfn_to_page(unsigned long pfn
)
4594 return __pfn_to_page(pfn
);
4596 unsigned long page_to_pfn(struct page
*page
)
4598 return __page_to_pfn(page
);
4600 EXPORT_SYMBOL(pfn_to_page
);
4601 EXPORT_SYMBOL(page_to_pfn
);
4602 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4604 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4605 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4608 #ifdef CONFIG_SPARSEMEM
4609 return __pfn_to_section(pfn
)->pageblock_flags
;
4611 return zone
->pageblock_flags
;
4612 #endif /* CONFIG_SPARSEMEM */
4615 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4617 #ifdef CONFIG_SPARSEMEM
4618 pfn
&= (PAGES_PER_SECTION
-1);
4619 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4621 pfn
= pfn
- zone
->zone_start_pfn
;
4622 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4623 #endif /* CONFIG_SPARSEMEM */
4627 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4628 * @page: The page within the block of interest
4629 * @start_bitidx: The first bit of interest to retrieve
4630 * @end_bitidx: The last bit of interest
4631 * returns pageblock_bits flags
4633 unsigned long get_pageblock_flags_group(struct page
*page
,
4634 int start_bitidx
, int end_bitidx
)
4637 unsigned long *bitmap
;
4638 unsigned long pfn
, bitidx
;
4639 unsigned long flags
= 0;
4640 unsigned long value
= 1;
4642 zone
= page_zone(page
);
4643 pfn
= page_to_pfn(page
);
4644 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4645 bitidx
= pfn_to_bitidx(zone
, pfn
);
4647 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4648 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4655 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4656 * @page: The page within the block of interest
4657 * @start_bitidx: The first bit of interest
4658 * @end_bitidx: The last bit of interest
4659 * @flags: The flags to set
4661 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4662 int start_bitidx
, int end_bitidx
)
4665 unsigned long *bitmap
;
4666 unsigned long pfn
, bitidx
;
4667 unsigned long value
= 1;
4669 zone
= page_zone(page
);
4670 pfn
= page_to_pfn(page
);
4671 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4672 bitidx
= pfn_to_bitidx(zone
, pfn
);
4673 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4674 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4676 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4678 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4680 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4684 * This is designed as sub function...plz see page_isolation.c also.
4685 * set/clear page block's type to be ISOLATE.
4686 * page allocater never alloc memory from ISOLATE block.
4689 int set_migratetype_isolate(struct page
*page
)
4692 unsigned long flags
;
4695 zone
= page_zone(page
);
4696 spin_lock_irqsave(&zone
->lock
, flags
);
4698 * In future, more migrate types will be able to be isolation target.
4700 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4702 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4703 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4706 spin_unlock_irqrestore(&zone
->lock
, flags
);
4712 void unset_migratetype_isolate(struct page
*page
)
4715 unsigned long flags
;
4716 zone
= page_zone(page
);
4717 spin_lock_irqsave(&zone
->lock
, flags
);
4718 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4720 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4721 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4723 spin_unlock_irqrestore(&zone
->lock
, flags
);
4726 #ifdef CONFIG_MEMORY_HOTREMOVE
4728 * All pages in the range must be isolated before calling this.
4731 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4737 unsigned long flags
;
4738 /* find the first valid pfn */
4739 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4744 zone
= page_zone(pfn_to_page(pfn
));
4745 spin_lock_irqsave(&zone
->lock
, flags
);
4747 while (pfn
< end_pfn
) {
4748 if (!pfn_valid(pfn
)) {
4752 page
= pfn_to_page(pfn
);
4753 BUG_ON(page_count(page
));
4754 BUG_ON(!PageBuddy(page
));
4755 order
= page_order(page
);
4756 #ifdef CONFIG_DEBUG_VM
4757 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4758 pfn
, 1 << order
, end_pfn
);
4760 list_del(&page
->lru
);
4761 rmv_page_order(page
);
4762 zone
->free_area
[order
].nr_free
--;
4763 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4765 for (i
= 0; i
< (1 << order
); i
++)
4766 SetPageReserved((page
+i
));
4767 pfn
+= (1 << order
);
4769 spin_unlock_irqrestore(&zone
->lock
, flags
);