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 "BUG: Bad page state: %lu messages suppressed\n",
248 resume
= jiffies
+ 60 * HZ
;
250 printk(KERN_ALERT
"BUG: 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
);
259 /* Leave bad fields for debug, except PageBuddy could make trouble */
260 __ClearPageBuddy(page
);
261 add_taint(TAINT_BAD_PAGE
);
265 * Higher-order pages are called "compound pages". They are structured thusly:
267 * The first PAGE_SIZE page is called the "head page".
269 * The remaining PAGE_SIZE pages are called "tail pages".
271 * All pages have PG_compound set. All pages have their ->private pointing at
272 * the head page (even the head page has this).
274 * The first tail page's ->lru.next holds the address of the compound page's
275 * put_page() function. Its ->lru.prev holds the order of allocation.
276 * This usage means that zero-order pages may not be compound.
279 static void free_compound_page(struct page
*page
)
281 __free_pages_ok(page
, compound_order(page
));
284 void prep_compound_page(struct page
*page
, unsigned long order
)
287 int nr_pages
= 1 << order
;
289 set_compound_page_dtor(page
, free_compound_page
);
290 set_compound_order(page
, order
);
292 for (i
= 1; i
< nr_pages
; i
++) {
293 struct page
*p
= page
+ i
;
296 p
->first_page
= page
;
300 #ifdef CONFIG_HUGETLBFS
301 void prep_compound_gigantic_page(struct page
*page
, unsigned long order
)
304 int nr_pages
= 1 << order
;
305 struct page
*p
= page
+ 1;
307 set_compound_page_dtor(page
, free_compound_page
);
308 set_compound_order(page
, order
);
310 for (i
= 1; i
< nr_pages
; i
++, p
= mem_map_next(p
, page
, i
)) {
312 p
->first_page
= page
;
317 static int destroy_compound_page(struct page
*page
, unsigned long order
)
320 int nr_pages
= 1 << order
;
323 if (unlikely(compound_order(page
) != order
) ||
324 unlikely(!PageHead(page
))) {
329 __ClearPageHead(page
);
331 for (i
= 1; i
< nr_pages
; i
++) {
332 struct page
*p
= page
+ i
;
334 if (unlikely(!PageTail(p
) | (p
->first_page
!= page
))) {
344 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
349 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
350 * and __GFP_HIGHMEM from hard or soft interrupt context.
352 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
353 for (i
= 0; i
< (1 << order
); i
++)
354 clear_highpage(page
+ i
);
357 static inline void set_page_order(struct page
*page
, int order
)
359 set_page_private(page
, order
);
360 __SetPageBuddy(page
);
363 static inline void rmv_page_order(struct page
*page
)
365 __ClearPageBuddy(page
);
366 set_page_private(page
, 0);
370 * Locate the struct page for both the matching buddy in our
371 * pair (buddy1) and the combined O(n+1) page they form (page).
373 * 1) Any buddy B1 will have an order O twin B2 which satisfies
374 * the following equation:
376 * For example, if the starting buddy (buddy2) is #8 its order
378 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
380 * 2) Any buddy B will have an order O+1 parent P which
381 * satisfies the following equation:
384 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
386 static inline struct page
*
387 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
389 unsigned long buddy_idx
= page_idx
^ (1 << order
);
391 return page
+ (buddy_idx
- page_idx
);
394 static inline unsigned long
395 __find_combined_index(unsigned long page_idx
, unsigned int order
)
397 return (page_idx
& ~(1 << order
));
401 * This function checks whether a page is free && is the buddy
402 * we can do coalesce a page and its buddy if
403 * (a) the buddy is not in a hole &&
404 * (b) the buddy is in the buddy system &&
405 * (c) a page and its buddy have the same order &&
406 * (d) a page and its buddy are in the same zone.
408 * For recording whether a page is in the buddy system, we use PG_buddy.
409 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
411 * For recording page's order, we use page_private(page).
413 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
416 if (!pfn_valid_within(page_to_pfn(buddy
)))
419 if (page_zone_id(page
) != page_zone_id(buddy
))
422 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
423 BUG_ON(page_count(buddy
) != 0);
430 * Freeing function for a buddy system allocator.
432 * The concept of a buddy system is to maintain direct-mapped table
433 * (containing bit values) for memory blocks of various "orders".
434 * The bottom level table contains the map for the smallest allocatable
435 * units of memory (here, pages), and each level above it describes
436 * pairs of units from the levels below, hence, "buddies".
437 * At a high level, all that happens here is marking the table entry
438 * at the bottom level available, and propagating the changes upward
439 * as necessary, plus some accounting needed to play nicely with other
440 * parts of the VM system.
441 * At each level, we keep a list of pages, which are heads of continuous
442 * free pages of length of (1 << order) and marked with PG_buddy. Page's
443 * order is recorded in page_private(page) field.
444 * So when we are allocating or freeing one, we can derive the state of the
445 * other. That is, if we allocate a small block, and both were
446 * free, the remainder of the region must be split into blocks.
447 * If a block is freed, and its buddy is also free, then this
448 * triggers coalescing into a block of larger size.
453 static inline void __free_one_page(struct page
*page
,
454 struct zone
*zone
, unsigned int order
)
456 unsigned long page_idx
;
457 int order_size
= 1 << order
;
458 int migratetype
= get_pageblock_migratetype(page
);
460 if (unlikely(PageCompound(page
)))
461 if (unlikely(destroy_compound_page(page
, order
)))
464 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
466 VM_BUG_ON(page_idx
& (order_size
- 1));
467 VM_BUG_ON(bad_range(zone
, page
));
469 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
470 while (order
< MAX_ORDER
-1) {
471 unsigned long combined_idx
;
474 buddy
= __page_find_buddy(page
, page_idx
, order
);
475 if (!page_is_buddy(page
, buddy
, order
))
478 /* Our buddy is free, merge with it and move up one order. */
479 list_del(&buddy
->lru
);
480 zone
->free_area
[order
].nr_free
--;
481 rmv_page_order(buddy
);
482 combined_idx
= __find_combined_index(page_idx
, order
);
483 page
= page
+ (combined_idx
- page_idx
);
484 page_idx
= combined_idx
;
487 set_page_order(page
, order
);
489 &zone
->free_area
[order
].free_list
[migratetype
]);
490 zone
->free_area
[order
].nr_free
++;
493 static inline int free_pages_check(struct page
*page
)
495 free_page_mlock(page
);
496 if (unlikely(page_mapcount(page
) |
497 (page
->mapping
!= NULL
) |
498 (page_count(page
) != 0) |
499 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
503 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
504 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
509 * Frees a list of pages.
510 * Assumes all pages on list are in same zone, and of same order.
511 * count is the number of pages to free.
513 * If the zone was previously in an "all pages pinned" state then look to
514 * see if this freeing clears that state.
516 * And clear the zone's pages_scanned counter, to hold off the "all pages are
517 * pinned" detection logic.
519 static void free_pages_bulk(struct zone
*zone
, int count
,
520 struct list_head
*list
, int order
)
522 spin_lock(&zone
->lock
);
523 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
524 zone
->pages_scanned
= 0;
528 VM_BUG_ON(list_empty(list
));
529 page
= list_entry(list
->prev
, struct page
, lru
);
530 /* have to delete it as __free_one_page list manipulates */
531 list_del(&page
->lru
);
532 __free_one_page(page
, zone
, order
);
534 spin_unlock(&zone
->lock
);
537 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
539 spin_lock(&zone
->lock
);
540 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
541 zone
->pages_scanned
= 0;
542 __free_one_page(page
, zone
, order
);
543 spin_unlock(&zone
->lock
);
546 static void __free_pages_ok(struct page
*page
, unsigned int order
)
552 for (i
= 0 ; i
< (1 << order
) ; ++i
)
553 bad
+= free_pages_check(page
+ i
);
557 if (!PageHighMem(page
)) {
558 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
559 debug_check_no_obj_freed(page_address(page
),
562 arch_free_page(page
, order
);
563 kernel_map_pages(page
, 1 << order
, 0);
565 local_irq_save(flags
);
566 __count_vm_events(PGFREE
, 1 << order
);
567 free_one_page(page_zone(page
), page
, order
);
568 local_irq_restore(flags
);
572 * permit the bootmem allocator to evade page validation on high-order frees
574 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
577 __ClearPageReserved(page
);
578 set_page_count(page
, 0);
579 set_page_refcounted(page
);
585 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
586 struct page
*p
= &page
[loop
];
588 if (loop
+ 1 < BITS_PER_LONG
)
590 __ClearPageReserved(p
);
591 set_page_count(p
, 0);
594 set_page_refcounted(page
);
595 __free_pages(page
, order
);
601 * The order of subdivision here is critical for the IO subsystem.
602 * Please do not alter this order without good reasons and regression
603 * testing. Specifically, as large blocks of memory are subdivided,
604 * the order in which smaller blocks are delivered depends on the order
605 * they're subdivided in this function. This is the primary factor
606 * influencing the order in which pages are delivered to the IO
607 * subsystem according to empirical testing, and this is also justified
608 * by considering the behavior of a buddy system containing a single
609 * large block of memory acted on by a series of small allocations.
610 * This behavior is a critical factor in sglist merging's success.
614 static inline void expand(struct zone
*zone
, struct page
*page
,
615 int low
, int high
, struct free_area
*area
,
618 unsigned long size
= 1 << high
;
624 VM_BUG_ON(bad_range(zone
, &page
[size
]));
625 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
627 set_page_order(&page
[size
], high
);
632 * This page is about to be returned from the page allocator
634 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
636 if (unlikely(page_mapcount(page
) |
637 (page
->mapping
!= NULL
) |
638 (page_count(page
) != 0) |
639 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
644 set_page_private(page
, 0);
645 set_page_refcounted(page
);
647 arch_alloc_page(page
, order
);
648 kernel_map_pages(page
, 1 << order
, 1);
650 if (gfp_flags
& __GFP_ZERO
)
651 prep_zero_page(page
, order
, gfp_flags
);
653 if (order
&& (gfp_flags
& __GFP_COMP
))
654 prep_compound_page(page
, order
);
660 * Go through the free lists for the given migratetype and remove
661 * the smallest available page from the freelists
663 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
666 unsigned int current_order
;
667 struct free_area
* area
;
670 /* Find a page of the appropriate size in the preferred list */
671 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
672 area
= &(zone
->free_area
[current_order
]);
673 if (list_empty(&area
->free_list
[migratetype
]))
676 page
= list_entry(area
->free_list
[migratetype
].next
,
678 list_del(&page
->lru
);
679 rmv_page_order(page
);
681 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
682 expand(zone
, page
, order
, current_order
, area
, migratetype
);
691 * This array describes the order lists are fallen back to when
692 * the free lists for the desirable migrate type are depleted
694 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
695 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
696 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
697 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
698 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
702 * Move the free pages in a range to the free lists of the requested type.
703 * Note that start_page and end_pages are not aligned on a pageblock
704 * boundary. If alignment is required, use move_freepages_block()
706 static int move_freepages(struct zone
*zone
,
707 struct page
*start_page
, struct page
*end_page
,
714 #ifndef CONFIG_HOLES_IN_ZONE
716 * page_zone is not safe to call in this context when
717 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
718 * anyway as we check zone boundaries in move_freepages_block().
719 * Remove at a later date when no bug reports exist related to
720 * grouping pages by mobility
722 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
725 for (page
= start_page
; page
<= end_page
;) {
726 /* Make sure we are not inadvertently changing nodes */
727 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
729 if (!pfn_valid_within(page_to_pfn(page
))) {
734 if (!PageBuddy(page
)) {
739 order
= page_order(page
);
740 list_del(&page
->lru
);
742 &zone
->free_area
[order
].free_list
[migratetype
]);
744 pages_moved
+= 1 << order
;
750 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
753 unsigned long start_pfn
, end_pfn
;
754 struct page
*start_page
, *end_page
;
756 start_pfn
= page_to_pfn(page
);
757 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
758 start_page
= pfn_to_page(start_pfn
);
759 end_page
= start_page
+ pageblock_nr_pages
- 1;
760 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
762 /* Do not cross zone boundaries */
763 if (start_pfn
< zone
->zone_start_pfn
)
765 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
768 return move_freepages(zone
, start_page
, end_page
, migratetype
);
771 /* Remove an element from the buddy allocator from the fallback list */
772 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
773 int start_migratetype
)
775 struct free_area
* area
;
780 /* Find the largest possible block of pages in the other list */
781 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
783 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
784 migratetype
= fallbacks
[start_migratetype
][i
];
786 /* MIGRATE_RESERVE handled later if necessary */
787 if (migratetype
== MIGRATE_RESERVE
)
790 area
= &(zone
->free_area
[current_order
]);
791 if (list_empty(&area
->free_list
[migratetype
]))
794 page
= list_entry(area
->free_list
[migratetype
].next
,
799 * If breaking a large block of pages, move all free
800 * pages to the preferred allocation list. If falling
801 * back for a reclaimable kernel allocation, be more
802 * agressive about taking ownership of free pages
804 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
805 start_migratetype
== MIGRATE_RECLAIMABLE
) {
807 pages
= move_freepages_block(zone
, page
,
810 /* Claim the whole block if over half of it is free */
811 if (pages
>= (1 << (pageblock_order
-1)))
812 set_pageblock_migratetype(page
,
815 migratetype
= start_migratetype
;
818 /* Remove the page from the freelists */
819 list_del(&page
->lru
);
820 rmv_page_order(page
);
821 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
824 if (current_order
== pageblock_order
)
825 set_pageblock_migratetype(page
,
828 expand(zone
, page
, order
, current_order
, area
, migratetype
);
833 /* Use MIGRATE_RESERVE rather than fail an allocation */
834 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
838 * Do the hard work of removing an element from the buddy allocator.
839 * Call me with the zone->lock already held.
841 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
846 page
= __rmqueue_smallest(zone
, order
, migratetype
);
849 page
= __rmqueue_fallback(zone
, order
, migratetype
);
855 * Obtain a specified number of elements from the buddy allocator, all under
856 * a single hold of the lock, for efficiency. Add them to the supplied list.
857 * Returns the number of new pages which were placed at *list.
859 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
860 unsigned long count
, struct list_head
*list
,
865 spin_lock(&zone
->lock
);
866 for (i
= 0; i
< count
; ++i
) {
867 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
868 if (unlikely(page
== NULL
))
872 * Split buddy pages returned by expand() are received here
873 * in physical page order. The page is added to the callers and
874 * list and the list head then moves forward. From the callers
875 * perspective, the linked list is ordered by page number in
876 * some conditions. This is useful for IO devices that can
877 * merge IO requests if the physical pages are ordered
880 list_add(&page
->lru
, list
);
881 set_page_private(page
, migratetype
);
884 spin_unlock(&zone
->lock
);
890 * Called from the vmstat counter updater to drain pagesets of this
891 * currently executing processor on remote nodes after they have
894 * Note that this function must be called with the thread pinned to
895 * a single processor.
897 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
902 local_irq_save(flags
);
903 if (pcp
->count
>= pcp
->batch
)
904 to_drain
= pcp
->batch
;
906 to_drain
= pcp
->count
;
907 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
908 pcp
->count
-= to_drain
;
909 local_irq_restore(flags
);
914 * Drain pages of the indicated processor.
916 * The processor must either be the current processor and the
917 * thread pinned to the current processor or a processor that
920 static void drain_pages(unsigned int cpu
)
925 for_each_populated_zone(zone
) {
926 struct per_cpu_pageset
*pset
;
927 struct per_cpu_pages
*pcp
;
929 pset
= zone_pcp(zone
, cpu
);
932 local_irq_save(flags
);
933 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
935 local_irq_restore(flags
);
940 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
942 void drain_local_pages(void *arg
)
944 drain_pages(smp_processor_id());
948 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
950 void drain_all_pages(void)
952 on_each_cpu(drain_local_pages
, NULL
, 1);
955 #ifdef CONFIG_HIBERNATION
957 void mark_free_pages(struct zone
*zone
)
959 unsigned long pfn
, max_zone_pfn
;
962 struct list_head
*curr
;
964 if (!zone
->spanned_pages
)
967 spin_lock_irqsave(&zone
->lock
, flags
);
969 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
970 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
971 if (pfn_valid(pfn
)) {
972 struct page
*page
= pfn_to_page(pfn
);
974 if (!swsusp_page_is_forbidden(page
))
975 swsusp_unset_page_free(page
);
978 for_each_migratetype_order(order
, t
) {
979 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
982 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
983 for (i
= 0; i
< (1UL << order
); i
++)
984 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
987 spin_unlock_irqrestore(&zone
->lock
, flags
);
989 #endif /* CONFIG_PM */
992 * Free a 0-order page
994 static void free_hot_cold_page(struct page
*page
, int cold
)
996 struct zone
*zone
= page_zone(page
);
997 struct per_cpu_pages
*pcp
;
1001 page
->mapping
= NULL
;
1002 if (free_pages_check(page
))
1005 if (!PageHighMem(page
)) {
1006 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1007 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1009 arch_free_page(page
, 0);
1010 kernel_map_pages(page
, 1, 0);
1012 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1013 local_irq_save(flags
);
1014 __count_vm_event(PGFREE
);
1016 list_add_tail(&page
->lru
, &pcp
->list
);
1018 list_add(&page
->lru
, &pcp
->list
);
1019 set_page_private(page
, get_pageblock_migratetype(page
));
1021 if (pcp
->count
>= pcp
->high
) {
1022 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1023 pcp
->count
-= pcp
->batch
;
1025 local_irq_restore(flags
);
1029 void free_hot_page(struct page
*page
)
1031 free_hot_cold_page(page
, 0);
1034 void free_cold_page(struct page
*page
)
1036 free_hot_cold_page(page
, 1);
1040 * split_page takes a non-compound higher-order page, and splits it into
1041 * n (1<<order) sub-pages: page[0..n]
1042 * Each sub-page must be freed individually.
1044 * Note: this is probably too low level an operation for use in drivers.
1045 * Please consult with lkml before using this in your driver.
1047 void split_page(struct page
*page
, unsigned int order
)
1051 VM_BUG_ON(PageCompound(page
));
1052 VM_BUG_ON(!page_count(page
));
1053 for (i
= 1; i
< (1 << order
); i
++)
1054 set_page_refcounted(page
+ i
);
1058 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1059 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1062 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1063 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1065 unsigned long flags
;
1067 int cold
= !!(gfp_flags
& __GFP_COLD
);
1069 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1073 if (likely(order
== 0)) {
1074 struct per_cpu_pages
*pcp
;
1076 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1077 local_irq_save(flags
);
1079 pcp
->count
= rmqueue_bulk(zone
, 0,
1080 pcp
->batch
, &pcp
->list
, migratetype
);
1081 if (unlikely(!pcp
->count
))
1085 /* Find a page of the appropriate migrate type */
1087 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1088 if (page_private(page
) == migratetype
)
1091 list_for_each_entry(page
, &pcp
->list
, lru
)
1092 if (page_private(page
) == migratetype
)
1096 /* Allocate more to the pcp list if necessary */
1097 if (unlikely(&page
->lru
== &pcp
->list
)) {
1098 pcp
->count
+= rmqueue_bulk(zone
, 0,
1099 pcp
->batch
, &pcp
->list
, migratetype
);
1100 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1103 list_del(&page
->lru
);
1106 spin_lock_irqsave(&zone
->lock
, flags
);
1107 page
= __rmqueue(zone
, order
, migratetype
);
1108 spin_unlock(&zone
->lock
);
1113 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1114 zone_statistics(preferred_zone
, zone
);
1115 local_irq_restore(flags
);
1118 VM_BUG_ON(bad_range(zone
, page
));
1119 if (prep_new_page(page
, order
, gfp_flags
))
1124 local_irq_restore(flags
);
1129 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1130 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1131 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1132 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1133 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1134 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1135 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1137 #ifdef CONFIG_FAIL_PAGE_ALLOC
1139 static struct fail_page_alloc_attr
{
1140 struct fault_attr attr
;
1142 u32 ignore_gfp_highmem
;
1143 u32 ignore_gfp_wait
;
1146 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1148 struct dentry
*ignore_gfp_highmem_file
;
1149 struct dentry
*ignore_gfp_wait_file
;
1150 struct dentry
*min_order_file
;
1152 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1154 } fail_page_alloc
= {
1155 .attr
= FAULT_ATTR_INITIALIZER
,
1156 .ignore_gfp_wait
= 1,
1157 .ignore_gfp_highmem
= 1,
1161 static int __init
setup_fail_page_alloc(char *str
)
1163 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1165 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1167 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1169 if (order
< fail_page_alloc
.min_order
)
1171 if (gfp_mask
& __GFP_NOFAIL
)
1173 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1175 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1178 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1181 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1183 static int __init
fail_page_alloc_debugfs(void)
1185 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1189 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1193 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1195 fail_page_alloc
.ignore_gfp_wait_file
=
1196 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1197 &fail_page_alloc
.ignore_gfp_wait
);
1199 fail_page_alloc
.ignore_gfp_highmem_file
=
1200 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1201 &fail_page_alloc
.ignore_gfp_highmem
);
1202 fail_page_alloc
.min_order_file
=
1203 debugfs_create_u32("min-order", mode
, dir
,
1204 &fail_page_alloc
.min_order
);
1206 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1207 !fail_page_alloc
.ignore_gfp_highmem_file
||
1208 !fail_page_alloc
.min_order_file
) {
1210 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1211 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1212 debugfs_remove(fail_page_alloc
.min_order_file
);
1213 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1219 late_initcall(fail_page_alloc_debugfs
);
1221 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1223 #else /* CONFIG_FAIL_PAGE_ALLOC */
1225 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1230 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1233 * Return 1 if free pages are above 'mark'. This takes into account the order
1234 * of the allocation.
1236 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1237 int classzone_idx
, int alloc_flags
)
1239 /* free_pages my go negative - that's OK */
1241 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1244 if (alloc_flags
& ALLOC_HIGH
)
1246 if (alloc_flags
& ALLOC_HARDER
)
1249 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1251 for (o
= 0; o
< order
; o
++) {
1252 /* At the next order, this order's pages become unavailable */
1253 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1255 /* Require fewer higher order pages to be free */
1258 if (free_pages
<= min
)
1266 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1267 * skip over zones that are not allowed by the cpuset, or that have
1268 * been recently (in last second) found to be nearly full. See further
1269 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1270 * that have to skip over a lot of full or unallowed zones.
1272 * If the zonelist cache is present in the passed in zonelist, then
1273 * returns a pointer to the allowed node mask (either the current
1274 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1276 * If the zonelist cache is not available for this zonelist, does
1277 * nothing and returns NULL.
1279 * If the fullzones BITMAP in the zonelist cache is stale (more than
1280 * a second since last zap'd) then we zap it out (clear its bits.)
1282 * We hold off even calling zlc_setup, until after we've checked the
1283 * first zone in the zonelist, on the theory that most allocations will
1284 * be satisfied from that first zone, so best to examine that zone as
1285 * quickly as we can.
1287 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1289 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1290 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1292 zlc
= zonelist
->zlcache_ptr
;
1296 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1297 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1298 zlc
->last_full_zap
= jiffies
;
1301 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1302 &cpuset_current_mems_allowed
:
1303 &node_states
[N_HIGH_MEMORY
];
1304 return allowednodes
;
1308 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1309 * if it is worth looking at further for free memory:
1310 * 1) Check that the zone isn't thought to be full (doesn't have its
1311 * bit set in the zonelist_cache fullzones BITMAP).
1312 * 2) Check that the zones node (obtained from the zonelist_cache
1313 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1314 * Return true (non-zero) if zone is worth looking at further, or
1315 * else return false (zero) if it is not.
1317 * This check -ignores- the distinction between various watermarks,
1318 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1319 * found to be full for any variation of these watermarks, it will
1320 * be considered full for up to one second by all requests, unless
1321 * we are so low on memory on all allowed nodes that we are forced
1322 * into the second scan of the zonelist.
1324 * In the second scan we ignore this zonelist cache and exactly
1325 * apply the watermarks to all zones, even it is slower to do so.
1326 * We are low on memory in the second scan, and should leave no stone
1327 * unturned looking for a free page.
1329 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1330 nodemask_t
*allowednodes
)
1332 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1333 int i
; /* index of *z in zonelist zones */
1334 int n
; /* node that zone *z is on */
1336 zlc
= zonelist
->zlcache_ptr
;
1340 i
= z
- zonelist
->_zonerefs
;
1343 /* This zone is worth trying if it is allowed but not full */
1344 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1348 * Given 'z' scanning a zonelist, set the corresponding bit in
1349 * zlc->fullzones, so that subsequent attempts to allocate a page
1350 * from that zone don't waste time re-examining it.
1352 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1354 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1355 int i
; /* index of *z in zonelist zones */
1357 zlc
= zonelist
->zlcache_ptr
;
1361 i
= z
- zonelist
->_zonerefs
;
1363 set_bit(i
, zlc
->fullzones
);
1366 #else /* CONFIG_NUMA */
1368 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1373 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1374 nodemask_t
*allowednodes
)
1379 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1382 #endif /* CONFIG_NUMA */
1385 * get_page_from_freelist goes through the zonelist trying to allocate
1388 static struct page
*
1389 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1390 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1393 struct page
*page
= NULL
;
1395 struct zone
*zone
, *preferred_zone
;
1396 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1397 int zlc_active
= 0; /* set if using zonelist_cache */
1398 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1400 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1402 if (!preferred_zone
)
1405 classzone_idx
= zone_idx(preferred_zone
);
1409 * Scan zonelist, looking for a zone with enough free.
1410 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1412 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1413 high_zoneidx
, nodemask
) {
1414 if (NUMA_BUILD
&& zlc_active
&&
1415 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1417 if ((alloc_flags
& ALLOC_CPUSET
) &&
1418 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1421 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1423 if (alloc_flags
& ALLOC_WMARK_MIN
)
1424 mark
= zone
->pages_min
;
1425 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1426 mark
= zone
->pages_low
;
1428 mark
= zone
->pages_high
;
1429 if (!zone_watermark_ok(zone
, order
, mark
,
1430 classzone_idx
, alloc_flags
)) {
1431 if (!zone_reclaim_mode
||
1432 !zone_reclaim(zone
, gfp_mask
, order
))
1433 goto this_zone_full
;
1437 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1442 zlc_mark_zone_full(zonelist
, z
);
1444 if (NUMA_BUILD
&& !did_zlc_setup
) {
1445 /* we do zlc_setup after the first zone is tried */
1446 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1452 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1453 /* Disable zlc cache for second zonelist scan */
1461 * This is the 'heart' of the zoned buddy allocator.
1464 __alloc_pages_internal(gfp_t gfp_mask
, unsigned int order
,
1465 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1467 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1468 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1472 struct reclaim_state reclaim_state
;
1473 struct task_struct
*p
= current
;
1476 unsigned long did_some_progress
;
1477 unsigned long pages_reclaimed
= 0;
1479 lockdep_trace_alloc(gfp_mask
);
1481 might_sleep_if(wait
);
1483 if (should_fail_alloc_page(gfp_mask
, order
))
1487 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1489 if (unlikely(!z
->zone
)) {
1491 * Happens if we have an empty zonelist as a result of
1492 * GFP_THISNODE being used on a memoryless node
1497 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1498 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1503 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1504 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1505 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1506 * using a larger set of nodes after it has established that the
1507 * allowed per node queues are empty and that nodes are
1510 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1513 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1514 wakeup_kswapd(zone
, order
);
1517 * OK, we're below the kswapd watermark and have kicked background
1518 * reclaim. Now things get more complex, so set up alloc_flags according
1519 * to how we want to proceed.
1521 * The caller may dip into page reserves a bit more if the caller
1522 * cannot run direct reclaim, or if the caller has realtime scheduling
1523 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1524 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1526 alloc_flags
= ALLOC_WMARK_MIN
;
1527 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1528 alloc_flags
|= ALLOC_HARDER
;
1529 if (gfp_mask
& __GFP_HIGH
)
1530 alloc_flags
|= ALLOC_HIGH
;
1532 alloc_flags
|= ALLOC_CPUSET
;
1535 * Go through the zonelist again. Let __GFP_HIGH and allocations
1536 * coming from realtime tasks go deeper into reserves.
1538 * This is the last chance, in general, before the goto nopage.
1539 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1540 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1542 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1543 high_zoneidx
, alloc_flags
);
1547 /* This allocation should allow future memory freeing. */
1550 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1551 && !in_interrupt()) {
1552 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1554 /* go through the zonelist yet again, ignoring mins */
1555 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1556 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1559 if (gfp_mask
& __GFP_NOFAIL
) {
1560 congestion_wait(WRITE
, HZ
/50);
1567 /* Atomic allocations - we can't balance anything */
1573 /* We now go into synchronous reclaim */
1574 cpuset_memory_pressure_bump();
1576 * The task's cpuset might have expanded its set of allowable nodes
1578 cpuset_update_task_memory_state();
1579 p
->flags
|= PF_MEMALLOC
;
1581 lockdep_set_current_reclaim_state(gfp_mask
);
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 lockdep_clear_current_reclaim_state();
1589 p
->flags
&= ~PF_MEMALLOC
;
1596 if (likely(did_some_progress
)) {
1597 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1598 zonelist
, high_zoneidx
, alloc_flags
);
1601 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1602 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1603 schedule_timeout_uninterruptible(1);
1608 * Go through the zonelist yet one more time, keep
1609 * very high watermark here, this is only to catch
1610 * a parallel oom killing, we must fail if we're still
1611 * under heavy pressure.
1613 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1614 order
, zonelist
, high_zoneidx
,
1615 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1617 clear_zonelist_oom(zonelist
, gfp_mask
);
1621 /* The OOM killer will not help higher order allocs so fail */
1622 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1623 clear_zonelist_oom(zonelist
, gfp_mask
);
1627 out_of_memory(zonelist
, gfp_mask
, order
);
1628 clear_zonelist_oom(zonelist
, gfp_mask
);
1633 * Don't let big-order allocations loop unless the caller explicitly
1634 * requests that. Wait for some write requests to complete then retry.
1636 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1637 * means __GFP_NOFAIL, but that may not be true in other
1640 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1641 * specified, then we retry until we no longer reclaim any pages
1642 * (above), or we've reclaimed an order of pages at least as
1643 * large as the allocation's order. In both cases, if the
1644 * allocation still fails, we stop retrying.
1646 pages_reclaimed
+= did_some_progress
;
1648 if (!(gfp_mask
& __GFP_NORETRY
)) {
1649 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1652 if (gfp_mask
& __GFP_REPEAT
&&
1653 pages_reclaimed
< (1 << order
))
1656 if (gfp_mask
& __GFP_NOFAIL
)
1660 congestion_wait(WRITE
, HZ
/50);
1665 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1666 printk(KERN_WARNING
"%s: page allocation failure."
1667 " order:%d, mode:0x%x\n",
1668 p
->comm
, order
, gfp_mask
);
1675 EXPORT_SYMBOL(__alloc_pages_internal
);
1678 * Common helper functions.
1680 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1683 page
= alloc_pages(gfp_mask
, order
);
1686 return (unsigned long) page_address(page
);
1689 EXPORT_SYMBOL(__get_free_pages
);
1691 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1696 * get_zeroed_page() returns a 32-bit address, which cannot represent
1699 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1701 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1703 return (unsigned long) page_address(page
);
1707 EXPORT_SYMBOL(get_zeroed_page
);
1709 void __pagevec_free(struct pagevec
*pvec
)
1711 int i
= pagevec_count(pvec
);
1714 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1717 void __free_pages(struct page
*page
, unsigned int order
)
1719 if (put_page_testzero(page
)) {
1721 free_hot_page(page
);
1723 __free_pages_ok(page
, order
);
1727 EXPORT_SYMBOL(__free_pages
);
1729 void free_pages(unsigned long addr
, unsigned int order
)
1732 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1733 __free_pages(virt_to_page((void *)addr
), order
);
1737 EXPORT_SYMBOL(free_pages
);
1740 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1741 * @size: the number of bytes to allocate
1742 * @gfp_mask: GFP flags for the allocation
1744 * This function is similar to alloc_pages(), except that it allocates the
1745 * minimum number of pages to satisfy the request. alloc_pages() can only
1746 * allocate memory in power-of-two pages.
1748 * This function is also limited by MAX_ORDER.
1750 * Memory allocated by this function must be released by free_pages_exact().
1752 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1754 unsigned int order
= get_order(size
);
1757 addr
= __get_free_pages(gfp_mask
, order
);
1759 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1760 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1762 split_page(virt_to_page(addr
), order
);
1763 while (used
< alloc_end
) {
1769 return (void *)addr
;
1771 EXPORT_SYMBOL(alloc_pages_exact
);
1774 * free_pages_exact - release memory allocated via alloc_pages_exact()
1775 * @virt: the value returned by alloc_pages_exact.
1776 * @size: size of allocation, same value as passed to alloc_pages_exact().
1778 * Release the memory allocated by a previous call to alloc_pages_exact.
1780 void free_pages_exact(void *virt
, size_t size
)
1782 unsigned long addr
= (unsigned long)virt
;
1783 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1785 while (addr
< end
) {
1790 EXPORT_SYMBOL(free_pages_exact
);
1792 static unsigned int nr_free_zone_pages(int offset
)
1797 /* Just pick one node, since fallback list is circular */
1798 unsigned int sum
= 0;
1800 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1802 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1803 unsigned long size
= zone
->present_pages
;
1804 unsigned long high
= zone
->pages_high
;
1813 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1815 unsigned int nr_free_buffer_pages(void)
1817 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1819 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1822 * Amount of free RAM allocatable within all zones
1824 unsigned int nr_free_pagecache_pages(void)
1826 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1829 static inline void show_node(struct zone
*zone
)
1832 printk("Node %d ", zone_to_nid(zone
));
1835 void si_meminfo(struct sysinfo
*val
)
1837 val
->totalram
= totalram_pages
;
1839 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1840 val
->bufferram
= nr_blockdev_pages();
1841 val
->totalhigh
= totalhigh_pages
;
1842 val
->freehigh
= nr_free_highpages();
1843 val
->mem_unit
= PAGE_SIZE
;
1846 EXPORT_SYMBOL(si_meminfo
);
1849 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1851 pg_data_t
*pgdat
= NODE_DATA(nid
);
1853 val
->totalram
= pgdat
->node_present_pages
;
1854 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1855 #ifdef CONFIG_HIGHMEM
1856 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1857 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1863 val
->mem_unit
= PAGE_SIZE
;
1867 #define K(x) ((x) << (PAGE_SHIFT-10))
1870 * Show free area list (used inside shift_scroll-lock stuff)
1871 * We also calculate the percentage fragmentation. We do this by counting the
1872 * memory on each free list with the exception of the first item on the list.
1874 void show_free_areas(void)
1879 for_each_populated_zone(zone
) {
1881 printk("%s per-cpu:\n", zone
->name
);
1883 for_each_online_cpu(cpu
) {
1884 struct per_cpu_pageset
*pageset
;
1886 pageset
= zone_pcp(zone
, cpu
);
1888 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1889 cpu
, pageset
->pcp
.high
,
1890 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1894 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1895 " inactive_file:%lu"
1896 //TODO: check/adjust line lengths
1897 #ifdef CONFIG_UNEVICTABLE_LRU
1900 " dirty:%lu writeback:%lu unstable:%lu\n"
1901 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1902 global_page_state(NR_ACTIVE_ANON
),
1903 global_page_state(NR_ACTIVE_FILE
),
1904 global_page_state(NR_INACTIVE_ANON
),
1905 global_page_state(NR_INACTIVE_FILE
),
1906 #ifdef CONFIG_UNEVICTABLE_LRU
1907 global_page_state(NR_UNEVICTABLE
),
1909 global_page_state(NR_FILE_DIRTY
),
1910 global_page_state(NR_WRITEBACK
),
1911 global_page_state(NR_UNSTABLE_NFS
),
1912 global_page_state(NR_FREE_PAGES
),
1913 global_page_state(NR_SLAB_RECLAIMABLE
) +
1914 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1915 global_page_state(NR_FILE_MAPPED
),
1916 global_page_state(NR_PAGETABLE
),
1917 global_page_state(NR_BOUNCE
));
1919 for_each_populated_zone(zone
) {
1928 " active_anon:%lukB"
1929 " inactive_anon:%lukB"
1930 " active_file:%lukB"
1931 " inactive_file:%lukB"
1932 #ifdef CONFIG_UNEVICTABLE_LRU
1933 " unevictable:%lukB"
1936 " pages_scanned:%lu"
1937 " all_unreclaimable? %s"
1940 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1943 K(zone
->pages_high
),
1944 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
1945 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
1946 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
1947 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
1948 #ifdef CONFIG_UNEVICTABLE_LRU
1949 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
1951 K(zone
->present_pages
),
1952 zone
->pages_scanned
,
1953 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1955 printk("lowmem_reserve[]:");
1956 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1957 printk(" %lu", zone
->lowmem_reserve
[i
]);
1961 for_each_populated_zone(zone
) {
1962 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1965 printk("%s: ", zone
->name
);
1967 spin_lock_irqsave(&zone
->lock
, flags
);
1968 for (order
= 0; order
< MAX_ORDER
; order
++) {
1969 nr
[order
] = zone
->free_area
[order
].nr_free
;
1970 total
+= nr
[order
] << order
;
1972 spin_unlock_irqrestore(&zone
->lock
, flags
);
1973 for (order
= 0; order
< MAX_ORDER
; order
++)
1974 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1975 printk("= %lukB\n", K(total
));
1978 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
1980 show_swap_cache_info();
1983 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
1985 zoneref
->zone
= zone
;
1986 zoneref
->zone_idx
= zone_idx(zone
);
1990 * Builds allocation fallback zone lists.
1992 * Add all populated zones of a node to the zonelist.
1994 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1995 int nr_zones
, enum zone_type zone_type
)
1999 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2004 zone
= pgdat
->node_zones
+ zone_type
;
2005 if (populated_zone(zone
)) {
2006 zoneref_set_zone(zone
,
2007 &zonelist
->_zonerefs
[nr_zones
++]);
2008 check_highest_zone(zone_type
);
2011 } while (zone_type
);
2018 * 0 = automatic detection of better ordering.
2019 * 1 = order by ([node] distance, -zonetype)
2020 * 2 = order by (-zonetype, [node] distance)
2022 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2023 * the same zonelist. So only NUMA can configure this param.
2025 #define ZONELIST_ORDER_DEFAULT 0
2026 #define ZONELIST_ORDER_NODE 1
2027 #define ZONELIST_ORDER_ZONE 2
2029 /* zonelist order in the kernel.
2030 * set_zonelist_order() will set this to NODE or ZONE.
2032 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2033 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2037 /* The value user specified ....changed by config */
2038 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2039 /* string for sysctl */
2040 #define NUMA_ZONELIST_ORDER_LEN 16
2041 char numa_zonelist_order
[16] = "default";
2044 * interface for configure zonelist ordering.
2045 * command line option "numa_zonelist_order"
2046 * = "[dD]efault - default, automatic configuration.
2047 * = "[nN]ode - order by node locality, then by zone within node
2048 * = "[zZ]one - order by zone, then by locality within zone
2051 static int __parse_numa_zonelist_order(char *s
)
2053 if (*s
== 'd' || *s
== 'D') {
2054 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2055 } else if (*s
== 'n' || *s
== 'N') {
2056 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2057 } else if (*s
== 'z' || *s
== 'Z') {
2058 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2061 "Ignoring invalid numa_zonelist_order value: "
2068 static __init
int setup_numa_zonelist_order(char *s
)
2071 return __parse_numa_zonelist_order(s
);
2074 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2077 * sysctl handler for numa_zonelist_order
2079 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2080 struct file
*file
, void __user
*buffer
, size_t *length
,
2083 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2087 strncpy(saved_string
, (char*)table
->data
,
2088 NUMA_ZONELIST_ORDER_LEN
);
2089 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2093 int oldval
= user_zonelist_order
;
2094 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2096 * bogus value. restore saved string
2098 strncpy((char*)table
->data
, saved_string
,
2099 NUMA_ZONELIST_ORDER_LEN
);
2100 user_zonelist_order
= oldval
;
2101 } else if (oldval
!= user_zonelist_order
)
2102 build_all_zonelists();
2108 #define MAX_NODE_LOAD (num_online_nodes())
2109 static int node_load
[MAX_NUMNODES
];
2112 * find_next_best_node - find the next node that should appear in a given node's fallback list
2113 * @node: node whose fallback list we're appending
2114 * @used_node_mask: nodemask_t of already used nodes
2116 * We use a number of factors to determine which is the next node that should
2117 * appear on a given node's fallback list. The node should not have appeared
2118 * already in @node's fallback list, and it should be the next closest node
2119 * according to the distance array (which contains arbitrary distance values
2120 * from each node to each node in the system), and should also prefer nodes
2121 * with no CPUs, since presumably they'll have very little allocation pressure
2122 * on them otherwise.
2123 * It returns -1 if no node is found.
2125 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2128 int min_val
= INT_MAX
;
2130 node_to_cpumask_ptr(tmp
, 0);
2132 /* Use the local node if we haven't already */
2133 if (!node_isset(node
, *used_node_mask
)) {
2134 node_set(node
, *used_node_mask
);
2138 for_each_node_state(n
, N_HIGH_MEMORY
) {
2140 /* Don't want a node to appear more than once */
2141 if (node_isset(n
, *used_node_mask
))
2144 /* Use the distance array to find the distance */
2145 val
= node_distance(node
, n
);
2147 /* Penalize nodes under us ("prefer the next node") */
2150 /* Give preference to headless and unused nodes */
2151 node_to_cpumask_ptr_next(tmp
, n
);
2152 if (!cpus_empty(*tmp
))
2153 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2155 /* Slight preference for less loaded node */
2156 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2157 val
+= node_load
[n
];
2159 if (val
< min_val
) {
2166 node_set(best_node
, *used_node_mask
);
2173 * Build zonelists ordered by node and zones within node.
2174 * This results in maximum locality--normal zone overflows into local
2175 * DMA zone, if any--but risks exhausting DMA zone.
2177 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2180 struct zonelist
*zonelist
;
2182 zonelist
= &pgdat
->node_zonelists
[0];
2183 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2185 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2187 zonelist
->_zonerefs
[j
].zone
= NULL
;
2188 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2192 * Build gfp_thisnode zonelists
2194 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2197 struct zonelist
*zonelist
;
2199 zonelist
= &pgdat
->node_zonelists
[1];
2200 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2201 zonelist
->_zonerefs
[j
].zone
= NULL
;
2202 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2206 * Build zonelists ordered by zone and nodes within zones.
2207 * This results in conserving DMA zone[s] until all Normal memory is
2208 * exhausted, but results in overflowing to remote node while memory
2209 * may still exist in local DMA zone.
2211 static int node_order
[MAX_NUMNODES
];
2213 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2216 int zone_type
; /* needs to be signed */
2218 struct zonelist
*zonelist
;
2220 zonelist
= &pgdat
->node_zonelists
[0];
2222 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2223 for (j
= 0; j
< nr_nodes
; j
++) {
2224 node
= node_order
[j
];
2225 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2226 if (populated_zone(z
)) {
2228 &zonelist
->_zonerefs
[pos
++]);
2229 check_highest_zone(zone_type
);
2233 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2234 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2237 static int default_zonelist_order(void)
2240 unsigned long low_kmem_size
,total_size
;
2244 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2245 * If they are really small and used heavily, the system can fall
2246 * into OOM very easily.
2247 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2249 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2252 for_each_online_node(nid
) {
2253 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2254 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2255 if (populated_zone(z
)) {
2256 if (zone_type
< ZONE_NORMAL
)
2257 low_kmem_size
+= z
->present_pages
;
2258 total_size
+= z
->present_pages
;
2262 if (!low_kmem_size
|| /* there are no DMA area. */
2263 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2264 return ZONELIST_ORDER_NODE
;
2266 * look into each node's config.
2267 * If there is a node whose DMA/DMA32 memory is very big area on
2268 * local memory, NODE_ORDER may be suitable.
2270 average_size
= total_size
/
2271 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2272 for_each_online_node(nid
) {
2275 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2276 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2277 if (populated_zone(z
)) {
2278 if (zone_type
< ZONE_NORMAL
)
2279 low_kmem_size
+= z
->present_pages
;
2280 total_size
+= z
->present_pages
;
2283 if (low_kmem_size
&&
2284 total_size
> average_size
&& /* ignore small node */
2285 low_kmem_size
> total_size
* 70/100)
2286 return ZONELIST_ORDER_NODE
;
2288 return ZONELIST_ORDER_ZONE
;
2291 static void set_zonelist_order(void)
2293 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2294 current_zonelist_order
= default_zonelist_order();
2296 current_zonelist_order
= user_zonelist_order
;
2299 static void build_zonelists(pg_data_t
*pgdat
)
2303 nodemask_t used_mask
;
2304 int local_node
, prev_node
;
2305 struct zonelist
*zonelist
;
2306 int order
= current_zonelist_order
;
2308 /* initialize zonelists */
2309 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2310 zonelist
= pgdat
->node_zonelists
+ i
;
2311 zonelist
->_zonerefs
[0].zone
= NULL
;
2312 zonelist
->_zonerefs
[0].zone_idx
= 0;
2315 /* NUMA-aware ordering of nodes */
2316 local_node
= pgdat
->node_id
;
2317 load
= num_online_nodes();
2318 prev_node
= local_node
;
2319 nodes_clear(used_mask
);
2321 memset(node_load
, 0, sizeof(node_load
));
2322 memset(node_order
, 0, sizeof(node_order
));
2325 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2326 int distance
= node_distance(local_node
, node
);
2329 * If another node is sufficiently far away then it is better
2330 * to reclaim pages in a zone before going off node.
2332 if (distance
> RECLAIM_DISTANCE
)
2333 zone_reclaim_mode
= 1;
2336 * We don't want to pressure a particular node.
2337 * So adding penalty to the first node in same
2338 * distance group to make it round-robin.
2340 if (distance
!= node_distance(local_node
, prev_node
))
2341 node_load
[node
] = load
;
2345 if (order
== ZONELIST_ORDER_NODE
)
2346 build_zonelists_in_node_order(pgdat
, node
);
2348 node_order
[j
++] = node
; /* remember order */
2351 if (order
== ZONELIST_ORDER_ZONE
) {
2352 /* calculate node order -- i.e., DMA last! */
2353 build_zonelists_in_zone_order(pgdat
, j
);
2356 build_thisnode_zonelists(pgdat
);
2359 /* Construct the zonelist performance cache - see further mmzone.h */
2360 static void build_zonelist_cache(pg_data_t
*pgdat
)
2362 struct zonelist
*zonelist
;
2363 struct zonelist_cache
*zlc
;
2366 zonelist
= &pgdat
->node_zonelists
[0];
2367 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2368 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2369 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2370 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2374 #else /* CONFIG_NUMA */
2376 static void set_zonelist_order(void)
2378 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2381 static void build_zonelists(pg_data_t
*pgdat
)
2383 int node
, local_node
;
2385 struct zonelist
*zonelist
;
2387 local_node
= pgdat
->node_id
;
2389 zonelist
= &pgdat
->node_zonelists
[0];
2390 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2393 * Now we build the zonelist so that it contains the zones
2394 * of all the other nodes.
2395 * We don't want to pressure a particular node, so when
2396 * building the zones for node N, we make sure that the
2397 * zones coming right after the local ones are those from
2398 * node N+1 (modulo N)
2400 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2401 if (!node_online(node
))
2403 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2406 for (node
= 0; node
< local_node
; node
++) {
2407 if (!node_online(node
))
2409 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2413 zonelist
->_zonerefs
[j
].zone
= NULL
;
2414 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2417 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2418 static void build_zonelist_cache(pg_data_t
*pgdat
)
2420 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2423 #endif /* CONFIG_NUMA */
2425 /* return values int ....just for stop_machine() */
2426 static int __build_all_zonelists(void *dummy
)
2430 for_each_online_node(nid
) {
2431 pg_data_t
*pgdat
= NODE_DATA(nid
);
2433 build_zonelists(pgdat
);
2434 build_zonelist_cache(pgdat
);
2439 void build_all_zonelists(void)
2441 set_zonelist_order();
2443 if (system_state
== SYSTEM_BOOTING
) {
2444 __build_all_zonelists(NULL
);
2445 mminit_verify_zonelist();
2446 cpuset_init_current_mems_allowed();
2448 /* we have to stop all cpus to guarantee there is no user
2450 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2451 /* cpuset refresh routine should be here */
2453 vm_total_pages
= nr_free_pagecache_pages();
2455 * Disable grouping by mobility if the number of pages in the
2456 * system is too low to allow the mechanism to work. It would be
2457 * more accurate, but expensive to check per-zone. This check is
2458 * made on memory-hotadd so a system can start with mobility
2459 * disabled and enable it later
2461 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2462 page_group_by_mobility_disabled
= 1;
2464 page_group_by_mobility_disabled
= 0;
2466 printk("Built %i zonelists in %s order, mobility grouping %s. "
2467 "Total pages: %ld\n",
2469 zonelist_order_name
[current_zonelist_order
],
2470 page_group_by_mobility_disabled
? "off" : "on",
2473 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2478 * Helper functions to size the waitqueue hash table.
2479 * Essentially these want to choose hash table sizes sufficiently
2480 * large so that collisions trying to wait on pages are rare.
2481 * But in fact, the number of active page waitqueues on typical
2482 * systems is ridiculously low, less than 200. So this is even
2483 * conservative, even though it seems large.
2485 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2486 * waitqueues, i.e. the size of the waitq table given the number of pages.
2488 #define PAGES_PER_WAITQUEUE 256
2490 #ifndef CONFIG_MEMORY_HOTPLUG
2491 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2493 unsigned long size
= 1;
2495 pages
/= PAGES_PER_WAITQUEUE
;
2497 while (size
< pages
)
2501 * Once we have dozens or even hundreds of threads sleeping
2502 * on IO we've got bigger problems than wait queue collision.
2503 * Limit the size of the wait table to a reasonable size.
2505 size
= min(size
, 4096UL);
2507 return max(size
, 4UL);
2511 * A zone's size might be changed by hot-add, so it is not possible to determine
2512 * a suitable size for its wait_table. So we use the maximum size now.
2514 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2516 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2517 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2518 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2520 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2521 * or more by the traditional way. (See above). It equals:
2523 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2524 * ia64(16K page size) : = ( 8G + 4M)byte.
2525 * powerpc (64K page size) : = (32G +16M)byte.
2527 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2534 * This is an integer logarithm so that shifts can be used later
2535 * to extract the more random high bits from the multiplicative
2536 * hash function before the remainder is taken.
2538 static inline unsigned long wait_table_bits(unsigned long size
)
2543 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2546 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2547 * of blocks reserved is based on zone->pages_min. The memory within the
2548 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2549 * higher will lead to a bigger reserve which will get freed as contiguous
2550 * blocks as reclaim kicks in
2552 static void setup_zone_migrate_reserve(struct zone
*zone
)
2554 unsigned long start_pfn
, pfn
, end_pfn
;
2556 unsigned long reserve
, block_migratetype
;
2558 /* Get the start pfn, end pfn and the number of blocks to reserve */
2559 start_pfn
= zone
->zone_start_pfn
;
2560 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2561 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2564 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2565 if (!pfn_valid(pfn
))
2567 page
= pfn_to_page(pfn
);
2569 /* Watch out for overlapping nodes */
2570 if (page_to_nid(page
) != zone_to_nid(zone
))
2573 /* Blocks with reserved pages will never free, skip them. */
2574 if (PageReserved(page
))
2577 block_migratetype
= get_pageblock_migratetype(page
);
2579 /* If this block is reserved, account for it */
2580 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2585 /* Suitable for reserving if this block is movable */
2586 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2587 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2588 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2594 * If the reserve is met and this is a previous reserved block,
2597 if (block_migratetype
== MIGRATE_RESERVE
) {
2598 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2599 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2605 * Initially all pages are reserved - free ones are freed
2606 * up by free_all_bootmem() once the early boot process is
2607 * done. Non-atomic initialization, single-pass.
2609 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2610 unsigned long start_pfn
, enum memmap_context context
)
2613 unsigned long end_pfn
= start_pfn
+ size
;
2617 if (highest_memmap_pfn
< end_pfn
- 1)
2618 highest_memmap_pfn
= end_pfn
- 1;
2620 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2621 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2623 * There can be holes in boot-time mem_map[]s
2624 * handed to this function. They do not
2625 * exist on hotplugged memory.
2627 if (context
== MEMMAP_EARLY
) {
2628 if (!early_pfn_valid(pfn
))
2630 if (!early_pfn_in_nid(pfn
, nid
))
2633 page
= pfn_to_page(pfn
);
2634 set_page_links(page
, zone
, nid
, pfn
);
2635 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2636 init_page_count(page
);
2637 reset_page_mapcount(page
);
2638 SetPageReserved(page
);
2640 * Mark the block movable so that blocks are reserved for
2641 * movable at startup. This will force kernel allocations
2642 * to reserve their blocks rather than leaking throughout
2643 * the address space during boot when many long-lived
2644 * kernel allocations are made. Later some blocks near
2645 * the start are marked MIGRATE_RESERVE by
2646 * setup_zone_migrate_reserve()
2648 * bitmap is created for zone's valid pfn range. but memmap
2649 * can be created for invalid pages (for alignment)
2650 * check here not to call set_pageblock_migratetype() against
2653 if ((z
->zone_start_pfn
<= pfn
)
2654 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2655 && !(pfn
& (pageblock_nr_pages
- 1)))
2656 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2658 INIT_LIST_HEAD(&page
->lru
);
2659 #ifdef WANT_PAGE_VIRTUAL
2660 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2661 if (!is_highmem_idx(zone
))
2662 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2667 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2670 for_each_migratetype_order(order
, t
) {
2671 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2672 zone
->free_area
[order
].nr_free
= 0;
2676 #ifndef __HAVE_ARCH_MEMMAP_INIT
2677 #define memmap_init(size, nid, zone, start_pfn) \
2678 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2681 static int zone_batchsize(struct zone
*zone
)
2686 * The per-cpu-pages pools are set to around 1000th of the
2687 * size of the zone. But no more than 1/2 of a meg.
2689 * OK, so we don't know how big the cache is. So guess.
2691 batch
= zone
->present_pages
/ 1024;
2692 if (batch
* PAGE_SIZE
> 512 * 1024)
2693 batch
= (512 * 1024) / PAGE_SIZE
;
2694 batch
/= 4; /* We effectively *= 4 below */
2699 * Clamp the batch to a 2^n - 1 value. Having a power
2700 * of 2 value was found to be more likely to have
2701 * suboptimal cache aliasing properties in some cases.
2703 * For example if 2 tasks are alternately allocating
2704 * batches of pages, one task can end up with a lot
2705 * of pages of one half of the possible page colors
2706 * and the other with pages of the other colors.
2708 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2713 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2715 struct per_cpu_pages
*pcp
;
2717 memset(p
, 0, sizeof(*p
));
2721 pcp
->high
= 6 * batch
;
2722 pcp
->batch
= max(1UL, 1 * batch
);
2723 INIT_LIST_HEAD(&pcp
->list
);
2727 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2728 * to the value high for the pageset p.
2731 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2734 struct per_cpu_pages
*pcp
;
2738 pcp
->batch
= max(1UL, high
/4);
2739 if ((high
/4) > (PAGE_SHIFT
* 8))
2740 pcp
->batch
= PAGE_SHIFT
* 8;
2746 * Boot pageset table. One per cpu which is going to be used for all
2747 * zones and all nodes. The parameters will be set in such a way
2748 * that an item put on a list will immediately be handed over to
2749 * the buddy list. This is safe since pageset manipulation is done
2750 * with interrupts disabled.
2752 * Some NUMA counter updates may also be caught by the boot pagesets.
2754 * The boot_pagesets must be kept even after bootup is complete for
2755 * unused processors and/or zones. They do play a role for bootstrapping
2756 * hotplugged processors.
2758 * zoneinfo_show() and maybe other functions do
2759 * not check if the processor is online before following the pageset pointer.
2760 * Other parts of the kernel may not check if the zone is available.
2762 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2765 * Dynamically allocate memory for the
2766 * per cpu pageset array in struct zone.
2768 static int __cpuinit
process_zones(int cpu
)
2770 struct zone
*zone
, *dzone
;
2771 int node
= cpu_to_node(cpu
);
2773 node_set_state(node
, N_CPU
); /* this node has a cpu */
2775 for_each_populated_zone(zone
) {
2776 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2778 if (!zone_pcp(zone
, cpu
))
2781 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2783 if (percpu_pagelist_fraction
)
2784 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2785 (zone
->present_pages
/ percpu_pagelist_fraction
));
2790 for_each_zone(dzone
) {
2791 if (!populated_zone(dzone
))
2795 kfree(zone_pcp(dzone
, cpu
));
2796 zone_pcp(dzone
, cpu
) = NULL
;
2801 static inline void free_zone_pagesets(int cpu
)
2805 for_each_zone(zone
) {
2806 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2808 /* Free per_cpu_pageset if it is slab allocated */
2809 if (pset
!= &boot_pageset
[cpu
])
2811 zone_pcp(zone
, cpu
) = NULL
;
2815 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2816 unsigned long action
,
2819 int cpu
= (long)hcpu
;
2820 int ret
= NOTIFY_OK
;
2823 case CPU_UP_PREPARE
:
2824 case CPU_UP_PREPARE_FROZEN
:
2825 if (process_zones(cpu
))
2828 case CPU_UP_CANCELED
:
2829 case CPU_UP_CANCELED_FROZEN
:
2831 case CPU_DEAD_FROZEN
:
2832 free_zone_pagesets(cpu
);
2840 static struct notifier_block __cpuinitdata pageset_notifier
=
2841 { &pageset_cpuup_callback
, NULL
, 0 };
2843 void __init
setup_per_cpu_pageset(void)
2847 /* Initialize per_cpu_pageset for cpu 0.
2848 * A cpuup callback will do this for every cpu
2849 * as it comes online
2851 err
= process_zones(smp_processor_id());
2853 register_cpu_notifier(&pageset_notifier
);
2858 static noinline __init_refok
2859 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2862 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2866 * The per-page waitqueue mechanism uses hashed waitqueues
2869 zone
->wait_table_hash_nr_entries
=
2870 wait_table_hash_nr_entries(zone_size_pages
);
2871 zone
->wait_table_bits
=
2872 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2873 alloc_size
= zone
->wait_table_hash_nr_entries
2874 * sizeof(wait_queue_head_t
);
2876 if (!slab_is_available()) {
2877 zone
->wait_table
= (wait_queue_head_t
*)
2878 alloc_bootmem_node(pgdat
, alloc_size
);
2881 * This case means that a zone whose size was 0 gets new memory
2882 * via memory hot-add.
2883 * But it may be the case that a new node was hot-added. In
2884 * this case vmalloc() will not be able to use this new node's
2885 * memory - this wait_table must be initialized to use this new
2886 * node itself as well.
2887 * To use this new node's memory, further consideration will be
2890 zone
->wait_table
= vmalloc(alloc_size
);
2892 if (!zone
->wait_table
)
2895 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2896 init_waitqueue_head(zone
->wait_table
+ i
);
2901 static __meminit
void zone_pcp_init(struct zone
*zone
)
2904 unsigned long batch
= zone_batchsize(zone
);
2906 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2908 /* Early boot. Slab allocator not functional yet */
2909 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2910 setup_pageset(&boot_pageset
[cpu
],0);
2912 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2915 if (zone
->present_pages
)
2916 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2917 zone
->name
, zone
->present_pages
, batch
);
2920 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2921 unsigned long zone_start_pfn
,
2923 enum memmap_context context
)
2925 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2927 ret
= zone_wait_table_init(zone
, size
);
2930 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2932 zone
->zone_start_pfn
= zone_start_pfn
;
2934 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
2935 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2937 (unsigned long)zone_idx(zone
),
2938 zone_start_pfn
, (zone_start_pfn
+ size
));
2940 zone_init_free_lists(zone
);
2945 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2947 * Basic iterator support. Return the first range of PFNs for a node
2948 * Note: nid == MAX_NUMNODES returns first region regardless of node
2950 static int __meminit
first_active_region_index_in_nid(int nid
)
2954 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2955 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2962 * Basic iterator support. Return the next active range of PFNs for a node
2963 * Note: nid == MAX_NUMNODES returns next region regardless of node
2965 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2967 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2968 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2974 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2976 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2977 * Architectures may implement their own version but if add_active_range()
2978 * was used and there are no special requirements, this is a convenient
2981 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
2985 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2986 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2987 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2989 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2990 return early_node_map
[i
].nid
;
2992 /* This is a memory hole */
2995 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2997 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3001 nid
= __early_pfn_to_nid(pfn
);
3004 /* just returns 0 */
3008 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3009 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3013 nid
= __early_pfn_to_nid(pfn
);
3014 if (nid
>= 0 && nid
!= node
)
3020 /* Basic iterator support to walk early_node_map[] */
3021 #define for_each_active_range_index_in_nid(i, nid) \
3022 for (i = first_active_region_index_in_nid(nid); i != -1; \
3023 i = next_active_region_index_in_nid(i, nid))
3026 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3027 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3028 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3030 * If an architecture guarantees that all ranges registered with
3031 * add_active_ranges() contain no holes and may be freed, this
3032 * this function may be used instead of calling free_bootmem() manually.
3034 void __init
free_bootmem_with_active_regions(int nid
,
3035 unsigned long max_low_pfn
)
3039 for_each_active_range_index_in_nid(i
, nid
) {
3040 unsigned long size_pages
= 0;
3041 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3043 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3046 if (end_pfn
> max_low_pfn
)
3047 end_pfn
= max_low_pfn
;
3049 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3050 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3051 PFN_PHYS(early_node_map
[i
].start_pfn
),
3052 size_pages
<< PAGE_SHIFT
);
3056 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3061 for_each_active_range_index_in_nid(i
, nid
) {
3062 ret
= work_fn(early_node_map
[i
].start_pfn
,
3063 early_node_map
[i
].end_pfn
, data
);
3069 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3070 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3072 * If an architecture guarantees that all ranges registered with
3073 * add_active_ranges() contain no holes and may be freed, this
3074 * function may be used instead of calling memory_present() manually.
3076 void __init
sparse_memory_present_with_active_regions(int nid
)
3080 for_each_active_range_index_in_nid(i
, nid
)
3081 memory_present(early_node_map
[i
].nid
,
3082 early_node_map
[i
].start_pfn
,
3083 early_node_map
[i
].end_pfn
);
3087 * push_node_boundaries - Push node boundaries to at least the requested boundary
3088 * @nid: The nid of the node to push the boundary for
3089 * @start_pfn: The start pfn of the node
3090 * @end_pfn: The end pfn of the node
3092 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3093 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3094 * be hotplugged even though no physical memory exists. This function allows
3095 * an arch to push out the node boundaries so mem_map is allocated that can
3098 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3099 void __init
push_node_boundaries(unsigned int nid
,
3100 unsigned long start_pfn
, unsigned long end_pfn
)
3102 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3103 "Entering push_node_boundaries(%u, %lu, %lu)\n",
3104 nid
, start_pfn
, end_pfn
);
3106 /* Initialise the boundary for this node if necessary */
3107 if (node_boundary_end_pfn
[nid
] == 0)
3108 node_boundary_start_pfn
[nid
] = -1UL;
3110 /* Update the boundaries */
3111 if (node_boundary_start_pfn
[nid
] > start_pfn
)
3112 node_boundary_start_pfn
[nid
] = start_pfn
;
3113 if (node_boundary_end_pfn
[nid
] < end_pfn
)
3114 node_boundary_end_pfn
[nid
] = end_pfn
;
3117 /* If necessary, push the node boundary out for reserve hotadd */
3118 static void __meminit
account_node_boundary(unsigned int nid
,
3119 unsigned long *start_pfn
, unsigned long *end_pfn
)
3121 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3122 "Entering account_node_boundary(%u, %lu, %lu)\n",
3123 nid
, *start_pfn
, *end_pfn
);
3125 /* Return if boundary information has not been provided */
3126 if (node_boundary_end_pfn
[nid
] == 0)
3129 /* Check the boundaries and update if necessary */
3130 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
3131 *start_pfn
= node_boundary_start_pfn
[nid
];
3132 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
3133 *end_pfn
= node_boundary_end_pfn
[nid
];
3136 void __init
push_node_boundaries(unsigned int nid
,
3137 unsigned long start_pfn
, unsigned long end_pfn
) {}
3139 static void __meminit
account_node_boundary(unsigned int nid
,
3140 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3145 * get_pfn_range_for_nid - Return the start and end page frames for a node
3146 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3147 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3148 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3150 * It returns the start and end page frame of a node based on information
3151 * provided by an arch calling add_active_range(). If called for a node
3152 * with no available memory, a warning is printed and the start and end
3155 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3156 unsigned long *start_pfn
, unsigned long *end_pfn
)
3162 for_each_active_range_index_in_nid(i
, nid
) {
3163 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3164 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3167 if (*start_pfn
== -1UL)
3170 /* Push the node boundaries out if requested */
3171 account_node_boundary(nid
, start_pfn
, end_pfn
);
3175 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3176 * assumption is made that zones within a node are ordered in monotonic
3177 * increasing memory addresses so that the "highest" populated zone is used
3179 static void __init
find_usable_zone_for_movable(void)
3182 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3183 if (zone_index
== ZONE_MOVABLE
)
3186 if (arch_zone_highest_possible_pfn
[zone_index
] >
3187 arch_zone_lowest_possible_pfn
[zone_index
])
3191 VM_BUG_ON(zone_index
== -1);
3192 movable_zone
= zone_index
;
3196 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3197 * because it is sized independant of architecture. Unlike the other zones,
3198 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3199 * in each node depending on the size of each node and how evenly kernelcore
3200 * is distributed. This helper function adjusts the zone ranges
3201 * provided by the architecture for a given node by using the end of the
3202 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3203 * zones within a node are in order of monotonic increases memory addresses
3205 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3206 unsigned long zone_type
,
3207 unsigned long node_start_pfn
,
3208 unsigned long node_end_pfn
,
3209 unsigned long *zone_start_pfn
,
3210 unsigned long *zone_end_pfn
)
3212 /* Only adjust if ZONE_MOVABLE is on this node */
3213 if (zone_movable_pfn
[nid
]) {
3214 /* Size ZONE_MOVABLE */
3215 if (zone_type
== ZONE_MOVABLE
) {
3216 *zone_start_pfn
= zone_movable_pfn
[nid
];
3217 *zone_end_pfn
= min(node_end_pfn
,
3218 arch_zone_highest_possible_pfn
[movable_zone
]);
3220 /* Adjust for ZONE_MOVABLE starting within this range */
3221 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3222 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3223 *zone_end_pfn
= zone_movable_pfn
[nid
];
3225 /* Check if this whole range is within ZONE_MOVABLE */
3226 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3227 *zone_start_pfn
= *zone_end_pfn
;
3232 * Return the number of pages a zone spans in a node, including holes
3233 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3235 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3236 unsigned long zone_type
,
3237 unsigned long *ignored
)
3239 unsigned long node_start_pfn
, node_end_pfn
;
3240 unsigned long zone_start_pfn
, zone_end_pfn
;
3242 /* Get the start and end of the node and zone */
3243 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3244 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3245 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3246 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3247 node_start_pfn
, node_end_pfn
,
3248 &zone_start_pfn
, &zone_end_pfn
);
3250 /* Check that this node has pages within the zone's required range */
3251 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3254 /* Move the zone boundaries inside the node if necessary */
3255 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3256 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3258 /* Return the spanned pages */
3259 return zone_end_pfn
- zone_start_pfn
;
3263 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3264 * then all holes in the requested range will be accounted for.
3266 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3267 unsigned long range_start_pfn
,
3268 unsigned long range_end_pfn
)
3271 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3272 unsigned long start_pfn
;
3274 /* Find the end_pfn of the first active range of pfns in the node */
3275 i
= first_active_region_index_in_nid(nid
);
3279 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3281 /* Account for ranges before physical memory on this node */
3282 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3283 hole_pages
= prev_end_pfn
- range_start_pfn
;
3285 /* Find all holes for the zone within the node */
3286 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3288 /* No need to continue if prev_end_pfn is outside the zone */
3289 if (prev_end_pfn
>= range_end_pfn
)
3292 /* Make sure the end of the zone is not within the hole */
3293 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3294 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3296 /* Update the hole size cound and move on */
3297 if (start_pfn
> range_start_pfn
) {
3298 BUG_ON(prev_end_pfn
> start_pfn
);
3299 hole_pages
+= start_pfn
- prev_end_pfn
;
3301 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3304 /* Account for ranges past physical memory on this node */
3305 if (range_end_pfn
> prev_end_pfn
)
3306 hole_pages
+= range_end_pfn
-
3307 max(range_start_pfn
, prev_end_pfn
);
3313 * absent_pages_in_range - Return number of page frames in holes within a range
3314 * @start_pfn: The start PFN to start searching for holes
3315 * @end_pfn: The end PFN to stop searching for holes
3317 * It returns the number of pages frames in memory holes within a range.
3319 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3320 unsigned long end_pfn
)
3322 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3325 /* Return the number of page frames in holes in a zone on a node */
3326 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3327 unsigned long zone_type
,
3328 unsigned long *ignored
)
3330 unsigned long node_start_pfn
, node_end_pfn
;
3331 unsigned long zone_start_pfn
, zone_end_pfn
;
3333 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3334 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3336 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3339 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3340 node_start_pfn
, node_end_pfn
,
3341 &zone_start_pfn
, &zone_end_pfn
);
3342 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3346 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3347 unsigned long zone_type
,
3348 unsigned long *zones_size
)
3350 return zones_size
[zone_type
];
3353 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3354 unsigned long zone_type
,
3355 unsigned long *zholes_size
)
3360 return zholes_size
[zone_type
];
3365 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3366 unsigned long *zones_size
, unsigned long *zholes_size
)
3368 unsigned long realtotalpages
, totalpages
= 0;
3371 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3372 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3374 pgdat
->node_spanned_pages
= totalpages
;
3376 realtotalpages
= totalpages
;
3377 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3379 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3381 pgdat
->node_present_pages
= realtotalpages
;
3382 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3386 #ifndef CONFIG_SPARSEMEM
3388 * Calculate the size of the zone->blockflags rounded to an unsigned long
3389 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3390 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3391 * round what is now in bits to nearest long in bits, then return it in
3394 static unsigned long __init
usemap_size(unsigned long zonesize
)
3396 unsigned long usemapsize
;
3398 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3399 usemapsize
= usemapsize
>> pageblock_order
;
3400 usemapsize
*= NR_PAGEBLOCK_BITS
;
3401 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3403 return usemapsize
/ 8;
3406 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3407 struct zone
*zone
, unsigned long zonesize
)
3409 unsigned long usemapsize
= usemap_size(zonesize
);
3410 zone
->pageblock_flags
= NULL
;
3412 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3415 static void inline setup_usemap(struct pglist_data
*pgdat
,
3416 struct zone
*zone
, unsigned long zonesize
) {}
3417 #endif /* CONFIG_SPARSEMEM */
3419 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3421 /* Return a sensible default order for the pageblock size. */
3422 static inline int pageblock_default_order(void)
3424 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3425 return HUGETLB_PAGE_ORDER
;
3430 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3431 static inline void __init
set_pageblock_order(unsigned int order
)
3433 /* Check that pageblock_nr_pages has not already been setup */
3434 if (pageblock_order
)
3438 * Assume the largest contiguous order of interest is a huge page.
3439 * This value may be variable depending on boot parameters on IA64
3441 pageblock_order
= order
;
3443 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3446 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3447 * and pageblock_default_order() are unused as pageblock_order is set
3448 * at compile-time. See include/linux/pageblock-flags.h for the values of
3449 * pageblock_order based on the kernel config
3451 static inline int pageblock_default_order(unsigned int order
)
3455 #define set_pageblock_order(x) do {} while (0)
3457 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3460 * Set up the zone data structures:
3461 * - mark all pages reserved
3462 * - mark all memory queues empty
3463 * - clear the memory bitmaps
3465 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3466 unsigned long *zones_size
, unsigned long *zholes_size
)
3469 int nid
= pgdat
->node_id
;
3470 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3473 pgdat_resize_init(pgdat
);
3474 pgdat
->nr_zones
= 0;
3475 init_waitqueue_head(&pgdat
->kswapd_wait
);
3476 pgdat
->kswapd_max_order
= 0;
3477 pgdat_page_cgroup_init(pgdat
);
3479 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3480 struct zone
*zone
= pgdat
->node_zones
+ j
;
3481 unsigned long size
, realsize
, memmap_pages
;
3484 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3485 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3489 * Adjust realsize so that it accounts for how much memory
3490 * is used by this zone for memmap. This affects the watermark
3491 * and per-cpu initialisations
3494 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3495 if (realsize
>= memmap_pages
) {
3496 realsize
-= memmap_pages
;
3499 " %s zone: %lu pages used for memmap\n",
3500 zone_names
[j
], memmap_pages
);
3503 " %s zone: %lu pages exceeds realsize %lu\n",
3504 zone_names
[j
], memmap_pages
, realsize
);
3506 /* Account for reserved pages */
3507 if (j
== 0 && realsize
> dma_reserve
) {
3508 realsize
-= dma_reserve
;
3509 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3510 zone_names
[0], dma_reserve
);
3513 if (!is_highmem_idx(j
))
3514 nr_kernel_pages
+= realsize
;
3515 nr_all_pages
+= realsize
;
3517 zone
->spanned_pages
= size
;
3518 zone
->present_pages
= realsize
;
3521 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3523 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3525 zone
->name
= zone_names
[j
];
3526 spin_lock_init(&zone
->lock
);
3527 spin_lock_init(&zone
->lru_lock
);
3528 zone_seqlock_init(zone
);
3529 zone
->zone_pgdat
= pgdat
;
3531 zone
->prev_priority
= DEF_PRIORITY
;
3533 zone_pcp_init(zone
);
3535 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3536 zone
->lru
[l
].nr_scan
= 0;
3538 zone
->reclaim_stat
.recent_rotated
[0] = 0;
3539 zone
->reclaim_stat
.recent_rotated
[1] = 0;
3540 zone
->reclaim_stat
.recent_scanned
[0] = 0;
3541 zone
->reclaim_stat
.recent_scanned
[1] = 0;
3542 zap_zone_vm_stats(zone
);
3547 set_pageblock_order(pageblock_default_order());
3548 setup_usemap(pgdat
, zone
, size
);
3549 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3550 size
, MEMMAP_EARLY
);
3552 memmap_init(size
, nid
, j
, zone_start_pfn
);
3553 zone_start_pfn
+= size
;
3557 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3559 /* Skip empty nodes */
3560 if (!pgdat
->node_spanned_pages
)
3563 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3564 /* ia64 gets its own node_mem_map, before this, without bootmem */
3565 if (!pgdat
->node_mem_map
) {
3566 unsigned long size
, start
, end
;
3570 * The zone's endpoints aren't required to be MAX_ORDER
3571 * aligned but the node_mem_map endpoints must be in order
3572 * for the buddy allocator to function correctly.
3574 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3575 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3576 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3577 size
= (end
- start
) * sizeof(struct page
);
3578 map
= alloc_remap(pgdat
->node_id
, size
);
3580 map
= alloc_bootmem_node(pgdat
, size
);
3581 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3583 #ifndef CONFIG_NEED_MULTIPLE_NODES
3585 * With no DISCONTIG, the global mem_map is just set as node 0's
3587 if (pgdat
== NODE_DATA(0)) {
3588 mem_map
= NODE_DATA(0)->node_mem_map
;
3589 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3590 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3591 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3592 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3595 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3598 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3599 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3601 pg_data_t
*pgdat
= NODE_DATA(nid
);
3603 pgdat
->node_id
= nid
;
3604 pgdat
->node_start_pfn
= node_start_pfn
;
3605 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3607 alloc_node_mem_map(pgdat
);
3608 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3609 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3610 nid
, (unsigned long)pgdat
,
3611 (unsigned long)pgdat
->node_mem_map
);
3614 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3617 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3619 #if MAX_NUMNODES > 1
3621 * Figure out the number of possible node ids.
3623 static void __init
setup_nr_node_ids(void)
3626 unsigned int highest
= 0;
3628 for_each_node_mask(node
, node_possible_map
)
3630 nr_node_ids
= highest
+ 1;
3633 static inline void setup_nr_node_ids(void)
3639 * add_active_range - Register a range of PFNs backed by physical memory
3640 * @nid: The node ID the range resides on
3641 * @start_pfn: The start PFN of the available physical memory
3642 * @end_pfn: The end PFN of the available physical memory
3644 * These ranges are stored in an early_node_map[] and later used by
3645 * free_area_init_nodes() to calculate zone sizes and holes. If the
3646 * range spans a memory hole, it is up to the architecture to ensure
3647 * the memory is not freed by the bootmem allocator. If possible
3648 * the range being registered will be merged with existing ranges.
3650 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3651 unsigned long end_pfn
)
3655 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3656 "Entering add_active_range(%d, %#lx, %#lx) "
3657 "%d entries of %d used\n",
3658 nid
, start_pfn
, end_pfn
,
3659 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3661 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3663 /* Merge with existing active regions if possible */
3664 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3665 if (early_node_map
[i
].nid
!= nid
)
3668 /* Skip if an existing region covers this new one */
3669 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3670 end_pfn
<= early_node_map
[i
].end_pfn
)
3673 /* Merge forward if suitable */
3674 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3675 end_pfn
> early_node_map
[i
].end_pfn
) {
3676 early_node_map
[i
].end_pfn
= end_pfn
;
3680 /* Merge backward if suitable */
3681 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3682 end_pfn
>= early_node_map
[i
].start_pfn
) {
3683 early_node_map
[i
].start_pfn
= start_pfn
;
3688 /* Check that early_node_map is large enough */
3689 if (i
>= MAX_ACTIVE_REGIONS
) {
3690 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3691 MAX_ACTIVE_REGIONS
);
3695 early_node_map
[i
].nid
= nid
;
3696 early_node_map
[i
].start_pfn
= start_pfn
;
3697 early_node_map
[i
].end_pfn
= end_pfn
;
3698 nr_nodemap_entries
= i
+ 1;
3702 * remove_active_range - Shrink an existing registered range of PFNs
3703 * @nid: The node id the range is on that should be shrunk
3704 * @start_pfn: The new PFN of the range
3705 * @end_pfn: The new PFN of the range
3707 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3708 * The map is kept near the end physical page range that has already been
3709 * registered. This function allows an arch to shrink an existing registered
3712 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3713 unsigned long end_pfn
)
3718 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3719 nid
, start_pfn
, end_pfn
);
3721 /* Find the old active region end and shrink */
3722 for_each_active_range_index_in_nid(i
, nid
) {
3723 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3724 early_node_map
[i
].end_pfn
<= end_pfn
) {
3726 early_node_map
[i
].start_pfn
= 0;
3727 early_node_map
[i
].end_pfn
= 0;
3731 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3732 early_node_map
[i
].end_pfn
> start_pfn
) {
3733 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3734 early_node_map
[i
].end_pfn
= start_pfn
;
3735 if (temp_end_pfn
> end_pfn
)
3736 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3739 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3740 early_node_map
[i
].end_pfn
> end_pfn
&&
3741 early_node_map
[i
].start_pfn
< end_pfn
) {
3742 early_node_map
[i
].start_pfn
= end_pfn
;
3750 /* remove the blank ones */
3751 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3752 if (early_node_map
[i
].nid
!= nid
)
3754 if (early_node_map
[i
].end_pfn
)
3756 /* we found it, get rid of it */
3757 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3758 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3759 sizeof(early_node_map
[j
]));
3760 j
= nr_nodemap_entries
- 1;
3761 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3762 nr_nodemap_entries
--;
3767 * remove_all_active_ranges - Remove all currently registered regions
3769 * During discovery, it may be found that a table like SRAT is invalid
3770 * and an alternative discovery method must be used. This function removes
3771 * all currently registered regions.
3773 void __init
remove_all_active_ranges(void)
3775 memset(early_node_map
, 0, sizeof(early_node_map
));
3776 nr_nodemap_entries
= 0;
3777 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3778 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3779 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3780 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3783 /* Compare two active node_active_regions */
3784 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3786 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3787 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3789 /* Done this way to avoid overflows */
3790 if (arange
->start_pfn
> brange
->start_pfn
)
3792 if (arange
->start_pfn
< brange
->start_pfn
)
3798 /* sort the node_map by start_pfn */
3799 static void __init
sort_node_map(void)
3801 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3802 sizeof(struct node_active_region
),
3803 cmp_node_active_region
, NULL
);
3806 /* Find the lowest pfn for a node */
3807 static unsigned long __init
find_min_pfn_for_node(int nid
)
3810 unsigned long min_pfn
= ULONG_MAX
;
3812 /* Assuming a sorted map, the first range found has the starting pfn */
3813 for_each_active_range_index_in_nid(i
, nid
)
3814 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3816 if (min_pfn
== ULONG_MAX
) {
3818 "Could not find start_pfn for node %d\n", nid
);
3826 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3828 * It returns the minimum PFN based on information provided via
3829 * add_active_range().
3831 unsigned long __init
find_min_pfn_with_active_regions(void)
3833 return find_min_pfn_for_node(MAX_NUMNODES
);
3837 * early_calculate_totalpages()
3838 * Sum pages in active regions for movable zone.
3839 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3841 static unsigned long __init
early_calculate_totalpages(void)
3844 unsigned long totalpages
= 0;
3846 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3847 unsigned long pages
= early_node_map
[i
].end_pfn
-
3848 early_node_map
[i
].start_pfn
;
3849 totalpages
+= pages
;
3851 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3857 * Find the PFN the Movable zone begins in each node. Kernel memory
3858 * is spread evenly between nodes as long as the nodes have enough
3859 * memory. When they don't, some nodes will have more kernelcore than
3862 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3865 unsigned long usable_startpfn
;
3866 unsigned long kernelcore_node
, kernelcore_remaining
;
3867 unsigned long totalpages
= early_calculate_totalpages();
3868 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3871 * If movablecore was specified, calculate what size of
3872 * kernelcore that corresponds so that memory usable for
3873 * any allocation type is evenly spread. If both kernelcore
3874 * and movablecore are specified, then the value of kernelcore
3875 * will be used for required_kernelcore if it's greater than
3876 * what movablecore would have allowed.
3878 if (required_movablecore
) {
3879 unsigned long corepages
;
3882 * Round-up so that ZONE_MOVABLE is at least as large as what
3883 * was requested by the user
3885 required_movablecore
=
3886 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3887 corepages
= totalpages
- required_movablecore
;
3889 required_kernelcore
= max(required_kernelcore
, corepages
);
3892 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3893 if (!required_kernelcore
)
3896 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3897 find_usable_zone_for_movable();
3898 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3901 /* Spread kernelcore memory as evenly as possible throughout nodes */
3902 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3903 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3905 * Recalculate kernelcore_node if the division per node
3906 * now exceeds what is necessary to satisfy the requested
3907 * amount of memory for the kernel
3909 if (required_kernelcore
< kernelcore_node
)
3910 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3913 * As the map is walked, we track how much memory is usable
3914 * by the kernel using kernelcore_remaining. When it is
3915 * 0, the rest of the node is usable by ZONE_MOVABLE
3917 kernelcore_remaining
= kernelcore_node
;
3919 /* Go through each range of PFNs within this node */
3920 for_each_active_range_index_in_nid(i
, nid
) {
3921 unsigned long start_pfn
, end_pfn
;
3922 unsigned long size_pages
;
3924 start_pfn
= max(early_node_map
[i
].start_pfn
,
3925 zone_movable_pfn
[nid
]);
3926 end_pfn
= early_node_map
[i
].end_pfn
;
3927 if (start_pfn
>= end_pfn
)
3930 /* Account for what is only usable for kernelcore */
3931 if (start_pfn
< usable_startpfn
) {
3932 unsigned long kernel_pages
;
3933 kernel_pages
= min(end_pfn
, usable_startpfn
)
3936 kernelcore_remaining
-= min(kernel_pages
,
3937 kernelcore_remaining
);
3938 required_kernelcore
-= min(kernel_pages
,
3939 required_kernelcore
);
3941 /* Continue if range is now fully accounted */
3942 if (end_pfn
<= usable_startpfn
) {
3945 * Push zone_movable_pfn to the end so
3946 * that if we have to rebalance
3947 * kernelcore across nodes, we will
3948 * not double account here
3950 zone_movable_pfn
[nid
] = end_pfn
;
3953 start_pfn
= usable_startpfn
;
3957 * The usable PFN range for ZONE_MOVABLE is from
3958 * start_pfn->end_pfn. Calculate size_pages as the
3959 * number of pages used as kernelcore
3961 size_pages
= end_pfn
- start_pfn
;
3962 if (size_pages
> kernelcore_remaining
)
3963 size_pages
= kernelcore_remaining
;
3964 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3967 * Some kernelcore has been met, update counts and
3968 * break if the kernelcore for this node has been
3971 required_kernelcore
-= min(required_kernelcore
,
3973 kernelcore_remaining
-= size_pages
;
3974 if (!kernelcore_remaining
)
3980 * If there is still required_kernelcore, we do another pass with one
3981 * less node in the count. This will push zone_movable_pfn[nid] further
3982 * along on the nodes that still have memory until kernelcore is
3986 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3989 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3990 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3991 zone_movable_pfn
[nid
] =
3992 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3995 /* Any regular memory on that node ? */
3996 static void check_for_regular_memory(pg_data_t
*pgdat
)
3998 #ifdef CONFIG_HIGHMEM
3999 enum zone_type zone_type
;
4001 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4002 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4003 if (zone
->present_pages
)
4004 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4010 * free_area_init_nodes - Initialise all pg_data_t and zone data
4011 * @max_zone_pfn: an array of max PFNs for each zone
4013 * This will call free_area_init_node() for each active node in the system.
4014 * Using the page ranges provided by add_active_range(), the size of each
4015 * zone in each node and their holes is calculated. If the maximum PFN
4016 * between two adjacent zones match, it is assumed that the zone is empty.
4017 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4018 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4019 * starts where the previous one ended. For example, ZONE_DMA32 starts
4020 * at arch_max_dma_pfn.
4022 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4027 /* Sort early_node_map as initialisation assumes it is sorted */
4030 /* Record where the zone boundaries are */
4031 memset(arch_zone_lowest_possible_pfn
, 0,
4032 sizeof(arch_zone_lowest_possible_pfn
));
4033 memset(arch_zone_highest_possible_pfn
, 0,
4034 sizeof(arch_zone_highest_possible_pfn
));
4035 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4036 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4037 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4038 if (i
== ZONE_MOVABLE
)
4040 arch_zone_lowest_possible_pfn
[i
] =
4041 arch_zone_highest_possible_pfn
[i
-1];
4042 arch_zone_highest_possible_pfn
[i
] =
4043 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4045 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4046 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4048 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4049 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4050 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4052 /* Print out the zone ranges */
4053 printk("Zone PFN ranges:\n");
4054 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4055 if (i
== ZONE_MOVABLE
)
4057 printk(" %-8s %0#10lx -> %0#10lx\n",
4059 arch_zone_lowest_possible_pfn
[i
],
4060 arch_zone_highest_possible_pfn
[i
]);
4063 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4064 printk("Movable zone start PFN for each node\n");
4065 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4066 if (zone_movable_pfn
[i
])
4067 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4070 /* Print out the early_node_map[] */
4071 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4072 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4073 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4074 early_node_map
[i
].start_pfn
,
4075 early_node_map
[i
].end_pfn
);
4077 /* Initialise every node */
4078 mminit_verify_pageflags_layout();
4079 setup_nr_node_ids();
4080 for_each_online_node(nid
) {
4081 pg_data_t
*pgdat
= NODE_DATA(nid
);
4082 free_area_init_node(nid
, NULL
,
4083 find_min_pfn_for_node(nid
), NULL
);
4085 /* Any memory on that node */
4086 if (pgdat
->node_present_pages
)
4087 node_set_state(nid
, N_HIGH_MEMORY
);
4088 check_for_regular_memory(pgdat
);
4092 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4094 unsigned long long coremem
;
4098 coremem
= memparse(p
, &p
);
4099 *core
= coremem
>> PAGE_SHIFT
;
4101 /* Paranoid check that UL is enough for the coremem value */
4102 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4108 * kernelcore=size sets the amount of memory for use for allocations that
4109 * cannot be reclaimed or migrated.
4111 static int __init
cmdline_parse_kernelcore(char *p
)
4113 return cmdline_parse_core(p
, &required_kernelcore
);
4117 * movablecore=size sets the amount of memory for use for allocations that
4118 * can be reclaimed or migrated.
4120 static int __init
cmdline_parse_movablecore(char *p
)
4122 return cmdline_parse_core(p
, &required_movablecore
);
4125 early_param("kernelcore", cmdline_parse_kernelcore
);
4126 early_param("movablecore", cmdline_parse_movablecore
);
4128 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4131 * set_dma_reserve - set the specified number of pages reserved in the first zone
4132 * @new_dma_reserve: The number of pages to mark reserved
4134 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4135 * In the DMA zone, a significant percentage may be consumed by kernel image
4136 * and other unfreeable allocations which can skew the watermarks badly. This
4137 * function may optionally be used to account for unfreeable pages in the
4138 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4139 * smaller per-cpu batchsize.
4141 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4143 dma_reserve
= new_dma_reserve
;
4146 #ifndef CONFIG_NEED_MULTIPLE_NODES
4147 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4148 EXPORT_SYMBOL(contig_page_data
);
4151 void __init
free_area_init(unsigned long *zones_size
)
4153 free_area_init_node(0, zones_size
,
4154 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4157 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4158 unsigned long action
, void *hcpu
)
4160 int cpu
= (unsigned long)hcpu
;
4162 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4166 * Spill the event counters of the dead processor
4167 * into the current processors event counters.
4168 * This artificially elevates the count of the current
4171 vm_events_fold_cpu(cpu
);
4174 * Zero the differential counters of the dead processor
4175 * so that the vm statistics are consistent.
4177 * This is only okay since the processor is dead and cannot
4178 * race with what we are doing.
4180 refresh_cpu_vm_stats(cpu
);
4185 void __init
page_alloc_init(void)
4187 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4191 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4192 * or min_free_kbytes changes.
4194 static void calculate_totalreserve_pages(void)
4196 struct pglist_data
*pgdat
;
4197 unsigned long reserve_pages
= 0;
4198 enum zone_type i
, j
;
4200 for_each_online_pgdat(pgdat
) {
4201 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4202 struct zone
*zone
= pgdat
->node_zones
+ i
;
4203 unsigned long max
= 0;
4205 /* Find valid and maximum lowmem_reserve in the zone */
4206 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4207 if (zone
->lowmem_reserve
[j
] > max
)
4208 max
= zone
->lowmem_reserve
[j
];
4211 /* we treat pages_high as reserved pages. */
4212 max
+= zone
->pages_high
;
4214 if (max
> zone
->present_pages
)
4215 max
= zone
->present_pages
;
4216 reserve_pages
+= max
;
4219 totalreserve_pages
= reserve_pages
;
4223 * setup_per_zone_lowmem_reserve - called whenever
4224 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4225 * has a correct pages reserved value, so an adequate number of
4226 * pages are left in the zone after a successful __alloc_pages().
4228 static void setup_per_zone_lowmem_reserve(void)
4230 struct pglist_data
*pgdat
;
4231 enum zone_type j
, idx
;
4233 for_each_online_pgdat(pgdat
) {
4234 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4235 struct zone
*zone
= pgdat
->node_zones
+ j
;
4236 unsigned long present_pages
= zone
->present_pages
;
4238 zone
->lowmem_reserve
[j
] = 0;
4242 struct zone
*lower_zone
;
4246 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4247 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4249 lower_zone
= pgdat
->node_zones
+ idx
;
4250 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4251 sysctl_lowmem_reserve_ratio
[idx
];
4252 present_pages
+= lower_zone
->present_pages
;
4257 /* update totalreserve_pages */
4258 calculate_totalreserve_pages();
4262 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4264 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4265 * with respect to min_free_kbytes.
4267 void setup_per_zone_pages_min(void)
4269 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4270 unsigned long lowmem_pages
= 0;
4272 unsigned long flags
;
4274 /* Calculate total number of !ZONE_HIGHMEM pages */
4275 for_each_zone(zone
) {
4276 if (!is_highmem(zone
))
4277 lowmem_pages
+= zone
->present_pages
;
4280 for_each_zone(zone
) {
4283 spin_lock_irqsave(&zone
->lock
, flags
);
4284 tmp
= (u64
)pages_min
* zone
->present_pages
;
4285 do_div(tmp
, lowmem_pages
);
4286 if (is_highmem(zone
)) {
4288 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4289 * need highmem pages, so cap pages_min to a small
4292 * The (pages_high-pages_low) and (pages_low-pages_min)
4293 * deltas controls asynch page reclaim, and so should
4294 * not be capped for highmem.
4298 min_pages
= zone
->present_pages
/ 1024;
4299 if (min_pages
< SWAP_CLUSTER_MAX
)
4300 min_pages
= SWAP_CLUSTER_MAX
;
4301 if (min_pages
> 128)
4303 zone
->pages_min
= min_pages
;
4306 * If it's a lowmem zone, reserve a number of pages
4307 * proportionate to the zone's size.
4309 zone
->pages_min
= tmp
;
4312 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4313 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4314 setup_zone_migrate_reserve(zone
);
4315 spin_unlock_irqrestore(&zone
->lock
, flags
);
4318 /* update totalreserve_pages */
4319 calculate_totalreserve_pages();
4323 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4325 * The inactive anon list should be small enough that the VM never has to
4326 * do too much work, but large enough that each inactive page has a chance
4327 * to be referenced again before it is swapped out.
4329 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4330 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4331 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4332 * the anonymous pages are kept on the inactive list.
4335 * memory ratio inactive anon
4336 * -------------------------------------
4345 static void setup_per_zone_inactive_ratio(void)
4349 for_each_zone(zone
) {
4350 unsigned int gb
, ratio
;
4352 /* Zone size in gigabytes */
4353 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4354 ratio
= int_sqrt(10 * gb
);
4358 zone
->inactive_ratio
= ratio
;
4363 * Initialise min_free_kbytes.
4365 * For small machines we want it small (128k min). For large machines
4366 * we want it large (64MB max). But it is not linear, because network
4367 * bandwidth does not increase linearly with machine size. We use
4369 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4370 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4386 static int __init
init_per_zone_pages_min(void)
4388 unsigned long lowmem_kbytes
;
4390 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4392 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4393 if (min_free_kbytes
< 128)
4394 min_free_kbytes
= 128;
4395 if (min_free_kbytes
> 65536)
4396 min_free_kbytes
= 65536;
4397 setup_per_zone_pages_min();
4398 setup_per_zone_lowmem_reserve();
4399 setup_per_zone_inactive_ratio();
4402 module_init(init_per_zone_pages_min
)
4405 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4406 * that we can call two helper functions whenever min_free_kbytes
4409 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4410 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4412 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4414 setup_per_zone_pages_min();
4419 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4420 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4425 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4430 zone
->min_unmapped_pages
= (zone
->present_pages
*
4431 sysctl_min_unmapped_ratio
) / 100;
4435 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4436 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4441 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4446 zone
->min_slab_pages
= (zone
->present_pages
*
4447 sysctl_min_slab_ratio
) / 100;
4453 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4454 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4455 * whenever sysctl_lowmem_reserve_ratio changes.
4457 * The reserve ratio obviously has absolutely no relation with the
4458 * pages_min watermarks. The lowmem reserve ratio can only make sense
4459 * if in function of the boot time zone sizes.
4461 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4462 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4464 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4465 setup_per_zone_lowmem_reserve();
4470 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4471 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4472 * can have before it gets flushed back to buddy allocator.
4475 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4476 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4482 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4483 if (!write
|| (ret
== -EINVAL
))
4485 for_each_zone(zone
) {
4486 for_each_online_cpu(cpu
) {
4488 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4489 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4495 int hashdist
= HASHDIST_DEFAULT
;
4498 static int __init
set_hashdist(char *str
)
4502 hashdist
= simple_strtoul(str
, &str
, 0);
4505 __setup("hashdist=", set_hashdist
);
4509 * allocate a large system hash table from bootmem
4510 * - it is assumed that the hash table must contain an exact power-of-2
4511 * quantity of entries
4512 * - limit is the number of hash buckets, not the total allocation size
4514 void *__init
alloc_large_system_hash(const char *tablename
,
4515 unsigned long bucketsize
,
4516 unsigned long numentries
,
4519 unsigned int *_hash_shift
,
4520 unsigned int *_hash_mask
,
4521 unsigned long limit
)
4523 unsigned long long max
= limit
;
4524 unsigned long log2qty
, size
;
4527 /* allow the kernel cmdline to have a say */
4529 /* round applicable memory size up to nearest megabyte */
4530 numentries
= nr_kernel_pages
;
4531 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4532 numentries
>>= 20 - PAGE_SHIFT
;
4533 numentries
<<= 20 - PAGE_SHIFT
;
4535 /* limit to 1 bucket per 2^scale bytes of low memory */
4536 if (scale
> PAGE_SHIFT
)
4537 numentries
>>= (scale
- PAGE_SHIFT
);
4539 numentries
<<= (PAGE_SHIFT
- scale
);
4541 /* Make sure we've got at least a 0-order allocation.. */
4542 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4543 numentries
= PAGE_SIZE
/ bucketsize
;
4545 numentries
= roundup_pow_of_two(numentries
);
4547 /* limit allocation size to 1/16 total memory by default */
4549 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4550 do_div(max
, bucketsize
);
4553 if (numentries
> max
)
4556 log2qty
= ilog2(numentries
);
4559 size
= bucketsize
<< log2qty
;
4560 if (flags
& HASH_EARLY
)
4561 table
= alloc_bootmem_nopanic(size
);
4563 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4565 unsigned long order
= get_order(size
);
4566 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4568 * If bucketsize is not a power-of-two, we may free
4569 * some pages at the end of hash table.
4572 unsigned long alloc_end
= (unsigned long)table
+
4573 (PAGE_SIZE
<< order
);
4574 unsigned long used
= (unsigned long)table
+
4576 split_page(virt_to_page(table
), order
);
4577 while (used
< alloc_end
) {
4583 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4586 panic("Failed to allocate %s hash table\n", tablename
);
4588 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4591 ilog2(size
) - PAGE_SHIFT
,
4595 *_hash_shift
= log2qty
;
4597 *_hash_mask
= (1 << log2qty
) - 1;
4602 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4603 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4606 #ifdef CONFIG_SPARSEMEM
4607 return __pfn_to_section(pfn
)->pageblock_flags
;
4609 return zone
->pageblock_flags
;
4610 #endif /* CONFIG_SPARSEMEM */
4613 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4615 #ifdef CONFIG_SPARSEMEM
4616 pfn
&= (PAGES_PER_SECTION
-1);
4617 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4619 pfn
= pfn
- zone
->zone_start_pfn
;
4620 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4621 #endif /* CONFIG_SPARSEMEM */
4625 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4626 * @page: The page within the block of interest
4627 * @start_bitidx: The first bit of interest to retrieve
4628 * @end_bitidx: The last bit of interest
4629 * returns pageblock_bits flags
4631 unsigned long get_pageblock_flags_group(struct page
*page
,
4632 int start_bitidx
, int end_bitidx
)
4635 unsigned long *bitmap
;
4636 unsigned long pfn
, bitidx
;
4637 unsigned long flags
= 0;
4638 unsigned long value
= 1;
4640 zone
= page_zone(page
);
4641 pfn
= page_to_pfn(page
);
4642 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4643 bitidx
= pfn_to_bitidx(zone
, pfn
);
4645 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4646 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4653 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4654 * @page: The page within the block of interest
4655 * @start_bitidx: The first bit of interest
4656 * @end_bitidx: The last bit of interest
4657 * @flags: The flags to set
4659 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4660 int start_bitidx
, int end_bitidx
)
4663 unsigned long *bitmap
;
4664 unsigned long pfn
, bitidx
;
4665 unsigned long value
= 1;
4667 zone
= page_zone(page
);
4668 pfn
= page_to_pfn(page
);
4669 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4670 bitidx
= pfn_to_bitidx(zone
, pfn
);
4671 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4672 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4674 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4676 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4678 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4682 * This is designed as sub function...plz see page_isolation.c also.
4683 * set/clear page block's type to be ISOLATE.
4684 * page allocater never alloc memory from ISOLATE block.
4687 int set_migratetype_isolate(struct page
*page
)
4690 unsigned long flags
;
4693 zone
= page_zone(page
);
4694 spin_lock_irqsave(&zone
->lock
, flags
);
4696 * In future, more migrate types will be able to be isolation target.
4698 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4700 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4701 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4704 spin_unlock_irqrestore(&zone
->lock
, flags
);
4710 void unset_migratetype_isolate(struct page
*page
)
4713 unsigned long flags
;
4714 zone
= page_zone(page
);
4715 spin_lock_irqsave(&zone
->lock
, flags
);
4716 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4718 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4719 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4721 spin_unlock_irqrestore(&zone
->lock
, flags
);
4724 #ifdef CONFIG_MEMORY_HOTREMOVE
4726 * All pages in the range must be isolated before calling this.
4729 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4735 unsigned long flags
;
4736 /* find the first valid pfn */
4737 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4742 zone
= page_zone(pfn_to_page(pfn
));
4743 spin_lock_irqsave(&zone
->lock
, flags
);
4745 while (pfn
< end_pfn
) {
4746 if (!pfn_valid(pfn
)) {
4750 page
= pfn_to_page(pfn
);
4751 BUG_ON(page_count(page
));
4752 BUG_ON(!PageBuddy(page
));
4753 order
= page_order(page
);
4754 #ifdef CONFIG_DEBUG_VM
4755 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4756 pfn
, 1 << order
, end_pfn
);
4758 list_del(&page
->lru
);
4759 rmv_page_order(page
);
4760 zone
->free_area
[order
].nr_free
--;
4761 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4763 for (i
= 0; i
< (1 << order
); i
++)
4764 SetPageReserved((page
+i
));
4765 pfn
+= (1 << order
);
4767 spin_unlock_irqrestore(&zone
->lock
, flags
);