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/kmemcheck.h>
27 #include <linux/module.h>
28 #include <linux/suspend.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/slab.h>
32 #include <linux/oom.h>
33 #include <linux/notifier.h>
34 #include <linux/topology.h>
35 #include <linux/sysctl.h>
36 #include <linux/cpu.h>
37 #include <linux/cpuset.h>
38 #include <linux/memory_hotplug.h>
39 #include <linux/nodemask.h>
40 #include <linux/vmalloc.h>
41 #include <linux/mempolicy.h>
42 #include <linux/stop_machine.h>
43 #include <linux/sort.h>
44 #include <linux/pfn.h>
45 #include <linux/backing-dev.h>
46 #include <linux/fault-inject.h>
47 #include <linux/page-isolation.h>
48 #include <linux/page_cgroup.h>
49 #include <linux/debugobjects.h>
50 #include <linux/kmemleak.h>
52 #include <asm/tlbflush.h>
53 #include <asm/div64.h>
57 * Array of node states.
59 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
60 [N_POSSIBLE
] = NODE_MASK_ALL
,
61 [N_ONLINE
] = { { [0] = 1UL } },
63 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
65 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
67 [N_CPU
] = { { [0] = 1UL } },
70 EXPORT_SYMBOL(node_states
);
72 unsigned long totalram_pages __read_mostly
;
73 unsigned long totalreserve_pages __read_mostly
;
74 unsigned long highest_memmap_pfn __read_mostly
;
75 int percpu_pagelist_fraction
;
76 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
78 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
79 int pageblock_order __read_mostly
;
82 static void __free_pages_ok(struct page
*page
, unsigned int order
);
85 * results with 256, 32 in the lowmem_reserve sysctl:
86 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
87 * 1G machine -> (16M dma, 784M normal, 224M high)
88 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
89 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
90 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
92 * TBD: should special case ZONE_DMA32 machines here - in those we normally
93 * don't need any ZONE_NORMAL reservation
95 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
96 #ifdef CONFIG_ZONE_DMA
99 #ifdef CONFIG_ZONE_DMA32
102 #ifdef CONFIG_HIGHMEM
108 EXPORT_SYMBOL(totalram_pages
);
110 static char * const zone_names
[MAX_NR_ZONES
] = {
111 #ifdef CONFIG_ZONE_DMA
114 #ifdef CONFIG_ZONE_DMA32
118 #ifdef CONFIG_HIGHMEM
124 int min_free_kbytes
= 1024;
126 unsigned long __meminitdata nr_kernel_pages
;
127 unsigned long __meminitdata nr_all_pages
;
128 static unsigned long __meminitdata dma_reserve
;
130 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
132 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
133 * ranges of memory (RAM) that may be registered with add_active_range().
134 * Ranges passed to add_active_range() will be merged if possible
135 * so the number of times add_active_range() can be called is
136 * related to the number of nodes and the number of holes
138 #ifdef CONFIG_MAX_ACTIVE_REGIONS
139 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
140 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
142 #if MAX_NUMNODES >= 32
143 /* If there can be many nodes, allow up to 50 holes per node */
144 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
146 /* By default, allow up to 256 distinct regions */
147 #define MAX_ACTIVE_REGIONS 256
151 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
152 static int __meminitdata nr_nodemap_entries
;
153 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
154 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
155 static unsigned long __initdata required_kernelcore
;
156 static unsigned long __initdata required_movablecore
;
157 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
159 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
161 EXPORT_SYMBOL(movable_zone
);
162 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
165 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
166 int nr_online_nodes __read_mostly
= 1;
167 EXPORT_SYMBOL(nr_node_ids
);
168 EXPORT_SYMBOL(nr_online_nodes
);
171 int page_group_by_mobility_disabled __read_mostly
;
173 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
176 if (unlikely(page_group_by_mobility_disabled
))
177 migratetype
= MIGRATE_UNMOVABLE
;
179 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
180 PB_migrate
, PB_migrate_end
);
183 bool oom_killer_disabled __read_mostly
;
185 #ifdef CONFIG_DEBUG_VM
186 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
190 unsigned long pfn
= page_to_pfn(page
);
193 seq
= zone_span_seqbegin(zone
);
194 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
196 else if (pfn
< zone
->zone_start_pfn
)
198 } while (zone_span_seqretry(zone
, seq
));
203 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
205 if (!pfn_valid_within(page_to_pfn(page
)))
207 if (zone
!= page_zone(page
))
213 * Temporary debugging check for pages not lying within a given zone.
215 static int bad_range(struct zone
*zone
, struct page
*page
)
217 if (page_outside_zone_boundaries(zone
, page
))
219 if (!page_is_consistent(zone
, page
))
225 static inline int bad_range(struct zone
*zone
, struct page
*page
)
231 static void bad_page(struct page
*page
)
233 static unsigned long resume
;
234 static unsigned long nr_shown
;
235 static unsigned long nr_unshown
;
238 * Allow a burst of 60 reports, then keep quiet for that minute;
239 * or allow a steady drip of one report per second.
241 if (nr_shown
== 60) {
242 if (time_before(jiffies
, resume
)) {
248 "BUG: Bad page state: %lu messages suppressed\n",
255 resume
= jiffies
+ 60 * HZ
;
257 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
258 current
->comm
, page_to_pfn(page
));
260 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
261 page
, (void *)page
->flags
, page_count(page
),
262 page_mapcount(page
), page
->mapping
, page
->index
);
266 /* Leave bad fields for debug, except PageBuddy could make trouble */
267 __ClearPageBuddy(page
);
268 add_taint(TAINT_BAD_PAGE
);
272 * Higher-order pages are called "compound pages". They are structured thusly:
274 * The first PAGE_SIZE page is called the "head page".
276 * The remaining PAGE_SIZE pages are called "tail pages".
278 * All pages have PG_compound set. All pages have their ->private pointing at
279 * the head page (even the head page has this).
281 * The first tail page's ->lru.next holds the address of the compound page's
282 * put_page() function. Its ->lru.prev holds the order of allocation.
283 * This usage means that zero-order pages may not be compound.
286 static void free_compound_page(struct page
*page
)
288 __free_pages_ok(page
, compound_order(page
));
291 void prep_compound_page(struct page
*page
, unsigned long order
)
294 int nr_pages
= 1 << order
;
296 set_compound_page_dtor(page
, free_compound_page
);
297 set_compound_order(page
, order
);
299 for (i
= 1; i
< nr_pages
; i
++) {
300 struct page
*p
= page
+ i
;
303 p
->first_page
= page
;
307 static int destroy_compound_page(struct page
*page
, unsigned long order
)
310 int nr_pages
= 1 << order
;
313 if (unlikely(compound_order(page
) != order
) ||
314 unlikely(!PageHead(page
))) {
319 __ClearPageHead(page
);
321 for (i
= 1; i
< nr_pages
; i
++) {
322 struct page
*p
= page
+ i
;
324 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
334 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
339 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
340 * and __GFP_HIGHMEM from hard or soft interrupt context.
342 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
343 for (i
= 0; i
< (1 << order
); i
++)
344 clear_highpage(page
+ i
);
347 static inline void set_page_order(struct page
*page
, int order
)
349 set_page_private(page
, order
);
350 __SetPageBuddy(page
);
353 static inline void rmv_page_order(struct page
*page
)
355 __ClearPageBuddy(page
);
356 set_page_private(page
, 0);
360 * Locate the struct page for both the matching buddy in our
361 * pair (buddy1) and the combined O(n+1) page they form (page).
363 * 1) Any buddy B1 will have an order O twin B2 which satisfies
364 * the following equation:
366 * For example, if the starting buddy (buddy2) is #8 its order
368 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
370 * 2) Any buddy B will have an order O+1 parent P which
371 * satisfies the following equation:
374 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
376 static inline struct page
*
377 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
379 unsigned long buddy_idx
= page_idx
^ (1 << order
);
381 return page
+ (buddy_idx
- page_idx
);
384 static inline unsigned long
385 __find_combined_index(unsigned long page_idx
, unsigned int order
)
387 return (page_idx
& ~(1 << order
));
391 * This function checks whether a page is free && is the buddy
392 * we can do coalesce a page and its buddy if
393 * (a) the buddy is not in a hole &&
394 * (b) the buddy is in the buddy system &&
395 * (c) a page and its buddy have the same order &&
396 * (d) a page and its buddy are in the same zone.
398 * For recording whether a page is in the buddy system, we use PG_buddy.
399 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
401 * For recording page's order, we use page_private(page).
403 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
406 if (!pfn_valid_within(page_to_pfn(buddy
)))
409 if (page_zone_id(page
) != page_zone_id(buddy
))
412 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
413 VM_BUG_ON(page_count(buddy
) != 0);
420 * Freeing function for a buddy system allocator.
422 * The concept of a buddy system is to maintain direct-mapped table
423 * (containing bit values) for memory blocks of various "orders".
424 * The bottom level table contains the map for the smallest allocatable
425 * units of memory (here, pages), and each level above it describes
426 * pairs of units from the levels below, hence, "buddies".
427 * At a high level, all that happens here is marking the table entry
428 * at the bottom level available, and propagating the changes upward
429 * as necessary, plus some accounting needed to play nicely with other
430 * parts of the VM system.
431 * At each level, we keep a list of pages, which are heads of continuous
432 * free pages of length of (1 << order) and marked with PG_buddy. Page's
433 * order is recorded in page_private(page) field.
434 * So when we are allocating or freeing one, we can derive the state of the
435 * other. That is, if we allocate a small block, and both were
436 * free, the remainder of the region must be split into blocks.
437 * If a block is freed, and its buddy is also free, then this
438 * triggers coalescing into a block of larger size.
443 static inline void __free_one_page(struct page
*page
,
444 struct zone
*zone
, unsigned int order
,
447 unsigned long page_idx
;
449 if (unlikely(PageCompound(page
)))
450 if (unlikely(destroy_compound_page(page
, order
)))
453 VM_BUG_ON(migratetype
== -1);
455 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
457 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
458 VM_BUG_ON(bad_range(zone
, page
));
460 while (order
< MAX_ORDER
-1) {
461 unsigned long combined_idx
;
464 buddy
= __page_find_buddy(page
, page_idx
, order
);
465 if (!page_is_buddy(page
, buddy
, order
))
468 /* Our buddy is free, merge with it and move up one order. */
469 list_del(&buddy
->lru
);
470 zone
->free_area
[order
].nr_free
--;
471 rmv_page_order(buddy
);
472 combined_idx
= __find_combined_index(page_idx
, order
);
473 page
= page
+ (combined_idx
- page_idx
);
474 page_idx
= combined_idx
;
477 set_page_order(page
, order
);
479 &zone
->free_area
[order
].free_list
[migratetype
]);
480 zone
->free_area
[order
].nr_free
++;
483 #ifdef CONFIG_HAVE_MLOCKED_PAGE_BIT
485 * free_page_mlock() -- clean up attempts to free and mlocked() page.
486 * Page should not be on lru, so no need to fix that up.
487 * free_pages_check() will verify...
489 static inline void free_page_mlock(struct page
*page
)
491 __dec_zone_page_state(page
, NR_MLOCK
);
492 __count_vm_event(UNEVICTABLE_MLOCKFREED
);
495 static void free_page_mlock(struct page
*page
) { }
498 static inline int free_pages_check(struct page
*page
)
500 if (unlikely(page_mapcount(page
) |
501 (page
->mapping
!= NULL
) |
502 (atomic_read(&page
->_count
) != 0) |
503 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
507 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
508 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
513 * Frees a list of pages.
514 * Assumes all pages on list are in same zone, and of same order.
515 * count is the number of pages to free.
517 * If the zone was previously in an "all pages pinned" state then look to
518 * see if this freeing clears that state.
520 * And clear the zone's pages_scanned counter, to hold off the "all pages are
521 * pinned" detection logic.
523 static void free_pages_bulk(struct zone
*zone
, int count
,
524 struct list_head
*list
, int order
)
526 spin_lock(&zone
->lock
);
527 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
528 zone
->pages_scanned
= 0;
530 __mod_zone_page_state(zone
, NR_FREE_PAGES
, count
<< order
);
534 VM_BUG_ON(list_empty(list
));
535 page
= list_entry(list
->prev
, struct page
, lru
);
536 /* have to delete it as __free_one_page list manipulates */
537 list_del(&page
->lru
);
538 __free_one_page(page
, zone
, order
, page_private(page
));
540 spin_unlock(&zone
->lock
);
543 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
546 spin_lock(&zone
->lock
);
547 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
548 zone
->pages_scanned
= 0;
550 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1 << order
);
551 __free_one_page(page
, zone
, order
, migratetype
);
552 spin_unlock(&zone
->lock
);
555 static void __free_pages_ok(struct page
*page
, unsigned int order
)
560 int wasMlocked
= TestClearPageMlocked(page
);
562 kmemcheck_free_shadow(page
, order
);
564 for (i
= 0 ; i
< (1 << order
) ; ++i
)
565 bad
+= free_pages_check(page
+ i
);
569 if (!PageHighMem(page
)) {
570 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
571 debug_check_no_obj_freed(page_address(page
),
574 arch_free_page(page
, order
);
575 kernel_map_pages(page
, 1 << order
, 0);
577 local_irq_save(flags
);
578 if (unlikely(wasMlocked
))
579 free_page_mlock(page
);
580 __count_vm_events(PGFREE
, 1 << order
);
581 free_one_page(page_zone(page
), page
, order
,
582 get_pageblock_migratetype(page
));
583 local_irq_restore(flags
);
587 * permit the bootmem allocator to evade page validation on high-order frees
589 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
592 __ClearPageReserved(page
);
593 set_page_count(page
, 0);
594 set_page_refcounted(page
);
600 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
601 struct page
*p
= &page
[loop
];
603 if (loop
+ 1 < BITS_PER_LONG
)
605 __ClearPageReserved(p
);
606 set_page_count(p
, 0);
609 set_page_refcounted(page
);
610 __free_pages(page
, order
);
616 * The order of subdivision here is critical for the IO subsystem.
617 * Please do not alter this order without good reasons and regression
618 * testing. Specifically, as large blocks of memory are subdivided,
619 * the order in which smaller blocks are delivered depends on the order
620 * they're subdivided in this function. This is the primary factor
621 * influencing the order in which pages are delivered to the IO
622 * subsystem according to empirical testing, and this is also justified
623 * by considering the behavior of a buddy system containing a single
624 * large block of memory acted on by a series of small allocations.
625 * This behavior is a critical factor in sglist merging's success.
629 static inline void expand(struct zone
*zone
, struct page
*page
,
630 int low
, int high
, struct free_area
*area
,
633 unsigned long size
= 1 << high
;
639 VM_BUG_ON(bad_range(zone
, &page
[size
]));
640 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
642 set_page_order(&page
[size
], high
);
647 * This page is about to be returned from the page allocator
649 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
651 if (unlikely(page_mapcount(page
) |
652 (page
->mapping
!= NULL
) |
653 (atomic_read(&page
->_count
) != 0) |
654 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
659 set_page_private(page
, 0);
660 set_page_refcounted(page
);
662 arch_alloc_page(page
, order
);
663 kernel_map_pages(page
, 1 << order
, 1);
665 if (gfp_flags
& __GFP_ZERO
)
666 prep_zero_page(page
, order
, gfp_flags
);
668 if (order
&& (gfp_flags
& __GFP_COMP
))
669 prep_compound_page(page
, order
);
675 * Go through the free lists for the given migratetype and remove
676 * the smallest available page from the freelists
679 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
682 unsigned int current_order
;
683 struct free_area
* area
;
686 /* Find a page of the appropriate size in the preferred list */
687 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
688 area
= &(zone
->free_area
[current_order
]);
689 if (list_empty(&area
->free_list
[migratetype
]))
692 page
= list_entry(area
->free_list
[migratetype
].next
,
694 list_del(&page
->lru
);
695 rmv_page_order(page
);
697 expand(zone
, page
, order
, current_order
, area
, migratetype
);
706 * This array describes the order lists are fallen back to when
707 * the free lists for the desirable migrate type are depleted
709 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
710 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
711 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
712 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
713 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
717 * Move the free pages in a range to the free lists of the requested type.
718 * Note that start_page and end_pages are not aligned on a pageblock
719 * boundary. If alignment is required, use move_freepages_block()
721 static int move_freepages(struct zone
*zone
,
722 struct page
*start_page
, struct page
*end_page
,
729 #ifndef CONFIG_HOLES_IN_ZONE
731 * page_zone is not safe to call in this context when
732 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
733 * anyway as we check zone boundaries in move_freepages_block().
734 * Remove at a later date when no bug reports exist related to
735 * grouping pages by mobility
737 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
740 for (page
= start_page
; page
<= end_page
;) {
741 /* Make sure we are not inadvertently changing nodes */
742 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
744 if (!pfn_valid_within(page_to_pfn(page
))) {
749 if (!PageBuddy(page
)) {
754 order
= page_order(page
);
755 list_del(&page
->lru
);
757 &zone
->free_area
[order
].free_list
[migratetype
]);
759 pages_moved
+= 1 << order
;
765 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
768 unsigned long start_pfn
, end_pfn
;
769 struct page
*start_page
, *end_page
;
771 start_pfn
= page_to_pfn(page
);
772 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
773 start_page
= pfn_to_page(start_pfn
);
774 end_page
= start_page
+ pageblock_nr_pages
- 1;
775 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
777 /* Do not cross zone boundaries */
778 if (start_pfn
< zone
->zone_start_pfn
)
780 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
783 return move_freepages(zone
, start_page
, end_page
, migratetype
);
786 /* Remove an element from the buddy allocator from the fallback list */
787 static inline struct page
*
788 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
790 struct free_area
* area
;
795 /* Find the largest possible block of pages in the other list */
796 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
798 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
799 migratetype
= fallbacks
[start_migratetype
][i
];
801 /* MIGRATE_RESERVE handled later if necessary */
802 if (migratetype
== MIGRATE_RESERVE
)
805 area
= &(zone
->free_area
[current_order
]);
806 if (list_empty(&area
->free_list
[migratetype
]))
809 page
= list_entry(area
->free_list
[migratetype
].next
,
814 * If breaking a large block of pages, move all free
815 * pages to the preferred allocation list. If falling
816 * back for a reclaimable kernel allocation, be more
817 * agressive about taking ownership of free pages
819 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
820 start_migratetype
== MIGRATE_RECLAIMABLE
||
821 page_group_by_mobility_disabled
) {
823 pages
= move_freepages_block(zone
, page
,
826 /* Claim the whole block if over half of it is free */
827 if (pages
>= (1 << (pageblock_order
-1)) ||
828 page_group_by_mobility_disabled
)
829 set_pageblock_migratetype(page
,
832 migratetype
= start_migratetype
;
835 /* Remove the page from the freelists */
836 list_del(&page
->lru
);
837 rmv_page_order(page
);
839 if (current_order
== pageblock_order
)
840 set_pageblock_migratetype(page
,
843 expand(zone
, page
, order
, current_order
, area
, migratetype
);
852 * Do the hard work of removing an element from the buddy allocator.
853 * Call me with the zone->lock already held.
855 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
861 page
= __rmqueue_smallest(zone
, order
, migratetype
);
863 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
864 page
= __rmqueue_fallback(zone
, order
, migratetype
);
867 * Use MIGRATE_RESERVE rather than fail an allocation. goto
868 * is used because __rmqueue_smallest is an inline function
869 * and we want just one call site
872 migratetype
= MIGRATE_RESERVE
;
881 * Obtain a specified number of elements from the buddy allocator, all under
882 * a single hold of the lock, for efficiency. Add them to the supplied list.
883 * Returns the number of new pages which were placed at *list.
885 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
886 unsigned long count
, struct list_head
*list
,
887 int migratetype
, int cold
)
891 spin_lock(&zone
->lock
);
892 for (i
= 0; i
< count
; ++i
) {
893 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
894 if (unlikely(page
== NULL
))
898 * Split buddy pages returned by expand() are received here
899 * in physical page order. The page is added to the callers and
900 * list and the list head then moves forward. From the callers
901 * perspective, the linked list is ordered by page number in
902 * some conditions. This is useful for IO devices that can
903 * merge IO requests if the physical pages are ordered
906 if (likely(cold
== 0))
907 list_add(&page
->lru
, list
);
909 list_add_tail(&page
->lru
, list
);
910 set_page_private(page
, migratetype
);
913 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
914 spin_unlock(&zone
->lock
);
920 * Called from the vmstat counter updater to drain pagesets of this
921 * currently executing processor on remote nodes after they have
924 * Note that this function must be called with the thread pinned to
925 * a single processor.
927 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
932 local_irq_save(flags
);
933 if (pcp
->count
>= pcp
->batch
)
934 to_drain
= pcp
->batch
;
936 to_drain
= pcp
->count
;
937 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
938 pcp
->count
-= to_drain
;
939 local_irq_restore(flags
);
944 * Drain pages of the indicated processor.
946 * The processor must either be the current processor and the
947 * thread pinned to the current processor or a processor that
950 static void drain_pages(unsigned int cpu
)
955 for_each_populated_zone(zone
) {
956 struct per_cpu_pageset
*pset
;
957 struct per_cpu_pages
*pcp
;
959 pset
= zone_pcp(zone
, cpu
);
962 local_irq_save(flags
);
963 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
965 local_irq_restore(flags
);
970 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
972 void drain_local_pages(void *arg
)
974 drain_pages(smp_processor_id());
978 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
980 void drain_all_pages(void)
982 on_each_cpu(drain_local_pages
, NULL
, 1);
985 #ifdef CONFIG_HIBERNATION
987 void mark_free_pages(struct zone
*zone
)
989 unsigned long pfn
, max_zone_pfn
;
992 struct list_head
*curr
;
994 if (!zone
->spanned_pages
)
997 spin_lock_irqsave(&zone
->lock
, flags
);
999 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1000 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1001 if (pfn_valid(pfn
)) {
1002 struct page
*page
= pfn_to_page(pfn
);
1004 if (!swsusp_page_is_forbidden(page
))
1005 swsusp_unset_page_free(page
);
1008 for_each_migratetype_order(order
, t
) {
1009 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1012 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1013 for (i
= 0; i
< (1UL << order
); i
++)
1014 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1017 spin_unlock_irqrestore(&zone
->lock
, flags
);
1019 #endif /* CONFIG_PM */
1022 * Free a 0-order page
1024 static void free_hot_cold_page(struct page
*page
, int cold
)
1026 struct zone
*zone
= page_zone(page
);
1027 struct per_cpu_pages
*pcp
;
1028 unsigned long flags
;
1029 int wasMlocked
= TestClearPageMlocked(page
);
1031 kmemcheck_free_shadow(page
, 0);
1034 page
->mapping
= NULL
;
1035 if (free_pages_check(page
))
1038 if (!PageHighMem(page
)) {
1039 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1040 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1042 arch_free_page(page
, 0);
1043 kernel_map_pages(page
, 1, 0);
1045 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1046 set_page_private(page
, get_pageblock_migratetype(page
));
1047 local_irq_save(flags
);
1048 if (unlikely(wasMlocked
))
1049 free_page_mlock(page
);
1050 __count_vm_event(PGFREE
);
1053 list_add_tail(&page
->lru
, &pcp
->list
);
1055 list_add(&page
->lru
, &pcp
->list
);
1057 if (pcp
->count
>= pcp
->high
) {
1058 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1059 pcp
->count
-= pcp
->batch
;
1061 local_irq_restore(flags
);
1065 void free_hot_page(struct page
*page
)
1067 free_hot_cold_page(page
, 0);
1070 void free_cold_page(struct page
*page
)
1072 free_hot_cold_page(page
, 1);
1076 * split_page takes a non-compound higher-order page, and splits it into
1077 * n (1<<order) sub-pages: page[0..n]
1078 * Each sub-page must be freed individually.
1080 * Note: this is probably too low level an operation for use in drivers.
1081 * Please consult with lkml before using this in your driver.
1083 void split_page(struct page
*page
, unsigned int order
)
1087 VM_BUG_ON(PageCompound(page
));
1088 VM_BUG_ON(!page_count(page
));
1090 #ifdef CONFIG_KMEMCHECK
1092 * Split shadow pages too, because free(page[0]) would
1093 * otherwise free the whole shadow.
1095 if (kmemcheck_page_is_tracked(page
))
1096 split_page(virt_to_page(page
[0].shadow
), order
);
1099 for (i
= 1; i
< (1 << order
); i
++)
1100 set_page_refcounted(page
+ i
);
1104 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1105 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1109 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1110 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1113 unsigned long flags
;
1115 int cold
= !!(gfp_flags
& __GFP_COLD
);
1120 if (likely(order
== 0)) {
1121 struct per_cpu_pages
*pcp
;
1123 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1124 local_irq_save(flags
);
1126 pcp
->count
= rmqueue_bulk(zone
, 0,
1127 pcp
->batch
, &pcp
->list
,
1129 if (unlikely(!pcp
->count
))
1133 /* Find a page of the appropriate migrate type */
1135 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1136 if (page_private(page
) == migratetype
)
1139 list_for_each_entry(page
, &pcp
->list
, lru
)
1140 if (page_private(page
) == migratetype
)
1144 /* Allocate more to the pcp list if necessary */
1145 if (unlikely(&page
->lru
== &pcp
->list
)) {
1146 pcp
->count
+= rmqueue_bulk(zone
, 0,
1147 pcp
->batch
, &pcp
->list
,
1149 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1152 list_del(&page
->lru
);
1155 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1157 * __GFP_NOFAIL is not to be used in new code.
1159 * All __GFP_NOFAIL callers should be fixed so that they
1160 * properly detect and handle allocation failures.
1162 * We most definitely don't want callers attempting to
1163 * allocate greater than order-1 page units with
1166 WARN_ON_ONCE(order
> 1);
1168 spin_lock_irqsave(&zone
->lock
, flags
);
1169 page
= __rmqueue(zone
, order
, migratetype
);
1170 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1 << order
));
1171 spin_unlock(&zone
->lock
);
1176 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1177 zone_statistics(preferred_zone
, zone
);
1178 local_irq_restore(flags
);
1181 VM_BUG_ON(bad_range(zone
, page
));
1182 if (prep_new_page(page
, order
, gfp_flags
))
1187 local_irq_restore(flags
);
1192 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1193 #define ALLOC_WMARK_MIN WMARK_MIN
1194 #define ALLOC_WMARK_LOW WMARK_LOW
1195 #define ALLOC_WMARK_HIGH WMARK_HIGH
1196 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1198 /* Mask to get the watermark bits */
1199 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1201 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1202 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1203 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1205 #ifdef CONFIG_FAIL_PAGE_ALLOC
1207 static struct fail_page_alloc_attr
{
1208 struct fault_attr attr
;
1210 u32 ignore_gfp_highmem
;
1211 u32 ignore_gfp_wait
;
1214 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1216 struct dentry
*ignore_gfp_highmem_file
;
1217 struct dentry
*ignore_gfp_wait_file
;
1218 struct dentry
*min_order_file
;
1220 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1222 } fail_page_alloc
= {
1223 .attr
= FAULT_ATTR_INITIALIZER
,
1224 .ignore_gfp_wait
= 1,
1225 .ignore_gfp_highmem
= 1,
1229 static int __init
setup_fail_page_alloc(char *str
)
1231 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1233 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1235 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1237 if (order
< fail_page_alloc
.min_order
)
1239 if (gfp_mask
& __GFP_NOFAIL
)
1241 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1243 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1246 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1249 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1251 static int __init
fail_page_alloc_debugfs(void)
1253 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1257 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1261 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1263 fail_page_alloc
.ignore_gfp_wait_file
=
1264 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1265 &fail_page_alloc
.ignore_gfp_wait
);
1267 fail_page_alloc
.ignore_gfp_highmem_file
=
1268 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1269 &fail_page_alloc
.ignore_gfp_highmem
);
1270 fail_page_alloc
.min_order_file
=
1271 debugfs_create_u32("min-order", mode
, dir
,
1272 &fail_page_alloc
.min_order
);
1274 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1275 !fail_page_alloc
.ignore_gfp_highmem_file
||
1276 !fail_page_alloc
.min_order_file
) {
1278 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1279 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1280 debugfs_remove(fail_page_alloc
.min_order_file
);
1281 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1287 late_initcall(fail_page_alloc_debugfs
);
1289 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1291 #else /* CONFIG_FAIL_PAGE_ALLOC */
1293 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1298 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1301 * Return 1 if free pages are above 'mark'. This takes into account the order
1302 * of the allocation.
1304 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1305 int classzone_idx
, int alloc_flags
)
1307 /* free_pages my go negative - that's OK */
1309 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1312 if (alloc_flags
& ALLOC_HIGH
)
1314 if (alloc_flags
& ALLOC_HARDER
)
1317 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1319 for (o
= 0; o
< order
; o
++) {
1320 /* At the next order, this order's pages become unavailable */
1321 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1323 /* Require fewer higher order pages to be free */
1326 if (free_pages
<= min
)
1334 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1335 * skip over zones that are not allowed by the cpuset, or that have
1336 * been recently (in last second) found to be nearly full. See further
1337 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1338 * that have to skip over a lot of full or unallowed zones.
1340 * If the zonelist cache is present in the passed in zonelist, then
1341 * returns a pointer to the allowed node mask (either the current
1342 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1344 * If the zonelist cache is not available for this zonelist, does
1345 * nothing and returns NULL.
1347 * If the fullzones BITMAP in the zonelist cache is stale (more than
1348 * a second since last zap'd) then we zap it out (clear its bits.)
1350 * We hold off even calling zlc_setup, until after we've checked the
1351 * first zone in the zonelist, on the theory that most allocations will
1352 * be satisfied from that first zone, so best to examine that zone as
1353 * quickly as we can.
1355 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1357 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1358 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1360 zlc
= zonelist
->zlcache_ptr
;
1364 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1365 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1366 zlc
->last_full_zap
= jiffies
;
1369 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1370 &cpuset_current_mems_allowed
:
1371 &node_states
[N_HIGH_MEMORY
];
1372 return allowednodes
;
1376 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1377 * if it is worth looking at further for free memory:
1378 * 1) Check that the zone isn't thought to be full (doesn't have its
1379 * bit set in the zonelist_cache fullzones BITMAP).
1380 * 2) Check that the zones node (obtained from the zonelist_cache
1381 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1382 * Return true (non-zero) if zone is worth looking at further, or
1383 * else return false (zero) if it is not.
1385 * This check -ignores- the distinction between various watermarks,
1386 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1387 * found to be full for any variation of these watermarks, it will
1388 * be considered full for up to one second by all requests, unless
1389 * we are so low on memory on all allowed nodes that we are forced
1390 * into the second scan of the zonelist.
1392 * In the second scan we ignore this zonelist cache and exactly
1393 * apply the watermarks to all zones, even it is slower to do so.
1394 * We are low on memory in the second scan, and should leave no stone
1395 * unturned looking for a free page.
1397 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1398 nodemask_t
*allowednodes
)
1400 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1401 int i
; /* index of *z in zonelist zones */
1402 int n
; /* node that zone *z is on */
1404 zlc
= zonelist
->zlcache_ptr
;
1408 i
= z
- zonelist
->_zonerefs
;
1411 /* This zone is worth trying if it is allowed but not full */
1412 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1416 * Given 'z' scanning a zonelist, set the corresponding bit in
1417 * zlc->fullzones, so that subsequent attempts to allocate a page
1418 * from that zone don't waste time re-examining it.
1420 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1422 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1423 int i
; /* index of *z in zonelist zones */
1425 zlc
= zonelist
->zlcache_ptr
;
1429 i
= z
- zonelist
->_zonerefs
;
1431 set_bit(i
, zlc
->fullzones
);
1434 #else /* CONFIG_NUMA */
1436 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1441 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1442 nodemask_t
*allowednodes
)
1447 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1450 #endif /* CONFIG_NUMA */
1453 * get_page_from_freelist goes through the zonelist trying to allocate
1456 static struct page
*
1457 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1458 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1459 struct zone
*preferred_zone
, int migratetype
)
1462 struct page
*page
= NULL
;
1465 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1466 int zlc_active
= 0; /* set if using zonelist_cache */
1467 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1469 classzone_idx
= zone_idx(preferred_zone
);
1472 * Scan zonelist, looking for a zone with enough free.
1473 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1475 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1476 high_zoneidx
, nodemask
) {
1477 if (NUMA_BUILD
&& zlc_active
&&
1478 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1480 if ((alloc_flags
& ALLOC_CPUSET
) &&
1481 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1484 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1485 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1489 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1490 if (zone_watermark_ok(zone
, order
, mark
,
1491 classzone_idx
, alloc_flags
))
1494 if (zone_reclaim_mode
== 0)
1495 goto this_zone_full
;
1497 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1499 case ZONE_RECLAIM_NOSCAN
:
1502 case ZONE_RECLAIM_FULL
:
1503 /* scanned but unreclaimable */
1504 goto this_zone_full
;
1506 /* did we reclaim enough */
1507 if (!zone_watermark_ok(zone
, order
, mark
,
1508 classzone_idx
, alloc_flags
))
1509 goto this_zone_full
;
1514 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1515 gfp_mask
, migratetype
);
1520 zlc_mark_zone_full(zonelist
, z
);
1522 if (NUMA_BUILD
&& !did_zlc_setup
&& nr_online_nodes
> 1) {
1524 * we do zlc_setup after the first zone is tried but only
1525 * if there are multiple nodes make it worthwhile
1527 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1533 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1534 /* Disable zlc cache for second zonelist scan */
1542 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1543 unsigned long pages_reclaimed
)
1545 /* Do not loop if specifically requested */
1546 if (gfp_mask
& __GFP_NORETRY
)
1550 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1551 * means __GFP_NOFAIL, but that may not be true in other
1554 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1558 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1559 * specified, then we retry until we no longer reclaim any pages
1560 * (above), or we've reclaimed an order of pages at least as
1561 * large as the allocation's order. In both cases, if the
1562 * allocation still fails, we stop retrying.
1564 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
1568 * Don't let big-order allocations loop unless the caller
1569 * explicitly requests that.
1571 if (gfp_mask
& __GFP_NOFAIL
)
1577 static inline struct page
*
1578 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
1579 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1580 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1585 /* Acquire the OOM killer lock for the zones in zonelist */
1586 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1587 schedule_timeout_uninterruptible(1);
1592 * Go through the zonelist yet one more time, keep very high watermark
1593 * here, this is only to catch a parallel oom killing, we must fail if
1594 * we're still under heavy pressure.
1596 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1597 order
, zonelist
, high_zoneidx
,
1598 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
1599 preferred_zone
, migratetype
);
1603 /* The OOM killer will not help higher order allocs */
1604 if (order
> PAGE_ALLOC_COSTLY_ORDER
&& !(gfp_mask
& __GFP_NOFAIL
))
1607 /* Exhausted what can be done so it's blamo time */
1608 out_of_memory(zonelist
, gfp_mask
, order
);
1611 clear_zonelist_oom(zonelist
, gfp_mask
);
1615 /* The really slow allocator path where we enter direct reclaim */
1616 static inline struct page
*
1617 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1618 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1619 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1620 int migratetype
, unsigned long *did_some_progress
)
1622 struct page
*page
= NULL
;
1623 struct reclaim_state reclaim_state
;
1624 struct task_struct
*p
= current
;
1628 /* We now go into synchronous reclaim */
1629 cpuset_memory_pressure_bump();
1630 p
->flags
|= PF_MEMALLOC
;
1631 lockdep_set_current_reclaim_state(gfp_mask
);
1632 reclaim_state
.reclaimed_slab
= 0;
1633 p
->reclaim_state
= &reclaim_state
;
1635 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1637 p
->reclaim_state
= NULL
;
1638 lockdep_clear_current_reclaim_state();
1639 p
->flags
&= ~PF_MEMALLOC
;
1646 if (likely(*did_some_progress
))
1647 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1648 zonelist
, high_zoneidx
,
1649 alloc_flags
, preferred_zone
,
1655 * This is called in the allocator slow-path if the allocation request is of
1656 * sufficient urgency to ignore watermarks and take other desperate measures
1658 static inline struct page
*
1659 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1660 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1661 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1667 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1668 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
1669 preferred_zone
, migratetype
);
1671 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1672 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1673 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
1679 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
1680 enum zone_type high_zoneidx
)
1685 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1686 wakeup_kswapd(zone
, order
);
1690 gfp_to_alloc_flags(gfp_t gfp_mask
)
1692 struct task_struct
*p
= current
;
1693 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
1694 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1696 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1697 BUILD_BUG_ON(__GFP_HIGH
!= ALLOC_HIGH
);
1700 * The caller may dip into page reserves a bit more if the caller
1701 * cannot run direct reclaim, or if the caller has realtime scheduling
1702 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1703 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1705 alloc_flags
|= (gfp_mask
& __GFP_HIGH
);
1708 alloc_flags
|= ALLOC_HARDER
;
1710 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1711 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1713 alloc_flags
&= ~ALLOC_CPUSET
;
1714 } else if (unlikely(rt_task(p
)))
1715 alloc_flags
|= ALLOC_HARDER
;
1717 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
1718 if (!in_interrupt() &&
1719 ((p
->flags
& PF_MEMALLOC
) ||
1720 unlikely(test_thread_flag(TIF_MEMDIE
))))
1721 alloc_flags
|= ALLOC_NO_WATERMARKS
;
1727 static inline struct page
*
1728 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
1729 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1730 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1733 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1734 struct page
*page
= NULL
;
1736 unsigned long pages_reclaimed
= 0;
1737 unsigned long did_some_progress
;
1738 struct task_struct
*p
= current
;
1741 * In the slowpath, we sanity check order to avoid ever trying to
1742 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1743 * be using allocators in order of preference for an area that is
1746 if (order
>= MAX_ORDER
) {
1747 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
1752 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1753 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1754 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1755 * using a larger set of nodes after it has established that the
1756 * allowed per node queues are empty and that nodes are
1759 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1762 wake_all_kswapd(order
, zonelist
, high_zoneidx
);
1765 * OK, we're below the kswapd watermark and have kicked background
1766 * reclaim. Now things get more complex, so set up alloc_flags according
1767 * to how we want to proceed.
1769 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
1772 /* This is the last chance, in general, before the goto nopage. */
1773 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1774 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
1775 preferred_zone
, migratetype
);
1780 /* Allocate without watermarks if the context allows */
1781 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
1782 page
= __alloc_pages_high_priority(gfp_mask
, order
,
1783 zonelist
, high_zoneidx
, nodemask
,
1784 preferred_zone
, migratetype
);
1789 /* Atomic allocations - we can't balance anything */
1793 /* Avoid recursion of direct reclaim */
1794 if (p
->flags
& PF_MEMALLOC
)
1797 /* Avoid allocations with no watermarks from looping endlessly */
1798 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
1801 /* Try direct reclaim and then allocating */
1802 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
1803 zonelist
, high_zoneidx
,
1805 alloc_flags
, preferred_zone
,
1806 migratetype
, &did_some_progress
);
1811 * If we failed to make any progress reclaiming, then we are
1812 * running out of options and have to consider going OOM
1814 if (!did_some_progress
) {
1815 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1816 if (oom_killer_disabled
)
1818 page
= __alloc_pages_may_oom(gfp_mask
, order
,
1819 zonelist
, high_zoneidx
,
1820 nodemask
, preferred_zone
,
1826 * The OOM killer does not trigger for high-order
1827 * ~__GFP_NOFAIL allocations so if no progress is being
1828 * made, there are no other options and retrying is
1831 if (order
> PAGE_ALLOC_COSTLY_ORDER
&&
1832 !(gfp_mask
& __GFP_NOFAIL
))
1839 /* Check if we should retry the allocation */
1840 pages_reclaimed
+= did_some_progress
;
1841 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
1842 /* Wait for some write requests to complete then retry */
1843 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1848 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1849 printk(KERN_WARNING
"%s: page allocation failure."
1850 " order:%d, mode:0x%x\n",
1851 p
->comm
, order
, gfp_mask
);
1857 if (kmemcheck_enabled
)
1858 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
1864 * This is the 'heart' of the zoned buddy allocator.
1867 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
1868 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1870 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1871 struct zone
*preferred_zone
;
1873 int migratetype
= allocflags_to_migratetype(gfp_mask
);
1875 gfp_mask
&= gfp_allowed_mask
;
1877 lockdep_trace_alloc(gfp_mask
);
1879 might_sleep_if(gfp_mask
& __GFP_WAIT
);
1881 if (should_fail_alloc_page(gfp_mask
, order
))
1885 * Check the zones suitable for the gfp_mask contain at least one
1886 * valid zone. It's possible to have an empty zonelist as a result
1887 * of GFP_THISNODE and a memoryless node
1889 if (unlikely(!zonelist
->_zonerefs
->zone
))
1892 /* The preferred zone is used for statistics later */
1893 first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
, &preferred_zone
);
1894 if (!preferred_zone
)
1897 /* First allocation attempt */
1898 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1899 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
1900 preferred_zone
, migratetype
);
1901 if (unlikely(!page
))
1902 page
= __alloc_pages_slowpath(gfp_mask
, order
,
1903 zonelist
, high_zoneidx
, nodemask
,
1904 preferred_zone
, migratetype
);
1908 EXPORT_SYMBOL(__alloc_pages_nodemask
);
1911 * Common helper functions.
1913 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1916 page
= alloc_pages(gfp_mask
, order
);
1919 return (unsigned long) page_address(page
);
1922 EXPORT_SYMBOL(__get_free_pages
);
1924 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1929 * get_zeroed_page() returns a 32-bit address, which cannot represent
1932 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1934 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1936 return (unsigned long) page_address(page
);
1940 EXPORT_SYMBOL(get_zeroed_page
);
1942 void __pagevec_free(struct pagevec
*pvec
)
1944 int i
= pagevec_count(pvec
);
1947 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1950 void __free_pages(struct page
*page
, unsigned int order
)
1952 if (put_page_testzero(page
)) {
1954 free_hot_page(page
);
1956 __free_pages_ok(page
, order
);
1960 EXPORT_SYMBOL(__free_pages
);
1962 void free_pages(unsigned long addr
, unsigned int order
)
1965 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1966 __free_pages(virt_to_page((void *)addr
), order
);
1970 EXPORT_SYMBOL(free_pages
);
1973 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1974 * @size: the number of bytes to allocate
1975 * @gfp_mask: GFP flags for the allocation
1977 * This function is similar to alloc_pages(), except that it allocates the
1978 * minimum number of pages to satisfy the request. alloc_pages() can only
1979 * allocate memory in power-of-two pages.
1981 * This function is also limited by MAX_ORDER.
1983 * Memory allocated by this function must be released by free_pages_exact().
1985 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1987 unsigned int order
= get_order(size
);
1990 addr
= __get_free_pages(gfp_mask
, order
);
1992 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1993 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1995 split_page(virt_to_page((void *)addr
), order
);
1996 while (used
< alloc_end
) {
2002 return (void *)addr
;
2004 EXPORT_SYMBOL(alloc_pages_exact
);
2007 * free_pages_exact - release memory allocated via alloc_pages_exact()
2008 * @virt: the value returned by alloc_pages_exact.
2009 * @size: size of allocation, same value as passed to alloc_pages_exact().
2011 * Release the memory allocated by a previous call to alloc_pages_exact.
2013 void free_pages_exact(void *virt
, size_t size
)
2015 unsigned long addr
= (unsigned long)virt
;
2016 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2018 while (addr
< end
) {
2023 EXPORT_SYMBOL(free_pages_exact
);
2025 static unsigned int nr_free_zone_pages(int offset
)
2030 /* Just pick one node, since fallback list is circular */
2031 unsigned int sum
= 0;
2033 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2035 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2036 unsigned long size
= zone
->present_pages
;
2037 unsigned long high
= high_wmark_pages(zone
);
2046 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2048 unsigned int nr_free_buffer_pages(void)
2050 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2052 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2055 * Amount of free RAM allocatable within all zones
2057 unsigned int nr_free_pagecache_pages(void)
2059 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2062 static inline void show_node(struct zone
*zone
)
2065 printk("Node %d ", zone_to_nid(zone
));
2068 void si_meminfo(struct sysinfo
*val
)
2070 val
->totalram
= totalram_pages
;
2072 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2073 val
->bufferram
= nr_blockdev_pages();
2074 val
->totalhigh
= totalhigh_pages
;
2075 val
->freehigh
= nr_free_highpages();
2076 val
->mem_unit
= PAGE_SIZE
;
2079 EXPORT_SYMBOL(si_meminfo
);
2082 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2084 pg_data_t
*pgdat
= NODE_DATA(nid
);
2086 val
->totalram
= pgdat
->node_present_pages
;
2087 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2088 #ifdef CONFIG_HIGHMEM
2089 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2090 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2096 val
->mem_unit
= PAGE_SIZE
;
2100 #define K(x) ((x) << (PAGE_SHIFT-10))
2103 * Show free area list (used inside shift_scroll-lock stuff)
2104 * We also calculate the percentage fragmentation. We do this by counting the
2105 * memory on each free list with the exception of the first item on the list.
2107 void show_free_areas(void)
2112 for_each_populated_zone(zone
) {
2114 printk("%s per-cpu:\n", zone
->name
);
2116 for_each_online_cpu(cpu
) {
2117 struct per_cpu_pageset
*pageset
;
2119 pageset
= zone_pcp(zone
, cpu
);
2121 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2122 cpu
, pageset
->pcp
.high
,
2123 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2127 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2128 " inactive_file:%lu"
2130 " dirty:%lu writeback:%lu unstable:%lu\n"
2131 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2132 global_page_state(NR_ACTIVE_ANON
),
2133 global_page_state(NR_ACTIVE_FILE
),
2134 global_page_state(NR_INACTIVE_ANON
),
2135 global_page_state(NR_INACTIVE_FILE
),
2136 global_page_state(NR_UNEVICTABLE
),
2137 global_page_state(NR_FILE_DIRTY
),
2138 global_page_state(NR_WRITEBACK
),
2139 global_page_state(NR_UNSTABLE_NFS
),
2140 global_page_state(NR_FREE_PAGES
),
2141 global_page_state(NR_SLAB_RECLAIMABLE
) +
2142 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2143 global_page_state(NR_FILE_MAPPED
),
2144 global_page_state(NR_PAGETABLE
),
2145 global_page_state(NR_BOUNCE
));
2147 for_each_populated_zone(zone
) {
2156 " active_anon:%lukB"
2157 " inactive_anon:%lukB"
2158 " active_file:%lukB"
2159 " inactive_file:%lukB"
2160 " unevictable:%lukB"
2162 " pages_scanned:%lu"
2163 " all_unreclaimable? %s"
2166 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2167 K(min_wmark_pages(zone
)),
2168 K(low_wmark_pages(zone
)),
2169 K(high_wmark_pages(zone
)),
2170 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2171 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2172 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2173 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2174 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2175 K(zone
->present_pages
),
2176 zone
->pages_scanned
,
2177 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
2179 printk("lowmem_reserve[]:");
2180 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2181 printk(" %lu", zone
->lowmem_reserve
[i
]);
2185 for_each_populated_zone(zone
) {
2186 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2189 printk("%s: ", zone
->name
);
2191 spin_lock_irqsave(&zone
->lock
, flags
);
2192 for (order
= 0; order
< MAX_ORDER
; order
++) {
2193 nr
[order
] = zone
->free_area
[order
].nr_free
;
2194 total
+= nr
[order
] << order
;
2196 spin_unlock_irqrestore(&zone
->lock
, flags
);
2197 for (order
= 0; order
< MAX_ORDER
; order
++)
2198 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2199 printk("= %lukB\n", K(total
));
2202 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2204 show_swap_cache_info();
2207 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2209 zoneref
->zone
= zone
;
2210 zoneref
->zone_idx
= zone_idx(zone
);
2214 * Builds allocation fallback zone lists.
2216 * Add all populated zones of a node to the zonelist.
2218 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2219 int nr_zones
, enum zone_type zone_type
)
2223 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2228 zone
= pgdat
->node_zones
+ zone_type
;
2229 if (populated_zone(zone
)) {
2230 zoneref_set_zone(zone
,
2231 &zonelist
->_zonerefs
[nr_zones
++]);
2232 check_highest_zone(zone_type
);
2235 } while (zone_type
);
2242 * 0 = automatic detection of better ordering.
2243 * 1 = order by ([node] distance, -zonetype)
2244 * 2 = order by (-zonetype, [node] distance)
2246 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2247 * the same zonelist. So only NUMA can configure this param.
2249 #define ZONELIST_ORDER_DEFAULT 0
2250 #define ZONELIST_ORDER_NODE 1
2251 #define ZONELIST_ORDER_ZONE 2
2253 /* zonelist order in the kernel.
2254 * set_zonelist_order() will set this to NODE or ZONE.
2256 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2257 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2261 /* The value user specified ....changed by config */
2262 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2263 /* string for sysctl */
2264 #define NUMA_ZONELIST_ORDER_LEN 16
2265 char numa_zonelist_order
[16] = "default";
2268 * interface for configure zonelist ordering.
2269 * command line option "numa_zonelist_order"
2270 * = "[dD]efault - default, automatic configuration.
2271 * = "[nN]ode - order by node locality, then by zone within node
2272 * = "[zZ]one - order by zone, then by locality within zone
2275 static int __parse_numa_zonelist_order(char *s
)
2277 if (*s
== 'd' || *s
== 'D') {
2278 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2279 } else if (*s
== 'n' || *s
== 'N') {
2280 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2281 } else if (*s
== 'z' || *s
== 'Z') {
2282 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2285 "Ignoring invalid numa_zonelist_order value: "
2292 static __init
int setup_numa_zonelist_order(char *s
)
2295 return __parse_numa_zonelist_order(s
);
2298 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2301 * sysctl handler for numa_zonelist_order
2303 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2304 struct file
*file
, void __user
*buffer
, size_t *length
,
2307 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2311 strncpy(saved_string
, (char*)table
->data
,
2312 NUMA_ZONELIST_ORDER_LEN
);
2313 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2317 int oldval
= user_zonelist_order
;
2318 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2320 * bogus value. restore saved string
2322 strncpy((char*)table
->data
, saved_string
,
2323 NUMA_ZONELIST_ORDER_LEN
);
2324 user_zonelist_order
= oldval
;
2325 } else if (oldval
!= user_zonelist_order
)
2326 build_all_zonelists();
2332 #define MAX_NODE_LOAD (nr_online_nodes)
2333 static int node_load
[MAX_NUMNODES
];
2336 * find_next_best_node - find the next node that should appear in a given node's fallback list
2337 * @node: node whose fallback list we're appending
2338 * @used_node_mask: nodemask_t of already used nodes
2340 * We use a number of factors to determine which is the next node that should
2341 * appear on a given node's fallback list. The node should not have appeared
2342 * already in @node's fallback list, and it should be the next closest node
2343 * according to the distance array (which contains arbitrary distance values
2344 * from each node to each node in the system), and should also prefer nodes
2345 * with no CPUs, since presumably they'll have very little allocation pressure
2346 * on them otherwise.
2347 * It returns -1 if no node is found.
2349 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2352 int min_val
= INT_MAX
;
2354 const struct cpumask
*tmp
= cpumask_of_node(0);
2356 /* Use the local node if we haven't already */
2357 if (!node_isset(node
, *used_node_mask
)) {
2358 node_set(node
, *used_node_mask
);
2362 for_each_node_state(n
, N_HIGH_MEMORY
) {
2364 /* Don't want a node to appear more than once */
2365 if (node_isset(n
, *used_node_mask
))
2368 /* Use the distance array to find the distance */
2369 val
= node_distance(node
, n
);
2371 /* Penalize nodes under us ("prefer the next node") */
2374 /* Give preference to headless and unused nodes */
2375 tmp
= cpumask_of_node(n
);
2376 if (!cpumask_empty(tmp
))
2377 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2379 /* Slight preference for less loaded node */
2380 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2381 val
+= node_load
[n
];
2383 if (val
< min_val
) {
2390 node_set(best_node
, *used_node_mask
);
2397 * Build zonelists ordered by node and zones within node.
2398 * This results in maximum locality--normal zone overflows into local
2399 * DMA zone, if any--but risks exhausting DMA zone.
2401 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2404 struct zonelist
*zonelist
;
2406 zonelist
= &pgdat
->node_zonelists
[0];
2407 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2409 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2411 zonelist
->_zonerefs
[j
].zone
= NULL
;
2412 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2416 * Build gfp_thisnode zonelists
2418 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2421 struct zonelist
*zonelist
;
2423 zonelist
= &pgdat
->node_zonelists
[1];
2424 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2425 zonelist
->_zonerefs
[j
].zone
= NULL
;
2426 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2430 * Build zonelists ordered by zone and nodes within zones.
2431 * This results in conserving DMA zone[s] until all Normal memory is
2432 * exhausted, but results in overflowing to remote node while memory
2433 * may still exist in local DMA zone.
2435 static int node_order
[MAX_NUMNODES
];
2437 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2440 int zone_type
; /* needs to be signed */
2442 struct zonelist
*zonelist
;
2444 zonelist
= &pgdat
->node_zonelists
[0];
2446 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2447 for (j
= 0; j
< nr_nodes
; j
++) {
2448 node
= node_order
[j
];
2449 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2450 if (populated_zone(z
)) {
2452 &zonelist
->_zonerefs
[pos
++]);
2453 check_highest_zone(zone_type
);
2457 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2458 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2461 static int default_zonelist_order(void)
2464 unsigned long low_kmem_size
,total_size
;
2468 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2469 * If they are really small and used heavily, the system can fall
2470 * into OOM very easily.
2471 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2473 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2476 for_each_online_node(nid
) {
2477 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2478 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2479 if (populated_zone(z
)) {
2480 if (zone_type
< ZONE_NORMAL
)
2481 low_kmem_size
+= z
->present_pages
;
2482 total_size
+= z
->present_pages
;
2486 if (!low_kmem_size
|| /* there are no DMA area. */
2487 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2488 return ZONELIST_ORDER_NODE
;
2490 * look into each node's config.
2491 * If there is a node whose DMA/DMA32 memory is very big area on
2492 * local memory, NODE_ORDER may be suitable.
2494 average_size
= total_size
/
2495 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2496 for_each_online_node(nid
) {
2499 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2500 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2501 if (populated_zone(z
)) {
2502 if (zone_type
< ZONE_NORMAL
)
2503 low_kmem_size
+= z
->present_pages
;
2504 total_size
+= z
->present_pages
;
2507 if (low_kmem_size
&&
2508 total_size
> average_size
&& /* ignore small node */
2509 low_kmem_size
> total_size
* 70/100)
2510 return ZONELIST_ORDER_NODE
;
2512 return ZONELIST_ORDER_ZONE
;
2515 static void set_zonelist_order(void)
2517 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2518 current_zonelist_order
= default_zonelist_order();
2520 current_zonelist_order
= user_zonelist_order
;
2523 static void build_zonelists(pg_data_t
*pgdat
)
2527 nodemask_t used_mask
;
2528 int local_node
, prev_node
;
2529 struct zonelist
*zonelist
;
2530 int order
= current_zonelist_order
;
2532 /* initialize zonelists */
2533 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2534 zonelist
= pgdat
->node_zonelists
+ i
;
2535 zonelist
->_zonerefs
[0].zone
= NULL
;
2536 zonelist
->_zonerefs
[0].zone_idx
= 0;
2539 /* NUMA-aware ordering of nodes */
2540 local_node
= pgdat
->node_id
;
2541 load
= nr_online_nodes
;
2542 prev_node
= local_node
;
2543 nodes_clear(used_mask
);
2545 memset(node_order
, 0, sizeof(node_order
));
2548 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2549 int distance
= node_distance(local_node
, node
);
2552 * If another node is sufficiently far away then it is better
2553 * to reclaim pages in a zone before going off node.
2555 if (distance
> RECLAIM_DISTANCE
)
2556 zone_reclaim_mode
= 1;
2559 * We don't want to pressure a particular node.
2560 * So adding penalty to the first node in same
2561 * distance group to make it round-robin.
2563 if (distance
!= node_distance(local_node
, prev_node
))
2564 node_load
[node
] = load
;
2568 if (order
== ZONELIST_ORDER_NODE
)
2569 build_zonelists_in_node_order(pgdat
, node
);
2571 node_order
[j
++] = node
; /* remember order */
2574 if (order
== ZONELIST_ORDER_ZONE
) {
2575 /* calculate node order -- i.e., DMA last! */
2576 build_zonelists_in_zone_order(pgdat
, j
);
2579 build_thisnode_zonelists(pgdat
);
2582 /* Construct the zonelist performance cache - see further mmzone.h */
2583 static void build_zonelist_cache(pg_data_t
*pgdat
)
2585 struct zonelist
*zonelist
;
2586 struct zonelist_cache
*zlc
;
2589 zonelist
= &pgdat
->node_zonelists
[0];
2590 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2591 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2592 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2593 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2597 #else /* CONFIG_NUMA */
2599 static void set_zonelist_order(void)
2601 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2604 static void build_zonelists(pg_data_t
*pgdat
)
2606 int node
, local_node
;
2608 struct zonelist
*zonelist
;
2610 local_node
= pgdat
->node_id
;
2612 zonelist
= &pgdat
->node_zonelists
[0];
2613 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2616 * Now we build the zonelist so that it contains the zones
2617 * of all the other nodes.
2618 * We don't want to pressure a particular node, so when
2619 * building the zones for node N, we make sure that the
2620 * zones coming right after the local ones are those from
2621 * node N+1 (modulo N)
2623 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2624 if (!node_online(node
))
2626 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2629 for (node
= 0; node
< local_node
; node
++) {
2630 if (!node_online(node
))
2632 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2636 zonelist
->_zonerefs
[j
].zone
= NULL
;
2637 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2640 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2641 static void build_zonelist_cache(pg_data_t
*pgdat
)
2643 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2646 #endif /* CONFIG_NUMA */
2648 /* return values int ....just for stop_machine() */
2649 static int __build_all_zonelists(void *dummy
)
2654 memset(node_load
, 0, sizeof(node_load
));
2656 for_each_online_node(nid
) {
2657 pg_data_t
*pgdat
= NODE_DATA(nid
);
2659 build_zonelists(pgdat
);
2660 build_zonelist_cache(pgdat
);
2665 void build_all_zonelists(void)
2667 set_zonelist_order();
2669 if (system_state
== SYSTEM_BOOTING
) {
2670 __build_all_zonelists(NULL
);
2671 mminit_verify_zonelist();
2672 cpuset_init_current_mems_allowed();
2674 /* we have to stop all cpus to guarantee there is no user
2676 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2677 /* cpuset refresh routine should be here */
2679 vm_total_pages
= nr_free_pagecache_pages();
2681 * Disable grouping by mobility if the number of pages in the
2682 * system is too low to allow the mechanism to work. It would be
2683 * more accurate, but expensive to check per-zone. This check is
2684 * made on memory-hotadd so a system can start with mobility
2685 * disabled and enable it later
2687 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2688 page_group_by_mobility_disabled
= 1;
2690 page_group_by_mobility_disabled
= 0;
2692 printk("Built %i zonelists in %s order, mobility grouping %s. "
2693 "Total pages: %ld\n",
2695 zonelist_order_name
[current_zonelist_order
],
2696 page_group_by_mobility_disabled
? "off" : "on",
2699 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2704 * Helper functions to size the waitqueue hash table.
2705 * Essentially these want to choose hash table sizes sufficiently
2706 * large so that collisions trying to wait on pages are rare.
2707 * But in fact, the number of active page waitqueues on typical
2708 * systems is ridiculously low, less than 200. So this is even
2709 * conservative, even though it seems large.
2711 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2712 * waitqueues, i.e. the size of the waitq table given the number of pages.
2714 #define PAGES_PER_WAITQUEUE 256
2716 #ifndef CONFIG_MEMORY_HOTPLUG
2717 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2719 unsigned long size
= 1;
2721 pages
/= PAGES_PER_WAITQUEUE
;
2723 while (size
< pages
)
2727 * Once we have dozens or even hundreds of threads sleeping
2728 * on IO we've got bigger problems than wait queue collision.
2729 * Limit the size of the wait table to a reasonable size.
2731 size
= min(size
, 4096UL);
2733 return max(size
, 4UL);
2737 * A zone's size might be changed by hot-add, so it is not possible to determine
2738 * a suitable size for its wait_table. So we use the maximum size now.
2740 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2742 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2743 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2744 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2746 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2747 * or more by the traditional way. (See above). It equals:
2749 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2750 * ia64(16K page size) : = ( 8G + 4M)byte.
2751 * powerpc (64K page size) : = (32G +16M)byte.
2753 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2760 * This is an integer logarithm so that shifts can be used later
2761 * to extract the more random high bits from the multiplicative
2762 * hash function before the remainder is taken.
2764 static inline unsigned long wait_table_bits(unsigned long size
)
2769 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2772 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2773 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2774 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
2775 * higher will lead to a bigger reserve which will get freed as contiguous
2776 * blocks as reclaim kicks in
2778 static void setup_zone_migrate_reserve(struct zone
*zone
)
2780 unsigned long start_pfn
, pfn
, end_pfn
;
2782 unsigned long reserve
, block_migratetype
;
2784 /* Get the start pfn, end pfn and the number of blocks to reserve */
2785 start_pfn
= zone
->zone_start_pfn
;
2786 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2787 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
2790 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2791 if (!pfn_valid(pfn
))
2793 page
= pfn_to_page(pfn
);
2795 /* Watch out for overlapping nodes */
2796 if (page_to_nid(page
) != zone_to_nid(zone
))
2799 /* Blocks with reserved pages will never free, skip them. */
2800 if (PageReserved(page
))
2803 block_migratetype
= get_pageblock_migratetype(page
);
2805 /* If this block is reserved, account for it */
2806 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2811 /* Suitable for reserving if this block is movable */
2812 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2813 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2814 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2820 * If the reserve is met and this is a previous reserved block,
2823 if (block_migratetype
== MIGRATE_RESERVE
) {
2824 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2825 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2831 * Initially all pages are reserved - free ones are freed
2832 * up by free_all_bootmem() once the early boot process is
2833 * done. Non-atomic initialization, single-pass.
2835 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2836 unsigned long start_pfn
, enum memmap_context context
)
2839 unsigned long end_pfn
= start_pfn
+ size
;
2843 if (highest_memmap_pfn
< end_pfn
- 1)
2844 highest_memmap_pfn
= end_pfn
- 1;
2846 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2847 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2849 * There can be holes in boot-time mem_map[]s
2850 * handed to this function. They do not
2851 * exist on hotplugged memory.
2853 if (context
== MEMMAP_EARLY
) {
2854 if (!early_pfn_valid(pfn
))
2856 if (!early_pfn_in_nid(pfn
, nid
))
2859 page
= pfn_to_page(pfn
);
2860 set_page_links(page
, zone
, nid
, pfn
);
2861 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2862 init_page_count(page
);
2863 reset_page_mapcount(page
);
2864 SetPageReserved(page
);
2866 * Mark the block movable so that blocks are reserved for
2867 * movable at startup. This will force kernel allocations
2868 * to reserve their blocks rather than leaking throughout
2869 * the address space during boot when many long-lived
2870 * kernel allocations are made. Later some blocks near
2871 * the start are marked MIGRATE_RESERVE by
2872 * setup_zone_migrate_reserve()
2874 * bitmap is created for zone's valid pfn range. but memmap
2875 * can be created for invalid pages (for alignment)
2876 * check here not to call set_pageblock_migratetype() against
2879 if ((z
->zone_start_pfn
<= pfn
)
2880 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2881 && !(pfn
& (pageblock_nr_pages
- 1)))
2882 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2884 INIT_LIST_HEAD(&page
->lru
);
2885 #ifdef WANT_PAGE_VIRTUAL
2886 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2887 if (!is_highmem_idx(zone
))
2888 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2893 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2896 for_each_migratetype_order(order
, t
) {
2897 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2898 zone
->free_area
[order
].nr_free
= 0;
2902 #ifndef __HAVE_ARCH_MEMMAP_INIT
2903 #define memmap_init(size, nid, zone, start_pfn) \
2904 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2907 static int zone_batchsize(struct zone
*zone
)
2913 * The per-cpu-pages pools are set to around 1000th of the
2914 * size of the zone. But no more than 1/2 of a meg.
2916 * OK, so we don't know how big the cache is. So guess.
2918 batch
= zone
->present_pages
/ 1024;
2919 if (batch
* PAGE_SIZE
> 512 * 1024)
2920 batch
= (512 * 1024) / PAGE_SIZE
;
2921 batch
/= 4; /* We effectively *= 4 below */
2926 * Clamp the batch to a 2^n - 1 value. Having a power
2927 * of 2 value was found to be more likely to have
2928 * suboptimal cache aliasing properties in some cases.
2930 * For example if 2 tasks are alternately allocating
2931 * batches of pages, one task can end up with a lot
2932 * of pages of one half of the possible page colors
2933 * and the other with pages of the other colors.
2935 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
2940 /* The deferral and batching of frees should be suppressed under NOMMU
2943 * The problem is that NOMMU needs to be able to allocate large chunks
2944 * of contiguous memory as there's no hardware page translation to
2945 * assemble apparent contiguous memory from discontiguous pages.
2947 * Queueing large contiguous runs of pages for batching, however,
2948 * causes the pages to actually be freed in smaller chunks. As there
2949 * can be a significant delay between the individual batches being
2950 * recycled, this leads to the once large chunks of space being
2951 * fragmented and becoming unavailable for high-order allocations.
2957 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2959 struct per_cpu_pages
*pcp
;
2961 memset(p
, 0, sizeof(*p
));
2965 pcp
->high
= 6 * batch
;
2966 pcp
->batch
= max(1UL, 1 * batch
);
2967 INIT_LIST_HEAD(&pcp
->list
);
2971 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2972 * to the value high for the pageset p.
2975 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2978 struct per_cpu_pages
*pcp
;
2982 pcp
->batch
= max(1UL, high
/4);
2983 if ((high
/4) > (PAGE_SHIFT
* 8))
2984 pcp
->batch
= PAGE_SHIFT
* 8;
2990 * Boot pageset table. One per cpu which is going to be used for all
2991 * zones and all nodes. The parameters will be set in such a way
2992 * that an item put on a list will immediately be handed over to
2993 * the buddy list. This is safe since pageset manipulation is done
2994 * with interrupts disabled.
2996 * Some NUMA counter updates may also be caught by the boot pagesets.
2998 * The boot_pagesets must be kept even after bootup is complete for
2999 * unused processors and/or zones. They do play a role for bootstrapping
3000 * hotplugged processors.
3002 * zoneinfo_show() and maybe other functions do
3003 * not check if the processor is online before following the pageset pointer.
3004 * Other parts of the kernel may not check if the zone is available.
3006 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
3009 * Dynamically allocate memory for the
3010 * per cpu pageset array in struct zone.
3012 static int __cpuinit
process_zones(int cpu
)
3014 struct zone
*zone
, *dzone
;
3015 int node
= cpu_to_node(cpu
);
3017 node_set_state(node
, N_CPU
); /* this node has a cpu */
3019 for_each_populated_zone(zone
) {
3020 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
3022 if (!zone_pcp(zone
, cpu
))
3025 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
3027 if (percpu_pagelist_fraction
)
3028 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
3029 (zone
->present_pages
/ percpu_pagelist_fraction
));
3034 for_each_zone(dzone
) {
3035 if (!populated_zone(dzone
))
3039 kfree(zone_pcp(dzone
, cpu
));
3040 zone_pcp(dzone
, cpu
) = &boot_pageset
[cpu
];
3045 static inline void free_zone_pagesets(int cpu
)
3049 for_each_zone(zone
) {
3050 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
3052 /* Free per_cpu_pageset if it is slab allocated */
3053 if (pset
!= &boot_pageset
[cpu
])
3055 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
3059 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
3060 unsigned long action
,
3063 int cpu
= (long)hcpu
;
3064 int ret
= NOTIFY_OK
;
3067 case CPU_UP_PREPARE
:
3068 case CPU_UP_PREPARE_FROZEN
:
3069 if (process_zones(cpu
))
3072 case CPU_UP_CANCELED
:
3073 case CPU_UP_CANCELED_FROZEN
:
3075 case CPU_DEAD_FROZEN
:
3076 free_zone_pagesets(cpu
);
3084 static struct notifier_block __cpuinitdata pageset_notifier
=
3085 { &pageset_cpuup_callback
, NULL
, 0 };
3087 void __init
setup_per_cpu_pageset(void)
3091 /* Initialize per_cpu_pageset for cpu 0.
3092 * A cpuup callback will do this for every cpu
3093 * as it comes online
3095 err
= process_zones(smp_processor_id());
3097 register_cpu_notifier(&pageset_notifier
);
3102 static noinline __init_refok
3103 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3106 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3110 * The per-page waitqueue mechanism uses hashed waitqueues
3113 zone
->wait_table_hash_nr_entries
=
3114 wait_table_hash_nr_entries(zone_size_pages
);
3115 zone
->wait_table_bits
=
3116 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3117 alloc_size
= zone
->wait_table_hash_nr_entries
3118 * sizeof(wait_queue_head_t
);
3120 if (!slab_is_available()) {
3121 zone
->wait_table
= (wait_queue_head_t
*)
3122 alloc_bootmem_node(pgdat
, alloc_size
);
3125 * This case means that a zone whose size was 0 gets new memory
3126 * via memory hot-add.
3127 * But it may be the case that a new node was hot-added. In
3128 * this case vmalloc() will not be able to use this new node's
3129 * memory - this wait_table must be initialized to use this new
3130 * node itself as well.
3131 * To use this new node's memory, further consideration will be
3134 zone
->wait_table
= vmalloc(alloc_size
);
3136 if (!zone
->wait_table
)
3139 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3140 init_waitqueue_head(zone
->wait_table
+ i
);
3145 static int __zone_pcp_update(void *data
)
3147 struct zone
*zone
= data
;
3149 unsigned long batch
= zone_batchsize(zone
), flags
;
3151 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
3152 struct per_cpu_pageset
*pset
;
3153 struct per_cpu_pages
*pcp
;
3155 pset
= zone_pcp(zone
, cpu
);
3158 local_irq_save(flags
);
3159 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
3160 setup_pageset(pset
, batch
);
3161 local_irq_restore(flags
);
3166 void zone_pcp_update(struct zone
*zone
)
3168 stop_machine(__zone_pcp_update
, zone
, NULL
);
3171 static __meminit
void zone_pcp_init(struct zone
*zone
)
3174 unsigned long batch
= zone_batchsize(zone
);
3176 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
3178 /* Early boot. Slab allocator not functional yet */
3179 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
3180 setup_pageset(&boot_pageset
[cpu
],0);
3182 setup_pageset(zone_pcp(zone
,cpu
), batch
);
3185 if (zone
->present_pages
)
3186 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
3187 zone
->name
, zone
->present_pages
, batch
);
3190 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3191 unsigned long zone_start_pfn
,
3193 enum memmap_context context
)
3195 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3197 ret
= zone_wait_table_init(zone
, size
);
3200 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3202 zone
->zone_start_pfn
= zone_start_pfn
;
3204 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3205 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3207 (unsigned long)zone_idx(zone
),
3208 zone_start_pfn
, (zone_start_pfn
+ size
));
3210 zone_init_free_lists(zone
);
3215 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3217 * Basic iterator support. Return the first range of PFNs for a node
3218 * Note: nid == MAX_NUMNODES returns first region regardless of node
3220 static int __meminit
first_active_region_index_in_nid(int nid
)
3224 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3225 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
3232 * Basic iterator support. Return the next active range of PFNs for a node
3233 * Note: nid == MAX_NUMNODES returns next region regardless of node
3235 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
3237 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
3238 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
3244 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3246 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3247 * Architectures may implement their own version but if add_active_range()
3248 * was used and there are no special requirements, this is a convenient
3251 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3255 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3256 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3257 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3259 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3260 return early_node_map
[i
].nid
;
3262 /* This is a memory hole */
3265 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3267 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3271 nid
= __early_pfn_to_nid(pfn
);
3274 /* just returns 0 */
3278 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3279 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3283 nid
= __early_pfn_to_nid(pfn
);
3284 if (nid
>= 0 && nid
!= node
)
3290 /* Basic iterator support to walk early_node_map[] */
3291 #define for_each_active_range_index_in_nid(i, nid) \
3292 for (i = first_active_region_index_in_nid(nid); i != -1; \
3293 i = next_active_region_index_in_nid(i, nid))
3296 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3297 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3298 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3300 * If an architecture guarantees that all ranges registered with
3301 * add_active_ranges() contain no holes and may be freed, this
3302 * this function may be used instead of calling free_bootmem() manually.
3304 void __init
free_bootmem_with_active_regions(int nid
,
3305 unsigned long max_low_pfn
)
3309 for_each_active_range_index_in_nid(i
, nid
) {
3310 unsigned long size_pages
= 0;
3311 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3313 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3316 if (end_pfn
> max_low_pfn
)
3317 end_pfn
= max_low_pfn
;
3319 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3320 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3321 PFN_PHYS(early_node_map
[i
].start_pfn
),
3322 size_pages
<< PAGE_SHIFT
);
3326 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3331 for_each_active_range_index_in_nid(i
, nid
) {
3332 ret
= work_fn(early_node_map
[i
].start_pfn
,
3333 early_node_map
[i
].end_pfn
, data
);
3339 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3340 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3342 * If an architecture guarantees that all ranges registered with
3343 * add_active_ranges() contain no holes and may be freed, this
3344 * function may be used instead of calling memory_present() manually.
3346 void __init
sparse_memory_present_with_active_regions(int nid
)
3350 for_each_active_range_index_in_nid(i
, nid
)
3351 memory_present(early_node_map
[i
].nid
,
3352 early_node_map
[i
].start_pfn
,
3353 early_node_map
[i
].end_pfn
);
3357 * get_pfn_range_for_nid - Return the start and end page frames for a node
3358 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3359 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3360 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3362 * It returns the start and end page frame of a node based on information
3363 * provided by an arch calling add_active_range(). If called for a node
3364 * with no available memory, a warning is printed and the start and end
3367 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3368 unsigned long *start_pfn
, unsigned long *end_pfn
)
3374 for_each_active_range_index_in_nid(i
, nid
) {
3375 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3376 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3379 if (*start_pfn
== -1UL)
3384 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3385 * assumption is made that zones within a node are ordered in monotonic
3386 * increasing memory addresses so that the "highest" populated zone is used
3388 static void __init
find_usable_zone_for_movable(void)
3391 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3392 if (zone_index
== ZONE_MOVABLE
)
3395 if (arch_zone_highest_possible_pfn
[zone_index
] >
3396 arch_zone_lowest_possible_pfn
[zone_index
])
3400 VM_BUG_ON(zone_index
== -1);
3401 movable_zone
= zone_index
;
3405 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3406 * because it is sized independant of architecture. Unlike the other zones,
3407 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3408 * in each node depending on the size of each node and how evenly kernelcore
3409 * is distributed. This helper function adjusts the zone ranges
3410 * provided by the architecture for a given node by using the end of the
3411 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3412 * zones within a node are in order of monotonic increases memory addresses
3414 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3415 unsigned long zone_type
,
3416 unsigned long node_start_pfn
,
3417 unsigned long node_end_pfn
,
3418 unsigned long *zone_start_pfn
,
3419 unsigned long *zone_end_pfn
)
3421 /* Only adjust if ZONE_MOVABLE is on this node */
3422 if (zone_movable_pfn
[nid
]) {
3423 /* Size ZONE_MOVABLE */
3424 if (zone_type
== ZONE_MOVABLE
) {
3425 *zone_start_pfn
= zone_movable_pfn
[nid
];
3426 *zone_end_pfn
= min(node_end_pfn
,
3427 arch_zone_highest_possible_pfn
[movable_zone
]);
3429 /* Adjust for ZONE_MOVABLE starting within this range */
3430 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3431 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3432 *zone_end_pfn
= zone_movable_pfn
[nid
];
3434 /* Check if this whole range is within ZONE_MOVABLE */
3435 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3436 *zone_start_pfn
= *zone_end_pfn
;
3441 * Return the number of pages a zone spans in a node, including holes
3442 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3444 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3445 unsigned long zone_type
,
3446 unsigned long *ignored
)
3448 unsigned long node_start_pfn
, node_end_pfn
;
3449 unsigned long zone_start_pfn
, zone_end_pfn
;
3451 /* Get the start and end of the node and zone */
3452 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3453 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3454 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3455 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3456 node_start_pfn
, node_end_pfn
,
3457 &zone_start_pfn
, &zone_end_pfn
);
3459 /* Check that this node has pages within the zone's required range */
3460 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3463 /* Move the zone boundaries inside the node if necessary */
3464 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3465 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3467 /* Return the spanned pages */
3468 return zone_end_pfn
- zone_start_pfn
;
3472 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3473 * then all holes in the requested range will be accounted for.
3475 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3476 unsigned long range_start_pfn
,
3477 unsigned long range_end_pfn
)
3480 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3481 unsigned long start_pfn
;
3483 /* Find the end_pfn of the first active range of pfns in the node */
3484 i
= first_active_region_index_in_nid(nid
);
3488 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3490 /* Account for ranges before physical memory on this node */
3491 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3492 hole_pages
= prev_end_pfn
- range_start_pfn
;
3494 /* Find all holes for the zone within the node */
3495 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3497 /* No need to continue if prev_end_pfn is outside the zone */
3498 if (prev_end_pfn
>= range_end_pfn
)
3501 /* Make sure the end of the zone is not within the hole */
3502 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3503 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3505 /* Update the hole size cound and move on */
3506 if (start_pfn
> range_start_pfn
) {
3507 BUG_ON(prev_end_pfn
> start_pfn
);
3508 hole_pages
+= start_pfn
- prev_end_pfn
;
3510 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3513 /* Account for ranges past physical memory on this node */
3514 if (range_end_pfn
> prev_end_pfn
)
3515 hole_pages
+= range_end_pfn
-
3516 max(range_start_pfn
, prev_end_pfn
);
3522 * absent_pages_in_range - Return number of page frames in holes within a range
3523 * @start_pfn: The start PFN to start searching for holes
3524 * @end_pfn: The end PFN to stop searching for holes
3526 * It returns the number of pages frames in memory holes within a range.
3528 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3529 unsigned long end_pfn
)
3531 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3534 /* Return the number of page frames in holes in a zone on a node */
3535 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3536 unsigned long zone_type
,
3537 unsigned long *ignored
)
3539 unsigned long node_start_pfn
, node_end_pfn
;
3540 unsigned long zone_start_pfn
, zone_end_pfn
;
3542 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3543 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3545 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3548 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3549 node_start_pfn
, node_end_pfn
,
3550 &zone_start_pfn
, &zone_end_pfn
);
3551 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3555 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3556 unsigned long zone_type
,
3557 unsigned long *zones_size
)
3559 return zones_size
[zone_type
];
3562 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3563 unsigned long zone_type
,
3564 unsigned long *zholes_size
)
3569 return zholes_size
[zone_type
];
3574 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3575 unsigned long *zones_size
, unsigned long *zholes_size
)
3577 unsigned long realtotalpages
, totalpages
= 0;
3580 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3581 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3583 pgdat
->node_spanned_pages
= totalpages
;
3585 realtotalpages
= totalpages
;
3586 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3588 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3590 pgdat
->node_present_pages
= realtotalpages
;
3591 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3595 #ifndef CONFIG_SPARSEMEM
3597 * Calculate the size of the zone->blockflags rounded to an unsigned long
3598 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3599 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3600 * round what is now in bits to nearest long in bits, then return it in
3603 static unsigned long __init
usemap_size(unsigned long zonesize
)
3605 unsigned long usemapsize
;
3607 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3608 usemapsize
= usemapsize
>> pageblock_order
;
3609 usemapsize
*= NR_PAGEBLOCK_BITS
;
3610 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3612 return usemapsize
/ 8;
3615 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3616 struct zone
*zone
, unsigned long zonesize
)
3618 unsigned long usemapsize
= usemap_size(zonesize
);
3619 zone
->pageblock_flags
= NULL
;
3621 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3624 static void inline setup_usemap(struct pglist_data
*pgdat
,
3625 struct zone
*zone
, unsigned long zonesize
) {}
3626 #endif /* CONFIG_SPARSEMEM */
3628 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3630 /* Return a sensible default order for the pageblock size. */
3631 static inline int pageblock_default_order(void)
3633 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3634 return HUGETLB_PAGE_ORDER
;
3639 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3640 static inline void __init
set_pageblock_order(unsigned int order
)
3642 /* Check that pageblock_nr_pages has not already been setup */
3643 if (pageblock_order
)
3647 * Assume the largest contiguous order of interest is a huge page.
3648 * This value may be variable depending on boot parameters on IA64
3650 pageblock_order
= order
;
3652 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3655 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3656 * and pageblock_default_order() are unused as pageblock_order is set
3657 * at compile-time. See include/linux/pageblock-flags.h for the values of
3658 * pageblock_order based on the kernel config
3660 static inline int pageblock_default_order(unsigned int order
)
3664 #define set_pageblock_order(x) do {} while (0)
3666 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3669 * Set up the zone data structures:
3670 * - mark all pages reserved
3671 * - mark all memory queues empty
3672 * - clear the memory bitmaps
3674 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3675 unsigned long *zones_size
, unsigned long *zholes_size
)
3678 int nid
= pgdat
->node_id
;
3679 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3682 pgdat_resize_init(pgdat
);
3683 pgdat
->nr_zones
= 0;
3684 init_waitqueue_head(&pgdat
->kswapd_wait
);
3685 pgdat
->kswapd_max_order
= 0;
3686 pgdat_page_cgroup_init(pgdat
);
3688 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3689 struct zone
*zone
= pgdat
->node_zones
+ j
;
3690 unsigned long size
, realsize
, memmap_pages
;
3693 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3694 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3698 * Adjust realsize so that it accounts for how much memory
3699 * is used by this zone for memmap. This affects the watermark
3700 * and per-cpu initialisations
3703 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3704 if (realsize
>= memmap_pages
) {
3705 realsize
-= memmap_pages
;
3708 " %s zone: %lu pages used for memmap\n",
3709 zone_names
[j
], memmap_pages
);
3712 " %s zone: %lu pages exceeds realsize %lu\n",
3713 zone_names
[j
], memmap_pages
, realsize
);
3715 /* Account for reserved pages */
3716 if (j
== 0 && realsize
> dma_reserve
) {
3717 realsize
-= dma_reserve
;
3718 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3719 zone_names
[0], dma_reserve
);
3722 if (!is_highmem_idx(j
))
3723 nr_kernel_pages
+= realsize
;
3724 nr_all_pages
+= realsize
;
3726 zone
->spanned_pages
= size
;
3727 zone
->present_pages
= realsize
;
3730 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3732 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3734 zone
->name
= zone_names
[j
];
3735 spin_lock_init(&zone
->lock
);
3736 spin_lock_init(&zone
->lru_lock
);
3737 zone_seqlock_init(zone
);
3738 zone
->zone_pgdat
= pgdat
;
3740 zone
->prev_priority
= DEF_PRIORITY
;
3742 zone_pcp_init(zone
);
3744 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3745 zone
->lru
[l
].nr_saved_scan
= 0;
3747 zone
->reclaim_stat
.recent_rotated
[0] = 0;
3748 zone
->reclaim_stat
.recent_rotated
[1] = 0;
3749 zone
->reclaim_stat
.recent_scanned
[0] = 0;
3750 zone
->reclaim_stat
.recent_scanned
[1] = 0;
3751 zap_zone_vm_stats(zone
);
3756 set_pageblock_order(pageblock_default_order());
3757 setup_usemap(pgdat
, zone
, size
);
3758 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3759 size
, MEMMAP_EARLY
);
3761 memmap_init(size
, nid
, j
, zone_start_pfn
);
3762 zone_start_pfn
+= size
;
3766 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3768 /* Skip empty nodes */
3769 if (!pgdat
->node_spanned_pages
)
3772 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3773 /* ia64 gets its own node_mem_map, before this, without bootmem */
3774 if (!pgdat
->node_mem_map
) {
3775 unsigned long size
, start
, end
;
3779 * The zone's endpoints aren't required to be MAX_ORDER
3780 * aligned but the node_mem_map endpoints must be in order
3781 * for the buddy allocator to function correctly.
3783 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3784 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3785 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3786 size
= (end
- start
) * sizeof(struct page
);
3787 map
= alloc_remap(pgdat
->node_id
, size
);
3789 map
= alloc_bootmem_node(pgdat
, size
);
3790 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3792 #ifndef CONFIG_NEED_MULTIPLE_NODES
3794 * With no DISCONTIG, the global mem_map is just set as node 0's
3796 if (pgdat
== NODE_DATA(0)) {
3797 mem_map
= NODE_DATA(0)->node_mem_map
;
3798 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3799 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3800 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3801 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3804 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3807 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3808 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3810 pg_data_t
*pgdat
= NODE_DATA(nid
);
3812 pgdat
->node_id
= nid
;
3813 pgdat
->node_start_pfn
= node_start_pfn
;
3814 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3816 alloc_node_mem_map(pgdat
);
3817 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3818 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3819 nid
, (unsigned long)pgdat
,
3820 (unsigned long)pgdat
->node_mem_map
);
3823 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3826 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3828 #if MAX_NUMNODES > 1
3830 * Figure out the number of possible node ids.
3832 static void __init
setup_nr_node_ids(void)
3835 unsigned int highest
= 0;
3837 for_each_node_mask(node
, node_possible_map
)
3839 nr_node_ids
= highest
+ 1;
3842 static inline void setup_nr_node_ids(void)
3848 * add_active_range - Register a range of PFNs backed by physical memory
3849 * @nid: The node ID the range resides on
3850 * @start_pfn: The start PFN of the available physical memory
3851 * @end_pfn: The end PFN of the available physical memory
3853 * These ranges are stored in an early_node_map[] and later used by
3854 * free_area_init_nodes() to calculate zone sizes and holes. If the
3855 * range spans a memory hole, it is up to the architecture to ensure
3856 * the memory is not freed by the bootmem allocator. If possible
3857 * the range being registered will be merged with existing ranges.
3859 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3860 unsigned long end_pfn
)
3864 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3865 "Entering add_active_range(%d, %#lx, %#lx) "
3866 "%d entries of %d used\n",
3867 nid
, start_pfn
, end_pfn
,
3868 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3870 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3872 /* Merge with existing active regions if possible */
3873 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3874 if (early_node_map
[i
].nid
!= nid
)
3877 /* Skip if an existing region covers this new one */
3878 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3879 end_pfn
<= early_node_map
[i
].end_pfn
)
3882 /* Merge forward if suitable */
3883 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3884 end_pfn
> early_node_map
[i
].end_pfn
) {
3885 early_node_map
[i
].end_pfn
= end_pfn
;
3889 /* Merge backward if suitable */
3890 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3891 end_pfn
>= early_node_map
[i
].start_pfn
) {
3892 early_node_map
[i
].start_pfn
= start_pfn
;
3897 /* Check that early_node_map is large enough */
3898 if (i
>= MAX_ACTIVE_REGIONS
) {
3899 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3900 MAX_ACTIVE_REGIONS
);
3904 early_node_map
[i
].nid
= nid
;
3905 early_node_map
[i
].start_pfn
= start_pfn
;
3906 early_node_map
[i
].end_pfn
= end_pfn
;
3907 nr_nodemap_entries
= i
+ 1;
3911 * remove_active_range - Shrink an existing registered range of PFNs
3912 * @nid: The node id the range is on that should be shrunk
3913 * @start_pfn: The new PFN of the range
3914 * @end_pfn: The new PFN of the range
3916 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3917 * The map is kept near the end physical page range that has already been
3918 * registered. This function allows an arch to shrink an existing registered
3921 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3922 unsigned long end_pfn
)
3927 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3928 nid
, start_pfn
, end_pfn
);
3930 /* Find the old active region end and shrink */
3931 for_each_active_range_index_in_nid(i
, nid
) {
3932 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3933 early_node_map
[i
].end_pfn
<= end_pfn
) {
3935 early_node_map
[i
].start_pfn
= 0;
3936 early_node_map
[i
].end_pfn
= 0;
3940 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3941 early_node_map
[i
].end_pfn
> start_pfn
) {
3942 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3943 early_node_map
[i
].end_pfn
= start_pfn
;
3944 if (temp_end_pfn
> end_pfn
)
3945 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3948 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3949 early_node_map
[i
].end_pfn
> end_pfn
&&
3950 early_node_map
[i
].start_pfn
< end_pfn
) {
3951 early_node_map
[i
].start_pfn
= end_pfn
;
3959 /* remove the blank ones */
3960 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3961 if (early_node_map
[i
].nid
!= nid
)
3963 if (early_node_map
[i
].end_pfn
)
3965 /* we found it, get rid of it */
3966 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3967 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3968 sizeof(early_node_map
[j
]));
3969 j
= nr_nodemap_entries
- 1;
3970 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3971 nr_nodemap_entries
--;
3976 * remove_all_active_ranges - Remove all currently registered regions
3978 * During discovery, it may be found that a table like SRAT is invalid
3979 * and an alternative discovery method must be used. This function removes
3980 * all currently registered regions.
3982 void __init
remove_all_active_ranges(void)
3984 memset(early_node_map
, 0, sizeof(early_node_map
));
3985 nr_nodemap_entries
= 0;
3988 /* Compare two active node_active_regions */
3989 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3991 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3992 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3994 /* Done this way to avoid overflows */
3995 if (arange
->start_pfn
> brange
->start_pfn
)
3997 if (arange
->start_pfn
< brange
->start_pfn
)
4003 /* sort the node_map by start_pfn */
4004 static void __init
sort_node_map(void)
4006 sort(early_node_map
, (size_t)nr_nodemap_entries
,
4007 sizeof(struct node_active_region
),
4008 cmp_node_active_region
, NULL
);
4011 /* Find the lowest pfn for a node */
4012 static unsigned long __init
find_min_pfn_for_node(int nid
)
4015 unsigned long min_pfn
= ULONG_MAX
;
4017 /* Assuming a sorted map, the first range found has the starting pfn */
4018 for_each_active_range_index_in_nid(i
, nid
)
4019 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
4021 if (min_pfn
== ULONG_MAX
) {
4023 "Could not find start_pfn for node %d\n", nid
);
4031 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4033 * It returns the minimum PFN based on information provided via
4034 * add_active_range().
4036 unsigned long __init
find_min_pfn_with_active_regions(void)
4038 return find_min_pfn_for_node(MAX_NUMNODES
);
4042 * early_calculate_totalpages()
4043 * Sum pages in active regions for movable zone.
4044 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4046 static unsigned long __init
early_calculate_totalpages(void)
4049 unsigned long totalpages
= 0;
4051 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4052 unsigned long pages
= early_node_map
[i
].end_pfn
-
4053 early_node_map
[i
].start_pfn
;
4054 totalpages
+= pages
;
4056 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
4062 * Find the PFN the Movable zone begins in each node. Kernel memory
4063 * is spread evenly between nodes as long as the nodes have enough
4064 * memory. When they don't, some nodes will have more kernelcore than
4067 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
4070 unsigned long usable_startpfn
;
4071 unsigned long kernelcore_node
, kernelcore_remaining
;
4072 /* save the state before borrow the nodemask */
4073 nodemask_t saved_node_state
= node_states
[N_HIGH_MEMORY
];
4074 unsigned long totalpages
= early_calculate_totalpages();
4075 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4078 * If movablecore was specified, calculate what size of
4079 * kernelcore that corresponds so that memory usable for
4080 * any allocation type is evenly spread. If both kernelcore
4081 * and movablecore are specified, then the value of kernelcore
4082 * will be used for required_kernelcore if it's greater than
4083 * what movablecore would have allowed.
4085 if (required_movablecore
) {
4086 unsigned long corepages
;
4089 * Round-up so that ZONE_MOVABLE is at least as large as what
4090 * was requested by the user
4092 required_movablecore
=
4093 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4094 corepages
= totalpages
- required_movablecore
;
4096 required_kernelcore
= max(required_kernelcore
, corepages
);
4099 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4100 if (!required_kernelcore
)
4103 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4104 find_usable_zone_for_movable();
4105 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4108 /* Spread kernelcore memory as evenly as possible throughout nodes */
4109 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4110 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4112 * Recalculate kernelcore_node if the division per node
4113 * now exceeds what is necessary to satisfy the requested
4114 * amount of memory for the kernel
4116 if (required_kernelcore
< kernelcore_node
)
4117 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4120 * As the map is walked, we track how much memory is usable
4121 * by the kernel using kernelcore_remaining. When it is
4122 * 0, the rest of the node is usable by ZONE_MOVABLE
4124 kernelcore_remaining
= kernelcore_node
;
4126 /* Go through each range of PFNs within this node */
4127 for_each_active_range_index_in_nid(i
, nid
) {
4128 unsigned long start_pfn
, end_pfn
;
4129 unsigned long size_pages
;
4131 start_pfn
= max(early_node_map
[i
].start_pfn
,
4132 zone_movable_pfn
[nid
]);
4133 end_pfn
= early_node_map
[i
].end_pfn
;
4134 if (start_pfn
>= end_pfn
)
4137 /* Account for what is only usable for kernelcore */
4138 if (start_pfn
< usable_startpfn
) {
4139 unsigned long kernel_pages
;
4140 kernel_pages
= min(end_pfn
, usable_startpfn
)
4143 kernelcore_remaining
-= min(kernel_pages
,
4144 kernelcore_remaining
);
4145 required_kernelcore
-= min(kernel_pages
,
4146 required_kernelcore
);
4148 /* Continue if range is now fully accounted */
4149 if (end_pfn
<= usable_startpfn
) {
4152 * Push zone_movable_pfn to the end so
4153 * that if we have to rebalance
4154 * kernelcore across nodes, we will
4155 * not double account here
4157 zone_movable_pfn
[nid
] = end_pfn
;
4160 start_pfn
= usable_startpfn
;
4164 * The usable PFN range for ZONE_MOVABLE is from
4165 * start_pfn->end_pfn. Calculate size_pages as the
4166 * number of pages used as kernelcore
4168 size_pages
= end_pfn
- start_pfn
;
4169 if (size_pages
> kernelcore_remaining
)
4170 size_pages
= kernelcore_remaining
;
4171 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4174 * Some kernelcore has been met, update counts and
4175 * break if the kernelcore for this node has been
4178 required_kernelcore
-= min(required_kernelcore
,
4180 kernelcore_remaining
-= size_pages
;
4181 if (!kernelcore_remaining
)
4187 * If there is still required_kernelcore, we do another pass with one
4188 * less node in the count. This will push zone_movable_pfn[nid] further
4189 * along on the nodes that still have memory until kernelcore is
4193 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4196 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4197 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4198 zone_movable_pfn
[nid
] =
4199 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4202 /* restore the node_state */
4203 node_states
[N_HIGH_MEMORY
] = saved_node_state
;
4206 /* Any regular memory on that node ? */
4207 static void check_for_regular_memory(pg_data_t
*pgdat
)
4209 #ifdef CONFIG_HIGHMEM
4210 enum zone_type zone_type
;
4212 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4213 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4214 if (zone
->present_pages
)
4215 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4221 * free_area_init_nodes - Initialise all pg_data_t and zone data
4222 * @max_zone_pfn: an array of max PFNs for each zone
4224 * This will call free_area_init_node() for each active node in the system.
4225 * Using the page ranges provided by add_active_range(), the size of each
4226 * zone in each node and their holes is calculated. If the maximum PFN
4227 * between two adjacent zones match, it is assumed that the zone is empty.
4228 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4229 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4230 * starts where the previous one ended. For example, ZONE_DMA32 starts
4231 * at arch_max_dma_pfn.
4233 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4238 /* Sort early_node_map as initialisation assumes it is sorted */
4241 /* Record where the zone boundaries are */
4242 memset(arch_zone_lowest_possible_pfn
, 0,
4243 sizeof(arch_zone_lowest_possible_pfn
));
4244 memset(arch_zone_highest_possible_pfn
, 0,
4245 sizeof(arch_zone_highest_possible_pfn
));
4246 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4247 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4248 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4249 if (i
== ZONE_MOVABLE
)
4251 arch_zone_lowest_possible_pfn
[i
] =
4252 arch_zone_highest_possible_pfn
[i
-1];
4253 arch_zone_highest_possible_pfn
[i
] =
4254 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4256 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4257 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4259 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4260 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4261 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4263 /* Print out the zone ranges */
4264 printk("Zone PFN ranges:\n");
4265 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4266 if (i
== ZONE_MOVABLE
)
4268 printk(" %-8s %0#10lx -> %0#10lx\n",
4270 arch_zone_lowest_possible_pfn
[i
],
4271 arch_zone_highest_possible_pfn
[i
]);
4274 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4275 printk("Movable zone start PFN for each node\n");
4276 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4277 if (zone_movable_pfn
[i
])
4278 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4281 /* Print out the early_node_map[] */
4282 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4283 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4284 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4285 early_node_map
[i
].start_pfn
,
4286 early_node_map
[i
].end_pfn
);
4288 /* Initialise every node */
4289 mminit_verify_pageflags_layout();
4290 setup_nr_node_ids();
4291 for_each_online_node(nid
) {
4292 pg_data_t
*pgdat
= NODE_DATA(nid
);
4293 free_area_init_node(nid
, NULL
,
4294 find_min_pfn_for_node(nid
), NULL
);
4296 /* Any memory on that node */
4297 if (pgdat
->node_present_pages
)
4298 node_set_state(nid
, N_HIGH_MEMORY
);
4299 check_for_regular_memory(pgdat
);
4303 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4305 unsigned long long coremem
;
4309 coremem
= memparse(p
, &p
);
4310 *core
= coremem
>> PAGE_SHIFT
;
4312 /* Paranoid check that UL is enough for the coremem value */
4313 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4319 * kernelcore=size sets the amount of memory for use for allocations that
4320 * cannot be reclaimed or migrated.
4322 static int __init
cmdline_parse_kernelcore(char *p
)
4324 return cmdline_parse_core(p
, &required_kernelcore
);
4328 * movablecore=size sets the amount of memory for use for allocations that
4329 * can be reclaimed or migrated.
4331 static int __init
cmdline_parse_movablecore(char *p
)
4333 return cmdline_parse_core(p
, &required_movablecore
);
4336 early_param("kernelcore", cmdline_parse_kernelcore
);
4337 early_param("movablecore", cmdline_parse_movablecore
);
4339 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4342 * set_dma_reserve - set the specified number of pages reserved in the first zone
4343 * @new_dma_reserve: The number of pages to mark reserved
4345 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4346 * In the DMA zone, a significant percentage may be consumed by kernel image
4347 * and other unfreeable allocations which can skew the watermarks badly. This
4348 * function may optionally be used to account for unfreeable pages in the
4349 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4350 * smaller per-cpu batchsize.
4352 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4354 dma_reserve
= new_dma_reserve
;
4357 #ifndef CONFIG_NEED_MULTIPLE_NODES
4358 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4359 EXPORT_SYMBOL(contig_page_data
);
4362 void __init
free_area_init(unsigned long *zones_size
)
4364 free_area_init_node(0, zones_size
,
4365 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4368 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4369 unsigned long action
, void *hcpu
)
4371 int cpu
= (unsigned long)hcpu
;
4373 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4377 * Spill the event counters of the dead processor
4378 * into the current processors event counters.
4379 * This artificially elevates the count of the current
4382 vm_events_fold_cpu(cpu
);
4385 * Zero the differential counters of the dead processor
4386 * so that the vm statistics are consistent.
4388 * This is only okay since the processor is dead and cannot
4389 * race with what we are doing.
4391 refresh_cpu_vm_stats(cpu
);
4396 void __init
page_alloc_init(void)
4398 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4402 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4403 * or min_free_kbytes changes.
4405 static void calculate_totalreserve_pages(void)
4407 struct pglist_data
*pgdat
;
4408 unsigned long reserve_pages
= 0;
4409 enum zone_type i
, j
;
4411 for_each_online_pgdat(pgdat
) {
4412 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4413 struct zone
*zone
= pgdat
->node_zones
+ i
;
4414 unsigned long max
= 0;
4416 /* Find valid and maximum lowmem_reserve in the zone */
4417 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4418 if (zone
->lowmem_reserve
[j
] > max
)
4419 max
= zone
->lowmem_reserve
[j
];
4422 /* we treat the high watermark as reserved pages. */
4423 max
+= high_wmark_pages(zone
);
4425 if (max
> zone
->present_pages
)
4426 max
= zone
->present_pages
;
4427 reserve_pages
+= max
;
4430 totalreserve_pages
= reserve_pages
;
4434 * setup_per_zone_lowmem_reserve - called whenever
4435 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4436 * has a correct pages reserved value, so an adequate number of
4437 * pages are left in the zone after a successful __alloc_pages().
4439 static void setup_per_zone_lowmem_reserve(void)
4441 struct pglist_data
*pgdat
;
4442 enum zone_type j
, idx
;
4444 for_each_online_pgdat(pgdat
) {
4445 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4446 struct zone
*zone
= pgdat
->node_zones
+ j
;
4447 unsigned long present_pages
= zone
->present_pages
;
4449 zone
->lowmem_reserve
[j
] = 0;
4453 struct zone
*lower_zone
;
4457 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4458 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4460 lower_zone
= pgdat
->node_zones
+ idx
;
4461 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4462 sysctl_lowmem_reserve_ratio
[idx
];
4463 present_pages
+= lower_zone
->present_pages
;
4468 /* update totalreserve_pages */
4469 calculate_totalreserve_pages();
4473 * setup_per_zone_wmarks - called when min_free_kbytes changes
4474 * or when memory is hot-{added|removed}
4476 * Ensures that the watermark[min,low,high] values for each zone are set
4477 * correctly with respect to min_free_kbytes.
4479 void setup_per_zone_wmarks(void)
4481 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4482 unsigned long lowmem_pages
= 0;
4484 unsigned long flags
;
4486 /* Calculate total number of !ZONE_HIGHMEM pages */
4487 for_each_zone(zone
) {
4488 if (!is_highmem(zone
))
4489 lowmem_pages
+= zone
->present_pages
;
4492 for_each_zone(zone
) {
4495 spin_lock_irqsave(&zone
->lock
, flags
);
4496 tmp
= (u64
)pages_min
* zone
->present_pages
;
4497 do_div(tmp
, lowmem_pages
);
4498 if (is_highmem(zone
)) {
4500 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4501 * need highmem pages, so cap pages_min to a small
4504 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4505 * deltas controls asynch page reclaim, and so should
4506 * not be capped for highmem.
4510 min_pages
= zone
->present_pages
/ 1024;
4511 if (min_pages
< SWAP_CLUSTER_MAX
)
4512 min_pages
= SWAP_CLUSTER_MAX
;
4513 if (min_pages
> 128)
4515 zone
->watermark
[WMARK_MIN
] = min_pages
;
4518 * If it's a lowmem zone, reserve a number of pages
4519 * proportionate to the zone's size.
4521 zone
->watermark
[WMARK_MIN
] = tmp
;
4524 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
4525 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
4526 setup_zone_migrate_reserve(zone
);
4527 spin_unlock_irqrestore(&zone
->lock
, flags
);
4530 /* update totalreserve_pages */
4531 calculate_totalreserve_pages();
4535 * The inactive anon list should be small enough that the VM never has to
4536 * do too much work, but large enough that each inactive page has a chance
4537 * to be referenced again before it is swapped out.
4539 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4540 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4541 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4542 * the anonymous pages are kept on the inactive list.
4545 * memory ratio inactive anon
4546 * -------------------------------------
4555 void calculate_zone_inactive_ratio(struct zone
*zone
)
4557 unsigned int gb
, ratio
;
4559 /* Zone size in gigabytes */
4560 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4562 ratio
= int_sqrt(10 * gb
);
4566 zone
->inactive_ratio
= ratio
;
4569 static void __init
setup_per_zone_inactive_ratio(void)
4574 calculate_zone_inactive_ratio(zone
);
4578 * Initialise min_free_kbytes.
4580 * For small machines we want it small (128k min). For large machines
4581 * we want it large (64MB max). But it is not linear, because network
4582 * bandwidth does not increase linearly with machine size. We use
4584 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4585 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4601 static int __init
init_per_zone_wmark_min(void)
4603 unsigned long lowmem_kbytes
;
4605 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4607 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4608 if (min_free_kbytes
< 128)
4609 min_free_kbytes
= 128;
4610 if (min_free_kbytes
> 65536)
4611 min_free_kbytes
= 65536;
4612 setup_per_zone_wmarks();
4613 setup_per_zone_lowmem_reserve();
4614 setup_per_zone_inactive_ratio();
4617 module_init(init_per_zone_wmark_min
)
4620 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4621 * that we can call two helper functions whenever min_free_kbytes
4624 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4625 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4627 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4629 setup_per_zone_wmarks();
4634 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4635 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4640 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4645 zone
->min_unmapped_pages
= (zone
->present_pages
*
4646 sysctl_min_unmapped_ratio
) / 100;
4650 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4651 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4656 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4661 zone
->min_slab_pages
= (zone
->present_pages
*
4662 sysctl_min_slab_ratio
) / 100;
4668 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4669 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4670 * whenever sysctl_lowmem_reserve_ratio changes.
4672 * The reserve ratio obviously has absolutely no relation with the
4673 * minimum watermarks. The lowmem reserve ratio can only make sense
4674 * if in function of the boot time zone sizes.
4676 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4677 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4679 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4680 setup_per_zone_lowmem_reserve();
4685 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4686 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4687 * can have before it gets flushed back to buddy allocator.
4690 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4691 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4697 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4698 if (!write
|| (ret
== -EINVAL
))
4700 for_each_populated_zone(zone
) {
4701 for_each_online_cpu(cpu
) {
4703 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4704 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4710 int hashdist
= HASHDIST_DEFAULT
;
4713 static int __init
set_hashdist(char *str
)
4717 hashdist
= simple_strtoul(str
, &str
, 0);
4720 __setup("hashdist=", set_hashdist
);
4724 * allocate a large system hash table from bootmem
4725 * - it is assumed that the hash table must contain an exact power-of-2
4726 * quantity of entries
4727 * - limit is the number of hash buckets, not the total allocation size
4729 void *__init
alloc_large_system_hash(const char *tablename
,
4730 unsigned long bucketsize
,
4731 unsigned long numentries
,
4734 unsigned int *_hash_shift
,
4735 unsigned int *_hash_mask
,
4736 unsigned long limit
)
4738 unsigned long long max
= limit
;
4739 unsigned long log2qty
, size
;
4742 /* allow the kernel cmdline to have a say */
4744 /* round applicable memory size up to nearest megabyte */
4745 numentries
= nr_kernel_pages
;
4746 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4747 numentries
>>= 20 - PAGE_SHIFT
;
4748 numentries
<<= 20 - PAGE_SHIFT
;
4750 /* limit to 1 bucket per 2^scale bytes of low memory */
4751 if (scale
> PAGE_SHIFT
)
4752 numentries
>>= (scale
- PAGE_SHIFT
);
4754 numentries
<<= (PAGE_SHIFT
- scale
);
4756 /* Make sure we've got at least a 0-order allocation.. */
4757 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4758 numentries
= PAGE_SIZE
/ bucketsize
;
4760 numentries
= roundup_pow_of_two(numentries
);
4762 /* limit allocation size to 1/16 total memory by default */
4764 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4765 do_div(max
, bucketsize
);
4768 if (numentries
> max
)
4771 log2qty
= ilog2(numentries
);
4774 size
= bucketsize
<< log2qty
;
4775 if (flags
& HASH_EARLY
)
4776 table
= alloc_bootmem_nopanic(size
);
4778 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4781 * If bucketsize is not a power-of-two, we may free
4782 * some pages at the end of hash table which
4783 * alloc_pages_exact() automatically does
4785 if (get_order(size
) < MAX_ORDER
) {
4786 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
4787 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
4790 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4793 panic("Failed to allocate %s hash table\n", tablename
);
4795 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4798 ilog2(size
) - PAGE_SHIFT
,
4802 *_hash_shift
= log2qty
;
4804 *_hash_mask
= (1 << log2qty
) - 1;
4809 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4810 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4813 #ifdef CONFIG_SPARSEMEM
4814 return __pfn_to_section(pfn
)->pageblock_flags
;
4816 return zone
->pageblock_flags
;
4817 #endif /* CONFIG_SPARSEMEM */
4820 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4822 #ifdef CONFIG_SPARSEMEM
4823 pfn
&= (PAGES_PER_SECTION
-1);
4824 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4826 pfn
= pfn
- zone
->zone_start_pfn
;
4827 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4828 #endif /* CONFIG_SPARSEMEM */
4832 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4833 * @page: The page within the block of interest
4834 * @start_bitidx: The first bit of interest to retrieve
4835 * @end_bitidx: The last bit of interest
4836 * returns pageblock_bits flags
4838 unsigned long get_pageblock_flags_group(struct page
*page
,
4839 int start_bitidx
, int end_bitidx
)
4842 unsigned long *bitmap
;
4843 unsigned long pfn
, bitidx
;
4844 unsigned long flags
= 0;
4845 unsigned long value
= 1;
4847 zone
= page_zone(page
);
4848 pfn
= page_to_pfn(page
);
4849 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4850 bitidx
= pfn_to_bitidx(zone
, pfn
);
4852 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4853 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4860 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4861 * @page: The page within the block of interest
4862 * @start_bitidx: The first bit of interest
4863 * @end_bitidx: The last bit of interest
4864 * @flags: The flags to set
4866 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4867 int start_bitidx
, int end_bitidx
)
4870 unsigned long *bitmap
;
4871 unsigned long pfn
, bitidx
;
4872 unsigned long value
= 1;
4874 zone
= page_zone(page
);
4875 pfn
= page_to_pfn(page
);
4876 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4877 bitidx
= pfn_to_bitidx(zone
, pfn
);
4878 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4879 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4881 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4883 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4885 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4889 * This is designed as sub function...plz see page_isolation.c also.
4890 * set/clear page block's type to be ISOLATE.
4891 * page allocater never alloc memory from ISOLATE block.
4894 int set_migratetype_isolate(struct page
*page
)
4897 unsigned long flags
;
4900 zone
= page_zone(page
);
4901 spin_lock_irqsave(&zone
->lock
, flags
);
4903 * In future, more migrate types will be able to be isolation target.
4905 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4907 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4908 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4911 spin_unlock_irqrestore(&zone
->lock
, flags
);
4917 void unset_migratetype_isolate(struct page
*page
)
4920 unsigned long flags
;
4921 zone
= page_zone(page
);
4922 spin_lock_irqsave(&zone
->lock
, flags
);
4923 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4925 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4926 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4928 spin_unlock_irqrestore(&zone
->lock
, flags
);
4931 #ifdef CONFIG_MEMORY_HOTREMOVE
4933 * All pages in the range must be isolated before calling this.
4936 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4942 unsigned long flags
;
4943 /* find the first valid pfn */
4944 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4949 zone
= page_zone(pfn_to_page(pfn
));
4950 spin_lock_irqsave(&zone
->lock
, flags
);
4952 while (pfn
< end_pfn
) {
4953 if (!pfn_valid(pfn
)) {
4957 page
= pfn_to_page(pfn
);
4958 BUG_ON(page_count(page
));
4959 BUG_ON(!PageBuddy(page
));
4960 order
= page_order(page
);
4961 #ifdef CONFIG_DEBUG_VM
4962 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4963 pfn
, 1 << order
, end_pfn
);
4965 list_del(&page
->lru
);
4966 rmv_page_order(page
);
4967 zone
->free_area
[order
].nr_free
--;
4968 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4970 for (i
= 0; i
< (1 << order
); i
++)
4971 SetPageReserved((page
+i
));
4972 pfn
+= (1 << order
);
4974 spin_unlock_irqrestore(&zone
->lock
, flags
);