2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/oom.h>
32 #include <linux/notifier.h>
33 #include <linux/topology.h>
34 #include <linux/sysctl.h>
35 #include <linux/cpu.h>
36 #include <linux/cpuset.h>
37 #include <linux/memory_hotplug.h>
38 #include <linux/nodemask.h>
39 #include <linux/vmalloc.h>
40 #include <linux/mempolicy.h>
41 #include <linux/stop_machine.h>
42 #include <linux/sort.h>
43 #include <linux/pfn.h>
44 #include <linux/backing-dev.h>
45 #include <linux/fault-inject.h>
46 #include <linux/page-isolation.h>
47 #include <linux/page_cgroup.h>
48 #include <linux/debugobjects.h>
49 #include <linux/kmemleak.h>
51 #include <asm/tlbflush.h>
52 #include <asm/div64.h>
56 * Array of node states.
58 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
59 [N_POSSIBLE
] = NODE_MASK_ALL
,
60 [N_ONLINE
] = { { [0] = 1UL } },
62 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
64 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
66 [N_CPU
] = { { [0] = 1UL } },
69 EXPORT_SYMBOL(node_states
);
71 unsigned long totalram_pages __read_mostly
;
72 unsigned long totalreserve_pages __read_mostly
;
73 unsigned long highest_memmap_pfn __read_mostly
;
74 int percpu_pagelist_fraction
;
76 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
77 int pageblock_order __read_mostly
;
80 static void __free_pages_ok(struct page
*page
, unsigned int order
);
83 * results with 256, 32 in the lowmem_reserve sysctl:
84 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
85 * 1G machine -> (16M dma, 784M normal, 224M high)
86 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
87 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
88 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
90 * TBD: should special case ZONE_DMA32 machines here - in those we normally
91 * don't need any ZONE_NORMAL reservation
93 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
94 #ifdef CONFIG_ZONE_DMA
97 #ifdef CONFIG_ZONE_DMA32
100 #ifdef CONFIG_HIGHMEM
106 EXPORT_SYMBOL(totalram_pages
);
108 static char * const zone_names
[MAX_NR_ZONES
] = {
109 #ifdef CONFIG_ZONE_DMA
112 #ifdef CONFIG_ZONE_DMA32
116 #ifdef CONFIG_HIGHMEM
122 int min_free_kbytes
= 1024;
124 unsigned long __meminitdata nr_kernel_pages
;
125 unsigned long __meminitdata nr_all_pages
;
126 static unsigned long __meminitdata dma_reserve
;
128 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
130 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
131 * ranges of memory (RAM) that may be registered with add_active_range().
132 * Ranges passed to add_active_range() will be merged if possible
133 * so the number of times add_active_range() can be called is
134 * related to the number of nodes and the number of holes
136 #ifdef CONFIG_MAX_ACTIVE_REGIONS
137 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
138 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
140 #if MAX_NUMNODES >= 32
141 /* If there can be many nodes, allow up to 50 holes per node */
142 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
144 /* By default, allow up to 256 distinct regions */
145 #define MAX_ACTIVE_REGIONS 256
149 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
150 static int __meminitdata nr_nodemap_entries
;
151 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
152 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
153 static unsigned long __initdata required_kernelcore
;
154 static unsigned long __initdata required_movablecore
;
155 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
157 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
159 EXPORT_SYMBOL(movable_zone
);
160 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
164 int nr_online_nodes __read_mostly
= 1;
165 EXPORT_SYMBOL(nr_node_ids
);
166 EXPORT_SYMBOL(nr_online_nodes
);
169 int page_group_by_mobility_disabled __read_mostly
;
171 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
174 if (unlikely(page_group_by_mobility_disabled
))
175 migratetype
= MIGRATE_UNMOVABLE
;
177 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
178 PB_migrate
, PB_migrate_end
);
181 bool oom_killer_disabled __read_mostly
;
183 #ifdef CONFIG_DEBUG_VM
184 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
188 unsigned long pfn
= page_to_pfn(page
);
191 seq
= zone_span_seqbegin(zone
);
192 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
194 else if (pfn
< zone
->zone_start_pfn
)
196 } while (zone_span_seqretry(zone
, seq
));
201 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
203 if (!pfn_valid_within(page_to_pfn(page
)))
205 if (zone
!= page_zone(page
))
211 * Temporary debugging check for pages not lying within a given zone.
213 static int bad_range(struct zone
*zone
, struct page
*page
)
215 if (page_outside_zone_boundaries(zone
, page
))
217 if (!page_is_consistent(zone
, page
))
223 static inline int bad_range(struct zone
*zone
, struct page
*page
)
229 static void bad_page(struct page
*page
)
231 static unsigned long resume
;
232 static unsigned long nr_shown
;
233 static unsigned long nr_unshown
;
236 * Allow a burst of 60 reports, then keep quiet for that minute;
237 * or allow a steady drip of one report per second.
239 if (nr_shown
== 60) {
240 if (time_before(jiffies
, resume
)) {
246 "BUG: Bad page state: %lu messages suppressed\n",
253 resume
= jiffies
+ 60 * HZ
;
255 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
256 current
->comm
, page_to_pfn(page
));
258 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
259 page
, (void *)page
->flags
, page_count(page
),
260 page_mapcount(page
), page
->mapping
, page
->index
);
264 /* Leave bad fields for debug, except PageBuddy could make trouble */
265 __ClearPageBuddy(page
);
266 add_taint(TAINT_BAD_PAGE
);
270 * Higher-order pages are called "compound pages". They are structured thusly:
272 * The first PAGE_SIZE page is called the "head page".
274 * The remaining PAGE_SIZE pages are called "tail pages".
276 * All pages have PG_compound set. All pages have their ->private pointing at
277 * the head page (even the head page has this).
279 * The first tail page's ->lru.next holds the address of the compound page's
280 * put_page() function. Its ->lru.prev holds the order of allocation.
281 * This usage means that zero-order pages may not be compound.
284 static void free_compound_page(struct page
*page
)
286 __free_pages_ok(page
, compound_order(page
));
289 void prep_compound_page(struct page
*page
, unsigned long order
)
292 int nr_pages
= 1 << order
;
294 set_compound_page_dtor(page
, free_compound_page
);
295 set_compound_order(page
, order
);
297 for (i
= 1; i
< nr_pages
; i
++) {
298 struct page
*p
= page
+ i
;
301 p
->first_page
= page
;
305 static int destroy_compound_page(struct page
*page
, unsigned long order
)
308 int nr_pages
= 1 << order
;
311 if (unlikely(compound_order(page
) != order
) ||
312 unlikely(!PageHead(page
))) {
317 __ClearPageHead(page
);
319 for (i
= 1; i
< nr_pages
; i
++) {
320 struct page
*p
= page
+ i
;
322 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
332 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
337 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
338 * and __GFP_HIGHMEM from hard or soft interrupt context.
340 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
341 for (i
= 0; i
< (1 << order
); i
++)
342 clear_highpage(page
+ i
);
345 static inline void set_page_order(struct page
*page
, int order
)
347 set_page_private(page
, order
);
348 __SetPageBuddy(page
);
351 static inline void rmv_page_order(struct page
*page
)
353 __ClearPageBuddy(page
);
354 set_page_private(page
, 0);
358 * Locate the struct page for both the matching buddy in our
359 * pair (buddy1) and the combined O(n+1) page they form (page).
361 * 1) Any buddy B1 will have an order O twin B2 which satisfies
362 * the following equation:
364 * For example, if the starting buddy (buddy2) is #8 its order
366 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
368 * 2) Any buddy B will have an order O+1 parent P which
369 * satisfies the following equation:
372 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
374 static inline struct page
*
375 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
377 unsigned long buddy_idx
= page_idx
^ (1 << order
);
379 return page
+ (buddy_idx
- page_idx
);
382 static inline unsigned long
383 __find_combined_index(unsigned long page_idx
, unsigned int order
)
385 return (page_idx
& ~(1 << order
));
389 * This function checks whether a page is free && is the buddy
390 * we can do coalesce a page and its buddy if
391 * (a) the buddy is not in a hole &&
392 * (b) the buddy is in the buddy system &&
393 * (c) a page and its buddy have the same order &&
394 * (d) a page and its buddy are in the same zone.
396 * For recording whether a page is in the buddy system, we use PG_buddy.
397 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
399 * For recording page's order, we use page_private(page).
401 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
404 if (!pfn_valid_within(page_to_pfn(buddy
)))
407 if (page_zone_id(page
) != page_zone_id(buddy
))
410 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
411 VM_BUG_ON(page_count(buddy
) != 0);
418 * Freeing function for a buddy system allocator.
420 * The concept of a buddy system is to maintain direct-mapped table
421 * (containing bit values) for memory blocks of various "orders".
422 * The bottom level table contains the map for the smallest allocatable
423 * units of memory (here, pages), and each level above it describes
424 * pairs of units from the levels below, hence, "buddies".
425 * At a high level, all that happens here is marking the table entry
426 * at the bottom level available, and propagating the changes upward
427 * as necessary, plus some accounting needed to play nicely with other
428 * parts of the VM system.
429 * At each level, we keep a list of pages, which are heads of continuous
430 * free pages of length of (1 << order) and marked with PG_buddy. Page's
431 * order is recorded in page_private(page) field.
432 * So when we are allocating or freeing one, we can derive the state of the
433 * other. That is, if we allocate a small block, and both were
434 * free, the remainder of the region must be split into blocks.
435 * If a block is freed, and its buddy is also free, then this
436 * triggers coalescing into a block of larger size.
441 static inline void __free_one_page(struct page
*page
,
442 struct zone
*zone
, unsigned int order
,
445 unsigned long page_idx
;
447 if (unlikely(PageCompound(page
)))
448 if (unlikely(destroy_compound_page(page
, order
)))
451 VM_BUG_ON(migratetype
== -1);
453 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
455 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
456 VM_BUG_ON(bad_range(zone
, page
));
458 while (order
< MAX_ORDER
-1) {
459 unsigned long combined_idx
;
462 buddy
= __page_find_buddy(page
, page_idx
, order
);
463 if (!page_is_buddy(page
, buddy
, order
))
466 /* Our buddy is free, merge with it and move up one order. */
467 list_del(&buddy
->lru
);
468 zone
->free_area
[order
].nr_free
--;
469 rmv_page_order(buddy
);
470 combined_idx
= __find_combined_index(page_idx
, order
);
471 page
= page
+ (combined_idx
- page_idx
);
472 page_idx
= combined_idx
;
475 set_page_order(page
, order
);
477 &zone
->free_area
[order
].free_list
[migratetype
]);
478 zone
->free_area
[order
].nr_free
++;
481 #ifdef CONFIG_HAVE_MLOCKED_PAGE_BIT
483 * free_page_mlock() -- clean up attempts to free and mlocked() page.
484 * Page should not be on lru, so no need to fix that up.
485 * free_pages_check() will verify...
487 static inline void free_page_mlock(struct page
*page
)
489 __ClearPageMlocked(page
);
490 __dec_zone_page_state(page
, NR_MLOCK
);
491 __count_vm_event(UNEVICTABLE_MLOCKFREED
);
494 static void free_page_mlock(struct page
*page
) { }
497 static inline int free_pages_check(struct page
*page
)
499 if (unlikely(page_mapcount(page
) |
500 (page
->mapping
!= NULL
) |
501 (atomic_read(&page
->_count
) != 0) |
502 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
506 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
507 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
512 * Frees a list of pages.
513 * Assumes all pages on list are in same zone, and of same order.
514 * count is the number of pages to free.
516 * If the zone was previously in an "all pages pinned" state then look to
517 * see if this freeing clears that state.
519 * And clear the zone's pages_scanned counter, to hold off the "all pages are
520 * pinned" detection logic.
522 static void free_pages_bulk(struct zone
*zone
, int count
,
523 struct list_head
*list
, int order
)
525 spin_lock(&zone
->lock
);
526 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
527 zone
->pages_scanned
= 0;
529 __mod_zone_page_state(zone
, NR_FREE_PAGES
, count
<< order
);
533 VM_BUG_ON(list_empty(list
));
534 page
= list_entry(list
->prev
, struct page
, lru
);
535 /* have to delete it as __free_one_page list manipulates */
536 list_del(&page
->lru
);
537 __free_one_page(page
, zone
, order
, page_private(page
));
539 spin_unlock(&zone
->lock
);
542 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
545 spin_lock(&zone
->lock
);
546 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
547 zone
->pages_scanned
= 0;
549 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1 << order
);
550 __free_one_page(page
, zone
, order
, migratetype
);
551 spin_unlock(&zone
->lock
);
554 static void __free_pages_ok(struct page
*page
, unsigned int order
)
559 int clearMlocked
= PageMlocked(page
);
561 for (i
= 0 ; i
< (1 << order
) ; ++i
)
562 bad
+= free_pages_check(page
+ i
);
566 if (!PageHighMem(page
)) {
567 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
568 debug_check_no_obj_freed(page_address(page
),
571 arch_free_page(page
, order
);
572 kernel_map_pages(page
, 1 << order
, 0);
574 local_irq_save(flags
);
575 if (unlikely(clearMlocked
))
576 free_page_mlock(page
);
577 __count_vm_events(PGFREE
, 1 << order
);
578 free_one_page(page_zone(page
), page
, order
,
579 get_pageblock_migratetype(page
));
580 local_irq_restore(flags
);
584 * permit the bootmem allocator to evade page validation on high-order frees
586 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
589 __ClearPageReserved(page
);
590 set_page_count(page
, 0);
591 set_page_refcounted(page
);
597 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
598 struct page
*p
= &page
[loop
];
600 if (loop
+ 1 < BITS_PER_LONG
)
602 __ClearPageReserved(p
);
603 set_page_count(p
, 0);
606 set_page_refcounted(page
);
607 __free_pages(page
, order
);
613 * The order of subdivision here is critical for the IO subsystem.
614 * Please do not alter this order without good reasons and regression
615 * testing. Specifically, as large blocks of memory are subdivided,
616 * the order in which smaller blocks are delivered depends on the order
617 * they're subdivided in this function. This is the primary factor
618 * influencing the order in which pages are delivered to the IO
619 * subsystem according to empirical testing, and this is also justified
620 * by considering the behavior of a buddy system containing a single
621 * large block of memory acted on by a series of small allocations.
622 * This behavior is a critical factor in sglist merging's success.
626 static inline void expand(struct zone
*zone
, struct page
*page
,
627 int low
, int high
, struct free_area
*area
,
630 unsigned long size
= 1 << high
;
636 VM_BUG_ON(bad_range(zone
, &page
[size
]));
637 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
639 set_page_order(&page
[size
], high
);
644 * This page is about to be returned from the page allocator
646 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
648 if (unlikely(page_mapcount(page
) |
649 (page
->mapping
!= NULL
) |
650 (atomic_read(&page
->_count
) != 0) |
651 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
656 set_page_private(page
, 0);
657 set_page_refcounted(page
);
659 arch_alloc_page(page
, order
);
660 kernel_map_pages(page
, 1 << order
, 1);
662 if (gfp_flags
& __GFP_ZERO
)
663 prep_zero_page(page
, order
, gfp_flags
);
665 if (order
&& (gfp_flags
& __GFP_COMP
))
666 prep_compound_page(page
, order
);
672 * Go through the free lists for the given migratetype and remove
673 * the smallest available page from the freelists
676 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
679 unsigned int current_order
;
680 struct free_area
* area
;
683 /* Find a page of the appropriate size in the preferred list */
684 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
685 area
= &(zone
->free_area
[current_order
]);
686 if (list_empty(&area
->free_list
[migratetype
]))
689 page
= list_entry(area
->free_list
[migratetype
].next
,
691 list_del(&page
->lru
);
692 rmv_page_order(page
);
694 expand(zone
, page
, order
, current_order
, area
, migratetype
);
703 * This array describes the order lists are fallen back to when
704 * the free lists for the desirable migrate type are depleted
706 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
707 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
708 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
709 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
710 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
714 * Move the free pages in a range to the free lists of the requested type.
715 * Note that start_page and end_pages are not aligned on a pageblock
716 * boundary. If alignment is required, use move_freepages_block()
718 static int move_freepages(struct zone
*zone
,
719 struct page
*start_page
, struct page
*end_page
,
726 #ifndef CONFIG_HOLES_IN_ZONE
728 * page_zone is not safe to call in this context when
729 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
730 * anyway as we check zone boundaries in move_freepages_block().
731 * Remove at a later date when no bug reports exist related to
732 * grouping pages by mobility
734 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
737 for (page
= start_page
; page
<= end_page
;) {
738 /* Make sure we are not inadvertently changing nodes */
739 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
741 if (!pfn_valid_within(page_to_pfn(page
))) {
746 if (!PageBuddy(page
)) {
751 order
= page_order(page
);
752 list_del(&page
->lru
);
754 &zone
->free_area
[order
].free_list
[migratetype
]);
756 pages_moved
+= 1 << order
;
762 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
765 unsigned long start_pfn
, end_pfn
;
766 struct page
*start_page
, *end_page
;
768 start_pfn
= page_to_pfn(page
);
769 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
770 start_page
= pfn_to_page(start_pfn
);
771 end_page
= start_page
+ pageblock_nr_pages
- 1;
772 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
774 /* Do not cross zone boundaries */
775 if (start_pfn
< zone
->zone_start_pfn
)
777 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
780 return move_freepages(zone
, start_page
, end_page
, migratetype
);
783 /* Remove an element from the buddy allocator from the fallback list */
784 static inline struct page
*
785 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
787 struct free_area
* area
;
792 /* Find the largest possible block of pages in the other list */
793 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
795 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
796 migratetype
= fallbacks
[start_migratetype
][i
];
798 /* MIGRATE_RESERVE handled later if necessary */
799 if (migratetype
== MIGRATE_RESERVE
)
802 area
= &(zone
->free_area
[current_order
]);
803 if (list_empty(&area
->free_list
[migratetype
]))
806 page
= list_entry(area
->free_list
[migratetype
].next
,
811 * If breaking a large block of pages, move all free
812 * pages to the preferred allocation list. If falling
813 * back for a reclaimable kernel allocation, be more
814 * agressive about taking ownership of free pages
816 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
817 start_migratetype
== MIGRATE_RECLAIMABLE
) {
819 pages
= move_freepages_block(zone
, page
,
822 /* Claim the whole block if over half of it is free */
823 if (pages
>= (1 << (pageblock_order
-1)))
824 set_pageblock_migratetype(page
,
827 migratetype
= start_migratetype
;
830 /* Remove the page from the freelists */
831 list_del(&page
->lru
);
832 rmv_page_order(page
);
834 if (current_order
== pageblock_order
)
835 set_pageblock_migratetype(page
,
838 expand(zone
, page
, order
, current_order
, area
, migratetype
);
847 * Do the hard work of removing an element from the buddy allocator.
848 * Call me with the zone->lock already held.
850 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
856 page
= __rmqueue_smallest(zone
, order
, migratetype
);
858 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
859 page
= __rmqueue_fallback(zone
, order
, migratetype
);
862 * Use MIGRATE_RESERVE rather than fail an allocation. goto
863 * is used because __rmqueue_smallest is an inline function
864 * and we want just one call site
867 migratetype
= MIGRATE_RESERVE
;
876 * Obtain a specified number of elements from the buddy allocator, all under
877 * a single hold of the lock, for efficiency. Add them to the supplied list.
878 * Returns the number of new pages which were placed at *list.
880 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
881 unsigned long count
, struct list_head
*list
,
886 spin_lock(&zone
->lock
);
887 for (i
= 0; i
< count
; ++i
) {
888 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
889 if (unlikely(page
== NULL
))
893 * Split buddy pages returned by expand() are received here
894 * in physical page order. The page is added to the callers and
895 * list and the list head then moves forward. From the callers
896 * perspective, the linked list is ordered by page number in
897 * some conditions. This is useful for IO devices that can
898 * merge IO requests if the physical pages are ordered
901 list_add(&page
->lru
, list
);
902 set_page_private(page
, migratetype
);
905 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
906 spin_unlock(&zone
->lock
);
912 * Called from the vmstat counter updater to drain pagesets of this
913 * currently executing processor on remote nodes after they have
916 * Note that this function must be called with the thread pinned to
917 * a single processor.
919 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
924 local_irq_save(flags
);
925 if (pcp
->count
>= pcp
->batch
)
926 to_drain
= pcp
->batch
;
928 to_drain
= pcp
->count
;
929 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
930 pcp
->count
-= to_drain
;
931 local_irq_restore(flags
);
936 * Drain pages of the indicated processor.
938 * The processor must either be the current processor and the
939 * thread pinned to the current processor or a processor that
942 static void drain_pages(unsigned int cpu
)
947 for_each_populated_zone(zone
) {
948 struct per_cpu_pageset
*pset
;
949 struct per_cpu_pages
*pcp
;
951 pset
= zone_pcp(zone
, cpu
);
954 local_irq_save(flags
);
955 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
957 local_irq_restore(flags
);
962 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
964 void drain_local_pages(void *arg
)
966 drain_pages(smp_processor_id());
970 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
972 void drain_all_pages(void)
974 on_each_cpu(drain_local_pages
, NULL
, 1);
977 #ifdef CONFIG_HIBERNATION
979 void mark_free_pages(struct zone
*zone
)
981 unsigned long pfn
, max_zone_pfn
;
984 struct list_head
*curr
;
986 if (!zone
->spanned_pages
)
989 spin_lock_irqsave(&zone
->lock
, flags
);
991 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
992 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
993 if (pfn_valid(pfn
)) {
994 struct page
*page
= pfn_to_page(pfn
);
996 if (!swsusp_page_is_forbidden(page
))
997 swsusp_unset_page_free(page
);
1000 for_each_migratetype_order(order
, t
) {
1001 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1004 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1005 for (i
= 0; i
< (1UL << order
); i
++)
1006 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1009 spin_unlock_irqrestore(&zone
->lock
, flags
);
1011 #endif /* CONFIG_PM */
1014 * Free a 0-order page
1016 static void free_hot_cold_page(struct page
*page
, int cold
)
1018 struct zone
*zone
= page_zone(page
);
1019 struct per_cpu_pages
*pcp
;
1020 unsigned long flags
;
1021 int clearMlocked
= PageMlocked(page
);
1024 page
->mapping
= NULL
;
1025 if (free_pages_check(page
))
1028 if (!PageHighMem(page
)) {
1029 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1030 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1032 arch_free_page(page
, 0);
1033 kernel_map_pages(page
, 1, 0);
1035 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1036 set_page_private(page
, get_pageblock_migratetype(page
));
1037 local_irq_save(flags
);
1038 if (unlikely(clearMlocked
))
1039 free_page_mlock(page
);
1040 __count_vm_event(PGFREE
);
1043 list_add_tail(&page
->lru
, &pcp
->list
);
1045 list_add(&page
->lru
, &pcp
->list
);
1047 if (pcp
->count
>= pcp
->high
) {
1048 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1049 pcp
->count
-= pcp
->batch
;
1051 local_irq_restore(flags
);
1055 void free_hot_page(struct page
*page
)
1057 free_hot_cold_page(page
, 0);
1060 void free_cold_page(struct page
*page
)
1062 free_hot_cold_page(page
, 1);
1066 * split_page takes a non-compound higher-order page, and splits it into
1067 * n (1<<order) sub-pages: page[0..n]
1068 * Each sub-page must be freed individually.
1070 * Note: this is probably too low level an operation for use in drivers.
1071 * Please consult with lkml before using this in your driver.
1073 void split_page(struct page
*page
, unsigned int order
)
1077 VM_BUG_ON(PageCompound(page
));
1078 VM_BUG_ON(!page_count(page
));
1079 for (i
= 1; i
< (1 << order
); i
++)
1080 set_page_refcounted(page
+ i
);
1084 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1085 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1089 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1090 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1093 unsigned long flags
;
1095 int cold
= !!(gfp_flags
& __GFP_COLD
);
1100 if (likely(order
== 0)) {
1101 struct per_cpu_pages
*pcp
;
1103 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1104 local_irq_save(flags
);
1106 pcp
->count
= rmqueue_bulk(zone
, 0,
1107 pcp
->batch
, &pcp
->list
, migratetype
);
1108 if (unlikely(!pcp
->count
))
1112 /* Find a page of the appropriate migrate type */
1114 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1115 if (page_private(page
) == migratetype
)
1118 list_for_each_entry(page
, &pcp
->list
, lru
)
1119 if (page_private(page
) == migratetype
)
1123 /* Allocate more to the pcp list if necessary */
1124 if (unlikely(&page
->lru
== &pcp
->list
)) {
1125 pcp
->count
+= rmqueue_bulk(zone
, 0,
1126 pcp
->batch
, &pcp
->list
, migratetype
);
1127 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1130 list_del(&page
->lru
);
1133 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1135 * __GFP_NOFAIL is not to be used in new code.
1137 * All __GFP_NOFAIL callers should be fixed so that they
1138 * properly detect and handle allocation failures.
1140 * We most definitely don't want callers attempting to
1141 * allocate greater than single-page units with
1144 WARN_ON_ONCE(order
> 0);
1146 spin_lock_irqsave(&zone
->lock
, flags
);
1147 page
= __rmqueue(zone
, order
, migratetype
);
1148 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1 << order
));
1149 spin_unlock(&zone
->lock
);
1154 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1155 zone_statistics(preferred_zone
, zone
);
1156 local_irq_restore(flags
);
1159 VM_BUG_ON(bad_range(zone
, page
));
1160 if (prep_new_page(page
, order
, gfp_flags
))
1165 local_irq_restore(flags
);
1170 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1171 #define ALLOC_WMARK_MIN WMARK_MIN
1172 #define ALLOC_WMARK_LOW WMARK_LOW
1173 #define ALLOC_WMARK_HIGH WMARK_HIGH
1174 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1176 /* Mask to get the watermark bits */
1177 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1179 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1180 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1181 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1183 #ifdef CONFIG_FAIL_PAGE_ALLOC
1185 static struct fail_page_alloc_attr
{
1186 struct fault_attr attr
;
1188 u32 ignore_gfp_highmem
;
1189 u32 ignore_gfp_wait
;
1192 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1194 struct dentry
*ignore_gfp_highmem_file
;
1195 struct dentry
*ignore_gfp_wait_file
;
1196 struct dentry
*min_order_file
;
1198 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1200 } fail_page_alloc
= {
1201 .attr
= FAULT_ATTR_INITIALIZER
,
1202 .ignore_gfp_wait
= 1,
1203 .ignore_gfp_highmem
= 1,
1207 static int __init
setup_fail_page_alloc(char *str
)
1209 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1211 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1213 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1215 if (order
< fail_page_alloc
.min_order
)
1217 if (gfp_mask
& __GFP_NOFAIL
)
1219 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1221 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1224 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1227 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1229 static int __init
fail_page_alloc_debugfs(void)
1231 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1235 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1239 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1241 fail_page_alloc
.ignore_gfp_wait_file
=
1242 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1243 &fail_page_alloc
.ignore_gfp_wait
);
1245 fail_page_alloc
.ignore_gfp_highmem_file
=
1246 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1247 &fail_page_alloc
.ignore_gfp_highmem
);
1248 fail_page_alloc
.min_order_file
=
1249 debugfs_create_u32("min-order", mode
, dir
,
1250 &fail_page_alloc
.min_order
);
1252 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1253 !fail_page_alloc
.ignore_gfp_highmem_file
||
1254 !fail_page_alloc
.min_order_file
) {
1256 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1257 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1258 debugfs_remove(fail_page_alloc
.min_order_file
);
1259 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1265 late_initcall(fail_page_alloc_debugfs
);
1267 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1269 #else /* CONFIG_FAIL_PAGE_ALLOC */
1271 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1276 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1279 * Return 1 if free pages are above 'mark'. This takes into account the order
1280 * of the allocation.
1282 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1283 int classzone_idx
, int alloc_flags
)
1285 /* free_pages my go negative - that's OK */
1287 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1290 if (alloc_flags
& ALLOC_HIGH
)
1292 if (alloc_flags
& ALLOC_HARDER
)
1295 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1297 for (o
= 0; o
< order
; o
++) {
1298 /* At the next order, this order's pages become unavailable */
1299 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1301 /* Require fewer higher order pages to be free */
1304 if (free_pages
<= min
)
1312 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1313 * skip over zones that are not allowed by the cpuset, or that have
1314 * been recently (in last second) found to be nearly full. See further
1315 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1316 * that have to skip over a lot of full or unallowed zones.
1318 * If the zonelist cache is present in the passed in zonelist, then
1319 * returns a pointer to the allowed node mask (either the current
1320 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1322 * If the zonelist cache is not available for this zonelist, does
1323 * nothing and returns NULL.
1325 * If the fullzones BITMAP in the zonelist cache is stale (more than
1326 * a second since last zap'd) then we zap it out (clear its bits.)
1328 * We hold off even calling zlc_setup, until after we've checked the
1329 * first zone in the zonelist, on the theory that most allocations will
1330 * be satisfied from that first zone, so best to examine that zone as
1331 * quickly as we can.
1333 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1335 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1336 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1338 zlc
= zonelist
->zlcache_ptr
;
1342 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1343 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1344 zlc
->last_full_zap
= jiffies
;
1347 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1348 &cpuset_current_mems_allowed
:
1349 &node_states
[N_HIGH_MEMORY
];
1350 return allowednodes
;
1354 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1355 * if it is worth looking at further for free memory:
1356 * 1) Check that the zone isn't thought to be full (doesn't have its
1357 * bit set in the zonelist_cache fullzones BITMAP).
1358 * 2) Check that the zones node (obtained from the zonelist_cache
1359 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1360 * Return true (non-zero) if zone is worth looking at further, or
1361 * else return false (zero) if it is not.
1363 * This check -ignores- the distinction between various watermarks,
1364 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1365 * found to be full for any variation of these watermarks, it will
1366 * be considered full for up to one second by all requests, unless
1367 * we are so low on memory on all allowed nodes that we are forced
1368 * into the second scan of the zonelist.
1370 * In the second scan we ignore this zonelist cache and exactly
1371 * apply the watermarks to all zones, even it is slower to do so.
1372 * We are low on memory in the second scan, and should leave no stone
1373 * unturned looking for a free page.
1375 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1376 nodemask_t
*allowednodes
)
1378 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1379 int i
; /* index of *z in zonelist zones */
1380 int n
; /* node that zone *z is on */
1382 zlc
= zonelist
->zlcache_ptr
;
1386 i
= z
- zonelist
->_zonerefs
;
1389 /* This zone is worth trying if it is allowed but not full */
1390 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1394 * Given 'z' scanning a zonelist, set the corresponding bit in
1395 * zlc->fullzones, so that subsequent attempts to allocate a page
1396 * from that zone don't waste time re-examining it.
1398 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1400 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1401 int i
; /* index of *z in zonelist zones */
1403 zlc
= zonelist
->zlcache_ptr
;
1407 i
= z
- zonelist
->_zonerefs
;
1409 set_bit(i
, zlc
->fullzones
);
1412 #else /* CONFIG_NUMA */
1414 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1419 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1420 nodemask_t
*allowednodes
)
1425 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1428 #endif /* CONFIG_NUMA */
1431 * get_page_from_freelist goes through the zonelist trying to allocate
1434 static struct page
*
1435 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1436 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1437 struct zone
*preferred_zone
, int migratetype
)
1440 struct page
*page
= NULL
;
1443 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1444 int zlc_active
= 0; /* set if using zonelist_cache */
1445 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1447 classzone_idx
= zone_idx(preferred_zone
);
1450 * Scan zonelist, looking for a zone with enough free.
1451 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1453 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1454 high_zoneidx
, nodemask
) {
1455 if (NUMA_BUILD
&& zlc_active
&&
1456 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1458 if ((alloc_flags
& ALLOC_CPUSET
) &&
1459 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1462 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1463 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1467 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1468 if (zone_watermark_ok(zone
, order
, mark
,
1469 classzone_idx
, alloc_flags
))
1472 if (zone_reclaim_mode
== 0)
1473 goto this_zone_full
;
1475 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1477 case ZONE_RECLAIM_NOSCAN
:
1480 case ZONE_RECLAIM_FULL
:
1481 /* scanned but unreclaimable */
1482 goto this_zone_full
;
1484 /* did we reclaim enough */
1485 if (!zone_watermark_ok(zone
, order
, mark
,
1486 classzone_idx
, alloc_flags
))
1487 goto this_zone_full
;
1492 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1493 gfp_mask
, migratetype
);
1498 zlc_mark_zone_full(zonelist
, z
);
1500 if (NUMA_BUILD
&& !did_zlc_setup
&& nr_online_nodes
> 1) {
1502 * we do zlc_setup after the first zone is tried but only
1503 * if there are multiple nodes make it worthwhile
1505 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1511 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1512 /* Disable zlc cache for second zonelist scan */
1520 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1521 unsigned long pages_reclaimed
)
1523 /* Do not loop if specifically requested */
1524 if (gfp_mask
& __GFP_NORETRY
)
1528 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1529 * means __GFP_NOFAIL, but that may not be true in other
1532 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1536 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1537 * specified, then we retry until we no longer reclaim any pages
1538 * (above), or we've reclaimed an order of pages at least as
1539 * large as the allocation's order. In both cases, if the
1540 * allocation still fails, we stop retrying.
1542 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
1546 * Don't let big-order allocations loop unless the caller
1547 * explicitly requests that.
1549 if (gfp_mask
& __GFP_NOFAIL
)
1555 static inline struct page
*
1556 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
1557 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1558 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1563 /* Acquire the OOM killer lock for the zones in zonelist */
1564 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1565 schedule_timeout_uninterruptible(1);
1570 * Go through the zonelist yet one more time, keep very high watermark
1571 * here, this is only to catch a parallel oom killing, we must fail if
1572 * we're still under heavy pressure.
1574 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1575 order
, zonelist
, high_zoneidx
,
1576 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
1577 preferred_zone
, migratetype
);
1581 /* The OOM killer will not help higher order allocs */
1582 if (order
> PAGE_ALLOC_COSTLY_ORDER
&& !(gfp_mask
& __GFP_NOFAIL
))
1585 /* Exhausted what can be done so it's blamo time */
1586 out_of_memory(zonelist
, gfp_mask
, order
);
1589 clear_zonelist_oom(zonelist
, gfp_mask
);
1593 /* The really slow allocator path where we enter direct reclaim */
1594 static inline struct page
*
1595 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1596 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1597 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1598 int migratetype
, unsigned long *did_some_progress
)
1600 struct page
*page
= NULL
;
1601 struct reclaim_state reclaim_state
;
1602 struct task_struct
*p
= current
;
1606 /* We now go into synchronous reclaim */
1607 cpuset_memory_pressure_bump();
1610 * The task's cpuset might have expanded its set of allowable nodes
1612 p
->flags
|= PF_MEMALLOC
;
1613 lockdep_set_current_reclaim_state(gfp_mask
);
1614 reclaim_state
.reclaimed_slab
= 0;
1615 p
->reclaim_state
= &reclaim_state
;
1617 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1619 p
->reclaim_state
= NULL
;
1620 lockdep_clear_current_reclaim_state();
1621 p
->flags
&= ~PF_MEMALLOC
;
1628 if (likely(*did_some_progress
))
1629 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1630 zonelist
, high_zoneidx
,
1631 alloc_flags
, preferred_zone
,
1637 * This is called in the allocator slow-path if the allocation request is of
1638 * sufficient urgency to ignore watermarks and take other desperate measures
1640 static inline struct page
*
1641 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1642 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1643 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1649 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1650 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
1651 preferred_zone
, migratetype
);
1653 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1654 congestion_wait(WRITE
, HZ
/50);
1655 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
1661 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
1662 enum zone_type high_zoneidx
)
1667 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1668 wakeup_kswapd(zone
, order
);
1672 gfp_to_alloc_flags(gfp_t gfp_mask
)
1674 struct task_struct
*p
= current
;
1675 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
1676 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1678 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1679 BUILD_BUG_ON(__GFP_HIGH
!= ALLOC_HIGH
);
1682 * The caller may dip into page reserves a bit more if the caller
1683 * cannot run direct reclaim, or if the caller has realtime scheduling
1684 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1685 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1687 alloc_flags
|= (gfp_mask
& __GFP_HIGH
);
1690 alloc_flags
|= ALLOC_HARDER
;
1692 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1693 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1695 alloc_flags
&= ~ALLOC_CPUSET
;
1696 } else if (unlikely(rt_task(p
)))
1697 alloc_flags
|= ALLOC_HARDER
;
1699 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
1700 if (!in_interrupt() &&
1701 ((p
->flags
& PF_MEMALLOC
) ||
1702 unlikely(test_thread_flag(TIF_MEMDIE
))))
1703 alloc_flags
|= ALLOC_NO_WATERMARKS
;
1709 static inline struct page
*
1710 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
1711 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1712 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1715 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1716 struct page
*page
= NULL
;
1718 unsigned long pages_reclaimed
= 0;
1719 unsigned long did_some_progress
;
1720 struct task_struct
*p
= current
;
1723 * In the slowpath, we sanity check order to avoid ever trying to
1724 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1725 * be using allocators in order of preference for an area that is
1728 if (WARN_ON_ONCE(order
>= MAX_ORDER
))
1732 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1733 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1734 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1735 * using a larger set of nodes after it has established that the
1736 * allowed per node queues are empty and that nodes are
1739 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1742 wake_all_kswapd(order
, zonelist
, high_zoneidx
);
1745 * OK, we're below the kswapd watermark and have kicked background
1746 * reclaim. Now things get more complex, so set up alloc_flags according
1747 * to how we want to proceed.
1749 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
1752 /* This is the last chance, in general, before the goto nopage. */
1753 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1754 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
1755 preferred_zone
, migratetype
);
1760 /* Allocate without watermarks if the context allows */
1761 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
1762 page
= __alloc_pages_high_priority(gfp_mask
, order
,
1763 zonelist
, high_zoneidx
, nodemask
,
1764 preferred_zone
, migratetype
);
1769 /* Atomic allocations - we can't balance anything */
1773 /* Avoid recursion of direct reclaim */
1774 if (p
->flags
& PF_MEMALLOC
)
1777 /* Try direct reclaim and then allocating */
1778 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
1779 zonelist
, high_zoneidx
,
1781 alloc_flags
, preferred_zone
,
1782 migratetype
, &did_some_progress
);
1787 * If we failed to make any progress reclaiming, then we are
1788 * running out of options and have to consider going OOM
1790 if (!did_some_progress
) {
1791 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1792 if (oom_killer_disabled
)
1794 page
= __alloc_pages_may_oom(gfp_mask
, order
,
1795 zonelist
, high_zoneidx
,
1796 nodemask
, preferred_zone
,
1802 * The OOM killer does not trigger for high-order
1803 * ~__GFP_NOFAIL allocations so if no progress is being
1804 * made, there are no other options and retrying is
1807 if (order
> PAGE_ALLOC_COSTLY_ORDER
&&
1808 !(gfp_mask
& __GFP_NOFAIL
))
1815 /* Check if we should retry the allocation */
1816 pages_reclaimed
+= did_some_progress
;
1817 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
1818 /* Wait for some write requests to complete then retry */
1819 congestion_wait(WRITE
, HZ
/50);
1824 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1825 printk(KERN_WARNING
"%s: page allocation failure."
1826 " order:%d, mode:0x%x\n",
1827 p
->comm
, order
, gfp_mask
);
1837 * This is the 'heart' of the zoned buddy allocator.
1840 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
1841 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1843 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1844 struct zone
*preferred_zone
;
1846 int migratetype
= allocflags_to_migratetype(gfp_mask
);
1848 lockdep_trace_alloc(gfp_mask
);
1850 might_sleep_if(gfp_mask
& __GFP_WAIT
);
1852 if (should_fail_alloc_page(gfp_mask
, order
))
1856 * Check the zones suitable for the gfp_mask contain at least one
1857 * valid zone. It's possible to have an empty zonelist as a result
1858 * of GFP_THISNODE and a memoryless node
1860 if (unlikely(!zonelist
->_zonerefs
->zone
))
1863 /* The preferred zone is used for statistics later */
1864 first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
, &preferred_zone
);
1865 if (!preferred_zone
)
1868 /* First allocation attempt */
1869 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1870 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
1871 preferred_zone
, migratetype
);
1872 if (unlikely(!page
))
1873 page
= __alloc_pages_slowpath(gfp_mask
, order
,
1874 zonelist
, high_zoneidx
, nodemask
,
1875 preferred_zone
, migratetype
);
1879 EXPORT_SYMBOL(__alloc_pages_nodemask
);
1882 * Common helper functions.
1884 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1887 page
= alloc_pages(gfp_mask
, order
);
1890 return (unsigned long) page_address(page
);
1893 EXPORT_SYMBOL(__get_free_pages
);
1895 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1900 * get_zeroed_page() returns a 32-bit address, which cannot represent
1903 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1905 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1907 return (unsigned long) page_address(page
);
1911 EXPORT_SYMBOL(get_zeroed_page
);
1913 void __pagevec_free(struct pagevec
*pvec
)
1915 int i
= pagevec_count(pvec
);
1918 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1921 void __free_pages(struct page
*page
, unsigned int order
)
1923 if (put_page_testzero(page
)) {
1925 free_hot_page(page
);
1927 __free_pages_ok(page
, order
);
1931 EXPORT_SYMBOL(__free_pages
);
1933 void free_pages(unsigned long addr
, unsigned int order
)
1936 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1937 __free_pages(virt_to_page((void *)addr
), order
);
1941 EXPORT_SYMBOL(free_pages
);
1944 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1945 * @size: the number of bytes to allocate
1946 * @gfp_mask: GFP flags for the allocation
1948 * This function is similar to alloc_pages(), except that it allocates the
1949 * minimum number of pages to satisfy the request. alloc_pages() can only
1950 * allocate memory in power-of-two pages.
1952 * This function is also limited by MAX_ORDER.
1954 * Memory allocated by this function must be released by free_pages_exact().
1956 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1958 unsigned int order
= get_order(size
);
1961 addr
= __get_free_pages(gfp_mask
, order
);
1963 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1964 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1966 split_page(virt_to_page(addr
), order
);
1967 while (used
< alloc_end
) {
1973 return (void *)addr
;
1975 EXPORT_SYMBOL(alloc_pages_exact
);
1978 * free_pages_exact - release memory allocated via alloc_pages_exact()
1979 * @virt: the value returned by alloc_pages_exact.
1980 * @size: size of allocation, same value as passed to alloc_pages_exact().
1982 * Release the memory allocated by a previous call to alloc_pages_exact.
1984 void free_pages_exact(void *virt
, size_t size
)
1986 unsigned long addr
= (unsigned long)virt
;
1987 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1989 while (addr
< end
) {
1994 EXPORT_SYMBOL(free_pages_exact
);
1996 static unsigned int nr_free_zone_pages(int offset
)
2001 /* Just pick one node, since fallback list is circular */
2002 unsigned int sum
= 0;
2004 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2006 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2007 unsigned long size
= zone
->present_pages
;
2008 unsigned long high
= high_wmark_pages(zone
);
2017 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2019 unsigned int nr_free_buffer_pages(void)
2021 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2023 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2026 * Amount of free RAM allocatable within all zones
2028 unsigned int nr_free_pagecache_pages(void)
2030 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2033 static inline void show_node(struct zone
*zone
)
2036 printk("Node %d ", zone_to_nid(zone
));
2039 void si_meminfo(struct sysinfo
*val
)
2041 val
->totalram
= totalram_pages
;
2043 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2044 val
->bufferram
= nr_blockdev_pages();
2045 val
->totalhigh
= totalhigh_pages
;
2046 val
->freehigh
= nr_free_highpages();
2047 val
->mem_unit
= PAGE_SIZE
;
2050 EXPORT_SYMBOL(si_meminfo
);
2053 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2055 pg_data_t
*pgdat
= NODE_DATA(nid
);
2057 val
->totalram
= pgdat
->node_present_pages
;
2058 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2059 #ifdef CONFIG_HIGHMEM
2060 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2061 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2067 val
->mem_unit
= PAGE_SIZE
;
2071 #define K(x) ((x) << (PAGE_SHIFT-10))
2074 * Show free area list (used inside shift_scroll-lock stuff)
2075 * We also calculate the percentage fragmentation. We do this by counting the
2076 * memory on each free list with the exception of the first item on the list.
2078 void show_free_areas(void)
2083 for_each_populated_zone(zone
) {
2085 printk("%s per-cpu:\n", zone
->name
);
2087 for_each_online_cpu(cpu
) {
2088 struct per_cpu_pageset
*pageset
;
2090 pageset
= zone_pcp(zone
, cpu
);
2092 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2093 cpu
, pageset
->pcp
.high
,
2094 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2098 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2099 " inactive_file:%lu"
2101 " dirty:%lu writeback:%lu unstable:%lu\n"
2102 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2103 global_page_state(NR_ACTIVE_ANON
),
2104 global_page_state(NR_ACTIVE_FILE
),
2105 global_page_state(NR_INACTIVE_ANON
),
2106 global_page_state(NR_INACTIVE_FILE
),
2107 global_page_state(NR_UNEVICTABLE
),
2108 global_page_state(NR_FILE_DIRTY
),
2109 global_page_state(NR_WRITEBACK
),
2110 global_page_state(NR_UNSTABLE_NFS
),
2111 global_page_state(NR_FREE_PAGES
),
2112 global_page_state(NR_SLAB_RECLAIMABLE
) +
2113 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2114 global_page_state(NR_FILE_MAPPED
),
2115 global_page_state(NR_PAGETABLE
),
2116 global_page_state(NR_BOUNCE
));
2118 for_each_populated_zone(zone
) {
2127 " active_anon:%lukB"
2128 " inactive_anon:%lukB"
2129 " active_file:%lukB"
2130 " inactive_file:%lukB"
2131 " unevictable:%lukB"
2133 " pages_scanned:%lu"
2134 " all_unreclaimable? %s"
2137 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2138 K(min_wmark_pages(zone
)),
2139 K(low_wmark_pages(zone
)),
2140 K(high_wmark_pages(zone
)),
2141 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2142 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2143 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2144 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2145 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2146 K(zone
->present_pages
),
2147 zone
->pages_scanned
,
2148 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
2150 printk("lowmem_reserve[]:");
2151 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2152 printk(" %lu", zone
->lowmem_reserve
[i
]);
2156 for_each_populated_zone(zone
) {
2157 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2160 printk("%s: ", zone
->name
);
2162 spin_lock_irqsave(&zone
->lock
, flags
);
2163 for (order
= 0; order
< MAX_ORDER
; order
++) {
2164 nr
[order
] = zone
->free_area
[order
].nr_free
;
2165 total
+= nr
[order
] << order
;
2167 spin_unlock_irqrestore(&zone
->lock
, flags
);
2168 for (order
= 0; order
< MAX_ORDER
; order
++)
2169 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2170 printk("= %lukB\n", K(total
));
2173 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2175 show_swap_cache_info();
2178 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2180 zoneref
->zone
= zone
;
2181 zoneref
->zone_idx
= zone_idx(zone
);
2185 * Builds allocation fallback zone lists.
2187 * Add all populated zones of a node to the zonelist.
2189 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2190 int nr_zones
, enum zone_type zone_type
)
2194 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2199 zone
= pgdat
->node_zones
+ zone_type
;
2200 if (populated_zone(zone
)) {
2201 zoneref_set_zone(zone
,
2202 &zonelist
->_zonerefs
[nr_zones
++]);
2203 check_highest_zone(zone_type
);
2206 } while (zone_type
);
2213 * 0 = automatic detection of better ordering.
2214 * 1 = order by ([node] distance, -zonetype)
2215 * 2 = order by (-zonetype, [node] distance)
2217 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2218 * the same zonelist. So only NUMA can configure this param.
2220 #define ZONELIST_ORDER_DEFAULT 0
2221 #define ZONELIST_ORDER_NODE 1
2222 #define ZONELIST_ORDER_ZONE 2
2224 /* zonelist order in the kernel.
2225 * set_zonelist_order() will set this to NODE or ZONE.
2227 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2228 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2232 /* The value user specified ....changed by config */
2233 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2234 /* string for sysctl */
2235 #define NUMA_ZONELIST_ORDER_LEN 16
2236 char numa_zonelist_order
[16] = "default";
2239 * interface for configure zonelist ordering.
2240 * command line option "numa_zonelist_order"
2241 * = "[dD]efault - default, automatic configuration.
2242 * = "[nN]ode - order by node locality, then by zone within node
2243 * = "[zZ]one - order by zone, then by locality within zone
2246 static int __parse_numa_zonelist_order(char *s
)
2248 if (*s
== 'd' || *s
== 'D') {
2249 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2250 } else if (*s
== 'n' || *s
== 'N') {
2251 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2252 } else if (*s
== 'z' || *s
== 'Z') {
2253 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2256 "Ignoring invalid numa_zonelist_order value: "
2263 static __init
int setup_numa_zonelist_order(char *s
)
2266 return __parse_numa_zonelist_order(s
);
2269 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2272 * sysctl handler for numa_zonelist_order
2274 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2275 struct file
*file
, void __user
*buffer
, size_t *length
,
2278 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2282 strncpy(saved_string
, (char*)table
->data
,
2283 NUMA_ZONELIST_ORDER_LEN
);
2284 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2288 int oldval
= user_zonelist_order
;
2289 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2291 * bogus value. restore saved string
2293 strncpy((char*)table
->data
, saved_string
,
2294 NUMA_ZONELIST_ORDER_LEN
);
2295 user_zonelist_order
= oldval
;
2296 } else if (oldval
!= user_zonelist_order
)
2297 build_all_zonelists();
2303 #define MAX_NODE_LOAD (nr_online_nodes)
2304 static int node_load
[MAX_NUMNODES
];
2307 * find_next_best_node - find the next node that should appear in a given node's fallback list
2308 * @node: node whose fallback list we're appending
2309 * @used_node_mask: nodemask_t of already used nodes
2311 * We use a number of factors to determine which is the next node that should
2312 * appear on a given node's fallback list. The node should not have appeared
2313 * already in @node's fallback list, and it should be the next closest node
2314 * according to the distance array (which contains arbitrary distance values
2315 * from each node to each node in the system), and should also prefer nodes
2316 * with no CPUs, since presumably they'll have very little allocation pressure
2317 * on them otherwise.
2318 * It returns -1 if no node is found.
2320 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2323 int min_val
= INT_MAX
;
2325 const struct cpumask
*tmp
= cpumask_of_node(0);
2327 /* Use the local node if we haven't already */
2328 if (!node_isset(node
, *used_node_mask
)) {
2329 node_set(node
, *used_node_mask
);
2333 for_each_node_state(n
, N_HIGH_MEMORY
) {
2335 /* Don't want a node to appear more than once */
2336 if (node_isset(n
, *used_node_mask
))
2339 /* Use the distance array to find the distance */
2340 val
= node_distance(node
, n
);
2342 /* Penalize nodes under us ("prefer the next node") */
2345 /* Give preference to headless and unused nodes */
2346 tmp
= cpumask_of_node(n
);
2347 if (!cpumask_empty(tmp
))
2348 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2350 /* Slight preference for less loaded node */
2351 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2352 val
+= node_load
[n
];
2354 if (val
< min_val
) {
2361 node_set(best_node
, *used_node_mask
);
2368 * Build zonelists ordered by node and zones within node.
2369 * This results in maximum locality--normal zone overflows into local
2370 * DMA zone, if any--but risks exhausting DMA zone.
2372 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2375 struct zonelist
*zonelist
;
2377 zonelist
= &pgdat
->node_zonelists
[0];
2378 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2380 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2382 zonelist
->_zonerefs
[j
].zone
= NULL
;
2383 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2387 * Build gfp_thisnode zonelists
2389 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2392 struct zonelist
*zonelist
;
2394 zonelist
= &pgdat
->node_zonelists
[1];
2395 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2396 zonelist
->_zonerefs
[j
].zone
= NULL
;
2397 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2401 * Build zonelists ordered by zone and nodes within zones.
2402 * This results in conserving DMA zone[s] until all Normal memory is
2403 * exhausted, but results in overflowing to remote node while memory
2404 * may still exist in local DMA zone.
2406 static int node_order
[MAX_NUMNODES
];
2408 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2411 int zone_type
; /* needs to be signed */
2413 struct zonelist
*zonelist
;
2415 zonelist
= &pgdat
->node_zonelists
[0];
2417 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2418 for (j
= 0; j
< nr_nodes
; j
++) {
2419 node
= node_order
[j
];
2420 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2421 if (populated_zone(z
)) {
2423 &zonelist
->_zonerefs
[pos
++]);
2424 check_highest_zone(zone_type
);
2428 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2429 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2432 static int default_zonelist_order(void)
2435 unsigned long low_kmem_size
,total_size
;
2439 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2440 * If they are really small and used heavily, the system can fall
2441 * into OOM very easily.
2442 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2444 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2447 for_each_online_node(nid
) {
2448 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2449 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2450 if (populated_zone(z
)) {
2451 if (zone_type
< ZONE_NORMAL
)
2452 low_kmem_size
+= z
->present_pages
;
2453 total_size
+= z
->present_pages
;
2457 if (!low_kmem_size
|| /* there are no DMA area. */
2458 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2459 return ZONELIST_ORDER_NODE
;
2461 * look into each node's config.
2462 * If there is a node whose DMA/DMA32 memory is very big area on
2463 * local memory, NODE_ORDER may be suitable.
2465 average_size
= total_size
/
2466 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2467 for_each_online_node(nid
) {
2470 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2471 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2472 if (populated_zone(z
)) {
2473 if (zone_type
< ZONE_NORMAL
)
2474 low_kmem_size
+= z
->present_pages
;
2475 total_size
+= z
->present_pages
;
2478 if (low_kmem_size
&&
2479 total_size
> average_size
&& /* ignore small node */
2480 low_kmem_size
> total_size
* 70/100)
2481 return ZONELIST_ORDER_NODE
;
2483 return ZONELIST_ORDER_ZONE
;
2486 static void set_zonelist_order(void)
2488 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2489 current_zonelist_order
= default_zonelist_order();
2491 current_zonelist_order
= user_zonelist_order
;
2494 static void build_zonelists(pg_data_t
*pgdat
)
2498 nodemask_t used_mask
;
2499 int local_node
, prev_node
;
2500 struct zonelist
*zonelist
;
2501 int order
= current_zonelist_order
;
2503 /* initialize zonelists */
2504 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2505 zonelist
= pgdat
->node_zonelists
+ i
;
2506 zonelist
->_zonerefs
[0].zone
= NULL
;
2507 zonelist
->_zonerefs
[0].zone_idx
= 0;
2510 /* NUMA-aware ordering of nodes */
2511 local_node
= pgdat
->node_id
;
2512 load
= nr_online_nodes
;
2513 prev_node
= local_node
;
2514 nodes_clear(used_mask
);
2516 memset(node_load
, 0, sizeof(node_load
));
2517 memset(node_order
, 0, sizeof(node_order
));
2520 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2521 int distance
= node_distance(local_node
, node
);
2524 * If another node is sufficiently far away then it is better
2525 * to reclaim pages in a zone before going off node.
2527 if (distance
> RECLAIM_DISTANCE
)
2528 zone_reclaim_mode
= 1;
2531 * We don't want to pressure a particular node.
2532 * So adding penalty to the first node in same
2533 * distance group to make it round-robin.
2535 if (distance
!= node_distance(local_node
, prev_node
))
2536 node_load
[node
] = load
;
2540 if (order
== ZONELIST_ORDER_NODE
)
2541 build_zonelists_in_node_order(pgdat
, node
);
2543 node_order
[j
++] = node
; /* remember order */
2546 if (order
== ZONELIST_ORDER_ZONE
) {
2547 /* calculate node order -- i.e., DMA last! */
2548 build_zonelists_in_zone_order(pgdat
, j
);
2551 build_thisnode_zonelists(pgdat
);
2554 /* Construct the zonelist performance cache - see further mmzone.h */
2555 static void build_zonelist_cache(pg_data_t
*pgdat
)
2557 struct zonelist
*zonelist
;
2558 struct zonelist_cache
*zlc
;
2561 zonelist
= &pgdat
->node_zonelists
[0];
2562 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2563 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2564 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2565 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2569 #else /* CONFIG_NUMA */
2571 static void set_zonelist_order(void)
2573 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2576 static void build_zonelists(pg_data_t
*pgdat
)
2578 int node
, local_node
;
2580 struct zonelist
*zonelist
;
2582 local_node
= pgdat
->node_id
;
2584 zonelist
= &pgdat
->node_zonelists
[0];
2585 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2588 * Now we build the zonelist so that it contains the zones
2589 * of all the other nodes.
2590 * We don't want to pressure a particular node, so when
2591 * building the zones for node N, we make sure that the
2592 * zones coming right after the local ones are those from
2593 * node N+1 (modulo N)
2595 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2596 if (!node_online(node
))
2598 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2601 for (node
= 0; node
< local_node
; node
++) {
2602 if (!node_online(node
))
2604 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2608 zonelist
->_zonerefs
[j
].zone
= NULL
;
2609 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2612 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2613 static void build_zonelist_cache(pg_data_t
*pgdat
)
2615 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2618 #endif /* CONFIG_NUMA */
2620 /* return values int ....just for stop_machine() */
2621 static int __build_all_zonelists(void *dummy
)
2625 for_each_online_node(nid
) {
2626 pg_data_t
*pgdat
= NODE_DATA(nid
);
2628 build_zonelists(pgdat
);
2629 build_zonelist_cache(pgdat
);
2634 void build_all_zonelists(void)
2636 set_zonelist_order();
2638 if (system_state
== SYSTEM_BOOTING
) {
2639 __build_all_zonelists(NULL
);
2640 mminit_verify_zonelist();
2641 cpuset_init_current_mems_allowed();
2643 /* we have to stop all cpus to guarantee there is no user
2645 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2646 /* cpuset refresh routine should be here */
2648 vm_total_pages
= nr_free_pagecache_pages();
2650 * Disable grouping by mobility if the number of pages in the
2651 * system is too low to allow the mechanism to work. It would be
2652 * more accurate, but expensive to check per-zone. This check is
2653 * made on memory-hotadd so a system can start with mobility
2654 * disabled and enable it later
2656 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2657 page_group_by_mobility_disabled
= 1;
2659 page_group_by_mobility_disabled
= 0;
2661 printk("Built %i zonelists in %s order, mobility grouping %s. "
2662 "Total pages: %ld\n",
2664 zonelist_order_name
[current_zonelist_order
],
2665 page_group_by_mobility_disabled
? "off" : "on",
2668 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2673 * Helper functions to size the waitqueue hash table.
2674 * Essentially these want to choose hash table sizes sufficiently
2675 * large so that collisions trying to wait on pages are rare.
2676 * But in fact, the number of active page waitqueues on typical
2677 * systems is ridiculously low, less than 200. So this is even
2678 * conservative, even though it seems large.
2680 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2681 * waitqueues, i.e. the size of the waitq table given the number of pages.
2683 #define PAGES_PER_WAITQUEUE 256
2685 #ifndef CONFIG_MEMORY_HOTPLUG
2686 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2688 unsigned long size
= 1;
2690 pages
/= PAGES_PER_WAITQUEUE
;
2692 while (size
< pages
)
2696 * Once we have dozens or even hundreds of threads sleeping
2697 * on IO we've got bigger problems than wait queue collision.
2698 * Limit the size of the wait table to a reasonable size.
2700 size
= min(size
, 4096UL);
2702 return max(size
, 4UL);
2706 * A zone's size might be changed by hot-add, so it is not possible to determine
2707 * a suitable size for its wait_table. So we use the maximum size now.
2709 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2711 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2712 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2713 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2715 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2716 * or more by the traditional way. (See above). It equals:
2718 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2719 * ia64(16K page size) : = ( 8G + 4M)byte.
2720 * powerpc (64K page size) : = (32G +16M)byte.
2722 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2729 * This is an integer logarithm so that shifts can be used later
2730 * to extract the more random high bits from the multiplicative
2731 * hash function before the remainder is taken.
2733 static inline unsigned long wait_table_bits(unsigned long size
)
2738 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2741 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2742 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2743 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
2744 * higher will lead to a bigger reserve which will get freed as contiguous
2745 * blocks as reclaim kicks in
2747 static void setup_zone_migrate_reserve(struct zone
*zone
)
2749 unsigned long start_pfn
, pfn
, end_pfn
;
2751 unsigned long reserve
, block_migratetype
;
2753 /* Get the start pfn, end pfn and the number of blocks to reserve */
2754 start_pfn
= zone
->zone_start_pfn
;
2755 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2756 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
2759 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2760 if (!pfn_valid(pfn
))
2762 page
= pfn_to_page(pfn
);
2764 /* Watch out for overlapping nodes */
2765 if (page_to_nid(page
) != zone_to_nid(zone
))
2768 /* Blocks with reserved pages will never free, skip them. */
2769 if (PageReserved(page
))
2772 block_migratetype
= get_pageblock_migratetype(page
);
2774 /* If this block is reserved, account for it */
2775 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2780 /* Suitable for reserving if this block is movable */
2781 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2782 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2783 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2789 * If the reserve is met and this is a previous reserved block,
2792 if (block_migratetype
== MIGRATE_RESERVE
) {
2793 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2794 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2800 * Initially all pages are reserved - free ones are freed
2801 * up by free_all_bootmem() once the early boot process is
2802 * done. Non-atomic initialization, single-pass.
2804 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2805 unsigned long start_pfn
, enum memmap_context context
)
2808 unsigned long end_pfn
= start_pfn
+ size
;
2812 if (highest_memmap_pfn
< end_pfn
- 1)
2813 highest_memmap_pfn
= end_pfn
- 1;
2815 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2816 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2818 * There can be holes in boot-time mem_map[]s
2819 * handed to this function. They do not
2820 * exist on hotplugged memory.
2822 if (context
== MEMMAP_EARLY
) {
2823 if (!early_pfn_valid(pfn
))
2825 if (!early_pfn_in_nid(pfn
, nid
))
2828 page
= pfn_to_page(pfn
);
2829 set_page_links(page
, zone
, nid
, pfn
);
2830 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2831 init_page_count(page
);
2832 reset_page_mapcount(page
);
2833 SetPageReserved(page
);
2835 * Mark the block movable so that blocks are reserved for
2836 * movable at startup. This will force kernel allocations
2837 * to reserve their blocks rather than leaking throughout
2838 * the address space during boot when many long-lived
2839 * kernel allocations are made. Later some blocks near
2840 * the start are marked MIGRATE_RESERVE by
2841 * setup_zone_migrate_reserve()
2843 * bitmap is created for zone's valid pfn range. but memmap
2844 * can be created for invalid pages (for alignment)
2845 * check here not to call set_pageblock_migratetype() against
2848 if ((z
->zone_start_pfn
<= pfn
)
2849 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2850 && !(pfn
& (pageblock_nr_pages
- 1)))
2851 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2853 INIT_LIST_HEAD(&page
->lru
);
2854 #ifdef WANT_PAGE_VIRTUAL
2855 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2856 if (!is_highmem_idx(zone
))
2857 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2862 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2865 for_each_migratetype_order(order
, t
) {
2866 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2867 zone
->free_area
[order
].nr_free
= 0;
2871 #ifndef __HAVE_ARCH_MEMMAP_INIT
2872 #define memmap_init(size, nid, zone, start_pfn) \
2873 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2876 static int zone_batchsize(struct zone
*zone
)
2882 * The per-cpu-pages pools are set to around 1000th of the
2883 * size of the zone. But no more than 1/2 of a meg.
2885 * OK, so we don't know how big the cache is. So guess.
2887 batch
= zone
->present_pages
/ 1024;
2888 if (batch
* PAGE_SIZE
> 512 * 1024)
2889 batch
= (512 * 1024) / PAGE_SIZE
;
2890 batch
/= 4; /* We effectively *= 4 below */
2895 * Clamp the batch to a 2^n - 1 value. Having a power
2896 * of 2 value was found to be more likely to have
2897 * suboptimal cache aliasing properties in some cases.
2899 * For example if 2 tasks are alternately allocating
2900 * batches of pages, one task can end up with a lot
2901 * of pages of one half of the possible page colors
2902 * and the other with pages of the other colors.
2904 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
2909 /* The deferral and batching of frees should be suppressed under NOMMU
2912 * The problem is that NOMMU needs to be able to allocate large chunks
2913 * of contiguous memory as there's no hardware page translation to
2914 * assemble apparent contiguous memory from discontiguous pages.
2916 * Queueing large contiguous runs of pages for batching, however,
2917 * causes the pages to actually be freed in smaller chunks. As there
2918 * can be a significant delay between the individual batches being
2919 * recycled, this leads to the once large chunks of space being
2920 * fragmented and becoming unavailable for high-order allocations.
2926 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2928 struct per_cpu_pages
*pcp
;
2930 memset(p
, 0, sizeof(*p
));
2934 pcp
->high
= 6 * batch
;
2935 pcp
->batch
= max(1UL, 1 * batch
);
2936 INIT_LIST_HEAD(&pcp
->list
);
2940 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2941 * to the value high for the pageset p.
2944 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2947 struct per_cpu_pages
*pcp
;
2951 pcp
->batch
= max(1UL, high
/4);
2952 if ((high
/4) > (PAGE_SHIFT
* 8))
2953 pcp
->batch
= PAGE_SHIFT
* 8;
2959 * Boot pageset table. One per cpu which is going to be used for all
2960 * zones and all nodes. The parameters will be set in such a way
2961 * that an item put on a list will immediately be handed over to
2962 * the buddy list. This is safe since pageset manipulation is done
2963 * with interrupts disabled.
2965 * Some NUMA counter updates may also be caught by the boot pagesets.
2967 * The boot_pagesets must be kept even after bootup is complete for
2968 * unused processors and/or zones. They do play a role for bootstrapping
2969 * hotplugged processors.
2971 * zoneinfo_show() and maybe other functions do
2972 * not check if the processor is online before following the pageset pointer.
2973 * Other parts of the kernel may not check if the zone is available.
2975 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2978 * Dynamically allocate memory for the
2979 * per cpu pageset array in struct zone.
2981 static int __cpuinit
process_zones(int cpu
)
2983 struct zone
*zone
, *dzone
;
2984 int node
= cpu_to_node(cpu
);
2986 node_set_state(node
, N_CPU
); /* this node has a cpu */
2988 for_each_populated_zone(zone
) {
2989 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2991 if (!zone_pcp(zone
, cpu
))
2994 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2996 if (percpu_pagelist_fraction
)
2997 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2998 (zone
->present_pages
/ percpu_pagelist_fraction
));
3003 for_each_zone(dzone
) {
3004 if (!populated_zone(dzone
))
3008 kfree(zone_pcp(dzone
, cpu
));
3009 zone_pcp(dzone
, cpu
) = NULL
;
3014 static inline void free_zone_pagesets(int cpu
)
3018 for_each_zone(zone
) {
3019 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
3021 /* Free per_cpu_pageset if it is slab allocated */
3022 if (pset
!= &boot_pageset
[cpu
])
3024 zone_pcp(zone
, cpu
) = NULL
;
3028 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
3029 unsigned long action
,
3032 int cpu
= (long)hcpu
;
3033 int ret
= NOTIFY_OK
;
3036 case CPU_UP_PREPARE
:
3037 case CPU_UP_PREPARE_FROZEN
:
3038 if (process_zones(cpu
))
3041 case CPU_UP_CANCELED
:
3042 case CPU_UP_CANCELED_FROZEN
:
3044 case CPU_DEAD_FROZEN
:
3045 free_zone_pagesets(cpu
);
3053 static struct notifier_block __cpuinitdata pageset_notifier
=
3054 { &pageset_cpuup_callback
, NULL
, 0 };
3056 void __init
setup_per_cpu_pageset(void)
3060 /* Initialize per_cpu_pageset for cpu 0.
3061 * A cpuup callback will do this for every cpu
3062 * as it comes online
3064 err
= process_zones(smp_processor_id());
3066 register_cpu_notifier(&pageset_notifier
);
3071 static noinline __init_refok
3072 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3075 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3079 * The per-page waitqueue mechanism uses hashed waitqueues
3082 zone
->wait_table_hash_nr_entries
=
3083 wait_table_hash_nr_entries(zone_size_pages
);
3084 zone
->wait_table_bits
=
3085 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3086 alloc_size
= zone
->wait_table_hash_nr_entries
3087 * sizeof(wait_queue_head_t
);
3089 if (!slab_is_available()) {
3090 zone
->wait_table
= (wait_queue_head_t
*)
3091 alloc_bootmem_node(pgdat
, alloc_size
);
3094 * This case means that a zone whose size was 0 gets new memory
3095 * via memory hot-add.
3096 * But it may be the case that a new node was hot-added. In
3097 * this case vmalloc() will not be able to use this new node's
3098 * memory - this wait_table must be initialized to use this new
3099 * node itself as well.
3100 * To use this new node's memory, further consideration will be
3103 zone
->wait_table
= vmalloc(alloc_size
);
3105 if (!zone
->wait_table
)
3108 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3109 init_waitqueue_head(zone
->wait_table
+ i
);
3114 static __meminit
void zone_pcp_init(struct zone
*zone
)
3117 unsigned long batch
= zone_batchsize(zone
);
3119 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
3121 /* Early boot. Slab allocator not functional yet */
3122 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
3123 setup_pageset(&boot_pageset
[cpu
],0);
3125 setup_pageset(zone_pcp(zone
,cpu
), batch
);
3128 if (zone
->present_pages
)
3129 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
3130 zone
->name
, zone
->present_pages
, batch
);
3133 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3134 unsigned long zone_start_pfn
,
3136 enum memmap_context context
)
3138 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3140 ret
= zone_wait_table_init(zone
, size
);
3143 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3145 zone
->zone_start_pfn
= zone_start_pfn
;
3147 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3148 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3150 (unsigned long)zone_idx(zone
),
3151 zone_start_pfn
, (zone_start_pfn
+ size
));
3153 zone_init_free_lists(zone
);
3158 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3160 * Basic iterator support. Return the first range of PFNs for a node
3161 * Note: nid == MAX_NUMNODES returns first region regardless of node
3163 static int __meminit
first_active_region_index_in_nid(int nid
)
3167 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3168 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
3175 * Basic iterator support. Return the next active range of PFNs for a node
3176 * Note: nid == MAX_NUMNODES returns next region regardless of node
3178 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
3180 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
3181 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
3187 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3189 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3190 * Architectures may implement their own version but if add_active_range()
3191 * was used and there are no special requirements, this is a convenient
3194 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3198 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3199 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3200 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3202 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3203 return early_node_map
[i
].nid
;
3205 /* This is a memory hole */
3208 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3210 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3214 nid
= __early_pfn_to_nid(pfn
);
3217 /* just returns 0 */
3221 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3222 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3226 nid
= __early_pfn_to_nid(pfn
);
3227 if (nid
>= 0 && nid
!= node
)
3233 /* Basic iterator support to walk early_node_map[] */
3234 #define for_each_active_range_index_in_nid(i, nid) \
3235 for (i = first_active_region_index_in_nid(nid); i != -1; \
3236 i = next_active_region_index_in_nid(i, nid))
3239 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3240 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3241 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3243 * If an architecture guarantees that all ranges registered with
3244 * add_active_ranges() contain no holes and may be freed, this
3245 * this function may be used instead of calling free_bootmem() manually.
3247 void __init
free_bootmem_with_active_regions(int nid
,
3248 unsigned long max_low_pfn
)
3252 for_each_active_range_index_in_nid(i
, nid
) {
3253 unsigned long size_pages
= 0;
3254 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3256 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3259 if (end_pfn
> max_low_pfn
)
3260 end_pfn
= max_low_pfn
;
3262 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3263 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3264 PFN_PHYS(early_node_map
[i
].start_pfn
),
3265 size_pages
<< PAGE_SHIFT
);
3269 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3274 for_each_active_range_index_in_nid(i
, nid
) {
3275 ret
= work_fn(early_node_map
[i
].start_pfn
,
3276 early_node_map
[i
].end_pfn
, data
);
3282 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3283 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3285 * If an architecture guarantees that all ranges registered with
3286 * add_active_ranges() contain no holes and may be freed, this
3287 * function may be used instead of calling memory_present() manually.
3289 void __init
sparse_memory_present_with_active_regions(int nid
)
3293 for_each_active_range_index_in_nid(i
, nid
)
3294 memory_present(early_node_map
[i
].nid
,
3295 early_node_map
[i
].start_pfn
,
3296 early_node_map
[i
].end_pfn
);
3300 * get_pfn_range_for_nid - Return the start and end page frames for a node
3301 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3302 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3303 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3305 * It returns the start and end page frame of a node based on information
3306 * provided by an arch calling add_active_range(). If called for a node
3307 * with no available memory, a warning is printed and the start and end
3310 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3311 unsigned long *start_pfn
, unsigned long *end_pfn
)
3317 for_each_active_range_index_in_nid(i
, nid
) {
3318 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3319 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3322 if (*start_pfn
== -1UL)
3327 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3328 * assumption is made that zones within a node are ordered in monotonic
3329 * increasing memory addresses so that the "highest" populated zone is used
3331 static void __init
find_usable_zone_for_movable(void)
3334 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3335 if (zone_index
== ZONE_MOVABLE
)
3338 if (arch_zone_highest_possible_pfn
[zone_index
] >
3339 arch_zone_lowest_possible_pfn
[zone_index
])
3343 VM_BUG_ON(zone_index
== -1);
3344 movable_zone
= zone_index
;
3348 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3349 * because it is sized independant of architecture. Unlike the other zones,
3350 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3351 * in each node depending on the size of each node and how evenly kernelcore
3352 * is distributed. This helper function adjusts the zone ranges
3353 * provided by the architecture for a given node by using the end of the
3354 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3355 * zones within a node are in order of monotonic increases memory addresses
3357 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3358 unsigned long zone_type
,
3359 unsigned long node_start_pfn
,
3360 unsigned long node_end_pfn
,
3361 unsigned long *zone_start_pfn
,
3362 unsigned long *zone_end_pfn
)
3364 /* Only adjust if ZONE_MOVABLE is on this node */
3365 if (zone_movable_pfn
[nid
]) {
3366 /* Size ZONE_MOVABLE */
3367 if (zone_type
== ZONE_MOVABLE
) {
3368 *zone_start_pfn
= zone_movable_pfn
[nid
];
3369 *zone_end_pfn
= min(node_end_pfn
,
3370 arch_zone_highest_possible_pfn
[movable_zone
]);
3372 /* Adjust for ZONE_MOVABLE starting within this range */
3373 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3374 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3375 *zone_end_pfn
= zone_movable_pfn
[nid
];
3377 /* Check if this whole range is within ZONE_MOVABLE */
3378 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3379 *zone_start_pfn
= *zone_end_pfn
;
3384 * Return the number of pages a zone spans in a node, including holes
3385 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3387 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3388 unsigned long zone_type
,
3389 unsigned long *ignored
)
3391 unsigned long node_start_pfn
, node_end_pfn
;
3392 unsigned long zone_start_pfn
, zone_end_pfn
;
3394 /* Get the start and end of the node and zone */
3395 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3396 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3397 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3398 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3399 node_start_pfn
, node_end_pfn
,
3400 &zone_start_pfn
, &zone_end_pfn
);
3402 /* Check that this node has pages within the zone's required range */
3403 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3406 /* Move the zone boundaries inside the node if necessary */
3407 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3408 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3410 /* Return the spanned pages */
3411 return zone_end_pfn
- zone_start_pfn
;
3415 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3416 * then all holes in the requested range will be accounted for.
3418 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3419 unsigned long range_start_pfn
,
3420 unsigned long range_end_pfn
)
3423 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3424 unsigned long start_pfn
;
3426 /* Find the end_pfn of the first active range of pfns in the node */
3427 i
= first_active_region_index_in_nid(nid
);
3431 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3433 /* Account for ranges before physical memory on this node */
3434 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3435 hole_pages
= prev_end_pfn
- range_start_pfn
;
3437 /* Find all holes for the zone within the node */
3438 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3440 /* No need to continue if prev_end_pfn is outside the zone */
3441 if (prev_end_pfn
>= range_end_pfn
)
3444 /* Make sure the end of the zone is not within the hole */
3445 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3446 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3448 /* Update the hole size cound and move on */
3449 if (start_pfn
> range_start_pfn
) {
3450 BUG_ON(prev_end_pfn
> start_pfn
);
3451 hole_pages
+= start_pfn
- prev_end_pfn
;
3453 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3456 /* Account for ranges past physical memory on this node */
3457 if (range_end_pfn
> prev_end_pfn
)
3458 hole_pages
+= range_end_pfn
-
3459 max(range_start_pfn
, prev_end_pfn
);
3465 * absent_pages_in_range - Return number of page frames in holes within a range
3466 * @start_pfn: The start PFN to start searching for holes
3467 * @end_pfn: The end PFN to stop searching for holes
3469 * It returns the number of pages frames in memory holes within a range.
3471 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3472 unsigned long end_pfn
)
3474 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3477 /* Return the number of page frames in holes in a zone on a node */
3478 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3479 unsigned long zone_type
,
3480 unsigned long *ignored
)
3482 unsigned long node_start_pfn
, node_end_pfn
;
3483 unsigned long zone_start_pfn
, zone_end_pfn
;
3485 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3486 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3488 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3491 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3492 node_start_pfn
, node_end_pfn
,
3493 &zone_start_pfn
, &zone_end_pfn
);
3494 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3498 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3499 unsigned long zone_type
,
3500 unsigned long *zones_size
)
3502 return zones_size
[zone_type
];
3505 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3506 unsigned long zone_type
,
3507 unsigned long *zholes_size
)
3512 return zholes_size
[zone_type
];
3517 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3518 unsigned long *zones_size
, unsigned long *zholes_size
)
3520 unsigned long realtotalpages
, totalpages
= 0;
3523 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3524 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3526 pgdat
->node_spanned_pages
= totalpages
;
3528 realtotalpages
= totalpages
;
3529 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3531 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3533 pgdat
->node_present_pages
= realtotalpages
;
3534 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3538 #ifndef CONFIG_SPARSEMEM
3540 * Calculate the size of the zone->blockflags rounded to an unsigned long
3541 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3542 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3543 * round what is now in bits to nearest long in bits, then return it in
3546 static unsigned long __init
usemap_size(unsigned long zonesize
)
3548 unsigned long usemapsize
;
3550 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3551 usemapsize
= usemapsize
>> pageblock_order
;
3552 usemapsize
*= NR_PAGEBLOCK_BITS
;
3553 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3555 return usemapsize
/ 8;
3558 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3559 struct zone
*zone
, unsigned long zonesize
)
3561 unsigned long usemapsize
= usemap_size(zonesize
);
3562 zone
->pageblock_flags
= NULL
;
3564 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3567 static void inline setup_usemap(struct pglist_data
*pgdat
,
3568 struct zone
*zone
, unsigned long zonesize
) {}
3569 #endif /* CONFIG_SPARSEMEM */
3571 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3573 /* Return a sensible default order for the pageblock size. */
3574 static inline int pageblock_default_order(void)
3576 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3577 return HUGETLB_PAGE_ORDER
;
3582 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3583 static inline void __init
set_pageblock_order(unsigned int order
)
3585 /* Check that pageblock_nr_pages has not already been setup */
3586 if (pageblock_order
)
3590 * Assume the largest contiguous order of interest is a huge page.
3591 * This value may be variable depending on boot parameters on IA64
3593 pageblock_order
= order
;
3595 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3598 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3599 * and pageblock_default_order() are unused as pageblock_order is set
3600 * at compile-time. See include/linux/pageblock-flags.h for the values of
3601 * pageblock_order based on the kernel config
3603 static inline int pageblock_default_order(unsigned int order
)
3607 #define set_pageblock_order(x) do {} while (0)
3609 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3612 * Set up the zone data structures:
3613 * - mark all pages reserved
3614 * - mark all memory queues empty
3615 * - clear the memory bitmaps
3617 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3618 unsigned long *zones_size
, unsigned long *zholes_size
)
3621 int nid
= pgdat
->node_id
;
3622 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3625 pgdat_resize_init(pgdat
);
3626 pgdat
->nr_zones
= 0;
3627 init_waitqueue_head(&pgdat
->kswapd_wait
);
3628 pgdat
->kswapd_max_order
= 0;
3629 pgdat_page_cgroup_init(pgdat
);
3631 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3632 struct zone
*zone
= pgdat
->node_zones
+ j
;
3633 unsigned long size
, realsize
, memmap_pages
;
3636 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3637 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3641 * Adjust realsize so that it accounts for how much memory
3642 * is used by this zone for memmap. This affects the watermark
3643 * and per-cpu initialisations
3646 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3647 if (realsize
>= memmap_pages
) {
3648 realsize
-= memmap_pages
;
3651 " %s zone: %lu pages used for memmap\n",
3652 zone_names
[j
], memmap_pages
);
3655 " %s zone: %lu pages exceeds realsize %lu\n",
3656 zone_names
[j
], memmap_pages
, realsize
);
3658 /* Account for reserved pages */
3659 if (j
== 0 && realsize
> dma_reserve
) {
3660 realsize
-= dma_reserve
;
3661 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3662 zone_names
[0], dma_reserve
);
3665 if (!is_highmem_idx(j
))
3666 nr_kernel_pages
+= realsize
;
3667 nr_all_pages
+= realsize
;
3669 zone
->spanned_pages
= size
;
3670 zone
->present_pages
= realsize
;
3673 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3675 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3677 zone
->name
= zone_names
[j
];
3678 spin_lock_init(&zone
->lock
);
3679 spin_lock_init(&zone
->lru_lock
);
3680 zone_seqlock_init(zone
);
3681 zone
->zone_pgdat
= pgdat
;
3683 zone
->prev_priority
= DEF_PRIORITY
;
3685 zone_pcp_init(zone
);
3687 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3688 zone
->lru
[l
].nr_saved_scan
= 0;
3690 zone
->reclaim_stat
.recent_rotated
[0] = 0;
3691 zone
->reclaim_stat
.recent_rotated
[1] = 0;
3692 zone
->reclaim_stat
.recent_scanned
[0] = 0;
3693 zone
->reclaim_stat
.recent_scanned
[1] = 0;
3694 zap_zone_vm_stats(zone
);
3699 set_pageblock_order(pageblock_default_order());
3700 setup_usemap(pgdat
, zone
, size
);
3701 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3702 size
, MEMMAP_EARLY
);
3704 memmap_init(size
, nid
, j
, zone_start_pfn
);
3705 zone_start_pfn
+= size
;
3709 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3711 /* Skip empty nodes */
3712 if (!pgdat
->node_spanned_pages
)
3715 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3716 /* ia64 gets its own node_mem_map, before this, without bootmem */
3717 if (!pgdat
->node_mem_map
) {
3718 unsigned long size
, start
, end
;
3722 * The zone's endpoints aren't required to be MAX_ORDER
3723 * aligned but the node_mem_map endpoints must be in order
3724 * for the buddy allocator to function correctly.
3726 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3727 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3728 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3729 size
= (end
- start
) * sizeof(struct page
);
3730 map
= alloc_remap(pgdat
->node_id
, size
);
3732 map
= alloc_bootmem_node(pgdat
, size
);
3733 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3735 #ifndef CONFIG_NEED_MULTIPLE_NODES
3737 * With no DISCONTIG, the global mem_map is just set as node 0's
3739 if (pgdat
== NODE_DATA(0)) {
3740 mem_map
= NODE_DATA(0)->node_mem_map
;
3741 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3742 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3743 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3744 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3747 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3750 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3751 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3753 pg_data_t
*pgdat
= NODE_DATA(nid
);
3755 pgdat
->node_id
= nid
;
3756 pgdat
->node_start_pfn
= node_start_pfn
;
3757 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3759 alloc_node_mem_map(pgdat
);
3760 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3761 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3762 nid
, (unsigned long)pgdat
,
3763 (unsigned long)pgdat
->node_mem_map
);
3766 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3769 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3771 #if MAX_NUMNODES > 1
3773 * Figure out the number of possible node ids.
3775 static void __init
setup_nr_node_ids(void)
3778 unsigned int highest
= 0;
3780 for_each_node_mask(node
, node_possible_map
)
3782 nr_node_ids
= highest
+ 1;
3785 static inline void setup_nr_node_ids(void)
3791 * add_active_range - Register a range of PFNs backed by physical memory
3792 * @nid: The node ID the range resides on
3793 * @start_pfn: The start PFN of the available physical memory
3794 * @end_pfn: The end PFN of the available physical memory
3796 * These ranges are stored in an early_node_map[] and later used by
3797 * free_area_init_nodes() to calculate zone sizes and holes. If the
3798 * range spans a memory hole, it is up to the architecture to ensure
3799 * the memory is not freed by the bootmem allocator. If possible
3800 * the range being registered will be merged with existing ranges.
3802 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3803 unsigned long end_pfn
)
3807 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3808 "Entering add_active_range(%d, %#lx, %#lx) "
3809 "%d entries of %d used\n",
3810 nid
, start_pfn
, end_pfn
,
3811 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3813 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3815 /* Merge with existing active regions if possible */
3816 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3817 if (early_node_map
[i
].nid
!= nid
)
3820 /* Skip if an existing region covers this new one */
3821 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3822 end_pfn
<= early_node_map
[i
].end_pfn
)
3825 /* Merge forward if suitable */
3826 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3827 end_pfn
> early_node_map
[i
].end_pfn
) {
3828 early_node_map
[i
].end_pfn
= end_pfn
;
3832 /* Merge backward if suitable */
3833 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3834 end_pfn
>= early_node_map
[i
].start_pfn
) {
3835 early_node_map
[i
].start_pfn
= start_pfn
;
3840 /* Check that early_node_map is large enough */
3841 if (i
>= MAX_ACTIVE_REGIONS
) {
3842 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3843 MAX_ACTIVE_REGIONS
);
3847 early_node_map
[i
].nid
= nid
;
3848 early_node_map
[i
].start_pfn
= start_pfn
;
3849 early_node_map
[i
].end_pfn
= end_pfn
;
3850 nr_nodemap_entries
= i
+ 1;
3854 * remove_active_range - Shrink an existing registered range of PFNs
3855 * @nid: The node id the range is on that should be shrunk
3856 * @start_pfn: The new PFN of the range
3857 * @end_pfn: The new PFN of the range
3859 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3860 * The map is kept near the end physical page range that has already been
3861 * registered. This function allows an arch to shrink an existing registered
3864 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3865 unsigned long end_pfn
)
3870 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3871 nid
, start_pfn
, end_pfn
);
3873 /* Find the old active region end and shrink */
3874 for_each_active_range_index_in_nid(i
, nid
) {
3875 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3876 early_node_map
[i
].end_pfn
<= end_pfn
) {
3878 early_node_map
[i
].start_pfn
= 0;
3879 early_node_map
[i
].end_pfn
= 0;
3883 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3884 early_node_map
[i
].end_pfn
> start_pfn
) {
3885 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3886 early_node_map
[i
].end_pfn
= start_pfn
;
3887 if (temp_end_pfn
> end_pfn
)
3888 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3891 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3892 early_node_map
[i
].end_pfn
> end_pfn
&&
3893 early_node_map
[i
].start_pfn
< end_pfn
) {
3894 early_node_map
[i
].start_pfn
= end_pfn
;
3902 /* remove the blank ones */
3903 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3904 if (early_node_map
[i
].nid
!= nid
)
3906 if (early_node_map
[i
].end_pfn
)
3908 /* we found it, get rid of it */
3909 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3910 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3911 sizeof(early_node_map
[j
]));
3912 j
= nr_nodemap_entries
- 1;
3913 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3914 nr_nodemap_entries
--;
3919 * remove_all_active_ranges - Remove all currently registered regions
3921 * During discovery, it may be found that a table like SRAT is invalid
3922 * and an alternative discovery method must be used. This function removes
3923 * all currently registered regions.
3925 void __init
remove_all_active_ranges(void)
3927 memset(early_node_map
, 0, sizeof(early_node_map
));
3928 nr_nodemap_entries
= 0;
3931 /* Compare two active node_active_regions */
3932 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3934 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3935 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3937 /* Done this way to avoid overflows */
3938 if (arange
->start_pfn
> brange
->start_pfn
)
3940 if (arange
->start_pfn
< brange
->start_pfn
)
3946 /* sort the node_map by start_pfn */
3947 static void __init
sort_node_map(void)
3949 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3950 sizeof(struct node_active_region
),
3951 cmp_node_active_region
, NULL
);
3954 /* Find the lowest pfn for a node */
3955 static unsigned long __init
find_min_pfn_for_node(int nid
)
3958 unsigned long min_pfn
= ULONG_MAX
;
3960 /* Assuming a sorted map, the first range found has the starting pfn */
3961 for_each_active_range_index_in_nid(i
, nid
)
3962 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3964 if (min_pfn
== ULONG_MAX
) {
3966 "Could not find start_pfn for node %d\n", nid
);
3974 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3976 * It returns the minimum PFN based on information provided via
3977 * add_active_range().
3979 unsigned long __init
find_min_pfn_with_active_regions(void)
3981 return find_min_pfn_for_node(MAX_NUMNODES
);
3985 * early_calculate_totalpages()
3986 * Sum pages in active regions for movable zone.
3987 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3989 static unsigned long __init
early_calculate_totalpages(void)
3992 unsigned long totalpages
= 0;
3994 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3995 unsigned long pages
= early_node_map
[i
].end_pfn
-
3996 early_node_map
[i
].start_pfn
;
3997 totalpages
+= pages
;
3999 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
4005 * Find the PFN the Movable zone begins in each node. Kernel memory
4006 * is spread evenly between nodes as long as the nodes have enough
4007 * memory. When they don't, some nodes will have more kernelcore than
4010 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
4013 unsigned long usable_startpfn
;
4014 unsigned long kernelcore_node
, kernelcore_remaining
;
4015 unsigned long totalpages
= early_calculate_totalpages();
4016 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4019 * If movablecore was specified, calculate what size of
4020 * kernelcore that corresponds so that memory usable for
4021 * any allocation type is evenly spread. If both kernelcore
4022 * and movablecore are specified, then the value of kernelcore
4023 * will be used for required_kernelcore if it's greater than
4024 * what movablecore would have allowed.
4026 if (required_movablecore
) {
4027 unsigned long corepages
;
4030 * Round-up so that ZONE_MOVABLE is at least as large as what
4031 * was requested by the user
4033 required_movablecore
=
4034 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4035 corepages
= totalpages
- required_movablecore
;
4037 required_kernelcore
= max(required_kernelcore
, corepages
);
4040 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4041 if (!required_kernelcore
)
4044 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4045 find_usable_zone_for_movable();
4046 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4049 /* Spread kernelcore memory as evenly as possible throughout nodes */
4050 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4051 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4053 * Recalculate kernelcore_node if the division per node
4054 * now exceeds what is necessary to satisfy the requested
4055 * amount of memory for the kernel
4057 if (required_kernelcore
< kernelcore_node
)
4058 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4061 * As the map is walked, we track how much memory is usable
4062 * by the kernel using kernelcore_remaining. When it is
4063 * 0, the rest of the node is usable by ZONE_MOVABLE
4065 kernelcore_remaining
= kernelcore_node
;
4067 /* Go through each range of PFNs within this node */
4068 for_each_active_range_index_in_nid(i
, nid
) {
4069 unsigned long start_pfn
, end_pfn
;
4070 unsigned long size_pages
;
4072 start_pfn
= max(early_node_map
[i
].start_pfn
,
4073 zone_movable_pfn
[nid
]);
4074 end_pfn
= early_node_map
[i
].end_pfn
;
4075 if (start_pfn
>= end_pfn
)
4078 /* Account for what is only usable for kernelcore */
4079 if (start_pfn
< usable_startpfn
) {
4080 unsigned long kernel_pages
;
4081 kernel_pages
= min(end_pfn
, usable_startpfn
)
4084 kernelcore_remaining
-= min(kernel_pages
,
4085 kernelcore_remaining
);
4086 required_kernelcore
-= min(kernel_pages
,
4087 required_kernelcore
);
4089 /* Continue if range is now fully accounted */
4090 if (end_pfn
<= usable_startpfn
) {
4093 * Push zone_movable_pfn to the end so
4094 * that if we have to rebalance
4095 * kernelcore across nodes, we will
4096 * not double account here
4098 zone_movable_pfn
[nid
] = end_pfn
;
4101 start_pfn
= usable_startpfn
;
4105 * The usable PFN range for ZONE_MOVABLE is from
4106 * start_pfn->end_pfn. Calculate size_pages as the
4107 * number of pages used as kernelcore
4109 size_pages
= end_pfn
- start_pfn
;
4110 if (size_pages
> kernelcore_remaining
)
4111 size_pages
= kernelcore_remaining
;
4112 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4115 * Some kernelcore has been met, update counts and
4116 * break if the kernelcore for this node has been
4119 required_kernelcore
-= min(required_kernelcore
,
4121 kernelcore_remaining
-= size_pages
;
4122 if (!kernelcore_remaining
)
4128 * If there is still required_kernelcore, we do another pass with one
4129 * less node in the count. This will push zone_movable_pfn[nid] further
4130 * along on the nodes that still have memory until kernelcore is
4134 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4137 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4138 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4139 zone_movable_pfn
[nid
] =
4140 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4143 /* Any regular memory on that node ? */
4144 static void check_for_regular_memory(pg_data_t
*pgdat
)
4146 #ifdef CONFIG_HIGHMEM
4147 enum zone_type zone_type
;
4149 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4150 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4151 if (zone
->present_pages
)
4152 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4158 * free_area_init_nodes - Initialise all pg_data_t and zone data
4159 * @max_zone_pfn: an array of max PFNs for each zone
4161 * This will call free_area_init_node() for each active node in the system.
4162 * Using the page ranges provided by add_active_range(), the size of each
4163 * zone in each node and their holes is calculated. If the maximum PFN
4164 * between two adjacent zones match, it is assumed that the zone is empty.
4165 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4166 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4167 * starts where the previous one ended. For example, ZONE_DMA32 starts
4168 * at arch_max_dma_pfn.
4170 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4175 /* Sort early_node_map as initialisation assumes it is sorted */
4178 /* Record where the zone boundaries are */
4179 memset(arch_zone_lowest_possible_pfn
, 0,
4180 sizeof(arch_zone_lowest_possible_pfn
));
4181 memset(arch_zone_highest_possible_pfn
, 0,
4182 sizeof(arch_zone_highest_possible_pfn
));
4183 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4184 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4185 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4186 if (i
== ZONE_MOVABLE
)
4188 arch_zone_lowest_possible_pfn
[i
] =
4189 arch_zone_highest_possible_pfn
[i
-1];
4190 arch_zone_highest_possible_pfn
[i
] =
4191 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4193 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4194 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4196 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4197 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4198 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4200 /* Print out the zone ranges */
4201 printk("Zone PFN ranges:\n");
4202 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4203 if (i
== ZONE_MOVABLE
)
4205 printk(" %-8s %0#10lx -> %0#10lx\n",
4207 arch_zone_lowest_possible_pfn
[i
],
4208 arch_zone_highest_possible_pfn
[i
]);
4211 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4212 printk("Movable zone start PFN for each node\n");
4213 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4214 if (zone_movable_pfn
[i
])
4215 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4218 /* Print out the early_node_map[] */
4219 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4220 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4221 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4222 early_node_map
[i
].start_pfn
,
4223 early_node_map
[i
].end_pfn
);
4226 * find_zone_movable_pfns_for_nodes/early_calculate_totalpages init
4227 * that node_mask, clear it at first
4229 nodes_clear(node_states
[N_HIGH_MEMORY
]);
4230 /* Initialise every node */
4231 mminit_verify_pageflags_layout();
4232 setup_nr_node_ids();
4233 for_each_online_node(nid
) {
4234 pg_data_t
*pgdat
= NODE_DATA(nid
);
4235 free_area_init_node(nid
, NULL
,
4236 find_min_pfn_for_node(nid
), NULL
);
4238 /* Any memory on that node */
4239 if (pgdat
->node_present_pages
)
4240 node_set_state(nid
, N_HIGH_MEMORY
);
4241 check_for_regular_memory(pgdat
);
4245 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4247 unsigned long long coremem
;
4251 coremem
= memparse(p
, &p
);
4252 *core
= coremem
>> PAGE_SHIFT
;
4254 /* Paranoid check that UL is enough for the coremem value */
4255 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4261 * kernelcore=size sets the amount of memory for use for allocations that
4262 * cannot be reclaimed or migrated.
4264 static int __init
cmdline_parse_kernelcore(char *p
)
4266 return cmdline_parse_core(p
, &required_kernelcore
);
4270 * movablecore=size sets the amount of memory for use for allocations that
4271 * can be reclaimed or migrated.
4273 static int __init
cmdline_parse_movablecore(char *p
)
4275 return cmdline_parse_core(p
, &required_movablecore
);
4278 early_param("kernelcore", cmdline_parse_kernelcore
);
4279 early_param("movablecore", cmdline_parse_movablecore
);
4281 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4284 * set_dma_reserve - set the specified number of pages reserved in the first zone
4285 * @new_dma_reserve: The number of pages to mark reserved
4287 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4288 * In the DMA zone, a significant percentage may be consumed by kernel image
4289 * and other unfreeable allocations which can skew the watermarks badly. This
4290 * function may optionally be used to account for unfreeable pages in the
4291 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4292 * smaller per-cpu batchsize.
4294 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4296 dma_reserve
= new_dma_reserve
;
4299 #ifndef CONFIG_NEED_MULTIPLE_NODES
4300 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4301 EXPORT_SYMBOL(contig_page_data
);
4304 void __init
free_area_init(unsigned long *zones_size
)
4306 free_area_init_node(0, zones_size
,
4307 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4310 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4311 unsigned long action
, void *hcpu
)
4313 int cpu
= (unsigned long)hcpu
;
4315 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4319 * Spill the event counters of the dead processor
4320 * into the current processors event counters.
4321 * This artificially elevates the count of the current
4324 vm_events_fold_cpu(cpu
);
4327 * Zero the differential counters of the dead processor
4328 * so that the vm statistics are consistent.
4330 * This is only okay since the processor is dead and cannot
4331 * race with what we are doing.
4333 refresh_cpu_vm_stats(cpu
);
4338 void __init
page_alloc_init(void)
4340 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4344 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4345 * or min_free_kbytes changes.
4347 static void calculate_totalreserve_pages(void)
4349 struct pglist_data
*pgdat
;
4350 unsigned long reserve_pages
= 0;
4351 enum zone_type i
, j
;
4353 for_each_online_pgdat(pgdat
) {
4354 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4355 struct zone
*zone
= pgdat
->node_zones
+ i
;
4356 unsigned long max
= 0;
4358 /* Find valid and maximum lowmem_reserve in the zone */
4359 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4360 if (zone
->lowmem_reserve
[j
] > max
)
4361 max
= zone
->lowmem_reserve
[j
];
4364 /* we treat the high watermark as reserved pages. */
4365 max
+= high_wmark_pages(zone
);
4367 if (max
> zone
->present_pages
)
4368 max
= zone
->present_pages
;
4369 reserve_pages
+= max
;
4372 totalreserve_pages
= reserve_pages
;
4376 * setup_per_zone_lowmem_reserve - called whenever
4377 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4378 * has a correct pages reserved value, so an adequate number of
4379 * pages are left in the zone after a successful __alloc_pages().
4381 static void setup_per_zone_lowmem_reserve(void)
4383 struct pglist_data
*pgdat
;
4384 enum zone_type j
, idx
;
4386 for_each_online_pgdat(pgdat
) {
4387 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4388 struct zone
*zone
= pgdat
->node_zones
+ j
;
4389 unsigned long present_pages
= zone
->present_pages
;
4391 zone
->lowmem_reserve
[j
] = 0;
4395 struct zone
*lower_zone
;
4399 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4400 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4402 lower_zone
= pgdat
->node_zones
+ idx
;
4403 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4404 sysctl_lowmem_reserve_ratio
[idx
];
4405 present_pages
+= lower_zone
->present_pages
;
4410 /* update totalreserve_pages */
4411 calculate_totalreserve_pages();
4415 * setup_per_zone_wmarks - called when min_free_kbytes changes
4416 * or when memory is hot-{added|removed}
4418 * Ensures that the watermark[min,low,high] values for each zone are set
4419 * correctly with respect to min_free_kbytes.
4421 void setup_per_zone_wmarks(void)
4423 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4424 unsigned long lowmem_pages
= 0;
4426 unsigned long flags
;
4428 /* Calculate total number of !ZONE_HIGHMEM pages */
4429 for_each_zone(zone
) {
4430 if (!is_highmem(zone
))
4431 lowmem_pages
+= zone
->present_pages
;
4434 for_each_zone(zone
) {
4437 spin_lock_irqsave(&zone
->lock
, flags
);
4438 tmp
= (u64
)pages_min
* zone
->present_pages
;
4439 do_div(tmp
, lowmem_pages
);
4440 if (is_highmem(zone
)) {
4442 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4443 * need highmem pages, so cap pages_min to a small
4446 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4447 * deltas controls asynch page reclaim, and so should
4448 * not be capped for highmem.
4452 min_pages
= zone
->present_pages
/ 1024;
4453 if (min_pages
< SWAP_CLUSTER_MAX
)
4454 min_pages
= SWAP_CLUSTER_MAX
;
4455 if (min_pages
> 128)
4457 zone
->watermark
[WMARK_MIN
] = min_pages
;
4460 * If it's a lowmem zone, reserve a number of pages
4461 * proportionate to the zone's size.
4463 zone
->watermark
[WMARK_MIN
] = tmp
;
4466 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
4467 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
4468 setup_zone_migrate_reserve(zone
);
4469 spin_unlock_irqrestore(&zone
->lock
, flags
);
4472 /* update totalreserve_pages */
4473 calculate_totalreserve_pages();
4477 * The inactive anon list should be small enough that the VM never has to
4478 * do too much work, but large enough that each inactive page has a chance
4479 * to be referenced again before it is swapped out.
4481 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4482 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4483 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4484 * the anonymous pages are kept on the inactive list.
4487 * memory ratio inactive anon
4488 * -------------------------------------
4497 void calculate_zone_inactive_ratio(struct zone
*zone
)
4499 unsigned int gb
, ratio
;
4501 /* Zone size in gigabytes */
4502 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4504 ratio
= int_sqrt(10 * gb
);
4508 zone
->inactive_ratio
= ratio
;
4511 static void __init
setup_per_zone_inactive_ratio(void)
4516 calculate_zone_inactive_ratio(zone
);
4520 * Initialise min_free_kbytes.
4522 * For small machines we want it small (128k min). For large machines
4523 * we want it large (64MB max). But it is not linear, because network
4524 * bandwidth does not increase linearly with machine size. We use
4526 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4527 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4543 static int __init
init_per_zone_wmark_min(void)
4545 unsigned long lowmem_kbytes
;
4547 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4549 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4550 if (min_free_kbytes
< 128)
4551 min_free_kbytes
= 128;
4552 if (min_free_kbytes
> 65536)
4553 min_free_kbytes
= 65536;
4554 setup_per_zone_wmarks();
4555 setup_per_zone_lowmem_reserve();
4556 setup_per_zone_inactive_ratio();
4559 module_init(init_per_zone_wmark_min
)
4562 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4563 * that we can call two helper functions whenever min_free_kbytes
4566 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4567 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4569 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4571 setup_per_zone_wmarks();
4576 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4577 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4582 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4587 zone
->min_unmapped_pages
= (zone
->present_pages
*
4588 sysctl_min_unmapped_ratio
) / 100;
4592 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4593 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4598 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4603 zone
->min_slab_pages
= (zone
->present_pages
*
4604 sysctl_min_slab_ratio
) / 100;
4610 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4611 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4612 * whenever sysctl_lowmem_reserve_ratio changes.
4614 * The reserve ratio obviously has absolutely no relation with the
4615 * minimum watermarks. The lowmem reserve ratio can only make sense
4616 * if in function of the boot time zone sizes.
4618 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4619 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4621 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4622 setup_per_zone_lowmem_reserve();
4627 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4628 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4629 * can have before it gets flushed back to buddy allocator.
4632 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4633 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4639 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4640 if (!write
|| (ret
== -EINVAL
))
4642 for_each_zone(zone
) {
4643 for_each_online_cpu(cpu
) {
4645 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4646 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4652 int hashdist
= HASHDIST_DEFAULT
;
4655 static int __init
set_hashdist(char *str
)
4659 hashdist
= simple_strtoul(str
, &str
, 0);
4662 __setup("hashdist=", set_hashdist
);
4666 * allocate a large system hash table from bootmem
4667 * - it is assumed that the hash table must contain an exact power-of-2
4668 * quantity of entries
4669 * - limit is the number of hash buckets, not the total allocation size
4671 void *__init
alloc_large_system_hash(const char *tablename
,
4672 unsigned long bucketsize
,
4673 unsigned long numentries
,
4676 unsigned int *_hash_shift
,
4677 unsigned int *_hash_mask
,
4678 unsigned long limit
)
4680 unsigned long long max
= limit
;
4681 unsigned long log2qty
, size
;
4684 /* allow the kernel cmdline to have a say */
4686 /* round applicable memory size up to nearest megabyte */
4687 numentries
= nr_kernel_pages
;
4688 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4689 numentries
>>= 20 - PAGE_SHIFT
;
4690 numentries
<<= 20 - PAGE_SHIFT
;
4692 /* limit to 1 bucket per 2^scale bytes of low memory */
4693 if (scale
> PAGE_SHIFT
)
4694 numentries
>>= (scale
- PAGE_SHIFT
);
4696 numentries
<<= (PAGE_SHIFT
- scale
);
4698 /* Make sure we've got at least a 0-order allocation.. */
4699 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4700 numentries
= PAGE_SIZE
/ bucketsize
;
4702 numentries
= roundup_pow_of_two(numentries
);
4704 /* limit allocation size to 1/16 total memory by default */
4706 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4707 do_div(max
, bucketsize
);
4710 if (numentries
> max
)
4713 log2qty
= ilog2(numentries
);
4716 size
= bucketsize
<< log2qty
;
4717 if (flags
& HASH_EARLY
)
4718 table
= alloc_bootmem_nopanic(size
);
4720 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4723 * If bucketsize is not a power-of-two, we may free
4724 * some pages at the end of hash table which
4725 * alloc_pages_exact() automatically does
4727 if (get_order(size
) < MAX_ORDER
)
4728 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
4730 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4733 panic("Failed to allocate %s hash table\n", tablename
);
4735 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4738 ilog2(size
) - PAGE_SHIFT
,
4742 *_hash_shift
= log2qty
;
4744 *_hash_mask
= (1 << log2qty
) - 1;
4747 * If hashdist is set, the table allocation is done with __vmalloc()
4748 * which invokes the kmemleak_alloc() callback. This function may also
4749 * be called before the slab and kmemleak are initialised when
4750 * kmemleak simply buffers the request to be executed later
4751 * (GFP_ATOMIC flag ignored in this case).
4754 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
4759 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4760 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4763 #ifdef CONFIG_SPARSEMEM
4764 return __pfn_to_section(pfn
)->pageblock_flags
;
4766 return zone
->pageblock_flags
;
4767 #endif /* CONFIG_SPARSEMEM */
4770 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4772 #ifdef CONFIG_SPARSEMEM
4773 pfn
&= (PAGES_PER_SECTION
-1);
4774 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4776 pfn
= pfn
- zone
->zone_start_pfn
;
4777 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4778 #endif /* CONFIG_SPARSEMEM */
4782 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4783 * @page: The page within the block of interest
4784 * @start_bitidx: The first bit of interest to retrieve
4785 * @end_bitidx: The last bit of interest
4786 * returns pageblock_bits flags
4788 unsigned long get_pageblock_flags_group(struct page
*page
,
4789 int start_bitidx
, int end_bitidx
)
4792 unsigned long *bitmap
;
4793 unsigned long pfn
, bitidx
;
4794 unsigned long flags
= 0;
4795 unsigned long value
= 1;
4797 zone
= page_zone(page
);
4798 pfn
= page_to_pfn(page
);
4799 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4800 bitidx
= pfn_to_bitidx(zone
, pfn
);
4802 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4803 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4810 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4811 * @page: The page within the block of interest
4812 * @start_bitidx: The first bit of interest
4813 * @end_bitidx: The last bit of interest
4814 * @flags: The flags to set
4816 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4817 int start_bitidx
, int end_bitidx
)
4820 unsigned long *bitmap
;
4821 unsigned long pfn
, bitidx
;
4822 unsigned long value
= 1;
4824 zone
= page_zone(page
);
4825 pfn
= page_to_pfn(page
);
4826 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4827 bitidx
= pfn_to_bitidx(zone
, pfn
);
4828 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4829 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4831 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4833 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4835 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4839 * This is designed as sub function...plz see page_isolation.c also.
4840 * set/clear page block's type to be ISOLATE.
4841 * page allocater never alloc memory from ISOLATE block.
4844 int set_migratetype_isolate(struct page
*page
)
4847 unsigned long flags
;
4850 zone
= page_zone(page
);
4851 spin_lock_irqsave(&zone
->lock
, flags
);
4853 * In future, more migrate types will be able to be isolation target.
4855 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4857 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4858 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4861 spin_unlock_irqrestore(&zone
->lock
, flags
);
4867 void unset_migratetype_isolate(struct page
*page
)
4870 unsigned long flags
;
4871 zone
= page_zone(page
);
4872 spin_lock_irqsave(&zone
->lock
, flags
);
4873 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4875 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4876 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4878 spin_unlock_irqrestore(&zone
->lock
, flags
);
4881 #ifdef CONFIG_MEMORY_HOTREMOVE
4883 * All pages in the range must be isolated before calling this.
4886 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4892 unsigned long flags
;
4893 /* find the first valid pfn */
4894 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4899 zone
= page_zone(pfn_to_page(pfn
));
4900 spin_lock_irqsave(&zone
->lock
, flags
);
4902 while (pfn
< end_pfn
) {
4903 if (!pfn_valid(pfn
)) {
4907 page
= pfn_to_page(pfn
);
4908 BUG_ON(page_count(page
));
4909 BUG_ON(!PageBuddy(page
));
4910 order
= page_order(page
);
4911 #ifdef CONFIG_DEBUG_VM
4912 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4913 pfn
, 1 << order
, end_pfn
);
4915 list_del(&page
->lru
);
4916 rmv_page_order(page
);
4917 zone
->free_area
[order
].nr_free
--;
4918 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4920 for (i
= 0; i
< (1 << order
); i
++)
4921 SetPageReserved((page
+i
));
4922 pfn
+= (1 << order
);
4924 spin_unlock_irqrestore(&zone
->lock
, flags
);