2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/oom.h>
32 #include <linux/notifier.h>
33 #include <linux/topology.h>
34 #include <linux/sysctl.h>
35 #include <linux/cpu.h>
36 #include <linux/cpuset.h>
37 #include <linux/memory_hotplug.h>
38 #include <linux/nodemask.h>
39 #include <linux/vmalloc.h>
40 #include <linux/mempolicy.h>
41 #include <linux/stop_machine.h>
42 #include <linux/sort.h>
43 #include <linux/pfn.h>
44 #include <linux/backing-dev.h>
45 #include <linux/fault-inject.h>
46 #include <linux/page-isolation.h>
47 #include <linux/page_cgroup.h>
48 #include <linux/debugobjects.h>
50 #include <asm/tlbflush.h>
51 #include <asm/div64.h>
55 * Array of node states.
57 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
58 [N_POSSIBLE
] = NODE_MASK_ALL
,
59 [N_ONLINE
] = { { [0] = 1UL } },
61 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
63 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
65 [N_CPU
] = { { [0] = 1UL } },
68 EXPORT_SYMBOL(node_states
);
70 unsigned long totalram_pages __read_mostly
;
71 unsigned long totalreserve_pages __read_mostly
;
73 int percpu_pagelist_fraction
;
75 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
76 int pageblock_order __read_mostly
;
79 static void __free_pages_ok(struct page
*page
, unsigned int order
);
82 * results with 256, 32 in the lowmem_reserve sysctl:
83 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
84 * 1G machine -> (16M dma, 784M normal, 224M high)
85 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
86 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
87 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
89 * TBD: should special case ZONE_DMA32 machines here - in those we normally
90 * don't need any ZONE_NORMAL reservation
92 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
93 #ifdef CONFIG_ZONE_DMA
96 #ifdef CONFIG_ZONE_DMA32
105 EXPORT_SYMBOL(totalram_pages
);
107 static char * const zone_names
[MAX_NR_ZONES
] = {
108 #ifdef CONFIG_ZONE_DMA
111 #ifdef CONFIG_ZONE_DMA32
115 #ifdef CONFIG_HIGHMEM
121 int min_free_kbytes
= 1024;
123 unsigned long __meminitdata nr_kernel_pages
;
124 unsigned long __meminitdata nr_all_pages
;
125 static unsigned long __meminitdata dma_reserve
;
127 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
129 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
130 * ranges of memory (RAM) that may be registered with add_active_range().
131 * Ranges passed to add_active_range() will be merged if possible
132 * so the number of times add_active_range() can be called is
133 * related to the number of nodes and the number of holes
135 #ifdef CONFIG_MAX_ACTIVE_REGIONS
136 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
137 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
139 #if MAX_NUMNODES >= 32
140 /* If there can be many nodes, allow up to 50 holes per node */
141 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
143 /* By default, allow up to 256 distinct regions */
144 #define MAX_ACTIVE_REGIONS 256
148 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
149 static int __meminitdata nr_nodemap_entries
;
150 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
151 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
152 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
153 static unsigned long __meminitdata node_boundary_start_pfn
[MAX_NUMNODES
];
154 static unsigned long __meminitdata node_boundary_end_pfn
[MAX_NUMNODES
];
155 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
156 static unsigned long __initdata required_kernelcore
;
157 static unsigned long __initdata required_movablecore
;
158 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
160 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
162 EXPORT_SYMBOL(movable_zone
);
163 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
166 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
167 EXPORT_SYMBOL(nr_node_ids
);
170 int page_group_by_mobility_disabled __read_mostly
;
172 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
174 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
175 PB_migrate
, PB_migrate_end
);
178 #ifdef CONFIG_DEBUG_VM
179 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
183 unsigned long pfn
= page_to_pfn(page
);
186 seq
= zone_span_seqbegin(zone
);
187 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
189 else if (pfn
< zone
->zone_start_pfn
)
191 } while (zone_span_seqretry(zone
, seq
));
196 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
198 if (!pfn_valid_within(page_to_pfn(page
)))
200 if (zone
!= page_zone(page
))
206 * Temporary debugging check for pages not lying within a given zone.
208 static int bad_range(struct zone
*zone
, struct page
*page
)
210 if (page_outside_zone_boundaries(zone
, page
))
212 if (!page_is_consistent(zone
, page
))
218 static inline int bad_range(struct zone
*zone
, struct page
*page
)
224 static void bad_page(struct page
*page
)
226 printk(KERN_EMERG
"Bad page state in process '%s'\n" KERN_EMERG
227 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
228 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
229 (unsigned long)page
->flags
, page
->mapping
,
230 page_mapcount(page
), page_count(page
));
232 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
233 KERN_EMERG
"Backtrace:\n");
235 page
->flags
&= ~PAGE_FLAGS_CLEAR_WHEN_BAD
;
236 set_page_count(page
, 0);
237 reset_page_mapcount(page
);
238 page
->mapping
= NULL
;
239 add_taint(TAINT_BAD_PAGE
);
243 * Higher-order pages are called "compound pages". They are structured thusly:
245 * The first PAGE_SIZE page is called the "head page".
247 * The remaining PAGE_SIZE pages are called "tail pages".
249 * All pages have PG_compound set. All pages have their ->private pointing at
250 * the head page (even the head page has this).
252 * The first tail page's ->lru.next holds the address of the compound page's
253 * put_page() function. Its ->lru.prev holds the order of allocation.
254 * This usage means that zero-order pages may not be compound.
257 static void free_compound_page(struct page
*page
)
259 __free_pages_ok(page
, compound_order(page
));
262 void prep_compound_page(struct page
*page
, unsigned long order
)
265 int nr_pages
= 1 << order
;
267 set_compound_page_dtor(page
, free_compound_page
);
268 set_compound_order(page
, order
);
270 for (i
= 1; i
< nr_pages
; i
++) {
271 struct page
*p
= page
+ i
;
274 p
->first_page
= page
;
278 #ifdef CONFIG_HUGETLBFS
279 void prep_compound_gigantic_page(struct page
*page
, unsigned long order
)
282 int nr_pages
= 1 << order
;
283 struct page
*p
= page
+ 1;
285 set_compound_page_dtor(page
, free_compound_page
);
286 set_compound_order(page
, order
);
288 for (i
= 1; i
< nr_pages
; i
++, p
= mem_map_next(p
, page
, i
)) {
290 p
->first_page
= page
;
295 static void destroy_compound_page(struct page
*page
, unsigned long order
)
298 int nr_pages
= 1 << order
;
300 if (unlikely(compound_order(page
) != order
))
303 if (unlikely(!PageHead(page
)))
305 __ClearPageHead(page
);
306 for (i
= 1; i
< nr_pages
; i
++) {
307 struct page
*p
= page
+ i
;
309 if (unlikely(!PageTail(p
) |
310 (p
->first_page
!= page
)))
316 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
321 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
322 * and __GFP_HIGHMEM from hard or soft interrupt context.
324 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
325 for (i
= 0; i
< (1 << order
); i
++)
326 clear_highpage(page
+ i
);
329 static inline void set_page_order(struct page
*page
, int order
)
331 set_page_private(page
, order
);
332 __SetPageBuddy(page
);
335 static inline void rmv_page_order(struct page
*page
)
337 __ClearPageBuddy(page
);
338 set_page_private(page
, 0);
342 * Locate the struct page for both the matching buddy in our
343 * pair (buddy1) and the combined O(n+1) page they form (page).
345 * 1) Any buddy B1 will have an order O twin B2 which satisfies
346 * the following equation:
348 * For example, if the starting buddy (buddy2) is #8 its order
350 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
352 * 2) Any buddy B will have an order O+1 parent P which
353 * satisfies the following equation:
356 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
358 static inline struct page
*
359 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
361 unsigned long buddy_idx
= page_idx
^ (1 << order
);
363 return page
+ (buddy_idx
- page_idx
);
366 static inline unsigned long
367 __find_combined_index(unsigned long page_idx
, unsigned int order
)
369 return (page_idx
& ~(1 << order
));
373 * This function checks whether a page is free && is the buddy
374 * we can do coalesce a page and its buddy if
375 * (a) the buddy is not in a hole &&
376 * (b) the buddy is in the buddy system &&
377 * (c) a page and its buddy have the same order &&
378 * (d) a page and its buddy are in the same zone.
380 * For recording whether a page is in the buddy system, we use PG_buddy.
381 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
383 * For recording page's order, we use page_private(page).
385 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
388 if (!pfn_valid_within(page_to_pfn(buddy
)))
391 if (page_zone_id(page
) != page_zone_id(buddy
))
394 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
395 BUG_ON(page_count(buddy
) != 0);
402 * Freeing function for a buddy system allocator.
404 * The concept of a buddy system is to maintain direct-mapped table
405 * (containing bit values) for memory blocks of various "orders".
406 * The bottom level table contains the map for the smallest allocatable
407 * units of memory (here, pages), and each level above it describes
408 * pairs of units from the levels below, hence, "buddies".
409 * At a high level, all that happens here is marking the table entry
410 * at the bottom level available, and propagating the changes upward
411 * as necessary, plus some accounting needed to play nicely with other
412 * parts of the VM system.
413 * At each level, we keep a list of pages, which are heads of continuous
414 * free pages of length of (1 << order) and marked with PG_buddy. Page's
415 * order is recorded in page_private(page) field.
416 * So when we are allocating or freeing one, we can derive the state of the
417 * other. That is, if we allocate a small block, and both were
418 * free, the remainder of the region must be split into blocks.
419 * If a block is freed, and its buddy is also free, then this
420 * triggers coalescing into a block of larger size.
425 static inline void __free_one_page(struct page
*page
,
426 struct zone
*zone
, unsigned int order
)
428 unsigned long page_idx
;
429 int order_size
= 1 << order
;
430 int migratetype
= get_pageblock_migratetype(page
);
432 if (unlikely(PageCompound(page
)))
433 destroy_compound_page(page
, order
);
435 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
437 VM_BUG_ON(page_idx
& (order_size
- 1));
438 VM_BUG_ON(bad_range(zone
, page
));
440 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
441 while (order
< MAX_ORDER
-1) {
442 unsigned long combined_idx
;
445 buddy
= __page_find_buddy(page
, page_idx
, order
);
446 if (!page_is_buddy(page
, buddy
, order
))
449 /* Our buddy is free, merge with it and move up one order. */
450 list_del(&buddy
->lru
);
451 zone
->free_area
[order
].nr_free
--;
452 rmv_page_order(buddy
);
453 combined_idx
= __find_combined_index(page_idx
, order
);
454 page
= page
+ (combined_idx
- page_idx
);
455 page_idx
= combined_idx
;
458 set_page_order(page
, order
);
460 &zone
->free_area
[order
].free_list
[migratetype
]);
461 zone
->free_area
[order
].nr_free
++;
464 static inline int free_pages_check(struct page
*page
)
466 free_page_mlock(page
);
467 if (unlikely(page_mapcount(page
) |
468 (page
->mapping
!= NULL
) |
469 (page_count(page
) != 0) |
470 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)))
473 __ClearPageDirty(page
);
474 if (PageSwapBacked(page
))
475 __ClearPageSwapBacked(page
);
477 * For now, we report if PG_reserved was found set, but do not
478 * clear it, and do not free the page. But we shall soon need
479 * to do more, for when the ZERO_PAGE count wraps negative.
481 return PageReserved(page
);
485 * Frees a list of pages.
486 * Assumes all pages on list are in same zone, and of same order.
487 * count is the number of pages to free.
489 * If the zone was previously in an "all pages pinned" state then look to
490 * see if this freeing clears that state.
492 * And clear the zone's pages_scanned counter, to hold off the "all pages are
493 * pinned" detection logic.
495 static void free_pages_bulk(struct zone
*zone
, int count
,
496 struct list_head
*list
, int order
)
498 spin_lock(&zone
->lock
);
499 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
500 zone
->pages_scanned
= 0;
504 VM_BUG_ON(list_empty(list
));
505 page
= list_entry(list
->prev
, struct page
, lru
);
506 /* have to delete it as __free_one_page list manipulates */
507 list_del(&page
->lru
);
508 __free_one_page(page
, zone
, order
);
510 spin_unlock(&zone
->lock
);
513 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
515 spin_lock(&zone
->lock
);
516 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
517 zone
->pages_scanned
= 0;
518 __free_one_page(page
, zone
, order
);
519 spin_unlock(&zone
->lock
);
522 static void __free_pages_ok(struct page
*page
, unsigned int order
)
528 for (i
= 0 ; i
< (1 << order
) ; ++i
)
529 reserved
+= free_pages_check(page
+ i
);
533 if (!PageHighMem(page
)) {
534 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
535 debug_check_no_obj_freed(page_address(page
),
538 arch_free_page(page
, order
);
539 kernel_map_pages(page
, 1 << order
, 0);
541 local_irq_save(flags
);
542 __count_vm_events(PGFREE
, 1 << order
);
543 free_one_page(page_zone(page
), page
, order
);
544 local_irq_restore(flags
);
548 * permit the bootmem allocator to evade page validation on high-order frees
550 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
553 __ClearPageReserved(page
);
554 set_page_count(page
, 0);
555 set_page_refcounted(page
);
561 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
562 struct page
*p
= &page
[loop
];
564 if (loop
+ 1 < BITS_PER_LONG
)
566 __ClearPageReserved(p
);
567 set_page_count(p
, 0);
570 set_page_refcounted(page
);
571 __free_pages(page
, order
);
577 * The order of subdivision here is critical for the IO subsystem.
578 * Please do not alter this order without good reasons and regression
579 * testing. Specifically, as large blocks of memory are subdivided,
580 * the order in which smaller blocks are delivered depends on the order
581 * they're subdivided in this function. This is the primary factor
582 * influencing the order in which pages are delivered to the IO
583 * subsystem according to empirical testing, and this is also justified
584 * by considering the behavior of a buddy system containing a single
585 * large block of memory acted on by a series of small allocations.
586 * This behavior is a critical factor in sglist merging's success.
590 static inline void expand(struct zone
*zone
, struct page
*page
,
591 int low
, int high
, struct free_area
*area
,
594 unsigned long size
= 1 << high
;
600 VM_BUG_ON(bad_range(zone
, &page
[size
]));
601 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
603 set_page_order(&page
[size
], high
);
608 * This page is about to be returned from the page allocator
610 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
612 if (unlikely(page_mapcount(page
) |
613 (page
->mapping
!= NULL
) |
614 (page_count(page
) != 0) |
615 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)))
619 * For now, we report if PG_reserved was found set, but do not
620 * clear it, and do not allocate the page: as a safety net.
622 if (PageReserved(page
))
625 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_reclaim
|
626 1 << PG_referenced
| 1 << PG_arch_1
|
627 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
628 #ifdef CONFIG_UNEVICTABLE_LRU
632 set_page_private(page
, 0);
633 set_page_refcounted(page
);
635 arch_alloc_page(page
, order
);
636 kernel_map_pages(page
, 1 << order
, 1);
638 if (gfp_flags
& __GFP_ZERO
)
639 prep_zero_page(page
, order
, gfp_flags
);
641 if (order
&& (gfp_flags
& __GFP_COMP
))
642 prep_compound_page(page
, order
);
648 * Go through the free lists for the given migratetype and remove
649 * the smallest available page from the freelists
651 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
654 unsigned int current_order
;
655 struct free_area
* area
;
658 /* Find a page of the appropriate size in the preferred list */
659 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
660 area
= &(zone
->free_area
[current_order
]);
661 if (list_empty(&area
->free_list
[migratetype
]))
664 page
= list_entry(area
->free_list
[migratetype
].next
,
666 list_del(&page
->lru
);
667 rmv_page_order(page
);
669 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
670 expand(zone
, page
, order
, current_order
, area
, migratetype
);
679 * This array describes the order lists are fallen back to when
680 * the free lists for the desirable migrate type are depleted
682 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
683 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
684 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
685 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
686 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
690 * Move the free pages in a range to the free lists of the requested type.
691 * Note that start_page and end_pages are not aligned on a pageblock
692 * boundary. If alignment is required, use move_freepages_block()
694 static int move_freepages(struct zone
*zone
,
695 struct page
*start_page
, struct page
*end_page
,
702 #ifndef CONFIG_HOLES_IN_ZONE
704 * page_zone is not safe to call in this context when
705 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
706 * anyway as we check zone boundaries in move_freepages_block().
707 * Remove at a later date when no bug reports exist related to
708 * grouping pages by mobility
710 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
713 for (page
= start_page
; page
<= end_page
;) {
714 /* Make sure we are not inadvertently changing nodes */
715 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
717 if (!pfn_valid_within(page_to_pfn(page
))) {
722 if (!PageBuddy(page
)) {
727 order
= page_order(page
);
728 list_del(&page
->lru
);
730 &zone
->free_area
[order
].free_list
[migratetype
]);
732 pages_moved
+= 1 << order
;
738 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
741 unsigned long start_pfn
, end_pfn
;
742 struct page
*start_page
, *end_page
;
744 start_pfn
= page_to_pfn(page
);
745 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
746 start_page
= pfn_to_page(start_pfn
);
747 end_page
= start_page
+ pageblock_nr_pages
- 1;
748 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
750 /* Do not cross zone boundaries */
751 if (start_pfn
< zone
->zone_start_pfn
)
753 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
756 return move_freepages(zone
, start_page
, end_page
, migratetype
);
759 /* Remove an element from the buddy allocator from the fallback list */
760 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
761 int start_migratetype
)
763 struct free_area
* area
;
768 /* Find the largest possible block of pages in the other list */
769 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
771 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
772 migratetype
= fallbacks
[start_migratetype
][i
];
774 /* MIGRATE_RESERVE handled later if necessary */
775 if (migratetype
== MIGRATE_RESERVE
)
778 area
= &(zone
->free_area
[current_order
]);
779 if (list_empty(&area
->free_list
[migratetype
]))
782 page
= list_entry(area
->free_list
[migratetype
].next
,
787 * If breaking a large block of pages, move all free
788 * pages to the preferred allocation list. If falling
789 * back for a reclaimable kernel allocation, be more
790 * agressive about taking ownership of free pages
792 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
793 start_migratetype
== MIGRATE_RECLAIMABLE
) {
795 pages
= move_freepages_block(zone
, page
,
798 /* Claim the whole block if over half of it is free */
799 if (pages
>= (1 << (pageblock_order
-1)))
800 set_pageblock_migratetype(page
,
803 migratetype
= start_migratetype
;
806 /* Remove the page from the freelists */
807 list_del(&page
->lru
);
808 rmv_page_order(page
);
809 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
812 if (current_order
== pageblock_order
)
813 set_pageblock_migratetype(page
,
816 expand(zone
, page
, order
, current_order
, area
, migratetype
);
821 /* Use MIGRATE_RESERVE rather than fail an allocation */
822 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
826 * Do the hard work of removing an element from the buddy allocator.
827 * Call me with the zone->lock already held.
829 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
834 page
= __rmqueue_smallest(zone
, order
, migratetype
);
837 page
= __rmqueue_fallback(zone
, order
, migratetype
);
843 * Obtain a specified number of elements from the buddy allocator, all under
844 * a single hold of the lock, for efficiency. Add them to the supplied list.
845 * Returns the number of new pages which were placed at *list.
847 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
848 unsigned long count
, struct list_head
*list
,
853 spin_lock(&zone
->lock
);
854 for (i
= 0; i
< count
; ++i
) {
855 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
856 if (unlikely(page
== NULL
))
860 * Split buddy pages returned by expand() are received here
861 * in physical page order. The page is added to the callers and
862 * list and the list head then moves forward. From the callers
863 * perspective, the linked list is ordered by page number in
864 * some conditions. This is useful for IO devices that can
865 * merge IO requests if the physical pages are ordered
868 list_add(&page
->lru
, list
);
869 set_page_private(page
, migratetype
);
872 spin_unlock(&zone
->lock
);
878 * Called from the vmstat counter updater to drain pagesets of this
879 * currently executing processor on remote nodes after they have
882 * Note that this function must be called with the thread pinned to
883 * a single processor.
885 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
890 local_irq_save(flags
);
891 if (pcp
->count
>= pcp
->batch
)
892 to_drain
= pcp
->batch
;
894 to_drain
= pcp
->count
;
895 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
896 pcp
->count
-= to_drain
;
897 local_irq_restore(flags
);
902 * Drain pages of the indicated processor.
904 * The processor must either be the current processor and the
905 * thread pinned to the current processor or a processor that
908 static void drain_pages(unsigned int cpu
)
913 for_each_zone(zone
) {
914 struct per_cpu_pageset
*pset
;
915 struct per_cpu_pages
*pcp
;
917 if (!populated_zone(zone
))
920 pset
= zone_pcp(zone
, cpu
);
923 local_irq_save(flags
);
924 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
926 local_irq_restore(flags
);
931 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
933 void drain_local_pages(void *arg
)
935 drain_pages(smp_processor_id());
939 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
941 void drain_all_pages(void)
943 on_each_cpu(drain_local_pages
, NULL
, 1);
946 #ifdef CONFIG_HIBERNATION
948 void mark_free_pages(struct zone
*zone
)
950 unsigned long pfn
, max_zone_pfn
;
953 struct list_head
*curr
;
955 if (!zone
->spanned_pages
)
958 spin_lock_irqsave(&zone
->lock
, flags
);
960 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
961 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
962 if (pfn_valid(pfn
)) {
963 struct page
*page
= pfn_to_page(pfn
);
965 if (!swsusp_page_is_forbidden(page
))
966 swsusp_unset_page_free(page
);
969 for_each_migratetype_order(order
, t
) {
970 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
973 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
974 for (i
= 0; i
< (1UL << order
); i
++)
975 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
978 spin_unlock_irqrestore(&zone
->lock
, flags
);
980 #endif /* CONFIG_PM */
983 * Free a 0-order page
985 static void free_hot_cold_page(struct page
*page
, int cold
)
987 struct zone
*zone
= page_zone(page
);
988 struct per_cpu_pages
*pcp
;
992 page
->mapping
= NULL
;
993 if (free_pages_check(page
))
996 if (!PageHighMem(page
)) {
997 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
998 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1000 arch_free_page(page
, 0);
1001 kernel_map_pages(page
, 1, 0);
1003 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1004 local_irq_save(flags
);
1005 __count_vm_event(PGFREE
);
1007 list_add_tail(&page
->lru
, &pcp
->list
);
1009 list_add(&page
->lru
, &pcp
->list
);
1010 set_page_private(page
, get_pageblock_migratetype(page
));
1012 if (pcp
->count
>= pcp
->high
) {
1013 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1014 pcp
->count
-= pcp
->batch
;
1016 local_irq_restore(flags
);
1020 void free_hot_page(struct page
*page
)
1022 free_hot_cold_page(page
, 0);
1025 void free_cold_page(struct page
*page
)
1027 free_hot_cold_page(page
, 1);
1031 * split_page takes a non-compound higher-order page, and splits it into
1032 * n (1<<order) sub-pages: page[0..n]
1033 * Each sub-page must be freed individually.
1035 * Note: this is probably too low level an operation for use in drivers.
1036 * Please consult with lkml before using this in your driver.
1038 void split_page(struct page
*page
, unsigned int order
)
1042 VM_BUG_ON(PageCompound(page
));
1043 VM_BUG_ON(!page_count(page
));
1044 for (i
= 1; i
< (1 << order
); i
++)
1045 set_page_refcounted(page
+ i
);
1049 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1050 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1053 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1054 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1056 unsigned long flags
;
1058 int cold
= !!(gfp_flags
& __GFP_COLD
);
1060 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1064 if (likely(order
== 0)) {
1065 struct per_cpu_pages
*pcp
;
1067 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1068 local_irq_save(flags
);
1070 pcp
->count
= rmqueue_bulk(zone
, 0,
1071 pcp
->batch
, &pcp
->list
, migratetype
);
1072 if (unlikely(!pcp
->count
))
1076 /* Find a page of the appropriate migrate type */
1078 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1079 if (page_private(page
) == migratetype
)
1082 list_for_each_entry(page
, &pcp
->list
, lru
)
1083 if (page_private(page
) == migratetype
)
1087 /* Allocate more to the pcp list if necessary */
1088 if (unlikely(&page
->lru
== &pcp
->list
)) {
1089 pcp
->count
+= rmqueue_bulk(zone
, 0,
1090 pcp
->batch
, &pcp
->list
, migratetype
);
1091 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1094 list_del(&page
->lru
);
1097 spin_lock_irqsave(&zone
->lock
, flags
);
1098 page
= __rmqueue(zone
, order
, migratetype
);
1099 spin_unlock(&zone
->lock
);
1104 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1105 zone_statistics(preferred_zone
, zone
);
1106 local_irq_restore(flags
);
1109 VM_BUG_ON(bad_range(zone
, page
));
1110 if (prep_new_page(page
, order
, gfp_flags
))
1115 local_irq_restore(flags
);
1120 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1121 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1122 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1123 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1124 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1125 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1126 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1128 #ifdef CONFIG_FAIL_PAGE_ALLOC
1130 static struct fail_page_alloc_attr
{
1131 struct fault_attr attr
;
1133 u32 ignore_gfp_highmem
;
1134 u32 ignore_gfp_wait
;
1137 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1139 struct dentry
*ignore_gfp_highmem_file
;
1140 struct dentry
*ignore_gfp_wait_file
;
1141 struct dentry
*min_order_file
;
1143 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1145 } fail_page_alloc
= {
1146 .attr
= FAULT_ATTR_INITIALIZER
,
1147 .ignore_gfp_wait
= 1,
1148 .ignore_gfp_highmem
= 1,
1152 static int __init
setup_fail_page_alloc(char *str
)
1154 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1156 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1158 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1160 if (order
< fail_page_alloc
.min_order
)
1162 if (gfp_mask
& __GFP_NOFAIL
)
1164 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1166 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1169 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1172 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1174 static int __init
fail_page_alloc_debugfs(void)
1176 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1180 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1184 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1186 fail_page_alloc
.ignore_gfp_wait_file
=
1187 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1188 &fail_page_alloc
.ignore_gfp_wait
);
1190 fail_page_alloc
.ignore_gfp_highmem_file
=
1191 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1192 &fail_page_alloc
.ignore_gfp_highmem
);
1193 fail_page_alloc
.min_order_file
=
1194 debugfs_create_u32("min-order", mode
, dir
,
1195 &fail_page_alloc
.min_order
);
1197 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1198 !fail_page_alloc
.ignore_gfp_highmem_file
||
1199 !fail_page_alloc
.min_order_file
) {
1201 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1202 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1203 debugfs_remove(fail_page_alloc
.min_order_file
);
1204 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1210 late_initcall(fail_page_alloc_debugfs
);
1212 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1214 #else /* CONFIG_FAIL_PAGE_ALLOC */
1216 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1221 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1224 * Return 1 if free pages are above 'mark'. This takes into account the order
1225 * of the allocation.
1227 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1228 int classzone_idx
, int alloc_flags
)
1230 /* free_pages my go negative - that's OK */
1232 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1235 if (alloc_flags
& ALLOC_HIGH
)
1237 if (alloc_flags
& ALLOC_HARDER
)
1240 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1242 for (o
= 0; o
< order
; o
++) {
1243 /* At the next order, this order's pages become unavailable */
1244 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1246 /* Require fewer higher order pages to be free */
1249 if (free_pages
<= min
)
1257 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1258 * skip over zones that are not allowed by the cpuset, or that have
1259 * been recently (in last second) found to be nearly full. See further
1260 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1261 * that have to skip over a lot of full or unallowed zones.
1263 * If the zonelist cache is present in the passed in zonelist, then
1264 * returns a pointer to the allowed node mask (either the current
1265 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1267 * If the zonelist cache is not available for this zonelist, does
1268 * nothing and returns NULL.
1270 * If the fullzones BITMAP in the zonelist cache is stale (more than
1271 * a second since last zap'd) then we zap it out (clear its bits.)
1273 * We hold off even calling zlc_setup, until after we've checked the
1274 * first zone in the zonelist, on the theory that most allocations will
1275 * be satisfied from that first zone, so best to examine that zone as
1276 * quickly as we can.
1278 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1280 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1281 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1283 zlc
= zonelist
->zlcache_ptr
;
1287 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1288 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1289 zlc
->last_full_zap
= jiffies
;
1292 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1293 &cpuset_current_mems_allowed
:
1294 &node_states
[N_HIGH_MEMORY
];
1295 return allowednodes
;
1299 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1300 * if it is worth looking at further for free memory:
1301 * 1) Check that the zone isn't thought to be full (doesn't have its
1302 * bit set in the zonelist_cache fullzones BITMAP).
1303 * 2) Check that the zones node (obtained from the zonelist_cache
1304 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1305 * Return true (non-zero) if zone is worth looking at further, or
1306 * else return false (zero) if it is not.
1308 * This check -ignores- the distinction between various watermarks,
1309 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1310 * found to be full for any variation of these watermarks, it will
1311 * be considered full for up to one second by all requests, unless
1312 * we are so low on memory on all allowed nodes that we are forced
1313 * into the second scan of the zonelist.
1315 * In the second scan we ignore this zonelist cache and exactly
1316 * apply the watermarks to all zones, even it is slower to do so.
1317 * We are low on memory in the second scan, and should leave no stone
1318 * unturned looking for a free page.
1320 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1321 nodemask_t
*allowednodes
)
1323 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1324 int i
; /* index of *z in zonelist zones */
1325 int n
; /* node that zone *z is on */
1327 zlc
= zonelist
->zlcache_ptr
;
1331 i
= z
- zonelist
->_zonerefs
;
1334 /* This zone is worth trying if it is allowed but not full */
1335 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1339 * Given 'z' scanning a zonelist, set the corresponding bit in
1340 * zlc->fullzones, so that subsequent attempts to allocate a page
1341 * from that zone don't waste time re-examining it.
1343 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1345 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1346 int i
; /* index of *z in zonelist zones */
1348 zlc
= zonelist
->zlcache_ptr
;
1352 i
= z
- zonelist
->_zonerefs
;
1354 set_bit(i
, zlc
->fullzones
);
1357 #else /* CONFIG_NUMA */
1359 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1364 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1365 nodemask_t
*allowednodes
)
1370 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1373 #endif /* CONFIG_NUMA */
1376 * get_page_from_freelist goes through the zonelist trying to allocate
1379 static struct page
*
1380 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1381 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1384 struct page
*page
= NULL
;
1386 struct zone
*zone
, *preferred_zone
;
1387 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1388 int zlc_active
= 0; /* set if using zonelist_cache */
1389 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1391 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1393 if (!preferred_zone
)
1396 classzone_idx
= zone_idx(preferred_zone
);
1400 * Scan zonelist, looking for a zone with enough free.
1401 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1403 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1404 high_zoneidx
, nodemask
) {
1405 if (NUMA_BUILD
&& zlc_active
&&
1406 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1408 if ((alloc_flags
& ALLOC_CPUSET
) &&
1409 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1412 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1414 if (alloc_flags
& ALLOC_WMARK_MIN
)
1415 mark
= zone
->pages_min
;
1416 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1417 mark
= zone
->pages_low
;
1419 mark
= zone
->pages_high
;
1420 if (!zone_watermark_ok(zone
, order
, mark
,
1421 classzone_idx
, alloc_flags
)) {
1422 if (!zone_reclaim_mode
||
1423 !zone_reclaim(zone
, gfp_mask
, order
))
1424 goto this_zone_full
;
1428 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1433 zlc_mark_zone_full(zonelist
, z
);
1435 if (NUMA_BUILD
&& !did_zlc_setup
) {
1436 /* we do zlc_setup after the first zone is tried */
1437 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1443 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1444 /* Disable zlc cache for second zonelist scan */
1452 * This is the 'heart' of the zoned buddy allocator.
1455 __alloc_pages_internal(gfp_t gfp_mask
, unsigned int order
,
1456 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1458 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1459 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1463 struct reclaim_state reclaim_state
;
1464 struct task_struct
*p
= current
;
1467 unsigned long did_some_progress
;
1468 unsigned long pages_reclaimed
= 0;
1470 might_sleep_if(wait
);
1472 if (should_fail_alloc_page(gfp_mask
, order
))
1476 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1478 if (unlikely(!z
->zone
)) {
1480 * Happens if we have an empty zonelist as a result of
1481 * GFP_THISNODE being used on a memoryless node
1486 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1487 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1492 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1493 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1494 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1495 * using a larger set of nodes after it has established that the
1496 * allowed per node queues are empty and that nodes are
1499 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1502 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1503 wakeup_kswapd(zone
, order
);
1506 * OK, we're below the kswapd watermark and have kicked background
1507 * reclaim. Now things get more complex, so set up alloc_flags according
1508 * to how we want to proceed.
1510 * The caller may dip into page reserves a bit more if the caller
1511 * cannot run direct reclaim, or if the caller has realtime scheduling
1512 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1513 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1515 alloc_flags
= ALLOC_WMARK_MIN
;
1516 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1517 alloc_flags
|= ALLOC_HARDER
;
1518 if (gfp_mask
& __GFP_HIGH
)
1519 alloc_flags
|= ALLOC_HIGH
;
1521 alloc_flags
|= ALLOC_CPUSET
;
1524 * Go through the zonelist again. Let __GFP_HIGH and allocations
1525 * coming from realtime tasks go deeper into reserves.
1527 * This is the last chance, in general, before the goto nopage.
1528 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1529 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1531 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1532 high_zoneidx
, alloc_flags
);
1536 /* This allocation should allow future memory freeing. */
1539 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1540 && !in_interrupt()) {
1541 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1543 /* go through the zonelist yet again, ignoring mins */
1544 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1545 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1548 if (gfp_mask
& __GFP_NOFAIL
) {
1549 congestion_wait(WRITE
, HZ
/50);
1556 /* Atomic allocations - we can't balance anything */
1562 /* We now go into synchronous reclaim */
1563 cpuset_memory_pressure_bump();
1565 * The task's cpuset might have expanded its set of allowable nodes
1567 cpuset_update_task_memory_state();
1568 p
->flags
|= PF_MEMALLOC
;
1569 reclaim_state
.reclaimed_slab
= 0;
1570 p
->reclaim_state
= &reclaim_state
;
1572 did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
);
1574 p
->reclaim_state
= NULL
;
1575 p
->flags
&= ~PF_MEMALLOC
;
1582 if (likely(did_some_progress
)) {
1583 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1584 zonelist
, high_zoneidx
, alloc_flags
);
1587 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1588 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1589 schedule_timeout_uninterruptible(1);
1594 * Go through the zonelist yet one more time, keep
1595 * very high watermark here, this is only to catch
1596 * a parallel oom killing, we must fail if we're still
1597 * under heavy pressure.
1599 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1600 order
, zonelist
, high_zoneidx
,
1601 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1603 clear_zonelist_oom(zonelist
, gfp_mask
);
1607 /* The OOM killer will not help higher order allocs so fail */
1608 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1609 clear_zonelist_oom(zonelist
, gfp_mask
);
1613 out_of_memory(zonelist
, gfp_mask
, order
);
1614 clear_zonelist_oom(zonelist
, gfp_mask
);
1619 * Don't let big-order allocations loop unless the caller explicitly
1620 * requests that. Wait for some write requests to complete then retry.
1622 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1623 * means __GFP_NOFAIL, but that may not be true in other
1626 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1627 * specified, then we retry until we no longer reclaim any pages
1628 * (above), or we've reclaimed an order of pages at least as
1629 * large as the allocation's order. In both cases, if the
1630 * allocation still fails, we stop retrying.
1632 pages_reclaimed
+= did_some_progress
;
1634 if (!(gfp_mask
& __GFP_NORETRY
)) {
1635 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1638 if (gfp_mask
& __GFP_REPEAT
&&
1639 pages_reclaimed
< (1 << order
))
1642 if (gfp_mask
& __GFP_NOFAIL
)
1646 congestion_wait(WRITE
, HZ
/50);
1651 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1652 printk(KERN_WARNING
"%s: page allocation failure."
1653 " order:%d, mode:0x%x\n",
1654 p
->comm
, order
, gfp_mask
);
1661 EXPORT_SYMBOL(__alloc_pages_internal
);
1664 * Common helper functions.
1666 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1669 page
= alloc_pages(gfp_mask
, order
);
1672 return (unsigned long) page_address(page
);
1675 EXPORT_SYMBOL(__get_free_pages
);
1677 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1682 * get_zeroed_page() returns a 32-bit address, which cannot represent
1685 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1687 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1689 return (unsigned long) page_address(page
);
1693 EXPORT_SYMBOL(get_zeroed_page
);
1695 void __pagevec_free(struct pagevec
*pvec
)
1697 int i
= pagevec_count(pvec
);
1700 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1703 void __free_pages(struct page
*page
, unsigned int order
)
1705 if (put_page_testzero(page
)) {
1707 free_hot_page(page
);
1709 __free_pages_ok(page
, order
);
1713 EXPORT_SYMBOL(__free_pages
);
1715 void free_pages(unsigned long addr
, unsigned int order
)
1718 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1719 __free_pages(virt_to_page((void *)addr
), order
);
1723 EXPORT_SYMBOL(free_pages
);
1726 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1727 * @size: the number of bytes to allocate
1728 * @gfp_mask: GFP flags for the allocation
1730 * This function is similar to alloc_pages(), except that it allocates the
1731 * minimum number of pages to satisfy the request. alloc_pages() can only
1732 * allocate memory in power-of-two pages.
1734 * This function is also limited by MAX_ORDER.
1736 * Memory allocated by this function must be released by free_pages_exact().
1738 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1740 unsigned int order
= get_order(size
);
1743 addr
= __get_free_pages(gfp_mask
, order
);
1745 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1746 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1748 split_page(virt_to_page(addr
), order
);
1749 while (used
< alloc_end
) {
1755 return (void *)addr
;
1757 EXPORT_SYMBOL(alloc_pages_exact
);
1760 * free_pages_exact - release memory allocated via alloc_pages_exact()
1761 * @virt: the value returned by alloc_pages_exact.
1762 * @size: size of allocation, same value as passed to alloc_pages_exact().
1764 * Release the memory allocated by a previous call to alloc_pages_exact.
1766 void free_pages_exact(void *virt
, size_t size
)
1768 unsigned long addr
= (unsigned long)virt
;
1769 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1771 while (addr
< end
) {
1776 EXPORT_SYMBOL(free_pages_exact
);
1778 static unsigned int nr_free_zone_pages(int offset
)
1783 /* Just pick one node, since fallback list is circular */
1784 unsigned int sum
= 0;
1786 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1788 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1789 unsigned long size
= zone
->present_pages
;
1790 unsigned long high
= zone
->pages_high
;
1798 static unsigned int nr_unallocated_zone_pages(int offset
)
1803 /* Just pick one node, since fallback list is circular */
1804 unsigned int sum
= 0;
1806 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1808 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1809 unsigned long high
= zone
->pages_high
;
1810 unsigned long left
= zone_page_state(zone
, NR_FREE_PAGES
);
1819 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1821 unsigned int nr_free_buffer_pages(void)
1823 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1825 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1828 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1830 unsigned int nr_unallocated_buffer_pages(void)
1832 return nr_unallocated_zone_pages(gfp_zone(GFP_USER
));
1834 EXPORT_IF_TOI_MODULAR(nr_unallocated_buffer_pages
);
1837 * Amount of free RAM allocatable within all zones
1839 unsigned int nr_free_pagecache_pages(void)
1841 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1844 static inline void show_node(struct zone
*zone
)
1847 printk("Node %d ", zone_to_nid(zone
));
1850 void si_meminfo(struct sysinfo
*val
)
1852 val
->totalram
= totalram_pages
;
1854 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1855 val
->bufferram
= nr_blockdev_pages();
1856 val
->totalhigh
= totalhigh_pages
;
1857 val
->freehigh
= nr_free_highpages();
1858 val
->mem_unit
= PAGE_SIZE
;
1861 EXPORT_SYMBOL(si_meminfo
);
1864 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1866 pg_data_t
*pgdat
= NODE_DATA(nid
);
1868 val
->totalram
= pgdat
->node_present_pages
;
1869 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1870 #ifdef CONFIG_HIGHMEM
1871 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1872 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1878 val
->mem_unit
= PAGE_SIZE
;
1882 #define K(x) ((x) << (PAGE_SHIFT-10))
1885 * Show free area list (used inside shift_scroll-lock stuff)
1886 * We also calculate the percentage fragmentation. We do this by counting the
1887 * memory on each free list with the exception of the first item on the list.
1889 void show_free_areas(void)
1894 for_each_zone(zone
) {
1895 if (!populated_zone(zone
))
1899 printk("%s per-cpu:\n", zone
->name
);
1901 for_each_online_cpu(cpu
) {
1902 struct per_cpu_pageset
*pageset
;
1904 pageset
= zone_pcp(zone
, cpu
);
1906 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1907 cpu
, pageset
->pcp
.high
,
1908 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1912 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1913 " inactive_file:%lu"
1914 //TODO: check/adjust line lengths
1915 #ifdef CONFIG_UNEVICTABLE_LRU
1918 " dirty:%lu writeback:%lu unstable:%lu\n"
1919 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1920 global_page_state(NR_ACTIVE_ANON
),
1921 global_page_state(NR_ACTIVE_FILE
),
1922 global_page_state(NR_INACTIVE_ANON
),
1923 global_page_state(NR_INACTIVE_FILE
),
1924 #ifdef CONFIG_UNEVICTABLE_LRU
1925 global_page_state(NR_UNEVICTABLE
),
1927 global_page_state(NR_FILE_DIRTY
),
1928 global_page_state(NR_WRITEBACK
),
1929 global_page_state(NR_UNSTABLE_NFS
),
1930 global_page_state(NR_FREE_PAGES
),
1931 global_page_state(NR_SLAB_RECLAIMABLE
) +
1932 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1933 global_page_state(NR_FILE_MAPPED
),
1934 global_page_state(NR_PAGETABLE
),
1935 global_page_state(NR_BOUNCE
));
1937 for_each_zone(zone
) {
1940 if (!populated_zone(zone
))
1949 " active_anon:%lukB"
1950 " inactive_anon:%lukB"
1951 " active_file:%lukB"
1952 " inactive_file:%lukB"
1953 #ifdef CONFIG_UNEVICTABLE_LRU
1954 " unevictable:%lukB"
1957 " pages_scanned:%lu"
1958 " all_unreclaimable? %s"
1961 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1964 K(zone
->pages_high
),
1965 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
1966 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
1967 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
1968 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
1969 #ifdef CONFIG_UNEVICTABLE_LRU
1970 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
1972 K(zone
->present_pages
),
1973 zone
->pages_scanned
,
1974 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1976 printk("lowmem_reserve[]:");
1977 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1978 printk(" %lu", zone
->lowmem_reserve
[i
]);
1982 for_each_zone(zone
) {
1983 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1985 if (!populated_zone(zone
))
1989 printk("%s: ", zone
->name
);
1991 spin_lock_irqsave(&zone
->lock
, flags
);
1992 for (order
= 0; order
< MAX_ORDER
; order
++) {
1993 nr
[order
] = zone
->free_area
[order
].nr_free
;
1994 total
+= nr
[order
] << order
;
1996 spin_unlock_irqrestore(&zone
->lock
, flags
);
1997 for (order
= 0; order
< MAX_ORDER
; order
++)
1998 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1999 printk("= %lukB\n", K(total
));
2002 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2004 show_swap_cache_info();
2007 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2009 zoneref
->zone
= zone
;
2010 zoneref
->zone_idx
= zone_idx(zone
);
2014 * Builds allocation fallback zone lists.
2016 * Add all populated zones of a node to the zonelist.
2018 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2019 int nr_zones
, enum zone_type zone_type
)
2023 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2028 zone
= pgdat
->node_zones
+ zone_type
;
2029 if (populated_zone(zone
)) {
2030 zoneref_set_zone(zone
,
2031 &zonelist
->_zonerefs
[nr_zones
++]);
2032 check_highest_zone(zone_type
);
2035 } while (zone_type
);
2042 * 0 = automatic detection of better ordering.
2043 * 1 = order by ([node] distance, -zonetype)
2044 * 2 = order by (-zonetype, [node] distance)
2046 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2047 * the same zonelist. So only NUMA can configure this param.
2049 #define ZONELIST_ORDER_DEFAULT 0
2050 #define ZONELIST_ORDER_NODE 1
2051 #define ZONELIST_ORDER_ZONE 2
2053 /* zonelist order in the kernel.
2054 * set_zonelist_order() will set this to NODE or ZONE.
2056 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2057 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2061 /* The value user specified ....changed by config */
2062 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2063 /* string for sysctl */
2064 #define NUMA_ZONELIST_ORDER_LEN 16
2065 char numa_zonelist_order
[16] = "default";
2068 * interface for configure zonelist ordering.
2069 * command line option "numa_zonelist_order"
2070 * = "[dD]efault - default, automatic configuration.
2071 * = "[nN]ode - order by node locality, then by zone within node
2072 * = "[zZ]one - order by zone, then by locality within zone
2075 static int __parse_numa_zonelist_order(char *s
)
2077 if (*s
== 'd' || *s
== 'D') {
2078 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2079 } else if (*s
== 'n' || *s
== 'N') {
2080 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2081 } else if (*s
== 'z' || *s
== 'Z') {
2082 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2085 "Ignoring invalid numa_zonelist_order value: "
2092 static __init
int setup_numa_zonelist_order(char *s
)
2095 return __parse_numa_zonelist_order(s
);
2098 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2101 * sysctl handler for numa_zonelist_order
2103 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2104 struct file
*file
, void __user
*buffer
, size_t *length
,
2107 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2111 strncpy(saved_string
, (char*)table
->data
,
2112 NUMA_ZONELIST_ORDER_LEN
);
2113 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2117 int oldval
= user_zonelist_order
;
2118 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2120 * bogus value. restore saved string
2122 strncpy((char*)table
->data
, saved_string
,
2123 NUMA_ZONELIST_ORDER_LEN
);
2124 user_zonelist_order
= oldval
;
2125 } else if (oldval
!= user_zonelist_order
)
2126 build_all_zonelists();
2132 #define MAX_NODE_LOAD (num_online_nodes())
2133 static int node_load
[MAX_NUMNODES
];
2136 * find_next_best_node - find the next node that should appear in a given node's fallback list
2137 * @node: node whose fallback list we're appending
2138 * @used_node_mask: nodemask_t of already used nodes
2140 * We use a number of factors to determine which is the next node that should
2141 * appear on a given node's fallback list. The node should not have appeared
2142 * already in @node's fallback list, and it should be the next closest node
2143 * according to the distance array (which contains arbitrary distance values
2144 * from each node to each node in the system), and should also prefer nodes
2145 * with no CPUs, since presumably they'll have very little allocation pressure
2146 * on them otherwise.
2147 * It returns -1 if no node is found.
2149 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2152 int min_val
= INT_MAX
;
2154 node_to_cpumask_ptr(tmp
, 0);
2156 /* Use the local node if we haven't already */
2157 if (!node_isset(node
, *used_node_mask
)) {
2158 node_set(node
, *used_node_mask
);
2162 for_each_node_state(n
, N_HIGH_MEMORY
) {
2164 /* Don't want a node to appear more than once */
2165 if (node_isset(n
, *used_node_mask
))
2168 /* Use the distance array to find the distance */
2169 val
= node_distance(node
, n
);
2171 /* Penalize nodes under us ("prefer the next node") */
2174 /* Give preference to headless and unused nodes */
2175 node_to_cpumask_ptr_next(tmp
, n
);
2176 if (!cpus_empty(*tmp
))
2177 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2179 /* Slight preference for less loaded node */
2180 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2181 val
+= node_load
[n
];
2183 if (val
< min_val
) {
2190 node_set(best_node
, *used_node_mask
);
2197 * Build zonelists ordered by node and zones within node.
2198 * This results in maximum locality--normal zone overflows into local
2199 * DMA zone, if any--but risks exhausting DMA zone.
2201 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2204 struct zonelist
*zonelist
;
2206 zonelist
= &pgdat
->node_zonelists
[0];
2207 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2209 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2211 zonelist
->_zonerefs
[j
].zone
= NULL
;
2212 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2216 * Build gfp_thisnode zonelists
2218 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2221 struct zonelist
*zonelist
;
2223 zonelist
= &pgdat
->node_zonelists
[1];
2224 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2225 zonelist
->_zonerefs
[j
].zone
= NULL
;
2226 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2230 * Build zonelists ordered by zone and nodes within zones.
2231 * This results in conserving DMA zone[s] until all Normal memory is
2232 * exhausted, but results in overflowing to remote node while memory
2233 * may still exist in local DMA zone.
2235 static int node_order
[MAX_NUMNODES
];
2237 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2240 int zone_type
; /* needs to be signed */
2242 struct zonelist
*zonelist
;
2244 zonelist
= &pgdat
->node_zonelists
[0];
2246 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2247 for (j
= 0; j
< nr_nodes
; j
++) {
2248 node
= node_order
[j
];
2249 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2250 if (populated_zone(z
)) {
2252 &zonelist
->_zonerefs
[pos
++]);
2253 check_highest_zone(zone_type
);
2257 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2258 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2261 static int default_zonelist_order(void)
2264 unsigned long low_kmem_size
,total_size
;
2268 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2269 * If they are really small and used heavily, the system can fall
2270 * into OOM very easily.
2271 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2273 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2276 for_each_online_node(nid
) {
2277 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2278 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2279 if (populated_zone(z
)) {
2280 if (zone_type
< ZONE_NORMAL
)
2281 low_kmem_size
+= z
->present_pages
;
2282 total_size
+= z
->present_pages
;
2286 if (!low_kmem_size
|| /* there are no DMA area. */
2287 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2288 return ZONELIST_ORDER_NODE
;
2290 * look into each node's config.
2291 * If there is a node whose DMA/DMA32 memory is very big area on
2292 * local memory, NODE_ORDER may be suitable.
2294 average_size
= total_size
/
2295 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2296 for_each_online_node(nid
) {
2299 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2300 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2301 if (populated_zone(z
)) {
2302 if (zone_type
< ZONE_NORMAL
)
2303 low_kmem_size
+= z
->present_pages
;
2304 total_size
+= z
->present_pages
;
2307 if (low_kmem_size
&&
2308 total_size
> average_size
&& /* ignore small node */
2309 low_kmem_size
> total_size
* 70/100)
2310 return ZONELIST_ORDER_NODE
;
2312 return ZONELIST_ORDER_ZONE
;
2315 static void set_zonelist_order(void)
2317 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2318 current_zonelist_order
= default_zonelist_order();
2320 current_zonelist_order
= user_zonelist_order
;
2323 static void build_zonelists(pg_data_t
*pgdat
)
2327 nodemask_t used_mask
;
2328 int local_node
, prev_node
;
2329 struct zonelist
*zonelist
;
2330 int order
= current_zonelist_order
;
2332 /* initialize zonelists */
2333 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2334 zonelist
= pgdat
->node_zonelists
+ i
;
2335 zonelist
->_zonerefs
[0].zone
= NULL
;
2336 zonelist
->_zonerefs
[0].zone_idx
= 0;
2339 /* NUMA-aware ordering of nodes */
2340 local_node
= pgdat
->node_id
;
2341 load
= num_online_nodes();
2342 prev_node
= local_node
;
2343 nodes_clear(used_mask
);
2345 memset(node_load
, 0, sizeof(node_load
));
2346 memset(node_order
, 0, sizeof(node_order
));
2349 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2350 int distance
= node_distance(local_node
, node
);
2353 * If another node is sufficiently far away then it is better
2354 * to reclaim pages in a zone before going off node.
2356 if (distance
> RECLAIM_DISTANCE
)
2357 zone_reclaim_mode
= 1;
2360 * We don't want to pressure a particular node.
2361 * So adding penalty to the first node in same
2362 * distance group to make it round-robin.
2364 if (distance
!= node_distance(local_node
, prev_node
))
2365 node_load
[node
] = load
;
2369 if (order
== ZONELIST_ORDER_NODE
)
2370 build_zonelists_in_node_order(pgdat
, node
);
2372 node_order
[j
++] = node
; /* remember order */
2375 if (order
== ZONELIST_ORDER_ZONE
) {
2376 /* calculate node order -- i.e., DMA last! */
2377 build_zonelists_in_zone_order(pgdat
, j
);
2380 build_thisnode_zonelists(pgdat
);
2383 /* Construct the zonelist performance cache - see further mmzone.h */
2384 static void build_zonelist_cache(pg_data_t
*pgdat
)
2386 struct zonelist
*zonelist
;
2387 struct zonelist_cache
*zlc
;
2390 zonelist
= &pgdat
->node_zonelists
[0];
2391 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2392 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2393 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2394 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2398 #else /* CONFIG_NUMA */
2400 static void set_zonelist_order(void)
2402 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2405 static void build_zonelists(pg_data_t
*pgdat
)
2407 int node
, local_node
;
2409 struct zonelist
*zonelist
;
2411 local_node
= pgdat
->node_id
;
2413 zonelist
= &pgdat
->node_zonelists
[0];
2414 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2417 * Now we build the zonelist so that it contains the zones
2418 * of all the other nodes.
2419 * We don't want to pressure a particular node, so when
2420 * building the zones for node N, we make sure that the
2421 * zones coming right after the local ones are those from
2422 * node N+1 (modulo N)
2424 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2425 if (!node_online(node
))
2427 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2430 for (node
= 0; node
< local_node
; node
++) {
2431 if (!node_online(node
))
2433 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2437 zonelist
->_zonerefs
[j
].zone
= NULL
;
2438 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2441 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2442 static void build_zonelist_cache(pg_data_t
*pgdat
)
2444 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2447 #endif /* CONFIG_NUMA */
2449 /* return values int ....just for stop_machine() */
2450 static int __build_all_zonelists(void *dummy
)
2454 for_each_online_node(nid
) {
2455 pg_data_t
*pgdat
= NODE_DATA(nid
);
2457 build_zonelists(pgdat
);
2458 build_zonelist_cache(pgdat
);
2463 void build_all_zonelists(void)
2465 set_zonelist_order();
2467 if (system_state
== SYSTEM_BOOTING
) {
2468 __build_all_zonelists(NULL
);
2469 mminit_verify_zonelist();
2470 cpuset_init_current_mems_allowed();
2472 /* we have to stop all cpus to guarantee there is no user
2474 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2475 /* cpuset refresh routine should be here */
2477 vm_total_pages
= nr_free_pagecache_pages();
2479 * Disable grouping by mobility if the number of pages in the
2480 * system is too low to allow the mechanism to work. It would be
2481 * more accurate, but expensive to check per-zone. This check is
2482 * made on memory-hotadd so a system can start with mobility
2483 * disabled and enable it later
2485 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2486 page_group_by_mobility_disabled
= 1;
2488 page_group_by_mobility_disabled
= 0;
2490 printk("Built %i zonelists in %s order, mobility grouping %s. "
2491 "Total pages: %ld\n",
2493 zonelist_order_name
[current_zonelist_order
],
2494 page_group_by_mobility_disabled
? "off" : "on",
2497 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2502 * Helper functions to size the waitqueue hash table.
2503 * Essentially these want to choose hash table sizes sufficiently
2504 * large so that collisions trying to wait on pages are rare.
2505 * But in fact, the number of active page waitqueues on typical
2506 * systems is ridiculously low, less than 200. So this is even
2507 * conservative, even though it seems large.
2509 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2510 * waitqueues, i.e. the size of the waitq table given the number of pages.
2512 #define PAGES_PER_WAITQUEUE 256
2514 #ifndef CONFIG_MEMORY_HOTPLUG
2515 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2517 unsigned long size
= 1;
2519 pages
/= PAGES_PER_WAITQUEUE
;
2521 while (size
< pages
)
2525 * Once we have dozens or even hundreds of threads sleeping
2526 * on IO we've got bigger problems than wait queue collision.
2527 * Limit the size of the wait table to a reasonable size.
2529 size
= min(size
, 4096UL);
2531 return max(size
, 4UL);
2535 * A zone's size might be changed by hot-add, so it is not possible to determine
2536 * a suitable size for its wait_table. So we use the maximum size now.
2538 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2540 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2541 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2542 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2544 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2545 * or more by the traditional way. (See above). It equals:
2547 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2548 * ia64(16K page size) : = ( 8G + 4M)byte.
2549 * powerpc (64K page size) : = (32G +16M)byte.
2551 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2558 * This is an integer logarithm so that shifts can be used later
2559 * to extract the more random high bits from the multiplicative
2560 * hash function before the remainder is taken.
2562 static inline unsigned long wait_table_bits(unsigned long size
)
2567 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2570 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2571 * of blocks reserved is based on zone->pages_min. The memory within the
2572 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2573 * higher will lead to a bigger reserve which will get freed as contiguous
2574 * blocks as reclaim kicks in
2576 static void setup_zone_migrate_reserve(struct zone
*zone
)
2578 unsigned long start_pfn
, pfn
, end_pfn
;
2580 unsigned long reserve
, block_migratetype
;
2582 /* Get the start pfn, end pfn and the number of blocks to reserve */
2583 start_pfn
= zone
->zone_start_pfn
;
2584 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2585 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2588 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2589 if (!pfn_valid(pfn
))
2591 page
= pfn_to_page(pfn
);
2593 /* Watch out for overlapping nodes */
2594 if (page_to_nid(page
) != zone_to_nid(zone
))
2597 /* Blocks with reserved pages will never free, skip them. */
2598 if (PageReserved(page
))
2601 block_migratetype
= get_pageblock_migratetype(page
);
2603 /* If this block is reserved, account for it */
2604 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2609 /* Suitable for reserving if this block is movable */
2610 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2611 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2612 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2618 * If the reserve is met and this is a previous reserved block,
2621 if (block_migratetype
== MIGRATE_RESERVE
) {
2622 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2623 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2629 * Initially all pages are reserved - free ones are freed
2630 * up by free_all_bootmem() once the early boot process is
2631 * done. Non-atomic initialization, single-pass.
2633 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2634 unsigned long start_pfn
, enum memmap_context context
)
2637 unsigned long end_pfn
= start_pfn
+ size
;
2641 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2642 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2644 * There can be holes in boot-time mem_map[]s
2645 * handed to this function. They do not
2646 * exist on hotplugged memory.
2648 if (context
== MEMMAP_EARLY
) {
2649 if (!early_pfn_valid(pfn
))
2651 if (!early_pfn_in_nid(pfn
, nid
))
2654 page
= pfn_to_page(pfn
);
2655 set_page_links(page
, zone
, nid
, pfn
);
2656 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2657 init_page_count(page
);
2658 reset_page_mapcount(page
);
2659 SetPageReserved(page
);
2661 * Mark the block movable so that blocks are reserved for
2662 * movable at startup. This will force kernel allocations
2663 * to reserve their blocks rather than leaking throughout
2664 * the address space during boot when many long-lived
2665 * kernel allocations are made. Later some blocks near
2666 * the start are marked MIGRATE_RESERVE by
2667 * setup_zone_migrate_reserve()
2669 * bitmap is created for zone's valid pfn range. but memmap
2670 * can be created for invalid pages (for alignment)
2671 * check here not to call set_pageblock_migratetype() against
2674 if ((z
->zone_start_pfn
<= pfn
)
2675 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2676 && !(pfn
& (pageblock_nr_pages
- 1)))
2677 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2679 INIT_LIST_HEAD(&page
->lru
);
2680 #ifdef WANT_PAGE_VIRTUAL
2681 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2682 if (!is_highmem_idx(zone
))
2683 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2688 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2691 for_each_migratetype_order(order
, t
) {
2692 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2693 zone
->free_area
[order
].nr_free
= 0;
2697 #ifndef __HAVE_ARCH_MEMMAP_INIT
2698 #define memmap_init(size, nid, zone, start_pfn) \
2699 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2702 static int zone_batchsize(struct zone
*zone
)
2707 * The per-cpu-pages pools are set to around 1000th of the
2708 * size of the zone. But no more than 1/2 of a meg.
2710 * OK, so we don't know how big the cache is. So guess.
2712 batch
= zone
->present_pages
/ 1024;
2713 if (batch
* PAGE_SIZE
> 512 * 1024)
2714 batch
= (512 * 1024) / PAGE_SIZE
;
2715 batch
/= 4; /* We effectively *= 4 below */
2720 * Clamp the batch to a 2^n - 1 value. Having a power
2721 * of 2 value was found to be more likely to have
2722 * suboptimal cache aliasing properties in some cases.
2724 * For example if 2 tasks are alternately allocating
2725 * batches of pages, one task can end up with a lot
2726 * of pages of one half of the possible page colors
2727 * and the other with pages of the other colors.
2729 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2734 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2736 struct per_cpu_pages
*pcp
;
2738 memset(p
, 0, sizeof(*p
));
2742 pcp
->high
= 6 * batch
;
2743 pcp
->batch
= max(1UL, 1 * batch
);
2744 INIT_LIST_HEAD(&pcp
->list
);
2748 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2749 * to the value high for the pageset p.
2752 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2755 struct per_cpu_pages
*pcp
;
2759 pcp
->batch
= max(1UL, high
/4);
2760 if ((high
/4) > (PAGE_SHIFT
* 8))
2761 pcp
->batch
= PAGE_SHIFT
* 8;
2767 * Boot pageset table. One per cpu which is going to be used for all
2768 * zones and all nodes. The parameters will be set in such a way
2769 * that an item put on a list will immediately be handed over to
2770 * the buddy list. This is safe since pageset manipulation is done
2771 * with interrupts disabled.
2773 * Some NUMA counter updates may also be caught by the boot pagesets.
2775 * The boot_pagesets must be kept even after bootup is complete for
2776 * unused processors and/or zones. They do play a role for bootstrapping
2777 * hotplugged processors.
2779 * zoneinfo_show() and maybe other functions do
2780 * not check if the processor is online before following the pageset pointer.
2781 * Other parts of the kernel may not check if the zone is available.
2783 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2786 * Dynamically allocate memory for the
2787 * per cpu pageset array in struct zone.
2789 static int __cpuinit
process_zones(int cpu
)
2791 struct zone
*zone
, *dzone
;
2792 int node
= cpu_to_node(cpu
);
2794 node_set_state(node
, N_CPU
); /* this node has a cpu */
2796 for_each_zone(zone
) {
2798 if (!populated_zone(zone
))
2801 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2803 if (!zone_pcp(zone
, cpu
))
2806 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2808 if (percpu_pagelist_fraction
)
2809 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2810 (zone
->present_pages
/ percpu_pagelist_fraction
));
2815 for_each_zone(dzone
) {
2816 if (!populated_zone(dzone
))
2820 kfree(zone_pcp(dzone
, cpu
));
2821 zone_pcp(dzone
, cpu
) = NULL
;
2826 static inline void free_zone_pagesets(int cpu
)
2830 for_each_zone(zone
) {
2831 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2833 /* Free per_cpu_pageset if it is slab allocated */
2834 if (pset
!= &boot_pageset
[cpu
])
2836 zone_pcp(zone
, cpu
) = NULL
;
2840 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2841 unsigned long action
,
2844 int cpu
= (long)hcpu
;
2845 int ret
= NOTIFY_OK
;
2848 case CPU_UP_PREPARE
:
2849 case CPU_UP_PREPARE_FROZEN
:
2850 if (process_zones(cpu
))
2853 case CPU_UP_CANCELED
:
2854 case CPU_UP_CANCELED_FROZEN
:
2856 case CPU_DEAD_FROZEN
:
2857 free_zone_pagesets(cpu
);
2865 static struct notifier_block __cpuinitdata pageset_notifier
=
2866 { &pageset_cpuup_callback
, NULL
, 0 };
2868 void __init
setup_per_cpu_pageset(void)
2872 /* Initialize per_cpu_pageset for cpu 0.
2873 * A cpuup callback will do this for every cpu
2874 * as it comes online
2876 err
= process_zones(smp_processor_id());
2878 register_cpu_notifier(&pageset_notifier
);
2883 static noinline __init_refok
2884 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2887 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2891 * The per-page waitqueue mechanism uses hashed waitqueues
2894 zone
->wait_table_hash_nr_entries
=
2895 wait_table_hash_nr_entries(zone_size_pages
);
2896 zone
->wait_table_bits
=
2897 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2898 alloc_size
= zone
->wait_table_hash_nr_entries
2899 * sizeof(wait_queue_head_t
);
2901 if (!slab_is_available()) {
2902 zone
->wait_table
= (wait_queue_head_t
*)
2903 alloc_bootmem_node(pgdat
, alloc_size
);
2906 * This case means that a zone whose size was 0 gets new memory
2907 * via memory hot-add.
2908 * But it may be the case that a new node was hot-added. In
2909 * this case vmalloc() will not be able to use this new node's
2910 * memory - this wait_table must be initialized to use this new
2911 * node itself as well.
2912 * To use this new node's memory, further consideration will be
2915 zone
->wait_table
= vmalloc(alloc_size
);
2917 if (!zone
->wait_table
)
2920 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2921 init_waitqueue_head(zone
->wait_table
+ i
);
2926 static __meminit
void zone_pcp_init(struct zone
*zone
)
2929 unsigned long batch
= zone_batchsize(zone
);
2931 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2933 /* Early boot. Slab allocator not functional yet */
2934 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2935 setup_pageset(&boot_pageset
[cpu
],0);
2937 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2940 if (zone
->present_pages
)
2941 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2942 zone
->name
, zone
->present_pages
, batch
);
2945 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2946 unsigned long zone_start_pfn
,
2948 enum memmap_context context
)
2950 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2952 ret
= zone_wait_table_init(zone
, size
);
2955 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2957 zone
->zone_start_pfn
= zone_start_pfn
;
2959 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
2960 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2962 (unsigned long)zone_idx(zone
),
2963 zone_start_pfn
, (zone_start_pfn
+ size
));
2965 zone_init_free_lists(zone
);
2970 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2972 * Basic iterator support. Return the first range of PFNs for a node
2973 * Note: nid == MAX_NUMNODES returns first region regardless of node
2975 static int __meminit
first_active_region_index_in_nid(int nid
)
2979 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2980 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2987 * Basic iterator support. Return the next active range of PFNs for a node
2988 * Note: nid == MAX_NUMNODES returns next region regardless of node
2990 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2992 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2993 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2999 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3001 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3002 * Architectures may implement their own version but if add_active_range()
3003 * was used and there are no special requirements, this is a convenient
3006 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3010 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3011 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3012 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3014 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3015 return early_node_map
[i
].nid
;
3020 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3022 /* Basic iterator support to walk early_node_map[] */
3023 #define for_each_active_range_index_in_nid(i, nid) \
3024 for (i = first_active_region_index_in_nid(nid); i != -1; \
3025 i = next_active_region_index_in_nid(i, nid))
3028 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3029 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3030 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3032 * If an architecture guarantees that all ranges registered with
3033 * add_active_ranges() contain no holes and may be freed, this
3034 * this function may be used instead of calling free_bootmem() manually.
3036 void __init
free_bootmem_with_active_regions(int nid
,
3037 unsigned long max_low_pfn
)
3041 for_each_active_range_index_in_nid(i
, nid
) {
3042 unsigned long size_pages
= 0;
3043 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3045 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3048 if (end_pfn
> max_low_pfn
)
3049 end_pfn
= max_low_pfn
;
3051 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3052 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3053 PFN_PHYS(early_node_map
[i
].start_pfn
),
3054 size_pages
<< PAGE_SHIFT
);
3058 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3063 for_each_active_range_index_in_nid(i
, nid
) {
3064 ret
= work_fn(early_node_map
[i
].start_pfn
,
3065 early_node_map
[i
].end_pfn
, data
);
3071 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3072 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3074 * If an architecture guarantees that all ranges registered with
3075 * add_active_ranges() contain no holes and may be freed, this
3076 * function may be used instead of calling memory_present() manually.
3078 void __init
sparse_memory_present_with_active_regions(int nid
)
3082 for_each_active_range_index_in_nid(i
, nid
)
3083 memory_present(early_node_map
[i
].nid
,
3084 early_node_map
[i
].start_pfn
,
3085 early_node_map
[i
].end_pfn
);
3089 * push_node_boundaries - Push node boundaries to at least the requested boundary
3090 * @nid: The nid of the node to push the boundary for
3091 * @start_pfn: The start pfn of the node
3092 * @end_pfn: The end pfn of the node
3094 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3095 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3096 * be hotplugged even though no physical memory exists. This function allows
3097 * an arch to push out the node boundaries so mem_map is allocated that can
3100 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3101 void __init
push_node_boundaries(unsigned int nid
,
3102 unsigned long start_pfn
, unsigned long end_pfn
)
3104 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3105 "Entering push_node_boundaries(%u, %lu, %lu)\n",
3106 nid
, start_pfn
, end_pfn
);
3108 /* Initialise the boundary for this node if necessary */
3109 if (node_boundary_end_pfn
[nid
] == 0)
3110 node_boundary_start_pfn
[nid
] = -1UL;
3112 /* Update the boundaries */
3113 if (node_boundary_start_pfn
[nid
] > start_pfn
)
3114 node_boundary_start_pfn
[nid
] = start_pfn
;
3115 if (node_boundary_end_pfn
[nid
] < end_pfn
)
3116 node_boundary_end_pfn
[nid
] = end_pfn
;
3119 /* If necessary, push the node boundary out for reserve hotadd */
3120 static void __meminit
account_node_boundary(unsigned int nid
,
3121 unsigned long *start_pfn
, unsigned long *end_pfn
)
3123 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3124 "Entering account_node_boundary(%u, %lu, %lu)\n",
3125 nid
, *start_pfn
, *end_pfn
);
3127 /* Return if boundary information has not been provided */
3128 if (node_boundary_end_pfn
[nid
] == 0)
3131 /* Check the boundaries and update if necessary */
3132 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
3133 *start_pfn
= node_boundary_start_pfn
[nid
];
3134 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
3135 *end_pfn
= node_boundary_end_pfn
[nid
];
3138 void __init
push_node_boundaries(unsigned int nid
,
3139 unsigned long start_pfn
, unsigned long end_pfn
) {}
3141 static void __meminit
account_node_boundary(unsigned int nid
,
3142 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3147 * get_pfn_range_for_nid - Return the start and end page frames for a node
3148 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3149 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3150 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3152 * It returns the start and end page frame of a node based on information
3153 * provided by an arch calling add_active_range(). If called for a node
3154 * with no available memory, a warning is printed and the start and end
3157 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3158 unsigned long *start_pfn
, unsigned long *end_pfn
)
3164 for_each_active_range_index_in_nid(i
, nid
) {
3165 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3166 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3169 if (*start_pfn
== -1UL)
3172 /* Push the node boundaries out if requested */
3173 account_node_boundary(nid
, start_pfn
, end_pfn
);
3177 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3178 * assumption is made that zones within a node are ordered in monotonic
3179 * increasing memory addresses so that the "highest" populated zone is used
3181 static void __init
find_usable_zone_for_movable(void)
3184 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3185 if (zone_index
== ZONE_MOVABLE
)
3188 if (arch_zone_highest_possible_pfn
[zone_index
] >
3189 arch_zone_lowest_possible_pfn
[zone_index
])
3193 VM_BUG_ON(zone_index
== -1);
3194 movable_zone
= zone_index
;
3198 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3199 * because it is sized independant of architecture. Unlike the other zones,
3200 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3201 * in each node depending on the size of each node and how evenly kernelcore
3202 * is distributed. This helper function adjusts the zone ranges
3203 * provided by the architecture for a given node by using the end of the
3204 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3205 * zones within a node are in order of monotonic increases memory addresses
3207 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3208 unsigned long zone_type
,
3209 unsigned long node_start_pfn
,
3210 unsigned long node_end_pfn
,
3211 unsigned long *zone_start_pfn
,
3212 unsigned long *zone_end_pfn
)
3214 /* Only adjust if ZONE_MOVABLE is on this node */
3215 if (zone_movable_pfn
[nid
]) {
3216 /* Size ZONE_MOVABLE */
3217 if (zone_type
== ZONE_MOVABLE
) {
3218 *zone_start_pfn
= zone_movable_pfn
[nid
];
3219 *zone_end_pfn
= min(node_end_pfn
,
3220 arch_zone_highest_possible_pfn
[movable_zone
]);
3222 /* Adjust for ZONE_MOVABLE starting within this range */
3223 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3224 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3225 *zone_end_pfn
= zone_movable_pfn
[nid
];
3227 /* Check if this whole range is within ZONE_MOVABLE */
3228 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3229 *zone_start_pfn
= *zone_end_pfn
;
3234 * Return the number of pages a zone spans in a node, including holes
3235 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3237 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3238 unsigned long zone_type
,
3239 unsigned long *ignored
)
3241 unsigned long node_start_pfn
, node_end_pfn
;
3242 unsigned long zone_start_pfn
, zone_end_pfn
;
3244 /* Get the start and end of the node and zone */
3245 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3246 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3247 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3248 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3249 node_start_pfn
, node_end_pfn
,
3250 &zone_start_pfn
, &zone_end_pfn
);
3252 /* Check that this node has pages within the zone's required range */
3253 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3256 /* Move the zone boundaries inside the node if necessary */
3257 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3258 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3260 /* Return the spanned pages */
3261 return zone_end_pfn
- zone_start_pfn
;
3265 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3266 * then all holes in the requested range will be accounted for.
3268 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3269 unsigned long range_start_pfn
,
3270 unsigned long range_end_pfn
)
3273 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3274 unsigned long start_pfn
;
3276 /* Find the end_pfn of the first active range of pfns in the node */
3277 i
= first_active_region_index_in_nid(nid
);
3281 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3283 /* Account for ranges before physical memory on this node */
3284 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3285 hole_pages
= prev_end_pfn
- range_start_pfn
;
3287 /* Find all holes for the zone within the node */
3288 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3290 /* No need to continue if prev_end_pfn is outside the zone */
3291 if (prev_end_pfn
>= range_end_pfn
)
3294 /* Make sure the end of the zone is not within the hole */
3295 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3296 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3298 /* Update the hole size cound and move on */
3299 if (start_pfn
> range_start_pfn
) {
3300 BUG_ON(prev_end_pfn
> start_pfn
);
3301 hole_pages
+= start_pfn
- prev_end_pfn
;
3303 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3306 /* Account for ranges past physical memory on this node */
3307 if (range_end_pfn
> prev_end_pfn
)
3308 hole_pages
+= range_end_pfn
-
3309 max(range_start_pfn
, prev_end_pfn
);
3315 * absent_pages_in_range - Return number of page frames in holes within a range
3316 * @start_pfn: The start PFN to start searching for holes
3317 * @end_pfn: The end PFN to stop searching for holes
3319 * It returns the number of pages frames in memory holes within a range.
3321 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3322 unsigned long end_pfn
)
3324 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3327 /* Return the number of page frames in holes in a zone on a node */
3328 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3329 unsigned long zone_type
,
3330 unsigned long *ignored
)
3332 unsigned long node_start_pfn
, node_end_pfn
;
3333 unsigned long zone_start_pfn
, zone_end_pfn
;
3335 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3336 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3338 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3341 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3342 node_start_pfn
, node_end_pfn
,
3343 &zone_start_pfn
, &zone_end_pfn
);
3344 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3348 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3349 unsigned long zone_type
,
3350 unsigned long *zones_size
)
3352 return zones_size
[zone_type
];
3355 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3356 unsigned long zone_type
,
3357 unsigned long *zholes_size
)
3362 return zholes_size
[zone_type
];
3367 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3368 unsigned long *zones_size
, unsigned long *zholes_size
)
3370 unsigned long realtotalpages
, totalpages
= 0;
3373 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3374 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3376 pgdat
->node_spanned_pages
= totalpages
;
3378 realtotalpages
= totalpages
;
3379 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3381 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3383 pgdat
->node_present_pages
= realtotalpages
;
3384 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3388 #ifndef CONFIG_SPARSEMEM
3390 * Calculate the size of the zone->blockflags rounded to an unsigned long
3391 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3392 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3393 * round what is now in bits to nearest long in bits, then return it in
3396 static unsigned long __init
usemap_size(unsigned long zonesize
)
3398 unsigned long usemapsize
;
3400 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3401 usemapsize
= usemapsize
>> pageblock_order
;
3402 usemapsize
*= NR_PAGEBLOCK_BITS
;
3403 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3405 return usemapsize
/ 8;
3408 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3409 struct zone
*zone
, unsigned long zonesize
)
3411 unsigned long usemapsize
= usemap_size(zonesize
);
3412 zone
->pageblock_flags
= NULL
;
3414 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3415 memset(zone
->pageblock_flags
, 0, usemapsize
);
3419 static void inline setup_usemap(struct pglist_data
*pgdat
,
3420 struct zone
*zone
, unsigned long zonesize
) {}
3421 #endif /* CONFIG_SPARSEMEM */
3423 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3425 /* Return a sensible default order for the pageblock size. */
3426 static inline int pageblock_default_order(void)
3428 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3429 return HUGETLB_PAGE_ORDER
;
3434 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3435 static inline void __init
set_pageblock_order(unsigned int order
)
3437 /* Check that pageblock_nr_pages has not already been setup */
3438 if (pageblock_order
)
3442 * Assume the largest contiguous order of interest is a huge page.
3443 * This value may be variable depending on boot parameters on IA64
3445 pageblock_order
= order
;
3447 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3450 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3451 * and pageblock_default_order() are unused as pageblock_order is set
3452 * at compile-time. See include/linux/pageblock-flags.h for the values of
3453 * pageblock_order based on the kernel config
3455 static inline int pageblock_default_order(unsigned int order
)
3459 #define set_pageblock_order(x) do {} while (0)
3461 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3464 * Set up the zone data structures:
3465 * - mark all pages reserved
3466 * - mark all memory queues empty
3467 * - clear the memory bitmaps
3469 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3470 unsigned long *zones_size
, unsigned long *zholes_size
)
3473 int nid
= pgdat
->node_id
;
3474 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3477 pgdat_resize_init(pgdat
);
3478 pgdat
->nr_zones
= 0;
3479 init_waitqueue_head(&pgdat
->kswapd_wait
);
3480 pgdat
->kswapd_max_order
= 0;
3481 pgdat_page_cgroup_init(pgdat
);
3483 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3484 struct zone
*zone
= pgdat
->node_zones
+ j
;
3485 unsigned long size
, realsize
, memmap_pages
;
3488 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3489 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3493 * Adjust realsize so that it accounts for how much memory
3494 * is used by this zone for memmap. This affects the watermark
3495 * and per-cpu initialisations
3498 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3499 if (realsize
>= memmap_pages
) {
3500 realsize
-= memmap_pages
;
3502 " %s zone: %lu pages used for memmap\n",
3503 zone_names
[j
], memmap_pages
);
3506 " %s zone: %lu pages exceeds realsize %lu\n",
3507 zone_names
[j
], memmap_pages
, realsize
);
3509 /* Account for reserved pages */
3510 if (j
== 0 && realsize
> dma_reserve
) {
3511 realsize
-= dma_reserve
;
3512 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3513 zone_names
[0], dma_reserve
);
3516 if (!is_highmem_idx(j
))
3517 nr_kernel_pages
+= realsize
;
3518 nr_all_pages
+= realsize
;
3520 zone
->spanned_pages
= size
;
3521 zone
->present_pages
= realsize
;
3524 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3526 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3528 zone
->name
= zone_names
[j
];
3529 spin_lock_init(&zone
->lock
);
3530 spin_lock_init(&zone
->lru_lock
);
3531 zone_seqlock_init(zone
);
3532 zone
->zone_pgdat
= pgdat
;
3534 zone
->prev_priority
= DEF_PRIORITY
;
3536 zone_pcp_init(zone
);
3538 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3539 zone
->lru
[l
].nr_scan
= 0;
3541 zone
->recent_rotated
[0] = 0;
3542 zone
->recent_rotated
[1] = 0;
3543 zone
->recent_scanned
[0] = 0;
3544 zone
->recent_scanned
[1] = 0;
3545 zap_zone_vm_stats(zone
);
3550 set_pageblock_order(pageblock_default_order());
3551 setup_usemap(pgdat
, zone
, size
);
3552 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3553 size
, MEMMAP_EARLY
);
3555 memmap_init(size
, nid
, j
, zone_start_pfn
);
3556 zone_start_pfn
+= size
;
3560 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3562 /* Skip empty nodes */
3563 if (!pgdat
->node_spanned_pages
)
3566 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3567 /* ia64 gets its own node_mem_map, before this, without bootmem */
3568 if (!pgdat
->node_mem_map
) {
3569 unsigned long size
, start
, end
;
3573 * The zone's endpoints aren't required to be MAX_ORDER
3574 * aligned but the node_mem_map endpoints must be in order
3575 * for the buddy allocator to function correctly.
3577 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3578 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3579 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3580 size
= (end
- start
) * sizeof(struct page
);
3581 map
= alloc_remap(pgdat
->node_id
, size
);
3583 map
= alloc_bootmem_node(pgdat
, size
);
3584 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3586 #ifndef CONFIG_NEED_MULTIPLE_NODES
3588 * With no DISCONTIG, the global mem_map is just set as node 0's
3590 if (pgdat
== NODE_DATA(0)) {
3591 mem_map
= NODE_DATA(0)->node_mem_map
;
3592 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3593 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3594 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3595 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3598 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3601 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3602 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3604 pg_data_t
*pgdat
= NODE_DATA(nid
);
3606 pgdat
->node_id
= nid
;
3607 pgdat
->node_start_pfn
= node_start_pfn
;
3608 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3610 alloc_node_mem_map(pgdat
);
3611 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3612 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3613 nid
, (unsigned long)pgdat
,
3614 (unsigned long)pgdat
->node_mem_map
);
3617 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3620 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3622 #if MAX_NUMNODES > 1
3624 * Figure out the number of possible node ids.
3626 static void __init
setup_nr_node_ids(void)
3629 unsigned int highest
= 0;
3631 for_each_node_mask(node
, node_possible_map
)
3633 nr_node_ids
= highest
+ 1;
3636 static inline void setup_nr_node_ids(void)
3642 * add_active_range - Register a range of PFNs backed by physical memory
3643 * @nid: The node ID the range resides on
3644 * @start_pfn: The start PFN of the available physical memory
3645 * @end_pfn: The end PFN of the available physical memory
3647 * These ranges are stored in an early_node_map[] and later used by
3648 * free_area_init_nodes() to calculate zone sizes and holes. If the
3649 * range spans a memory hole, it is up to the architecture to ensure
3650 * the memory is not freed by the bootmem allocator. If possible
3651 * the range being registered will be merged with existing ranges.
3653 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3654 unsigned long end_pfn
)
3658 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3659 "Entering add_active_range(%d, %#lx, %#lx) "
3660 "%d entries of %d used\n",
3661 nid
, start_pfn
, end_pfn
,
3662 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3664 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3666 /* Merge with existing active regions if possible */
3667 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3668 if (early_node_map
[i
].nid
!= nid
)
3671 /* Skip if an existing region covers this new one */
3672 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3673 end_pfn
<= early_node_map
[i
].end_pfn
)
3676 /* Merge forward if suitable */
3677 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3678 end_pfn
> early_node_map
[i
].end_pfn
) {
3679 early_node_map
[i
].end_pfn
= end_pfn
;
3683 /* Merge backward if suitable */
3684 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3685 end_pfn
>= early_node_map
[i
].start_pfn
) {
3686 early_node_map
[i
].start_pfn
= start_pfn
;
3691 /* Check that early_node_map is large enough */
3692 if (i
>= MAX_ACTIVE_REGIONS
) {
3693 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3694 MAX_ACTIVE_REGIONS
);
3698 early_node_map
[i
].nid
= nid
;
3699 early_node_map
[i
].start_pfn
= start_pfn
;
3700 early_node_map
[i
].end_pfn
= end_pfn
;
3701 nr_nodemap_entries
= i
+ 1;
3705 * remove_active_range - Shrink an existing registered range of PFNs
3706 * @nid: The node id the range is on that should be shrunk
3707 * @start_pfn: The new PFN of the range
3708 * @end_pfn: The new PFN of the range
3710 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3711 * The map is kept near the end physical page range that has already been
3712 * registered. This function allows an arch to shrink an existing registered
3715 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3716 unsigned long end_pfn
)
3721 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3722 nid
, start_pfn
, end_pfn
);
3724 /* Find the old active region end and shrink */
3725 for_each_active_range_index_in_nid(i
, nid
) {
3726 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3727 early_node_map
[i
].end_pfn
<= end_pfn
) {
3729 early_node_map
[i
].start_pfn
= 0;
3730 early_node_map
[i
].end_pfn
= 0;
3734 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3735 early_node_map
[i
].end_pfn
> start_pfn
) {
3736 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3737 early_node_map
[i
].end_pfn
= start_pfn
;
3738 if (temp_end_pfn
> end_pfn
)
3739 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3742 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3743 early_node_map
[i
].end_pfn
> end_pfn
&&
3744 early_node_map
[i
].start_pfn
< end_pfn
) {
3745 early_node_map
[i
].start_pfn
= end_pfn
;
3753 /* remove the blank ones */
3754 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3755 if (early_node_map
[i
].nid
!= nid
)
3757 if (early_node_map
[i
].end_pfn
)
3759 /* we found it, get rid of it */
3760 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3761 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3762 sizeof(early_node_map
[j
]));
3763 j
= nr_nodemap_entries
- 1;
3764 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3765 nr_nodemap_entries
--;
3770 * remove_all_active_ranges - Remove all currently registered regions
3772 * During discovery, it may be found that a table like SRAT is invalid
3773 * and an alternative discovery method must be used. This function removes
3774 * all currently registered regions.
3776 void __init
remove_all_active_ranges(void)
3778 memset(early_node_map
, 0, sizeof(early_node_map
));
3779 nr_nodemap_entries
= 0;
3780 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3781 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3782 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3783 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3786 /* Compare two active node_active_regions */
3787 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3789 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3790 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3792 /* Done this way to avoid overflows */
3793 if (arange
->start_pfn
> brange
->start_pfn
)
3795 if (arange
->start_pfn
< brange
->start_pfn
)
3801 /* sort the node_map by start_pfn */
3802 static void __init
sort_node_map(void)
3804 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3805 sizeof(struct node_active_region
),
3806 cmp_node_active_region
, NULL
);
3809 /* Find the lowest pfn for a node */
3810 static unsigned long __init
find_min_pfn_for_node(int nid
)
3813 unsigned long min_pfn
= ULONG_MAX
;
3815 /* Assuming a sorted map, the first range found has the starting pfn */
3816 for_each_active_range_index_in_nid(i
, nid
)
3817 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3819 if (min_pfn
== ULONG_MAX
) {
3821 "Could not find start_pfn for node %d\n", nid
);
3829 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3831 * It returns the minimum PFN based on information provided via
3832 * add_active_range().
3834 unsigned long __init
find_min_pfn_with_active_regions(void)
3836 return find_min_pfn_for_node(MAX_NUMNODES
);
3840 * early_calculate_totalpages()
3841 * Sum pages in active regions for movable zone.
3842 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3844 static unsigned long __init
early_calculate_totalpages(void)
3847 unsigned long totalpages
= 0;
3849 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3850 unsigned long pages
= early_node_map
[i
].end_pfn
-
3851 early_node_map
[i
].start_pfn
;
3852 totalpages
+= pages
;
3854 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3860 * Find the PFN the Movable zone begins in each node. Kernel memory
3861 * is spread evenly between nodes as long as the nodes have enough
3862 * memory. When they don't, some nodes will have more kernelcore than
3865 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3868 unsigned long usable_startpfn
;
3869 unsigned long kernelcore_node
, kernelcore_remaining
;
3870 unsigned long totalpages
= early_calculate_totalpages();
3871 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3874 * If movablecore was specified, calculate what size of
3875 * kernelcore that corresponds so that memory usable for
3876 * any allocation type is evenly spread. If both kernelcore
3877 * and movablecore are specified, then the value of kernelcore
3878 * will be used for required_kernelcore if it's greater than
3879 * what movablecore would have allowed.
3881 if (required_movablecore
) {
3882 unsigned long corepages
;
3885 * Round-up so that ZONE_MOVABLE is at least as large as what
3886 * was requested by the user
3888 required_movablecore
=
3889 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3890 corepages
= totalpages
- required_movablecore
;
3892 required_kernelcore
= max(required_kernelcore
, corepages
);
3895 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3896 if (!required_kernelcore
)
3899 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3900 find_usable_zone_for_movable();
3901 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3904 /* Spread kernelcore memory as evenly as possible throughout nodes */
3905 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3906 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3908 * Recalculate kernelcore_node if the division per node
3909 * now exceeds what is necessary to satisfy the requested
3910 * amount of memory for the kernel
3912 if (required_kernelcore
< kernelcore_node
)
3913 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3916 * As the map is walked, we track how much memory is usable
3917 * by the kernel using kernelcore_remaining. When it is
3918 * 0, the rest of the node is usable by ZONE_MOVABLE
3920 kernelcore_remaining
= kernelcore_node
;
3922 /* Go through each range of PFNs within this node */
3923 for_each_active_range_index_in_nid(i
, nid
) {
3924 unsigned long start_pfn
, end_pfn
;
3925 unsigned long size_pages
;
3927 start_pfn
= max(early_node_map
[i
].start_pfn
,
3928 zone_movable_pfn
[nid
]);
3929 end_pfn
= early_node_map
[i
].end_pfn
;
3930 if (start_pfn
>= end_pfn
)
3933 /* Account for what is only usable for kernelcore */
3934 if (start_pfn
< usable_startpfn
) {
3935 unsigned long kernel_pages
;
3936 kernel_pages
= min(end_pfn
, usable_startpfn
)
3939 kernelcore_remaining
-= min(kernel_pages
,
3940 kernelcore_remaining
);
3941 required_kernelcore
-= min(kernel_pages
,
3942 required_kernelcore
);
3944 /* Continue if range is now fully accounted */
3945 if (end_pfn
<= usable_startpfn
) {
3948 * Push zone_movable_pfn to the end so
3949 * that if we have to rebalance
3950 * kernelcore across nodes, we will
3951 * not double account here
3953 zone_movable_pfn
[nid
] = end_pfn
;
3956 start_pfn
= usable_startpfn
;
3960 * The usable PFN range for ZONE_MOVABLE is from
3961 * start_pfn->end_pfn. Calculate size_pages as the
3962 * number of pages used as kernelcore
3964 size_pages
= end_pfn
- start_pfn
;
3965 if (size_pages
> kernelcore_remaining
)
3966 size_pages
= kernelcore_remaining
;
3967 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3970 * Some kernelcore has been met, update counts and
3971 * break if the kernelcore for this node has been
3974 required_kernelcore
-= min(required_kernelcore
,
3976 kernelcore_remaining
-= size_pages
;
3977 if (!kernelcore_remaining
)
3983 * If there is still required_kernelcore, we do another pass with one
3984 * less node in the count. This will push zone_movable_pfn[nid] further
3985 * along on the nodes that still have memory until kernelcore is
3989 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3992 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3993 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3994 zone_movable_pfn
[nid
] =
3995 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3998 /* Any regular memory on that node ? */
3999 static void check_for_regular_memory(pg_data_t
*pgdat
)
4001 #ifdef CONFIG_HIGHMEM
4002 enum zone_type zone_type
;
4004 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4005 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4006 if (zone
->present_pages
)
4007 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4013 * free_area_init_nodes - Initialise all pg_data_t and zone data
4014 * @max_zone_pfn: an array of max PFNs for each zone
4016 * This will call free_area_init_node() for each active node in the system.
4017 * Using the page ranges provided by add_active_range(), the size of each
4018 * zone in each node and their holes is calculated. If the maximum PFN
4019 * between two adjacent zones match, it is assumed that the zone is empty.
4020 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4021 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4022 * starts where the previous one ended. For example, ZONE_DMA32 starts
4023 * at arch_max_dma_pfn.
4025 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4030 /* Sort early_node_map as initialisation assumes it is sorted */
4033 /* Record where the zone boundaries are */
4034 memset(arch_zone_lowest_possible_pfn
, 0,
4035 sizeof(arch_zone_lowest_possible_pfn
));
4036 memset(arch_zone_highest_possible_pfn
, 0,
4037 sizeof(arch_zone_highest_possible_pfn
));
4038 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4039 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4040 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4041 if (i
== ZONE_MOVABLE
)
4043 arch_zone_lowest_possible_pfn
[i
] =
4044 arch_zone_highest_possible_pfn
[i
-1];
4045 arch_zone_highest_possible_pfn
[i
] =
4046 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4048 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4049 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4051 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4052 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4053 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4055 /* Print out the zone ranges */
4056 printk("Zone PFN ranges:\n");
4057 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4058 if (i
== ZONE_MOVABLE
)
4060 printk(" %-8s %0#10lx -> %0#10lx\n",
4062 arch_zone_lowest_possible_pfn
[i
],
4063 arch_zone_highest_possible_pfn
[i
]);
4066 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4067 printk("Movable zone start PFN for each node\n");
4068 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4069 if (zone_movable_pfn
[i
])
4070 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4073 /* Print out the early_node_map[] */
4074 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4075 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4076 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4077 early_node_map
[i
].start_pfn
,
4078 early_node_map
[i
].end_pfn
);
4080 /* Initialise every node */
4081 mminit_verify_pageflags_layout();
4082 setup_nr_node_ids();
4083 for_each_online_node(nid
) {
4084 pg_data_t
*pgdat
= NODE_DATA(nid
);
4085 free_area_init_node(nid
, NULL
,
4086 find_min_pfn_for_node(nid
), NULL
);
4088 /* Any memory on that node */
4089 if (pgdat
->node_present_pages
)
4090 node_set_state(nid
, N_HIGH_MEMORY
);
4091 check_for_regular_memory(pgdat
);
4095 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4097 unsigned long long coremem
;
4101 coremem
= memparse(p
, &p
);
4102 *core
= coremem
>> PAGE_SHIFT
;
4104 /* Paranoid check that UL is enough for the coremem value */
4105 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4111 * kernelcore=size sets the amount of memory for use for allocations that
4112 * cannot be reclaimed or migrated.
4114 static int __init
cmdline_parse_kernelcore(char *p
)
4116 return cmdline_parse_core(p
, &required_kernelcore
);
4120 * movablecore=size sets the amount of memory for use for allocations that
4121 * can be reclaimed or migrated.
4123 static int __init
cmdline_parse_movablecore(char *p
)
4125 return cmdline_parse_core(p
, &required_movablecore
);
4128 early_param("kernelcore", cmdline_parse_kernelcore
);
4129 early_param("movablecore", cmdline_parse_movablecore
);
4131 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4134 * set_dma_reserve - set the specified number of pages reserved in the first zone
4135 * @new_dma_reserve: The number of pages to mark reserved
4137 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4138 * In the DMA zone, a significant percentage may be consumed by kernel image
4139 * and other unfreeable allocations which can skew the watermarks badly. This
4140 * function may optionally be used to account for unfreeable pages in the
4141 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4142 * smaller per-cpu batchsize.
4144 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4146 dma_reserve
= new_dma_reserve
;
4149 #ifndef CONFIG_NEED_MULTIPLE_NODES
4150 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4151 EXPORT_SYMBOL(contig_page_data
);
4154 void __init
free_area_init(unsigned long *zones_size
)
4156 free_area_init_node(0, zones_size
,
4157 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4160 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4161 unsigned long action
, void *hcpu
)
4163 int cpu
= (unsigned long)hcpu
;
4165 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4169 * Spill the event counters of the dead processor
4170 * into the current processors event counters.
4171 * This artificially elevates the count of the current
4174 vm_events_fold_cpu(cpu
);
4177 * Zero the differential counters of the dead processor
4178 * so that the vm statistics are consistent.
4180 * This is only okay since the processor is dead and cannot
4181 * race with what we are doing.
4183 refresh_cpu_vm_stats(cpu
);
4188 void __init
page_alloc_init(void)
4190 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4194 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4195 * or min_free_kbytes changes.
4197 static void calculate_totalreserve_pages(void)
4199 struct pglist_data
*pgdat
;
4200 unsigned long reserve_pages
= 0;
4201 enum zone_type i
, j
;
4203 for_each_online_pgdat(pgdat
) {
4204 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4205 struct zone
*zone
= pgdat
->node_zones
+ i
;
4206 unsigned long max
= 0;
4208 /* Find valid and maximum lowmem_reserve in the zone */
4209 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4210 if (zone
->lowmem_reserve
[j
] > max
)
4211 max
= zone
->lowmem_reserve
[j
];
4214 /* we treat pages_high as reserved pages. */
4215 max
+= zone
->pages_high
;
4217 if (max
> zone
->present_pages
)
4218 max
= zone
->present_pages
;
4219 reserve_pages
+= max
;
4222 totalreserve_pages
= reserve_pages
;
4226 * setup_per_zone_lowmem_reserve - called whenever
4227 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4228 * has a correct pages reserved value, so an adequate number of
4229 * pages are left in the zone after a successful __alloc_pages().
4231 static void setup_per_zone_lowmem_reserve(void)
4233 struct pglist_data
*pgdat
;
4234 enum zone_type j
, idx
;
4236 for_each_online_pgdat(pgdat
) {
4237 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4238 struct zone
*zone
= pgdat
->node_zones
+ j
;
4239 unsigned long present_pages
= zone
->present_pages
;
4241 zone
->lowmem_reserve
[j
] = 0;
4245 struct zone
*lower_zone
;
4249 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4250 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4252 lower_zone
= pgdat
->node_zones
+ idx
;
4253 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4254 sysctl_lowmem_reserve_ratio
[idx
];
4255 present_pages
+= lower_zone
->present_pages
;
4260 /* update totalreserve_pages */
4261 calculate_totalreserve_pages();
4265 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4267 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4268 * with respect to min_free_kbytes.
4270 void setup_per_zone_pages_min(void)
4272 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4273 unsigned long lowmem_pages
= 0;
4275 unsigned long flags
;
4277 /* Calculate total number of !ZONE_HIGHMEM pages */
4278 for_each_zone(zone
) {
4279 if (!is_highmem(zone
))
4280 lowmem_pages
+= zone
->present_pages
;
4283 for_each_zone(zone
) {
4286 spin_lock_irqsave(&zone
->lock
, flags
);
4287 tmp
= (u64
)pages_min
* zone
->present_pages
;
4288 do_div(tmp
, lowmem_pages
);
4289 if (is_highmem(zone
)) {
4291 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4292 * need highmem pages, so cap pages_min to a small
4295 * The (pages_high-pages_low) and (pages_low-pages_min)
4296 * deltas controls asynch page reclaim, and so should
4297 * not be capped for highmem.
4301 min_pages
= zone
->present_pages
/ 1024;
4302 if (min_pages
< SWAP_CLUSTER_MAX
)
4303 min_pages
= SWAP_CLUSTER_MAX
;
4304 if (min_pages
> 128)
4306 zone
->pages_min
= min_pages
;
4309 * If it's a lowmem zone, reserve a number of pages
4310 * proportionate to the zone's size.
4312 zone
->pages_min
= tmp
;
4315 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4316 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4317 setup_zone_migrate_reserve(zone
);
4318 spin_unlock_irqrestore(&zone
->lock
, flags
);
4321 /* update totalreserve_pages */
4322 calculate_totalreserve_pages();
4326 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4328 * The inactive anon list should be small enough that the VM never has to
4329 * do too much work, but large enough that each inactive page has a chance
4330 * to be referenced again before it is swapped out.
4332 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4333 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4334 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4335 * the anonymous pages are kept on the inactive list.
4338 * memory ratio inactive anon
4339 * -------------------------------------
4348 void setup_per_zone_inactive_ratio(void)
4352 for_each_zone(zone
) {
4353 unsigned int gb
, ratio
;
4355 /* Zone size in gigabytes */
4356 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4357 ratio
= int_sqrt(10 * gb
);
4361 zone
->inactive_ratio
= ratio
;
4366 * Initialise min_free_kbytes.
4368 * For small machines we want it small (128k min). For large machines
4369 * we want it large (64MB max). But it is not linear, because network
4370 * bandwidth does not increase linearly with machine size. We use
4372 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4373 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4389 static int __init
init_per_zone_pages_min(void)
4391 unsigned long lowmem_kbytes
;
4393 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4395 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4396 if (min_free_kbytes
< 128)
4397 min_free_kbytes
= 128;
4398 if (min_free_kbytes
> 65536)
4399 min_free_kbytes
= 65536;
4400 setup_per_zone_pages_min();
4401 setup_per_zone_lowmem_reserve();
4402 setup_per_zone_inactive_ratio();
4405 module_init(init_per_zone_pages_min
)
4408 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4409 * that we can call two helper functions whenever min_free_kbytes
4412 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4413 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4415 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4417 setup_per_zone_pages_min();
4422 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4423 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4428 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4433 zone
->min_unmapped_pages
= (zone
->present_pages
*
4434 sysctl_min_unmapped_ratio
) / 100;
4438 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4439 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4444 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4449 zone
->min_slab_pages
= (zone
->present_pages
*
4450 sysctl_min_slab_ratio
) / 100;
4456 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4457 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4458 * whenever sysctl_lowmem_reserve_ratio changes.
4460 * The reserve ratio obviously has absolutely no relation with the
4461 * pages_min watermarks. The lowmem reserve ratio can only make sense
4462 * if in function of the boot time zone sizes.
4464 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4465 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4467 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4468 setup_per_zone_lowmem_reserve();
4473 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4474 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4475 * can have before it gets flushed back to buddy allocator.
4478 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4479 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4485 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4486 if (!write
|| (ret
== -EINVAL
))
4488 for_each_zone(zone
) {
4489 for_each_online_cpu(cpu
) {
4491 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4492 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4498 int hashdist
= HASHDIST_DEFAULT
;
4501 static int __init
set_hashdist(char *str
)
4505 hashdist
= simple_strtoul(str
, &str
, 0);
4508 __setup("hashdist=", set_hashdist
);
4512 * allocate a large system hash table from bootmem
4513 * - it is assumed that the hash table must contain an exact power-of-2
4514 * quantity of entries
4515 * - limit is the number of hash buckets, not the total allocation size
4517 void *__init
alloc_large_system_hash(const char *tablename
,
4518 unsigned long bucketsize
,
4519 unsigned long numentries
,
4522 unsigned int *_hash_shift
,
4523 unsigned int *_hash_mask
,
4524 unsigned long limit
)
4526 unsigned long long max
= limit
;
4527 unsigned long log2qty
, size
;
4530 /* allow the kernel cmdline to have a say */
4532 /* round applicable memory size up to nearest megabyte */
4533 numentries
= nr_kernel_pages
;
4534 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4535 numentries
>>= 20 - PAGE_SHIFT
;
4536 numentries
<<= 20 - PAGE_SHIFT
;
4538 /* limit to 1 bucket per 2^scale bytes of low memory */
4539 if (scale
> PAGE_SHIFT
)
4540 numentries
>>= (scale
- PAGE_SHIFT
);
4542 numentries
<<= (PAGE_SHIFT
- scale
);
4544 /* Make sure we've got at least a 0-order allocation.. */
4545 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4546 numentries
= PAGE_SIZE
/ bucketsize
;
4548 numentries
= roundup_pow_of_two(numentries
);
4550 /* limit allocation size to 1/16 total memory by default */
4552 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4553 do_div(max
, bucketsize
);
4556 if (numentries
> max
)
4559 log2qty
= ilog2(numentries
);
4562 size
= bucketsize
<< log2qty
;
4563 if (flags
& HASH_EARLY
)
4564 table
= alloc_bootmem_nopanic(size
);
4566 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4568 unsigned long order
= get_order(size
);
4569 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4571 * If bucketsize is not a power-of-two, we may free
4572 * some pages at the end of hash table.
4575 unsigned long alloc_end
= (unsigned long)table
+
4576 (PAGE_SIZE
<< order
);
4577 unsigned long used
= (unsigned long)table
+
4579 split_page(virt_to_page(table
), order
);
4580 while (used
< alloc_end
) {
4586 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4589 panic("Failed to allocate %s hash table\n", tablename
);
4591 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4594 ilog2(size
) - PAGE_SHIFT
,
4598 *_hash_shift
= log2qty
;
4600 *_hash_mask
= (1 << log2qty
) - 1;
4605 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4606 struct page
*pfn_to_page(unsigned long pfn
)
4608 return __pfn_to_page(pfn
);
4610 unsigned long page_to_pfn(struct page
*page
)
4612 return __page_to_pfn(page
);
4614 EXPORT_SYMBOL(pfn_to_page
);
4615 EXPORT_SYMBOL(page_to_pfn
);
4616 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4618 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4619 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4622 #ifdef CONFIG_SPARSEMEM
4623 return __pfn_to_section(pfn
)->pageblock_flags
;
4625 return zone
->pageblock_flags
;
4626 #endif /* CONFIG_SPARSEMEM */
4629 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4631 #ifdef CONFIG_SPARSEMEM
4632 pfn
&= (PAGES_PER_SECTION
-1);
4633 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4635 pfn
= pfn
- zone
->zone_start_pfn
;
4636 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4637 #endif /* CONFIG_SPARSEMEM */
4641 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4642 * @page: The page within the block of interest
4643 * @start_bitidx: The first bit of interest to retrieve
4644 * @end_bitidx: The last bit of interest
4645 * returns pageblock_bits flags
4647 unsigned long get_pageblock_flags_group(struct page
*page
,
4648 int start_bitidx
, int end_bitidx
)
4651 unsigned long *bitmap
;
4652 unsigned long pfn
, bitidx
;
4653 unsigned long flags
= 0;
4654 unsigned long value
= 1;
4656 zone
= page_zone(page
);
4657 pfn
= page_to_pfn(page
);
4658 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4659 bitidx
= pfn_to_bitidx(zone
, pfn
);
4661 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4662 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4669 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4670 * @page: The page within the block of interest
4671 * @start_bitidx: The first bit of interest
4672 * @end_bitidx: The last bit of interest
4673 * @flags: The flags to set
4675 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4676 int start_bitidx
, int end_bitidx
)
4679 unsigned long *bitmap
;
4680 unsigned long pfn
, bitidx
;
4681 unsigned long value
= 1;
4683 zone
= page_zone(page
);
4684 pfn
= page_to_pfn(page
);
4685 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4686 bitidx
= pfn_to_bitidx(zone
, pfn
);
4687 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4688 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4690 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4692 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4694 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4698 * This is designed as sub function...plz see page_isolation.c also.
4699 * set/clear page block's type to be ISOLATE.
4700 * page allocater never alloc memory from ISOLATE block.
4703 int set_migratetype_isolate(struct page
*page
)
4706 unsigned long flags
;
4709 zone
= page_zone(page
);
4710 spin_lock_irqsave(&zone
->lock
, flags
);
4712 * In future, more migrate types will be able to be isolation target.
4714 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4716 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4717 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4720 spin_unlock_irqrestore(&zone
->lock
, flags
);
4726 void unset_migratetype_isolate(struct page
*page
)
4729 unsigned long flags
;
4730 zone
= page_zone(page
);
4731 spin_lock_irqsave(&zone
->lock
, flags
);
4732 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4734 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4735 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4737 spin_unlock_irqrestore(&zone
->lock
, flags
);
4740 #ifdef CONFIG_MEMORY_HOTREMOVE
4742 * All pages in the range must be isolated before calling this.
4745 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4751 unsigned long flags
;
4752 /* find the first valid pfn */
4753 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4758 zone
= page_zone(pfn_to_page(pfn
));
4759 spin_lock_irqsave(&zone
->lock
, flags
);
4761 while (pfn
< end_pfn
) {
4762 if (!pfn_valid(pfn
)) {
4766 page
= pfn_to_page(pfn
);
4767 BUG_ON(page_count(page
));
4768 BUG_ON(!PageBuddy(page
));
4769 order
= page_order(page
);
4770 #ifdef CONFIG_DEBUG_VM
4771 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4772 pfn
, 1 << order
, end_pfn
);
4774 list_del(&page
->lru
);
4775 rmv_page_order(page
);
4776 zone
->free_area
[order
].nr_free
--;
4777 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4779 for (i
= 0; i
< (1 << order
); i
++)
4780 SetPageReserved((page
+i
));
4781 pfn
+= (1 << order
);
4783 spin_unlock_irqrestore(&zone
->lock
, flags
);