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/memcontrol.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 unsigned long __initdata required_kernelcore
;
157 static unsigned long __initdata required_movablecore
;
158 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 void *pc
= page_get_page_cgroup(page
);
228 printk(KERN_EMERG
"Bad page state in process '%s'\n" KERN_EMERG
229 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
230 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
231 (unsigned long)page
->flags
, page
->mapping
,
232 page_mapcount(page
), page_count(page
));
234 printk(KERN_EMERG
"cgroup:%p\n", pc
);
235 page_reset_bad_cgroup(page
);
237 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
238 KERN_EMERG
"Backtrace:\n");
240 page
->flags
&= ~PAGE_FLAGS_CLEAR_WHEN_BAD
;
241 set_page_count(page
, 0);
242 reset_page_mapcount(page
);
243 page
->mapping
= NULL
;
244 add_taint(TAINT_BAD_PAGE
);
248 * Higher-order pages are called "compound pages". They are structured thusly:
250 * The first PAGE_SIZE page is called the "head page".
252 * The remaining PAGE_SIZE pages are called "tail pages".
254 * All pages have PG_compound set. All pages have their ->private pointing at
255 * the head page (even the head page has this).
257 * The first tail page's ->lru.next holds the address of the compound page's
258 * put_page() function. Its ->lru.prev holds the order of allocation.
259 * This usage means that zero-order pages may not be compound.
262 static void free_compound_page(struct page
*page
)
264 __free_pages_ok(page
, compound_order(page
));
267 static void prep_compound_page(struct page
*page
, unsigned long order
)
270 int nr_pages
= 1 << order
;
272 set_compound_page_dtor(page
, free_compound_page
);
273 set_compound_order(page
, order
);
275 for (i
= 1; i
< nr_pages
; i
++) {
276 struct page
*p
= page
+ i
;
279 p
->first_page
= page
;
283 static void destroy_compound_page(struct page
*page
, unsigned long order
)
286 int nr_pages
= 1 << order
;
288 if (unlikely(compound_order(page
) != order
))
291 if (unlikely(!PageHead(page
)))
293 __ClearPageHead(page
);
294 for (i
= 1; i
< nr_pages
; i
++) {
295 struct page
*p
= page
+ i
;
297 if (unlikely(!PageTail(p
) |
298 (p
->first_page
!= page
)))
304 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
309 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
310 * and __GFP_HIGHMEM from hard or soft interrupt context.
312 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
313 for (i
= 0; i
< (1 << order
); i
++)
314 clear_highpage(page
+ i
);
317 static inline void set_page_order(struct page
*page
, int order
)
319 set_page_private(page
, order
);
320 __SetPageBuddy(page
);
323 static inline void rmv_page_order(struct page
*page
)
325 __ClearPageBuddy(page
);
326 set_page_private(page
, 0);
330 * Locate the struct page for both the matching buddy in our
331 * pair (buddy1) and the combined O(n+1) page they form (page).
333 * 1) Any buddy B1 will have an order O twin B2 which satisfies
334 * the following equation:
336 * For example, if the starting buddy (buddy2) is #8 its order
338 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
340 * 2) Any buddy B will have an order O+1 parent P which
341 * satisfies the following equation:
344 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
346 static inline struct page
*
347 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
349 unsigned long buddy_idx
= page_idx
^ (1 << order
);
351 return page
+ (buddy_idx
- page_idx
);
354 static inline unsigned long
355 __find_combined_index(unsigned long page_idx
, unsigned int order
)
357 return (page_idx
& ~(1 << order
));
361 * This function checks whether a page is free && is the buddy
362 * we can do coalesce a page and its buddy if
363 * (a) the buddy is not in a hole &&
364 * (b) the buddy is in the buddy system &&
365 * (c) a page and its buddy have the same order &&
366 * (d) a page and its buddy are in the same zone.
368 * For recording whether a page is in the buddy system, we use PG_buddy.
369 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
371 * For recording page's order, we use page_private(page).
373 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
376 if (!pfn_valid_within(page_to_pfn(buddy
)))
379 if (page_zone_id(page
) != page_zone_id(buddy
))
382 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
383 BUG_ON(page_count(buddy
) != 0);
390 * Freeing function for a buddy system allocator.
392 * The concept of a buddy system is to maintain direct-mapped table
393 * (containing bit values) for memory blocks of various "orders".
394 * The bottom level table contains the map for the smallest allocatable
395 * units of memory (here, pages), and each level above it describes
396 * pairs of units from the levels below, hence, "buddies".
397 * At a high level, all that happens here is marking the table entry
398 * at the bottom level available, and propagating the changes upward
399 * as necessary, plus some accounting needed to play nicely with other
400 * parts of the VM system.
401 * At each level, we keep a list of pages, which are heads of continuous
402 * free pages of length of (1 << order) and marked with PG_buddy. Page's
403 * order is recorded in page_private(page) field.
404 * So when we are allocating or freeing one, we can derive the state of the
405 * other. That is, if we allocate a small block, and both were
406 * free, the remainder of the region must be split into blocks.
407 * If a block is freed, and its buddy is also free, then this
408 * triggers coalescing into a block of larger size.
413 static inline void __free_one_page(struct page
*page
,
414 struct zone
*zone
, unsigned int order
)
416 unsigned long page_idx
;
417 int order_size
= 1 << order
;
418 int migratetype
= get_pageblock_migratetype(page
);
420 if (unlikely(PageCompound(page
)))
421 destroy_compound_page(page
, order
);
423 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
425 VM_BUG_ON(page_idx
& (order_size
- 1));
426 VM_BUG_ON(bad_range(zone
, page
));
428 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
429 while (order
< MAX_ORDER
-1) {
430 unsigned long combined_idx
;
433 buddy
= __page_find_buddy(page
, page_idx
, order
);
434 if (!page_is_buddy(page
, buddy
, order
))
435 break; /* Move the buddy up one level. */
437 list_del(&buddy
->lru
);
438 zone
->free_area
[order
].nr_free
--;
439 rmv_page_order(buddy
);
440 combined_idx
= __find_combined_index(page_idx
, order
);
441 page
= page
+ (combined_idx
- page_idx
);
442 page_idx
= combined_idx
;
445 set_page_order(page
, order
);
447 &zone
->free_area
[order
].free_list
[migratetype
]);
448 zone
->free_area
[order
].nr_free
++;
451 static inline int free_pages_check(struct page
*page
)
453 if (unlikely(page_mapcount(page
) |
454 (page
->mapping
!= NULL
) |
455 (page_get_page_cgroup(page
) != NULL
) |
456 (page_count(page
) != 0) |
457 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)))
460 __ClearPageDirty(page
);
462 * For now, we report if PG_reserved was found set, but do not
463 * clear it, and do not free the page. But we shall soon need
464 * to do more, for when the ZERO_PAGE count wraps negative.
466 return PageReserved(page
);
470 * Frees a list of pages.
471 * Assumes all pages on list are in same zone, and of same order.
472 * count is the number of pages to free.
474 * If the zone was previously in an "all pages pinned" state then look to
475 * see if this freeing clears that state.
477 * And clear the zone's pages_scanned counter, to hold off the "all pages are
478 * pinned" detection logic.
480 static void free_pages_bulk(struct zone
*zone
, int count
,
481 struct list_head
*list
, int order
)
483 spin_lock(&zone
->lock
);
484 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
485 zone
->pages_scanned
= 0;
489 VM_BUG_ON(list_empty(list
));
490 page
= list_entry(list
->prev
, struct page
, lru
);
491 /* have to delete it as __free_one_page list manipulates */
492 list_del(&page
->lru
);
493 __free_one_page(page
, zone
, order
);
495 spin_unlock(&zone
->lock
);
498 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
500 spin_lock(&zone
->lock
);
501 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
502 zone
->pages_scanned
= 0;
503 __free_one_page(page
, zone
, order
);
504 spin_unlock(&zone
->lock
);
507 static void __free_pages_ok(struct page
*page
, unsigned int order
)
513 for (i
= 0 ; i
< (1 << order
) ; ++i
)
514 reserved
+= free_pages_check(page
+ i
);
518 if (!PageHighMem(page
)) {
519 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
520 debug_check_no_obj_freed(page_address(page
),
523 arch_free_page(page
, order
);
524 kernel_map_pages(page
, 1 << order
, 0);
526 local_irq_save(flags
);
527 __count_vm_events(PGFREE
, 1 << order
);
528 free_one_page(page_zone(page
), page
, order
);
529 local_irq_restore(flags
);
533 * permit the bootmem allocator to evade page validation on high-order frees
535 void __free_pages_bootmem(struct page
*page
, unsigned int order
)
538 __ClearPageReserved(page
);
539 set_page_count(page
, 0);
540 set_page_refcounted(page
);
546 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
547 struct page
*p
= &page
[loop
];
549 if (loop
+ 1 < BITS_PER_LONG
)
551 __ClearPageReserved(p
);
552 set_page_count(p
, 0);
555 set_page_refcounted(page
);
556 __free_pages(page
, order
);
562 * The order of subdivision here is critical for the IO subsystem.
563 * Please do not alter this order without good reasons and regression
564 * testing. Specifically, as large blocks of memory are subdivided,
565 * the order in which smaller blocks are delivered depends on the order
566 * they're subdivided in this function. This is the primary factor
567 * influencing the order in which pages are delivered to the IO
568 * subsystem according to empirical testing, and this is also justified
569 * by considering the behavior of a buddy system containing a single
570 * large block of memory acted on by a series of small allocations.
571 * This behavior is a critical factor in sglist merging's success.
575 static inline void expand(struct zone
*zone
, struct page
*page
,
576 int low
, int high
, struct free_area
*area
,
579 unsigned long size
= 1 << high
;
585 VM_BUG_ON(bad_range(zone
, &page
[size
]));
586 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
588 set_page_order(&page
[size
], high
);
593 * This page is about to be returned from the page allocator
595 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
597 if (unlikely(page_mapcount(page
) |
598 (page
->mapping
!= NULL
) |
599 (page_get_page_cgroup(page
) != NULL
) |
600 (page_count(page
) != 0) |
601 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)))
605 * For now, we report if PG_reserved was found set, but do not
606 * clear it, and do not allocate the page: as a safety net.
608 if (PageReserved(page
))
611 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_reclaim
|
612 1 << PG_referenced
| 1 << PG_arch_1
|
613 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
);
614 set_page_private(page
, 0);
615 set_page_refcounted(page
);
617 arch_alloc_page(page
, order
);
618 kernel_map_pages(page
, 1 << order
, 1);
620 if (gfp_flags
& __GFP_ZERO
)
621 prep_zero_page(page
, order
, gfp_flags
);
623 if (order
&& (gfp_flags
& __GFP_COMP
))
624 prep_compound_page(page
, order
);
630 * Go through the free lists for the given migratetype and remove
631 * the smallest available page from the freelists
633 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
636 unsigned int current_order
;
637 struct free_area
* area
;
640 /* Find a page of the appropriate size in the preferred list */
641 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
642 area
= &(zone
->free_area
[current_order
]);
643 if (list_empty(&area
->free_list
[migratetype
]))
646 page
= list_entry(area
->free_list
[migratetype
].next
,
648 list_del(&page
->lru
);
649 rmv_page_order(page
);
651 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
652 expand(zone
, page
, order
, current_order
, area
, migratetype
);
661 * This array describes the order lists are fallen back to when
662 * the free lists for the desirable migrate type are depleted
664 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
665 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
666 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
667 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
668 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
672 * Move the free pages in a range to the free lists of the requested type.
673 * Note that start_page and end_pages are not aligned on a pageblock
674 * boundary. If alignment is required, use move_freepages_block()
676 int move_freepages(struct zone
*zone
,
677 struct page
*start_page
, struct page
*end_page
,
684 #ifndef CONFIG_HOLES_IN_ZONE
686 * page_zone is not safe to call in this context when
687 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
688 * anyway as we check zone boundaries in move_freepages_block().
689 * Remove at a later date when no bug reports exist related to
690 * grouping pages by mobility
692 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
695 for (page
= start_page
; page
<= end_page
;) {
696 /* Make sure we are not inadvertently changing nodes */
697 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
699 if (!pfn_valid_within(page_to_pfn(page
))) {
704 if (!PageBuddy(page
)) {
709 order
= page_order(page
);
710 list_del(&page
->lru
);
712 &zone
->free_area
[order
].free_list
[migratetype
]);
714 pages_moved
+= 1 << order
;
720 int move_freepages_block(struct zone
*zone
, struct page
*page
, int migratetype
)
722 unsigned long start_pfn
, end_pfn
;
723 struct page
*start_page
, *end_page
;
725 start_pfn
= page_to_pfn(page
);
726 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
727 start_page
= pfn_to_page(start_pfn
);
728 end_page
= start_page
+ pageblock_nr_pages
- 1;
729 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
731 /* Do not cross zone boundaries */
732 if (start_pfn
< zone
->zone_start_pfn
)
734 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
737 return move_freepages(zone
, start_page
, end_page
, migratetype
);
740 /* Remove an element from the buddy allocator from the fallback list */
741 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
742 int start_migratetype
)
744 struct free_area
* area
;
749 /* Find the largest possible block of pages in the other list */
750 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
752 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
753 migratetype
= fallbacks
[start_migratetype
][i
];
755 /* MIGRATE_RESERVE handled later if necessary */
756 if (migratetype
== MIGRATE_RESERVE
)
759 area
= &(zone
->free_area
[current_order
]);
760 if (list_empty(&area
->free_list
[migratetype
]))
763 page
= list_entry(area
->free_list
[migratetype
].next
,
768 * If breaking a large block of pages, move all free
769 * pages to the preferred allocation list. If falling
770 * back for a reclaimable kernel allocation, be more
771 * agressive about taking ownership of free pages
773 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
774 start_migratetype
== MIGRATE_RECLAIMABLE
) {
776 pages
= move_freepages_block(zone
, page
,
779 /* Claim the whole block if over half of it is free */
780 if (pages
>= (1 << (pageblock_order
-1)))
781 set_pageblock_migratetype(page
,
784 migratetype
= start_migratetype
;
787 /* Remove the page from the freelists */
788 list_del(&page
->lru
);
789 rmv_page_order(page
);
790 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
793 if (current_order
== pageblock_order
)
794 set_pageblock_migratetype(page
,
797 expand(zone
, page
, order
, current_order
, area
, migratetype
);
802 /* Use MIGRATE_RESERVE rather than fail an allocation */
803 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
807 * Do the hard work of removing an element from the buddy allocator.
808 * Call me with the zone->lock already held.
810 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
815 page
= __rmqueue_smallest(zone
, order
, migratetype
);
818 page
= __rmqueue_fallback(zone
, order
, migratetype
);
824 * Obtain a specified number of elements from the buddy allocator, all under
825 * a single hold of the lock, for efficiency. Add them to the supplied list.
826 * Returns the number of new pages which were placed at *list.
828 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
829 unsigned long count
, struct list_head
*list
,
834 spin_lock(&zone
->lock
);
835 for (i
= 0; i
< count
; ++i
) {
836 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
837 if (unlikely(page
== NULL
))
841 * Split buddy pages returned by expand() are received here
842 * in physical page order. The page is added to the callers and
843 * list and the list head then moves forward. From the callers
844 * perspective, the linked list is ordered by page number in
845 * some conditions. This is useful for IO devices that can
846 * merge IO requests if the physical pages are ordered
849 list_add(&page
->lru
, list
);
850 set_page_private(page
, migratetype
);
853 spin_unlock(&zone
->lock
);
859 * Called from the vmstat counter updater to drain pagesets of this
860 * currently executing processor on remote nodes after they have
863 * Note that this function must be called with the thread pinned to
864 * a single processor.
866 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
871 local_irq_save(flags
);
872 if (pcp
->count
>= pcp
->batch
)
873 to_drain
= pcp
->batch
;
875 to_drain
= pcp
->count
;
876 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
877 pcp
->count
-= to_drain
;
878 local_irq_restore(flags
);
883 * Drain pages of the indicated processor.
885 * The processor must either be the current processor and the
886 * thread pinned to the current processor or a processor that
889 static void drain_pages(unsigned int cpu
)
894 for_each_zone(zone
) {
895 struct per_cpu_pageset
*pset
;
896 struct per_cpu_pages
*pcp
;
898 if (!populated_zone(zone
))
901 pset
= zone_pcp(zone
, cpu
);
904 local_irq_save(flags
);
905 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
907 local_irq_restore(flags
);
912 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
914 void drain_local_pages(void *arg
)
916 drain_pages(smp_processor_id());
920 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
922 void drain_all_pages(void)
924 on_each_cpu(drain_local_pages
, NULL
, 0, 1);
927 #ifdef CONFIG_HIBERNATION
929 void mark_free_pages(struct zone
*zone
)
931 unsigned long pfn
, max_zone_pfn
;
934 struct list_head
*curr
;
936 if (!zone
->spanned_pages
)
939 spin_lock_irqsave(&zone
->lock
, flags
);
941 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
942 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
943 if (pfn_valid(pfn
)) {
944 struct page
*page
= pfn_to_page(pfn
);
946 if (!swsusp_page_is_forbidden(page
))
947 swsusp_unset_page_free(page
);
950 for_each_migratetype_order(order
, t
) {
951 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
954 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
955 for (i
= 0; i
< (1UL << order
); i
++)
956 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
959 spin_unlock_irqrestore(&zone
->lock
, flags
);
961 #endif /* CONFIG_PM */
964 * Free a 0-order page
966 static void free_hot_cold_page(struct page
*page
, int cold
)
968 struct zone
*zone
= page_zone(page
);
969 struct per_cpu_pages
*pcp
;
973 page
->mapping
= NULL
;
974 if (free_pages_check(page
))
977 if (!PageHighMem(page
)) {
978 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
979 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
981 arch_free_page(page
, 0);
982 kernel_map_pages(page
, 1, 0);
984 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
985 local_irq_save(flags
);
986 __count_vm_event(PGFREE
);
988 list_add_tail(&page
->lru
, &pcp
->list
);
990 list_add(&page
->lru
, &pcp
->list
);
991 set_page_private(page
, get_pageblock_migratetype(page
));
993 if (pcp
->count
>= pcp
->high
) {
994 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
995 pcp
->count
-= pcp
->batch
;
997 local_irq_restore(flags
);
1001 void free_hot_page(struct page
*page
)
1003 free_hot_cold_page(page
, 0);
1006 void free_cold_page(struct page
*page
)
1008 free_hot_cold_page(page
, 1);
1012 * split_page takes a non-compound higher-order page, and splits it into
1013 * n (1<<order) sub-pages: page[0..n]
1014 * Each sub-page must be freed individually.
1016 * Note: this is probably too low level an operation for use in drivers.
1017 * Please consult with lkml before using this in your driver.
1019 void split_page(struct page
*page
, unsigned int order
)
1023 VM_BUG_ON(PageCompound(page
));
1024 VM_BUG_ON(!page_count(page
));
1025 for (i
= 1; i
< (1 << order
); i
++)
1026 set_page_refcounted(page
+ i
);
1030 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1031 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1034 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1035 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1037 unsigned long flags
;
1039 int cold
= !!(gfp_flags
& __GFP_COLD
);
1041 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1045 if (likely(order
== 0)) {
1046 struct per_cpu_pages
*pcp
;
1048 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1049 local_irq_save(flags
);
1051 pcp
->count
= rmqueue_bulk(zone
, 0,
1052 pcp
->batch
, &pcp
->list
, migratetype
);
1053 if (unlikely(!pcp
->count
))
1057 /* Find a page of the appropriate migrate type */
1059 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1060 if (page_private(page
) == migratetype
)
1063 list_for_each_entry(page
, &pcp
->list
, lru
)
1064 if (page_private(page
) == migratetype
)
1068 /* Allocate more to the pcp list if necessary */
1069 if (unlikely(&page
->lru
== &pcp
->list
)) {
1070 pcp
->count
+= rmqueue_bulk(zone
, 0,
1071 pcp
->batch
, &pcp
->list
, migratetype
);
1072 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1075 list_del(&page
->lru
);
1078 spin_lock_irqsave(&zone
->lock
, flags
);
1079 page
= __rmqueue(zone
, order
, migratetype
);
1080 spin_unlock(&zone
->lock
);
1085 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1086 zone_statistics(preferred_zone
, zone
);
1087 local_irq_restore(flags
);
1090 VM_BUG_ON(bad_range(zone
, page
));
1091 if (prep_new_page(page
, order
, gfp_flags
))
1096 local_irq_restore(flags
);
1101 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1102 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1103 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1104 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1105 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1106 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1107 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1109 #ifdef CONFIG_FAIL_PAGE_ALLOC
1111 static struct fail_page_alloc_attr
{
1112 struct fault_attr attr
;
1114 u32 ignore_gfp_highmem
;
1115 u32 ignore_gfp_wait
;
1118 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1120 struct dentry
*ignore_gfp_highmem_file
;
1121 struct dentry
*ignore_gfp_wait_file
;
1122 struct dentry
*min_order_file
;
1124 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1126 } fail_page_alloc
= {
1127 .attr
= FAULT_ATTR_INITIALIZER
,
1128 .ignore_gfp_wait
= 1,
1129 .ignore_gfp_highmem
= 1,
1133 static int __init
setup_fail_page_alloc(char *str
)
1135 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1137 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1139 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1141 if (order
< fail_page_alloc
.min_order
)
1143 if (gfp_mask
& __GFP_NOFAIL
)
1145 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1147 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1150 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1153 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1155 static int __init
fail_page_alloc_debugfs(void)
1157 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1161 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1165 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1167 fail_page_alloc
.ignore_gfp_wait_file
=
1168 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1169 &fail_page_alloc
.ignore_gfp_wait
);
1171 fail_page_alloc
.ignore_gfp_highmem_file
=
1172 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1173 &fail_page_alloc
.ignore_gfp_highmem
);
1174 fail_page_alloc
.min_order_file
=
1175 debugfs_create_u32("min-order", mode
, dir
,
1176 &fail_page_alloc
.min_order
);
1178 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1179 !fail_page_alloc
.ignore_gfp_highmem_file
||
1180 !fail_page_alloc
.min_order_file
) {
1182 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1183 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1184 debugfs_remove(fail_page_alloc
.min_order_file
);
1185 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1191 late_initcall(fail_page_alloc_debugfs
);
1193 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1195 #else /* CONFIG_FAIL_PAGE_ALLOC */
1197 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1202 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1205 * Return 1 if free pages are above 'mark'. This takes into account the order
1206 * of the allocation.
1208 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1209 int classzone_idx
, int alloc_flags
)
1211 /* free_pages my go negative - that's OK */
1213 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1216 if (alloc_flags
& ALLOC_HIGH
)
1218 if (alloc_flags
& ALLOC_HARDER
)
1221 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1223 for (o
= 0; o
< order
; o
++) {
1224 /* At the next order, this order's pages become unavailable */
1225 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1227 /* Require fewer higher order pages to be free */
1230 if (free_pages
<= min
)
1238 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1239 * skip over zones that are not allowed by the cpuset, or that have
1240 * been recently (in last second) found to be nearly full. See further
1241 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1242 * that have to skip over a lot of full or unallowed zones.
1244 * If the zonelist cache is present in the passed in zonelist, then
1245 * returns a pointer to the allowed node mask (either the current
1246 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1248 * If the zonelist cache is not available for this zonelist, does
1249 * nothing and returns NULL.
1251 * If the fullzones BITMAP in the zonelist cache is stale (more than
1252 * a second since last zap'd) then we zap it out (clear its bits.)
1254 * We hold off even calling zlc_setup, until after we've checked the
1255 * first zone in the zonelist, on the theory that most allocations will
1256 * be satisfied from that first zone, so best to examine that zone as
1257 * quickly as we can.
1259 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1261 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1262 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1264 zlc
= zonelist
->zlcache_ptr
;
1268 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1269 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1270 zlc
->last_full_zap
= jiffies
;
1273 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1274 &cpuset_current_mems_allowed
:
1275 &node_states
[N_HIGH_MEMORY
];
1276 return allowednodes
;
1280 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1281 * if it is worth looking at further for free memory:
1282 * 1) Check that the zone isn't thought to be full (doesn't have its
1283 * bit set in the zonelist_cache fullzones BITMAP).
1284 * 2) Check that the zones node (obtained from the zonelist_cache
1285 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1286 * Return true (non-zero) if zone is worth looking at further, or
1287 * else return false (zero) if it is not.
1289 * This check -ignores- the distinction between various watermarks,
1290 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1291 * found to be full for any variation of these watermarks, it will
1292 * be considered full for up to one second by all requests, unless
1293 * we are so low on memory on all allowed nodes that we are forced
1294 * into the second scan of the zonelist.
1296 * In the second scan we ignore this zonelist cache and exactly
1297 * apply the watermarks to all zones, even it is slower to do so.
1298 * We are low on memory in the second scan, and should leave no stone
1299 * unturned looking for a free page.
1301 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1302 nodemask_t
*allowednodes
)
1304 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1305 int i
; /* index of *z in zonelist zones */
1306 int n
; /* node that zone *z is on */
1308 zlc
= zonelist
->zlcache_ptr
;
1312 i
= z
- zonelist
->_zonerefs
;
1315 /* This zone is worth trying if it is allowed but not full */
1316 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1320 * Given 'z' scanning a zonelist, set the corresponding bit in
1321 * zlc->fullzones, so that subsequent attempts to allocate a page
1322 * from that zone don't waste time re-examining it.
1324 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1326 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1327 int i
; /* index of *z in zonelist zones */
1329 zlc
= zonelist
->zlcache_ptr
;
1333 i
= z
- zonelist
->_zonerefs
;
1335 set_bit(i
, zlc
->fullzones
);
1338 #else /* CONFIG_NUMA */
1340 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1345 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1346 nodemask_t
*allowednodes
)
1351 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1354 #endif /* CONFIG_NUMA */
1357 * get_page_from_freelist goes through the zonelist trying to allocate
1360 static struct page
*
1361 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1362 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1365 struct page
*page
= NULL
;
1367 struct zone
*zone
, *preferred_zone
;
1368 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1369 int zlc_active
= 0; /* set if using zonelist_cache */
1370 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1372 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1374 if (!preferred_zone
)
1377 classzone_idx
= zone_idx(preferred_zone
);
1381 * Scan zonelist, looking for a zone with enough free.
1382 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1384 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1385 high_zoneidx
, nodemask
) {
1386 if (NUMA_BUILD
&& zlc_active
&&
1387 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1389 if ((alloc_flags
& ALLOC_CPUSET
) &&
1390 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1393 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1395 if (alloc_flags
& ALLOC_WMARK_MIN
)
1396 mark
= zone
->pages_min
;
1397 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1398 mark
= zone
->pages_low
;
1400 mark
= zone
->pages_high
;
1401 if (!zone_watermark_ok(zone
, order
, mark
,
1402 classzone_idx
, alloc_flags
)) {
1403 if (!zone_reclaim_mode
||
1404 !zone_reclaim(zone
, gfp_mask
, order
))
1405 goto this_zone_full
;
1409 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1414 zlc_mark_zone_full(zonelist
, z
);
1416 if (NUMA_BUILD
&& !did_zlc_setup
) {
1417 /* we do zlc_setup after the first zone is tried */
1418 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1424 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1425 /* Disable zlc cache for second zonelist scan */
1433 * This is the 'heart' of the zoned buddy allocator.
1435 static struct page
*
1436 __alloc_pages_internal(gfp_t gfp_mask
, unsigned int order
,
1437 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1439 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1440 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1444 struct reclaim_state reclaim_state
;
1445 struct task_struct
*p
= current
;
1448 unsigned long did_some_progress
;
1449 unsigned long pages_reclaimed
= 0;
1451 might_sleep_if(wait
);
1453 if (should_fail_alloc_page(gfp_mask
, order
))
1457 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1459 if (unlikely(!z
->zone
)) {
1461 * Happens if we have an empty zonelist as a result of
1462 * GFP_THISNODE being used on a memoryless node
1467 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1468 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1473 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1474 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1475 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1476 * using a larger set of nodes after it has established that the
1477 * allowed per node queues are empty and that nodes are
1480 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1483 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1484 wakeup_kswapd(zone
, order
);
1487 * OK, we're below the kswapd watermark and have kicked background
1488 * reclaim. Now things get more complex, so set up alloc_flags according
1489 * to how we want to proceed.
1491 * The caller may dip into page reserves a bit more if the caller
1492 * cannot run direct reclaim, or if the caller has realtime scheduling
1493 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1494 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1496 alloc_flags
= ALLOC_WMARK_MIN
;
1497 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1498 alloc_flags
|= ALLOC_HARDER
;
1499 if (gfp_mask
& __GFP_HIGH
)
1500 alloc_flags
|= ALLOC_HIGH
;
1502 alloc_flags
|= ALLOC_CPUSET
;
1505 * Go through the zonelist again. Let __GFP_HIGH and allocations
1506 * coming from realtime tasks go deeper into reserves.
1508 * This is the last chance, in general, before the goto nopage.
1509 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1510 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1512 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1513 high_zoneidx
, alloc_flags
);
1517 /* This allocation should allow future memory freeing. */
1520 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1521 && !in_interrupt()) {
1522 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1524 /* go through the zonelist yet again, ignoring mins */
1525 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1526 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1529 if (gfp_mask
& __GFP_NOFAIL
) {
1530 congestion_wait(WRITE
, HZ
/50);
1537 /* Atomic allocations - we can't balance anything */
1543 /* We now go into synchronous reclaim */
1544 cpuset_memory_pressure_bump();
1545 p
->flags
|= PF_MEMALLOC
;
1546 reclaim_state
.reclaimed_slab
= 0;
1547 p
->reclaim_state
= &reclaim_state
;
1549 did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
);
1551 p
->reclaim_state
= NULL
;
1552 p
->flags
&= ~PF_MEMALLOC
;
1559 if (likely(did_some_progress
)) {
1560 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1561 zonelist
, high_zoneidx
, alloc_flags
);
1564 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1565 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1566 schedule_timeout_uninterruptible(1);
1571 * Go through the zonelist yet one more time, keep
1572 * very high watermark here, this is only to catch
1573 * a parallel oom killing, we must fail if we're still
1574 * under heavy pressure.
1576 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1577 order
, zonelist
, high_zoneidx
,
1578 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1580 clear_zonelist_oom(zonelist
, gfp_mask
);
1584 /* The OOM killer will not help higher order allocs so fail */
1585 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1586 clear_zonelist_oom(zonelist
, gfp_mask
);
1590 out_of_memory(zonelist
, gfp_mask
, order
);
1591 clear_zonelist_oom(zonelist
, gfp_mask
);
1596 * Don't let big-order allocations loop unless the caller explicitly
1597 * requests that. Wait for some write requests to complete then retry.
1599 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1600 * means __GFP_NOFAIL, but that may not be true in other
1603 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1604 * specified, then we retry until we no longer reclaim any pages
1605 * (above), or we've reclaimed an order of pages at least as
1606 * large as the allocation's order. In both cases, if the
1607 * allocation still fails, we stop retrying.
1609 pages_reclaimed
+= did_some_progress
;
1611 if (!(gfp_mask
& __GFP_NORETRY
)) {
1612 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1615 if (gfp_mask
& __GFP_REPEAT
&&
1616 pages_reclaimed
< (1 << order
))
1619 if (gfp_mask
& __GFP_NOFAIL
)
1623 congestion_wait(WRITE
, HZ
/50);
1628 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1629 printk(KERN_WARNING
"%s: page allocation failure."
1630 " order:%d, mode:0x%x\n",
1631 p
->comm
, order
, gfp_mask
);
1640 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
1641 struct zonelist
*zonelist
)
1643 return __alloc_pages_internal(gfp_mask
, order
, zonelist
, NULL
);
1647 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
1648 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1650 return __alloc_pages_internal(gfp_mask
, order
, zonelist
, nodemask
);
1653 EXPORT_SYMBOL(__alloc_pages
);
1656 * Common helper functions.
1658 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1661 page
= alloc_pages(gfp_mask
, order
);
1664 return (unsigned long) page_address(page
);
1667 EXPORT_SYMBOL(__get_free_pages
);
1669 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1674 * get_zeroed_page() returns a 32-bit address, which cannot represent
1677 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1679 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1681 return (unsigned long) page_address(page
);
1685 EXPORT_SYMBOL(get_zeroed_page
);
1687 void __pagevec_free(struct pagevec
*pvec
)
1689 int i
= pagevec_count(pvec
);
1692 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1695 void __free_pages(struct page
*page
, unsigned int order
)
1697 if (put_page_testzero(page
)) {
1699 free_hot_page(page
);
1701 __free_pages_ok(page
, order
);
1705 EXPORT_SYMBOL(__free_pages
);
1707 void free_pages(unsigned long addr
, unsigned int order
)
1710 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1711 __free_pages(virt_to_page((void *)addr
), order
);
1715 EXPORT_SYMBOL(free_pages
);
1717 static unsigned int nr_free_zone_pages(int offset
)
1722 /* Just pick one node, since fallback list is circular */
1723 unsigned int sum
= 0;
1725 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1727 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1728 unsigned long size
= zone
->present_pages
;
1729 unsigned long high
= zone
->pages_high
;
1738 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1740 unsigned int nr_free_buffer_pages(void)
1742 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1744 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1747 * Amount of free RAM allocatable within all zones
1749 unsigned int nr_free_pagecache_pages(void)
1751 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1754 static inline void show_node(struct zone
*zone
)
1757 printk("Node %d ", zone_to_nid(zone
));
1760 void si_meminfo(struct sysinfo
*val
)
1762 val
->totalram
= totalram_pages
;
1764 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1765 val
->bufferram
= nr_blockdev_pages();
1766 val
->totalhigh
= totalhigh_pages
;
1767 val
->freehigh
= nr_free_highpages();
1768 val
->mem_unit
= PAGE_SIZE
;
1771 EXPORT_SYMBOL(si_meminfo
);
1774 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1776 pg_data_t
*pgdat
= NODE_DATA(nid
);
1778 val
->totalram
= pgdat
->node_present_pages
;
1779 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1780 #ifdef CONFIG_HIGHMEM
1781 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1782 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1788 val
->mem_unit
= PAGE_SIZE
;
1792 #define K(x) ((x) << (PAGE_SHIFT-10))
1795 * Show free area list (used inside shift_scroll-lock stuff)
1796 * We also calculate the percentage fragmentation. We do this by counting the
1797 * memory on each free list with the exception of the first item on the list.
1799 void show_free_areas(void)
1804 for_each_zone(zone
) {
1805 if (!populated_zone(zone
))
1809 printk("%s per-cpu:\n", zone
->name
);
1811 for_each_online_cpu(cpu
) {
1812 struct per_cpu_pageset
*pageset
;
1814 pageset
= zone_pcp(zone
, cpu
);
1816 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1817 cpu
, pageset
->pcp
.high
,
1818 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1822 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1823 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1824 global_page_state(NR_ACTIVE
),
1825 global_page_state(NR_INACTIVE
),
1826 global_page_state(NR_FILE_DIRTY
),
1827 global_page_state(NR_WRITEBACK
),
1828 global_page_state(NR_UNSTABLE_NFS
),
1829 global_page_state(NR_FREE_PAGES
),
1830 global_page_state(NR_SLAB_RECLAIMABLE
) +
1831 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1832 global_page_state(NR_FILE_MAPPED
),
1833 global_page_state(NR_PAGETABLE
),
1834 global_page_state(NR_BOUNCE
));
1836 for_each_zone(zone
) {
1839 if (!populated_zone(zone
))
1851 " pages_scanned:%lu"
1852 " all_unreclaimable? %s"
1855 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1858 K(zone
->pages_high
),
1859 K(zone_page_state(zone
, NR_ACTIVE
)),
1860 K(zone_page_state(zone
, NR_INACTIVE
)),
1861 K(zone
->present_pages
),
1862 zone
->pages_scanned
,
1863 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1865 printk("lowmem_reserve[]:");
1866 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1867 printk(" %lu", zone
->lowmem_reserve
[i
]);
1871 for_each_zone(zone
) {
1872 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1874 if (!populated_zone(zone
))
1878 printk("%s: ", zone
->name
);
1880 spin_lock_irqsave(&zone
->lock
, flags
);
1881 for (order
= 0; order
< MAX_ORDER
; order
++) {
1882 nr
[order
] = zone
->free_area
[order
].nr_free
;
1883 total
+= nr
[order
] << order
;
1885 spin_unlock_irqrestore(&zone
->lock
, flags
);
1886 for (order
= 0; order
< MAX_ORDER
; order
++)
1887 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1888 printk("= %lukB\n", K(total
));
1891 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
1893 show_swap_cache_info();
1896 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
1898 zoneref
->zone
= zone
;
1899 zoneref
->zone_idx
= zone_idx(zone
);
1903 * Builds allocation fallback zone lists.
1905 * Add all populated zones of a node to the zonelist.
1907 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1908 int nr_zones
, enum zone_type zone_type
)
1912 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1917 zone
= pgdat
->node_zones
+ zone_type
;
1918 if (populated_zone(zone
)) {
1919 zoneref_set_zone(zone
,
1920 &zonelist
->_zonerefs
[nr_zones
++]);
1921 check_highest_zone(zone_type
);
1924 } while (zone_type
);
1931 * 0 = automatic detection of better ordering.
1932 * 1 = order by ([node] distance, -zonetype)
1933 * 2 = order by (-zonetype, [node] distance)
1935 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1936 * the same zonelist. So only NUMA can configure this param.
1938 #define ZONELIST_ORDER_DEFAULT 0
1939 #define ZONELIST_ORDER_NODE 1
1940 #define ZONELIST_ORDER_ZONE 2
1942 /* zonelist order in the kernel.
1943 * set_zonelist_order() will set this to NODE or ZONE.
1945 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1946 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
1950 /* The value user specified ....changed by config */
1951 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1952 /* string for sysctl */
1953 #define NUMA_ZONELIST_ORDER_LEN 16
1954 char numa_zonelist_order
[16] = "default";
1957 * interface for configure zonelist ordering.
1958 * command line option "numa_zonelist_order"
1959 * = "[dD]efault - default, automatic configuration.
1960 * = "[nN]ode - order by node locality, then by zone within node
1961 * = "[zZ]one - order by zone, then by locality within zone
1964 static int __parse_numa_zonelist_order(char *s
)
1966 if (*s
== 'd' || *s
== 'D') {
1967 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1968 } else if (*s
== 'n' || *s
== 'N') {
1969 user_zonelist_order
= ZONELIST_ORDER_NODE
;
1970 } else if (*s
== 'z' || *s
== 'Z') {
1971 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
1974 "Ignoring invalid numa_zonelist_order value: "
1981 static __init
int setup_numa_zonelist_order(char *s
)
1984 return __parse_numa_zonelist_order(s
);
1987 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
1990 * sysctl handler for numa_zonelist_order
1992 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
1993 struct file
*file
, void __user
*buffer
, size_t *length
,
1996 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2000 strncpy(saved_string
, (char*)table
->data
,
2001 NUMA_ZONELIST_ORDER_LEN
);
2002 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2006 int oldval
= user_zonelist_order
;
2007 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2009 * bogus value. restore saved string
2011 strncpy((char*)table
->data
, saved_string
,
2012 NUMA_ZONELIST_ORDER_LEN
);
2013 user_zonelist_order
= oldval
;
2014 } else if (oldval
!= user_zonelist_order
)
2015 build_all_zonelists();
2021 #define MAX_NODE_LOAD (num_online_nodes())
2022 static int node_load
[MAX_NUMNODES
];
2025 * find_next_best_node - find the next node that should appear in a given node's fallback list
2026 * @node: node whose fallback list we're appending
2027 * @used_node_mask: nodemask_t of already used nodes
2029 * We use a number of factors to determine which is the next node that should
2030 * appear on a given node's fallback list. The node should not have appeared
2031 * already in @node's fallback list, and it should be the next closest node
2032 * according to the distance array (which contains arbitrary distance values
2033 * from each node to each node in the system), and should also prefer nodes
2034 * with no CPUs, since presumably they'll have very little allocation pressure
2035 * on them otherwise.
2036 * It returns -1 if no node is found.
2038 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2041 int min_val
= INT_MAX
;
2043 node_to_cpumask_ptr(tmp
, 0);
2045 /* Use the local node if we haven't already */
2046 if (!node_isset(node
, *used_node_mask
)) {
2047 node_set(node
, *used_node_mask
);
2051 for_each_node_state(n
, N_HIGH_MEMORY
) {
2053 /* Don't want a node to appear more than once */
2054 if (node_isset(n
, *used_node_mask
))
2057 /* Use the distance array to find the distance */
2058 val
= node_distance(node
, n
);
2060 /* Penalize nodes under us ("prefer the next node") */
2063 /* Give preference to headless and unused nodes */
2064 node_to_cpumask_ptr_next(tmp
, n
);
2065 if (!cpus_empty(*tmp
))
2066 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2068 /* Slight preference for less loaded node */
2069 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2070 val
+= node_load
[n
];
2072 if (val
< min_val
) {
2079 node_set(best_node
, *used_node_mask
);
2086 * Build zonelists ordered by node and zones within node.
2087 * This results in maximum locality--normal zone overflows into local
2088 * DMA zone, if any--but risks exhausting DMA zone.
2090 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2093 struct zonelist
*zonelist
;
2095 zonelist
= &pgdat
->node_zonelists
[0];
2096 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2098 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2100 zonelist
->_zonerefs
[j
].zone
= NULL
;
2101 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2105 * Build gfp_thisnode zonelists
2107 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2110 struct zonelist
*zonelist
;
2112 zonelist
= &pgdat
->node_zonelists
[1];
2113 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2114 zonelist
->_zonerefs
[j
].zone
= NULL
;
2115 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2119 * Build zonelists ordered by zone and nodes within zones.
2120 * This results in conserving DMA zone[s] until all Normal memory is
2121 * exhausted, but results in overflowing to remote node while memory
2122 * may still exist in local DMA zone.
2124 static int node_order
[MAX_NUMNODES
];
2126 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2129 int zone_type
; /* needs to be signed */
2131 struct zonelist
*zonelist
;
2133 zonelist
= &pgdat
->node_zonelists
[0];
2135 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2136 for (j
= 0; j
< nr_nodes
; j
++) {
2137 node
= node_order
[j
];
2138 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2139 if (populated_zone(z
)) {
2141 &zonelist
->_zonerefs
[pos
++]);
2142 check_highest_zone(zone_type
);
2146 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2147 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2150 static int default_zonelist_order(void)
2153 unsigned long low_kmem_size
,total_size
;
2157 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2158 * If they are really small and used heavily, the system can fall
2159 * into OOM very easily.
2160 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2162 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2165 for_each_online_node(nid
) {
2166 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2167 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2168 if (populated_zone(z
)) {
2169 if (zone_type
< ZONE_NORMAL
)
2170 low_kmem_size
+= z
->present_pages
;
2171 total_size
+= z
->present_pages
;
2175 if (!low_kmem_size
|| /* there are no DMA area. */
2176 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2177 return ZONELIST_ORDER_NODE
;
2179 * look into each node's config.
2180 * If there is a node whose DMA/DMA32 memory is very big area on
2181 * local memory, NODE_ORDER may be suitable.
2183 average_size
= total_size
/
2184 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2185 for_each_online_node(nid
) {
2188 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2189 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2190 if (populated_zone(z
)) {
2191 if (zone_type
< ZONE_NORMAL
)
2192 low_kmem_size
+= z
->present_pages
;
2193 total_size
+= z
->present_pages
;
2196 if (low_kmem_size
&&
2197 total_size
> average_size
&& /* ignore small node */
2198 low_kmem_size
> total_size
* 70/100)
2199 return ZONELIST_ORDER_NODE
;
2201 return ZONELIST_ORDER_ZONE
;
2204 static void set_zonelist_order(void)
2206 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2207 current_zonelist_order
= default_zonelist_order();
2209 current_zonelist_order
= user_zonelist_order
;
2212 static void build_zonelists(pg_data_t
*pgdat
)
2216 nodemask_t used_mask
;
2217 int local_node
, prev_node
;
2218 struct zonelist
*zonelist
;
2219 int order
= current_zonelist_order
;
2221 /* initialize zonelists */
2222 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2223 zonelist
= pgdat
->node_zonelists
+ i
;
2224 zonelist
->_zonerefs
[0].zone
= NULL
;
2225 zonelist
->_zonerefs
[0].zone_idx
= 0;
2228 /* NUMA-aware ordering of nodes */
2229 local_node
= pgdat
->node_id
;
2230 load
= num_online_nodes();
2231 prev_node
= local_node
;
2232 nodes_clear(used_mask
);
2234 memset(node_load
, 0, sizeof(node_load
));
2235 memset(node_order
, 0, sizeof(node_order
));
2238 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2239 int distance
= node_distance(local_node
, node
);
2242 * If another node is sufficiently far away then it is better
2243 * to reclaim pages in a zone before going off node.
2245 if (distance
> RECLAIM_DISTANCE
)
2246 zone_reclaim_mode
= 1;
2249 * We don't want to pressure a particular node.
2250 * So adding penalty to the first node in same
2251 * distance group to make it round-robin.
2253 if (distance
!= node_distance(local_node
, prev_node
))
2254 node_load
[node
] = load
;
2258 if (order
== ZONELIST_ORDER_NODE
)
2259 build_zonelists_in_node_order(pgdat
, node
);
2261 node_order
[j
++] = node
; /* remember order */
2264 if (order
== ZONELIST_ORDER_ZONE
) {
2265 /* calculate node order -- i.e., DMA last! */
2266 build_zonelists_in_zone_order(pgdat
, j
);
2269 build_thisnode_zonelists(pgdat
);
2272 /* Construct the zonelist performance cache - see further mmzone.h */
2273 static void build_zonelist_cache(pg_data_t
*pgdat
)
2275 struct zonelist
*zonelist
;
2276 struct zonelist_cache
*zlc
;
2279 zonelist
= &pgdat
->node_zonelists
[0];
2280 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2281 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2282 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2283 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2287 #else /* CONFIG_NUMA */
2289 static void set_zonelist_order(void)
2291 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2294 static void build_zonelists(pg_data_t
*pgdat
)
2296 int node
, local_node
;
2298 struct zonelist
*zonelist
;
2300 local_node
= pgdat
->node_id
;
2302 zonelist
= &pgdat
->node_zonelists
[0];
2303 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2306 * Now we build the zonelist so that it contains the zones
2307 * of all the other nodes.
2308 * We don't want to pressure a particular node, so when
2309 * building the zones for node N, we make sure that the
2310 * zones coming right after the local ones are those from
2311 * node N+1 (modulo N)
2313 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2314 if (!node_online(node
))
2316 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2319 for (node
= 0; node
< local_node
; node
++) {
2320 if (!node_online(node
))
2322 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2326 zonelist
->_zonerefs
[j
].zone
= NULL
;
2327 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2330 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2331 static void build_zonelist_cache(pg_data_t
*pgdat
)
2333 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2336 #endif /* CONFIG_NUMA */
2338 /* return values int ....just for stop_machine_run() */
2339 static int __build_all_zonelists(void *dummy
)
2343 for_each_online_node(nid
) {
2344 pg_data_t
*pgdat
= NODE_DATA(nid
);
2346 build_zonelists(pgdat
);
2347 build_zonelist_cache(pgdat
);
2352 void build_all_zonelists(void)
2354 set_zonelist_order();
2356 if (system_state
== SYSTEM_BOOTING
) {
2357 __build_all_zonelists(NULL
);
2358 cpuset_init_current_mems_allowed();
2360 /* we have to stop all cpus to guarantee there is no user
2362 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
2363 /* cpuset refresh routine should be here */
2365 vm_total_pages
= nr_free_pagecache_pages();
2367 * Disable grouping by mobility if the number of pages in the
2368 * system is too low to allow the mechanism to work. It would be
2369 * more accurate, but expensive to check per-zone. This check is
2370 * made on memory-hotadd so a system can start with mobility
2371 * disabled and enable it later
2373 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2374 page_group_by_mobility_disabled
= 1;
2376 page_group_by_mobility_disabled
= 0;
2378 printk("Built %i zonelists in %s order, mobility grouping %s. "
2379 "Total pages: %ld\n",
2381 zonelist_order_name
[current_zonelist_order
],
2382 page_group_by_mobility_disabled
? "off" : "on",
2385 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2390 * Helper functions to size the waitqueue hash table.
2391 * Essentially these want to choose hash table sizes sufficiently
2392 * large so that collisions trying to wait on pages are rare.
2393 * But in fact, the number of active page waitqueues on typical
2394 * systems is ridiculously low, less than 200. So this is even
2395 * conservative, even though it seems large.
2397 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2398 * waitqueues, i.e. the size of the waitq table given the number of pages.
2400 #define PAGES_PER_WAITQUEUE 256
2402 #ifndef CONFIG_MEMORY_HOTPLUG
2403 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2405 unsigned long size
= 1;
2407 pages
/= PAGES_PER_WAITQUEUE
;
2409 while (size
< pages
)
2413 * Once we have dozens or even hundreds of threads sleeping
2414 * on IO we've got bigger problems than wait queue collision.
2415 * Limit the size of the wait table to a reasonable size.
2417 size
= min(size
, 4096UL);
2419 return max(size
, 4UL);
2423 * A zone's size might be changed by hot-add, so it is not possible to determine
2424 * a suitable size for its wait_table. So we use the maximum size now.
2426 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2428 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2429 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2430 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2432 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2433 * or more by the traditional way. (See above). It equals:
2435 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2436 * ia64(16K page size) : = ( 8G + 4M)byte.
2437 * powerpc (64K page size) : = (32G +16M)byte.
2439 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2446 * This is an integer logarithm so that shifts can be used later
2447 * to extract the more random high bits from the multiplicative
2448 * hash function before the remainder is taken.
2450 static inline unsigned long wait_table_bits(unsigned long size
)
2455 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2458 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2459 * of blocks reserved is based on zone->pages_min. The memory within the
2460 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2461 * higher will lead to a bigger reserve which will get freed as contiguous
2462 * blocks as reclaim kicks in
2464 static void setup_zone_migrate_reserve(struct zone
*zone
)
2466 unsigned long start_pfn
, pfn
, end_pfn
;
2468 unsigned long reserve
, block_migratetype
;
2470 /* Get the start pfn, end pfn and the number of blocks to reserve */
2471 start_pfn
= zone
->zone_start_pfn
;
2472 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2473 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2476 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2477 if (!pfn_valid(pfn
))
2479 page
= pfn_to_page(pfn
);
2481 /* Watch out for overlapping nodes */
2482 if (page_to_nid(page
) != zone_to_nid(zone
))
2485 /* Blocks with reserved pages will never free, skip them. */
2486 if (PageReserved(page
))
2489 block_migratetype
= get_pageblock_migratetype(page
);
2491 /* If this block is reserved, account for it */
2492 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2497 /* Suitable for reserving if this block is movable */
2498 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2499 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2500 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2506 * If the reserve is met and this is a previous reserved block,
2509 if (block_migratetype
== MIGRATE_RESERVE
) {
2510 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2511 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2517 * Initially all pages are reserved - free ones are freed
2518 * up by free_all_bootmem() once the early boot process is
2519 * done. Non-atomic initialization, single-pass.
2521 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2522 unsigned long start_pfn
, enum memmap_context context
)
2525 unsigned long end_pfn
= start_pfn
+ size
;
2529 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2530 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2532 * There can be holes in boot-time mem_map[]s
2533 * handed to this function. They do not
2534 * exist on hotplugged memory.
2536 if (context
== MEMMAP_EARLY
) {
2537 if (!early_pfn_valid(pfn
))
2539 if (!early_pfn_in_nid(pfn
, nid
))
2542 page
= pfn_to_page(pfn
);
2543 set_page_links(page
, zone
, nid
, pfn
);
2544 init_page_count(page
);
2545 reset_page_mapcount(page
);
2546 SetPageReserved(page
);
2548 * Mark the block movable so that blocks are reserved for
2549 * movable at startup. This will force kernel allocations
2550 * to reserve their blocks rather than leaking throughout
2551 * the address space during boot when many long-lived
2552 * kernel allocations are made. Later some blocks near
2553 * the start are marked MIGRATE_RESERVE by
2554 * setup_zone_migrate_reserve()
2556 * bitmap is created for zone's valid pfn range. but memmap
2557 * can be created for invalid pages (for alignment)
2558 * check here not to call set_pageblock_migratetype() against
2561 if ((z
->zone_start_pfn
<= pfn
)
2562 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2563 && !(pfn
& (pageblock_nr_pages
- 1)))
2564 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2566 INIT_LIST_HEAD(&page
->lru
);
2567 #ifdef WANT_PAGE_VIRTUAL
2568 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2569 if (!is_highmem_idx(zone
))
2570 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2575 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2578 for_each_migratetype_order(order
, t
) {
2579 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2580 zone
->free_area
[order
].nr_free
= 0;
2584 #ifndef __HAVE_ARCH_MEMMAP_INIT
2585 #define memmap_init(size, nid, zone, start_pfn) \
2586 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2589 static int zone_batchsize(struct zone
*zone
)
2594 * The per-cpu-pages pools are set to around 1000th of the
2595 * size of the zone. But no more than 1/2 of a meg.
2597 * OK, so we don't know how big the cache is. So guess.
2599 batch
= zone
->present_pages
/ 1024;
2600 if (batch
* PAGE_SIZE
> 512 * 1024)
2601 batch
= (512 * 1024) / PAGE_SIZE
;
2602 batch
/= 4; /* We effectively *= 4 below */
2607 * Clamp the batch to a 2^n - 1 value. Having a power
2608 * of 2 value was found to be more likely to have
2609 * suboptimal cache aliasing properties in some cases.
2611 * For example if 2 tasks are alternately allocating
2612 * batches of pages, one task can end up with a lot
2613 * of pages of one half of the possible page colors
2614 * and the other with pages of the other colors.
2616 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2621 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2623 struct per_cpu_pages
*pcp
;
2625 memset(p
, 0, sizeof(*p
));
2629 pcp
->high
= 6 * batch
;
2630 pcp
->batch
= max(1UL, 1 * batch
);
2631 INIT_LIST_HEAD(&pcp
->list
);
2635 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2636 * to the value high for the pageset p.
2639 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2642 struct per_cpu_pages
*pcp
;
2646 pcp
->batch
= max(1UL, high
/4);
2647 if ((high
/4) > (PAGE_SHIFT
* 8))
2648 pcp
->batch
= PAGE_SHIFT
* 8;
2654 * Boot pageset table. One per cpu which is going to be used for all
2655 * zones and all nodes. The parameters will be set in such a way
2656 * that an item put on a list will immediately be handed over to
2657 * the buddy list. This is safe since pageset manipulation is done
2658 * with interrupts disabled.
2660 * Some NUMA counter updates may also be caught by the boot pagesets.
2662 * The boot_pagesets must be kept even after bootup is complete for
2663 * unused processors and/or zones. They do play a role for bootstrapping
2664 * hotplugged processors.
2666 * zoneinfo_show() and maybe other functions do
2667 * not check if the processor is online before following the pageset pointer.
2668 * Other parts of the kernel may not check if the zone is available.
2670 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2673 * Dynamically allocate memory for the
2674 * per cpu pageset array in struct zone.
2676 static int __cpuinit
process_zones(int cpu
)
2678 struct zone
*zone
, *dzone
;
2679 int node
= cpu_to_node(cpu
);
2681 node_set_state(node
, N_CPU
); /* this node has a cpu */
2683 for_each_zone(zone
) {
2685 if (!populated_zone(zone
))
2688 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2690 if (!zone_pcp(zone
, cpu
))
2693 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2695 if (percpu_pagelist_fraction
)
2696 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2697 (zone
->present_pages
/ percpu_pagelist_fraction
));
2702 for_each_zone(dzone
) {
2703 if (!populated_zone(dzone
))
2707 kfree(zone_pcp(dzone
, cpu
));
2708 zone_pcp(dzone
, cpu
) = NULL
;
2713 static inline void free_zone_pagesets(int cpu
)
2717 for_each_zone(zone
) {
2718 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2720 /* Free per_cpu_pageset if it is slab allocated */
2721 if (pset
!= &boot_pageset
[cpu
])
2723 zone_pcp(zone
, cpu
) = NULL
;
2727 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2728 unsigned long action
,
2731 int cpu
= (long)hcpu
;
2732 int ret
= NOTIFY_OK
;
2735 case CPU_UP_PREPARE
:
2736 case CPU_UP_PREPARE_FROZEN
:
2737 if (process_zones(cpu
))
2740 case CPU_UP_CANCELED
:
2741 case CPU_UP_CANCELED_FROZEN
:
2743 case CPU_DEAD_FROZEN
:
2744 free_zone_pagesets(cpu
);
2752 static struct notifier_block __cpuinitdata pageset_notifier
=
2753 { &pageset_cpuup_callback
, NULL
, 0 };
2755 void __init
setup_per_cpu_pageset(void)
2759 /* Initialize per_cpu_pageset for cpu 0.
2760 * A cpuup callback will do this for every cpu
2761 * as it comes online
2763 err
= process_zones(smp_processor_id());
2765 register_cpu_notifier(&pageset_notifier
);
2770 static noinline __init_refok
2771 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2774 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2778 * The per-page waitqueue mechanism uses hashed waitqueues
2781 zone
->wait_table_hash_nr_entries
=
2782 wait_table_hash_nr_entries(zone_size_pages
);
2783 zone
->wait_table_bits
=
2784 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2785 alloc_size
= zone
->wait_table_hash_nr_entries
2786 * sizeof(wait_queue_head_t
);
2788 if (!slab_is_available()) {
2789 zone
->wait_table
= (wait_queue_head_t
*)
2790 alloc_bootmem_node(pgdat
, alloc_size
);
2793 * This case means that a zone whose size was 0 gets new memory
2794 * via memory hot-add.
2795 * But it may be the case that a new node was hot-added. In
2796 * this case vmalloc() will not be able to use this new node's
2797 * memory - this wait_table must be initialized to use this new
2798 * node itself as well.
2799 * To use this new node's memory, further consideration will be
2802 zone
->wait_table
= vmalloc(alloc_size
);
2804 if (!zone
->wait_table
)
2807 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2808 init_waitqueue_head(zone
->wait_table
+ i
);
2813 static __meminit
void zone_pcp_init(struct zone
*zone
)
2816 unsigned long batch
= zone_batchsize(zone
);
2818 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2820 /* Early boot. Slab allocator not functional yet */
2821 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2822 setup_pageset(&boot_pageset
[cpu
],0);
2824 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2827 if (zone
->present_pages
)
2828 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2829 zone
->name
, zone
->present_pages
, batch
);
2832 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2833 unsigned long zone_start_pfn
,
2835 enum memmap_context context
)
2837 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2839 ret
= zone_wait_table_init(zone
, size
);
2842 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2844 zone
->zone_start_pfn
= zone_start_pfn
;
2846 zone_init_free_lists(zone
);
2851 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2853 * Basic iterator support. Return the first range of PFNs for a node
2854 * Note: nid == MAX_NUMNODES returns first region regardless of node
2856 static int __meminit
first_active_region_index_in_nid(int nid
)
2860 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2861 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2868 * Basic iterator support. Return the next active range of PFNs for a node
2869 * Note: nid == MAX_NUMNODES returns next region regardless of node
2871 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2873 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2874 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2880 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2882 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2883 * Architectures may implement their own version but if add_active_range()
2884 * was used and there are no special requirements, this is a convenient
2887 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2891 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2892 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2893 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2895 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2896 return early_node_map
[i
].nid
;
2901 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2903 /* Basic iterator support to walk early_node_map[] */
2904 #define for_each_active_range_index_in_nid(i, nid) \
2905 for (i = first_active_region_index_in_nid(nid); i != -1; \
2906 i = next_active_region_index_in_nid(i, nid))
2909 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2910 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2911 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2913 * If an architecture guarantees that all ranges registered with
2914 * add_active_ranges() contain no holes and may be freed, this
2915 * this function may be used instead of calling free_bootmem() manually.
2917 void __init
free_bootmem_with_active_regions(int nid
,
2918 unsigned long max_low_pfn
)
2922 for_each_active_range_index_in_nid(i
, nid
) {
2923 unsigned long size_pages
= 0;
2924 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2926 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2929 if (end_pfn
> max_low_pfn
)
2930 end_pfn
= max_low_pfn
;
2932 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2933 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2934 PFN_PHYS(early_node_map
[i
].start_pfn
),
2935 size_pages
<< PAGE_SHIFT
);
2940 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2941 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2943 * If an architecture guarantees that all ranges registered with
2944 * add_active_ranges() contain no holes and may be freed, this
2945 * function may be used instead of calling memory_present() manually.
2947 void __init
sparse_memory_present_with_active_regions(int nid
)
2951 for_each_active_range_index_in_nid(i
, nid
)
2952 memory_present(early_node_map
[i
].nid
,
2953 early_node_map
[i
].start_pfn
,
2954 early_node_map
[i
].end_pfn
);
2958 * push_node_boundaries - Push node boundaries to at least the requested boundary
2959 * @nid: The nid of the node to push the boundary for
2960 * @start_pfn: The start pfn of the node
2961 * @end_pfn: The end pfn of the node
2963 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2964 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2965 * be hotplugged even though no physical memory exists. This function allows
2966 * an arch to push out the node boundaries so mem_map is allocated that can
2969 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2970 void __init
push_node_boundaries(unsigned int nid
,
2971 unsigned long start_pfn
, unsigned long end_pfn
)
2973 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2974 nid
, start_pfn
, end_pfn
);
2976 /* Initialise the boundary for this node if necessary */
2977 if (node_boundary_end_pfn
[nid
] == 0)
2978 node_boundary_start_pfn
[nid
] = -1UL;
2980 /* Update the boundaries */
2981 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2982 node_boundary_start_pfn
[nid
] = start_pfn
;
2983 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2984 node_boundary_end_pfn
[nid
] = end_pfn
;
2987 /* If necessary, push the node boundary out for reserve hotadd */
2988 static void __meminit
account_node_boundary(unsigned int nid
,
2989 unsigned long *start_pfn
, unsigned long *end_pfn
)
2991 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2992 nid
, *start_pfn
, *end_pfn
);
2994 /* Return if boundary information has not been provided */
2995 if (node_boundary_end_pfn
[nid
] == 0)
2998 /* Check the boundaries and update if necessary */
2999 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
3000 *start_pfn
= node_boundary_start_pfn
[nid
];
3001 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
3002 *end_pfn
= node_boundary_end_pfn
[nid
];
3005 void __init
push_node_boundaries(unsigned int nid
,
3006 unsigned long start_pfn
, unsigned long end_pfn
) {}
3008 static void __meminit
account_node_boundary(unsigned int nid
,
3009 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3014 * get_pfn_range_for_nid - Return the start and end page frames for a node
3015 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3016 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3017 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3019 * It returns the start and end page frame of a node based on information
3020 * provided by an arch calling add_active_range(). If called for a node
3021 * with no available memory, a warning is printed and the start and end
3024 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3025 unsigned long *start_pfn
, unsigned long *end_pfn
)
3031 for_each_active_range_index_in_nid(i
, nid
) {
3032 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3033 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3036 if (*start_pfn
== -1UL)
3039 /* Push the node boundaries out if requested */
3040 account_node_boundary(nid
, start_pfn
, end_pfn
);
3044 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3045 * assumption is made that zones within a node are ordered in monotonic
3046 * increasing memory addresses so that the "highest" populated zone is used
3048 void __init
find_usable_zone_for_movable(void)
3051 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3052 if (zone_index
== ZONE_MOVABLE
)
3055 if (arch_zone_highest_possible_pfn
[zone_index
] >
3056 arch_zone_lowest_possible_pfn
[zone_index
])
3060 VM_BUG_ON(zone_index
== -1);
3061 movable_zone
= zone_index
;
3065 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3066 * because it is sized independant of architecture. Unlike the other zones,
3067 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3068 * in each node depending on the size of each node and how evenly kernelcore
3069 * is distributed. This helper function adjusts the zone ranges
3070 * provided by the architecture for a given node by using the end of the
3071 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3072 * zones within a node are in order of monotonic increases memory addresses
3074 void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3075 unsigned long zone_type
,
3076 unsigned long node_start_pfn
,
3077 unsigned long node_end_pfn
,
3078 unsigned long *zone_start_pfn
,
3079 unsigned long *zone_end_pfn
)
3081 /* Only adjust if ZONE_MOVABLE is on this node */
3082 if (zone_movable_pfn
[nid
]) {
3083 /* Size ZONE_MOVABLE */
3084 if (zone_type
== ZONE_MOVABLE
) {
3085 *zone_start_pfn
= zone_movable_pfn
[nid
];
3086 *zone_end_pfn
= min(node_end_pfn
,
3087 arch_zone_highest_possible_pfn
[movable_zone
]);
3089 /* Adjust for ZONE_MOVABLE starting within this range */
3090 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3091 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3092 *zone_end_pfn
= zone_movable_pfn
[nid
];
3094 /* Check if this whole range is within ZONE_MOVABLE */
3095 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3096 *zone_start_pfn
= *zone_end_pfn
;
3101 * Return the number of pages a zone spans in a node, including holes
3102 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3104 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3105 unsigned long zone_type
,
3106 unsigned long *ignored
)
3108 unsigned long node_start_pfn
, node_end_pfn
;
3109 unsigned long zone_start_pfn
, zone_end_pfn
;
3111 /* Get the start and end of the node and zone */
3112 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3113 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3114 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3115 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3116 node_start_pfn
, node_end_pfn
,
3117 &zone_start_pfn
, &zone_end_pfn
);
3119 /* Check that this node has pages within the zone's required range */
3120 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3123 /* Move the zone boundaries inside the node if necessary */
3124 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3125 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3127 /* Return the spanned pages */
3128 return zone_end_pfn
- zone_start_pfn
;
3132 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3133 * then all holes in the requested range will be accounted for.
3135 unsigned long __meminit
__absent_pages_in_range(int nid
,
3136 unsigned long range_start_pfn
,
3137 unsigned long range_end_pfn
)
3140 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3141 unsigned long start_pfn
;
3143 /* Find the end_pfn of the first active range of pfns in the node */
3144 i
= first_active_region_index_in_nid(nid
);
3148 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3150 /* Account for ranges before physical memory on this node */
3151 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3152 hole_pages
= prev_end_pfn
- range_start_pfn
;
3154 /* Find all holes for the zone within the node */
3155 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3157 /* No need to continue if prev_end_pfn is outside the zone */
3158 if (prev_end_pfn
>= range_end_pfn
)
3161 /* Make sure the end of the zone is not within the hole */
3162 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3163 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3165 /* Update the hole size cound and move on */
3166 if (start_pfn
> range_start_pfn
) {
3167 BUG_ON(prev_end_pfn
> start_pfn
);
3168 hole_pages
+= start_pfn
- prev_end_pfn
;
3170 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3173 /* Account for ranges past physical memory on this node */
3174 if (range_end_pfn
> prev_end_pfn
)
3175 hole_pages
+= range_end_pfn
-
3176 max(range_start_pfn
, prev_end_pfn
);
3182 * absent_pages_in_range - Return number of page frames in holes within a range
3183 * @start_pfn: The start PFN to start searching for holes
3184 * @end_pfn: The end PFN to stop searching for holes
3186 * It returns the number of pages frames in memory holes within a range.
3188 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3189 unsigned long end_pfn
)
3191 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3194 /* Return the number of page frames in holes in a zone on a node */
3195 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3196 unsigned long zone_type
,
3197 unsigned long *ignored
)
3199 unsigned long node_start_pfn
, node_end_pfn
;
3200 unsigned long zone_start_pfn
, zone_end_pfn
;
3202 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3203 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3205 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3208 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3209 node_start_pfn
, node_end_pfn
,
3210 &zone_start_pfn
, &zone_end_pfn
);
3211 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3215 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3216 unsigned long zone_type
,
3217 unsigned long *zones_size
)
3219 return zones_size
[zone_type
];
3222 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3223 unsigned long zone_type
,
3224 unsigned long *zholes_size
)
3229 return zholes_size
[zone_type
];
3234 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3235 unsigned long *zones_size
, unsigned long *zholes_size
)
3237 unsigned long realtotalpages
, totalpages
= 0;
3240 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3241 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3243 pgdat
->node_spanned_pages
= totalpages
;
3245 realtotalpages
= totalpages
;
3246 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3248 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3250 pgdat
->node_present_pages
= realtotalpages
;
3251 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3255 #ifndef CONFIG_SPARSEMEM
3257 * Calculate the size of the zone->blockflags rounded to an unsigned long
3258 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3259 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3260 * round what is now in bits to nearest long in bits, then return it in
3263 static unsigned long __init
usemap_size(unsigned long zonesize
)
3265 unsigned long usemapsize
;
3267 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3268 usemapsize
= usemapsize
>> pageblock_order
;
3269 usemapsize
*= NR_PAGEBLOCK_BITS
;
3270 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3272 return usemapsize
/ 8;
3275 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3276 struct zone
*zone
, unsigned long zonesize
)
3278 unsigned long usemapsize
= usemap_size(zonesize
);
3279 zone
->pageblock_flags
= NULL
;
3281 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3282 memset(zone
->pageblock_flags
, 0, usemapsize
);
3286 static void inline setup_usemap(struct pglist_data
*pgdat
,
3287 struct zone
*zone
, unsigned long zonesize
) {}
3288 #endif /* CONFIG_SPARSEMEM */
3290 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3292 /* Return a sensible default order for the pageblock size. */
3293 static inline int pageblock_default_order(void)
3295 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3296 return HUGETLB_PAGE_ORDER
;
3301 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3302 static inline void __init
set_pageblock_order(unsigned int order
)
3304 /* Check that pageblock_nr_pages has not already been setup */
3305 if (pageblock_order
)
3309 * Assume the largest contiguous order of interest is a huge page.
3310 * This value may be variable depending on boot parameters on IA64
3312 pageblock_order
= order
;
3314 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3317 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3318 * and pageblock_default_order() are unused as pageblock_order is set
3319 * at compile-time. See include/linux/pageblock-flags.h for the values of
3320 * pageblock_order based on the kernel config
3322 static inline int pageblock_default_order(unsigned int order
)
3326 #define set_pageblock_order(x) do {} while (0)
3328 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3331 * Set up the zone data structures:
3332 * - mark all pages reserved
3333 * - mark all memory queues empty
3334 * - clear the memory bitmaps
3336 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3337 unsigned long *zones_size
, unsigned long *zholes_size
)
3340 int nid
= pgdat
->node_id
;
3341 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3344 pgdat_resize_init(pgdat
);
3345 pgdat
->nr_zones
= 0;
3346 init_waitqueue_head(&pgdat
->kswapd_wait
);
3347 pgdat
->kswapd_max_order
= 0;
3349 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3350 struct zone
*zone
= pgdat
->node_zones
+ j
;
3351 unsigned long size
, realsize
, memmap_pages
;
3353 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3354 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3358 * Adjust realsize so that it accounts for how much memory
3359 * is used by this zone for memmap. This affects the watermark
3360 * and per-cpu initialisations
3363 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3364 if (realsize
>= memmap_pages
) {
3365 realsize
-= memmap_pages
;
3367 " %s zone: %lu pages used for memmap\n",
3368 zone_names
[j
], memmap_pages
);
3371 " %s zone: %lu pages exceeds realsize %lu\n",
3372 zone_names
[j
], memmap_pages
, realsize
);
3374 /* Account for reserved pages */
3375 if (j
== 0 && realsize
> dma_reserve
) {
3376 realsize
-= dma_reserve
;
3377 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3378 zone_names
[0], dma_reserve
);
3381 if (!is_highmem_idx(j
))
3382 nr_kernel_pages
+= realsize
;
3383 nr_all_pages
+= realsize
;
3385 zone
->spanned_pages
= size
;
3386 zone
->present_pages
= realsize
;
3389 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3391 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3393 zone
->name
= zone_names
[j
];
3394 spin_lock_init(&zone
->lock
);
3395 spin_lock_init(&zone
->lru_lock
);
3396 zone_seqlock_init(zone
);
3397 zone
->zone_pgdat
= pgdat
;
3399 zone
->prev_priority
= DEF_PRIORITY
;
3401 zone_pcp_init(zone
);
3402 INIT_LIST_HEAD(&zone
->active_list
);
3403 INIT_LIST_HEAD(&zone
->inactive_list
);
3404 zone
->nr_scan_active
= 0;
3405 zone
->nr_scan_inactive
= 0;
3406 zap_zone_vm_stats(zone
);
3411 set_pageblock_order(pageblock_default_order());
3412 setup_usemap(pgdat
, zone
, size
);
3413 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3414 size
, MEMMAP_EARLY
);
3416 memmap_init(size
, nid
, j
, zone_start_pfn
);
3417 zone_start_pfn
+= size
;
3421 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3423 /* Skip empty nodes */
3424 if (!pgdat
->node_spanned_pages
)
3427 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3428 /* ia64 gets its own node_mem_map, before this, without bootmem */
3429 if (!pgdat
->node_mem_map
) {
3430 unsigned long size
, start
, end
;
3434 * The zone's endpoints aren't required to be MAX_ORDER
3435 * aligned but the node_mem_map endpoints must be in order
3436 * for the buddy allocator to function correctly.
3438 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3439 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3440 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3441 size
= (end
- start
) * sizeof(struct page
);
3442 map
= alloc_remap(pgdat
->node_id
, size
);
3444 map
= alloc_bootmem_node(pgdat
, size
);
3445 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3447 #ifndef CONFIG_NEED_MULTIPLE_NODES
3449 * With no DISCONTIG, the global mem_map is just set as node 0's
3451 if (pgdat
== NODE_DATA(0)) {
3452 mem_map
= NODE_DATA(0)->node_mem_map
;
3453 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3454 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3455 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3456 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3459 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3462 void __paginginit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
3463 unsigned long *zones_size
, unsigned long node_start_pfn
,
3464 unsigned long *zholes_size
)
3466 pgdat
->node_id
= nid
;
3467 pgdat
->node_start_pfn
= node_start_pfn
;
3468 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3470 alloc_node_mem_map(pgdat
);
3472 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3475 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3477 #if MAX_NUMNODES > 1
3479 * Figure out the number of possible node ids.
3481 static void __init
setup_nr_node_ids(void)
3484 unsigned int highest
= 0;
3486 for_each_node_mask(node
, node_possible_map
)
3488 nr_node_ids
= highest
+ 1;
3491 static inline void setup_nr_node_ids(void)
3497 * add_active_range - Register a range of PFNs backed by physical memory
3498 * @nid: The node ID the range resides on
3499 * @start_pfn: The start PFN of the available physical memory
3500 * @end_pfn: The end PFN of the available physical memory
3502 * These ranges are stored in an early_node_map[] and later used by
3503 * free_area_init_nodes() to calculate zone sizes and holes. If the
3504 * range spans a memory hole, it is up to the architecture to ensure
3505 * the memory is not freed by the bootmem allocator. If possible
3506 * the range being registered will be merged with existing ranges.
3508 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3509 unsigned long end_pfn
)
3513 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
3514 "%d entries of %d used\n",
3515 nid
, start_pfn
, end_pfn
,
3516 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3518 /* Merge with existing active regions if possible */
3519 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3520 if (early_node_map
[i
].nid
!= nid
)
3523 /* Skip if an existing region covers this new one */
3524 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3525 end_pfn
<= early_node_map
[i
].end_pfn
)
3528 /* Merge forward if suitable */
3529 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3530 end_pfn
> early_node_map
[i
].end_pfn
) {
3531 early_node_map
[i
].end_pfn
= end_pfn
;
3535 /* Merge backward if suitable */
3536 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3537 end_pfn
>= early_node_map
[i
].start_pfn
) {
3538 early_node_map
[i
].start_pfn
= start_pfn
;
3543 /* Check that early_node_map is large enough */
3544 if (i
>= MAX_ACTIVE_REGIONS
) {
3545 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3546 MAX_ACTIVE_REGIONS
);
3550 early_node_map
[i
].nid
= nid
;
3551 early_node_map
[i
].start_pfn
= start_pfn
;
3552 early_node_map
[i
].end_pfn
= end_pfn
;
3553 nr_nodemap_entries
= i
+ 1;
3557 * shrink_active_range - Shrink an existing registered range of PFNs
3558 * @nid: The node id the range is on that should be shrunk
3559 * @old_end_pfn: The old end PFN of the range
3560 * @new_end_pfn: The new PFN of the range
3562 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3563 * The map is kept at the end physical page range that has already been
3564 * registered with add_active_range(). This function allows an arch to shrink
3565 * an existing registered range.
3567 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
3568 unsigned long new_end_pfn
)
3572 /* Find the old active region end and shrink */
3573 for_each_active_range_index_in_nid(i
, nid
)
3574 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
3575 early_node_map
[i
].end_pfn
= new_end_pfn
;
3581 * remove_all_active_ranges - Remove all currently registered regions
3583 * During discovery, it may be found that a table like SRAT is invalid
3584 * and an alternative discovery method must be used. This function removes
3585 * all currently registered regions.
3587 void __init
remove_all_active_ranges(void)
3589 memset(early_node_map
, 0, sizeof(early_node_map
));
3590 nr_nodemap_entries
= 0;
3591 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3592 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3593 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3594 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3597 /* Compare two active node_active_regions */
3598 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3600 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3601 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3603 /* Done this way to avoid overflows */
3604 if (arange
->start_pfn
> brange
->start_pfn
)
3606 if (arange
->start_pfn
< brange
->start_pfn
)
3612 /* sort the node_map by start_pfn */
3613 static void __init
sort_node_map(void)
3615 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3616 sizeof(struct node_active_region
),
3617 cmp_node_active_region
, NULL
);
3620 /* Find the lowest pfn for a node */
3621 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
3624 unsigned long min_pfn
= ULONG_MAX
;
3626 /* Assuming a sorted map, the first range found has the starting pfn */
3627 for_each_active_range_index_in_nid(i
, nid
)
3628 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3630 if (min_pfn
== ULONG_MAX
) {
3632 "Could not find start_pfn for node %lu\n", nid
);
3640 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3642 * It returns the minimum PFN based on information provided via
3643 * add_active_range().
3645 unsigned long __init
find_min_pfn_with_active_regions(void)
3647 return find_min_pfn_for_node(MAX_NUMNODES
);
3651 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3653 * It returns the maximum PFN based on information provided via
3654 * add_active_range().
3656 unsigned long __init
find_max_pfn_with_active_regions(void)
3659 unsigned long max_pfn
= 0;
3661 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3662 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
3668 * early_calculate_totalpages()
3669 * Sum pages in active regions for movable zone.
3670 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3672 static unsigned long __init
early_calculate_totalpages(void)
3675 unsigned long totalpages
= 0;
3677 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3678 unsigned long pages
= early_node_map
[i
].end_pfn
-
3679 early_node_map
[i
].start_pfn
;
3680 totalpages
+= pages
;
3682 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3688 * Find the PFN the Movable zone begins in each node. Kernel memory
3689 * is spread evenly between nodes as long as the nodes have enough
3690 * memory. When they don't, some nodes will have more kernelcore than
3693 void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3696 unsigned long usable_startpfn
;
3697 unsigned long kernelcore_node
, kernelcore_remaining
;
3698 unsigned long totalpages
= early_calculate_totalpages();
3699 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3702 * If movablecore was specified, calculate what size of
3703 * kernelcore that corresponds so that memory usable for
3704 * any allocation type is evenly spread. If both kernelcore
3705 * and movablecore are specified, then the value of kernelcore
3706 * will be used for required_kernelcore if it's greater than
3707 * what movablecore would have allowed.
3709 if (required_movablecore
) {
3710 unsigned long corepages
;
3713 * Round-up so that ZONE_MOVABLE is at least as large as what
3714 * was requested by the user
3716 required_movablecore
=
3717 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3718 corepages
= totalpages
- required_movablecore
;
3720 required_kernelcore
= max(required_kernelcore
, corepages
);
3723 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3724 if (!required_kernelcore
)
3727 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3728 find_usable_zone_for_movable();
3729 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3732 /* Spread kernelcore memory as evenly as possible throughout nodes */
3733 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3734 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3736 * Recalculate kernelcore_node if the division per node
3737 * now exceeds what is necessary to satisfy the requested
3738 * amount of memory for the kernel
3740 if (required_kernelcore
< kernelcore_node
)
3741 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3744 * As the map is walked, we track how much memory is usable
3745 * by the kernel using kernelcore_remaining. When it is
3746 * 0, the rest of the node is usable by ZONE_MOVABLE
3748 kernelcore_remaining
= kernelcore_node
;
3750 /* Go through each range of PFNs within this node */
3751 for_each_active_range_index_in_nid(i
, nid
) {
3752 unsigned long start_pfn
, end_pfn
;
3753 unsigned long size_pages
;
3755 start_pfn
= max(early_node_map
[i
].start_pfn
,
3756 zone_movable_pfn
[nid
]);
3757 end_pfn
= early_node_map
[i
].end_pfn
;
3758 if (start_pfn
>= end_pfn
)
3761 /* Account for what is only usable for kernelcore */
3762 if (start_pfn
< usable_startpfn
) {
3763 unsigned long kernel_pages
;
3764 kernel_pages
= min(end_pfn
, usable_startpfn
)
3767 kernelcore_remaining
-= min(kernel_pages
,
3768 kernelcore_remaining
);
3769 required_kernelcore
-= min(kernel_pages
,
3770 required_kernelcore
);
3772 /* Continue if range is now fully accounted */
3773 if (end_pfn
<= usable_startpfn
) {
3776 * Push zone_movable_pfn to the end so
3777 * that if we have to rebalance
3778 * kernelcore across nodes, we will
3779 * not double account here
3781 zone_movable_pfn
[nid
] = end_pfn
;
3784 start_pfn
= usable_startpfn
;
3788 * The usable PFN range for ZONE_MOVABLE is from
3789 * start_pfn->end_pfn. Calculate size_pages as the
3790 * number of pages used as kernelcore
3792 size_pages
= end_pfn
- start_pfn
;
3793 if (size_pages
> kernelcore_remaining
)
3794 size_pages
= kernelcore_remaining
;
3795 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3798 * Some kernelcore has been met, update counts and
3799 * break if the kernelcore for this node has been
3802 required_kernelcore
-= min(required_kernelcore
,
3804 kernelcore_remaining
-= size_pages
;
3805 if (!kernelcore_remaining
)
3811 * If there is still required_kernelcore, we do another pass with one
3812 * less node in the count. This will push zone_movable_pfn[nid] further
3813 * along on the nodes that still have memory until kernelcore is
3817 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3820 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3821 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3822 zone_movable_pfn
[nid
] =
3823 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3826 /* Any regular memory on that node ? */
3827 static void check_for_regular_memory(pg_data_t
*pgdat
)
3829 #ifdef CONFIG_HIGHMEM
3830 enum zone_type zone_type
;
3832 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3833 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3834 if (zone
->present_pages
)
3835 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3841 * free_area_init_nodes - Initialise all pg_data_t and zone data
3842 * @max_zone_pfn: an array of max PFNs for each zone
3844 * This will call free_area_init_node() for each active node in the system.
3845 * Using the page ranges provided by add_active_range(), the size of each
3846 * zone in each node and their holes is calculated. If the maximum PFN
3847 * between two adjacent zones match, it is assumed that the zone is empty.
3848 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3849 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3850 * starts where the previous one ended. For example, ZONE_DMA32 starts
3851 * at arch_max_dma_pfn.
3853 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3858 /* Sort early_node_map as initialisation assumes it is sorted */
3861 /* Record where the zone boundaries are */
3862 memset(arch_zone_lowest_possible_pfn
, 0,
3863 sizeof(arch_zone_lowest_possible_pfn
));
3864 memset(arch_zone_highest_possible_pfn
, 0,
3865 sizeof(arch_zone_highest_possible_pfn
));
3866 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3867 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3868 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3869 if (i
== ZONE_MOVABLE
)
3871 arch_zone_lowest_possible_pfn
[i
] =
3872 arch_zone_highest_possible_pfn
[i
-1];
3873 arch_zone_highest_possible_pfn
[i
] =
3874 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3876 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3877 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3879 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3880 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3881 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3883 /* Print out the zone ranges */
3884 printk("Zone PFN ranges:\n");
3885 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3886 if (i
== ZONE_MOVABLE
)
3888 printk(" %-8s %8lu -> %8lu\n",
3890 arch_zone_lowest_possible_pfn
[i
],
3891 arch_zone_highest_possible_pfn
[i
]);
3894 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3895 printk("Movable zone start PFN for each node\n");
3896 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
3897 if (zone_movable_pfn
[i
])
3898 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
3901 /* Print out the early_node_map[] */
3902 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
3903 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3904 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
3905 early_node_map
[i
].start_pfn
,
3906 early_node_map
[i
].end_pfn
);
3908 /* Initialise every node */
3909 setup_nr_node_ids();
3910 for_each_online_node(nid
) {
3911 pg_data_t
*pgdat
= NODE_DATA(nid
);
3912 free_area_init_node(nid
, pgdat
, NULL
,
3913 find_min_pfn_for_node(nid
), NULL
);
3915 /* Any memory on that node */
3916 if (pgdat
->node_present_pages
)
3917 node_set_state(nid
, N_HIGH_MEMORY
);
3918 check_for_regular_memory(pgdat
);
3922 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
3924 unsigned long long coremem
;
3928 coremem
= memparse(p
, &p
);
3929 *core
= coremem
>> PAGE_SHIFT
;
3931 /* Paranoid check that UL is enough for the coremem value */
3932 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
3938 * kernelcore=size sets the amount of memory for use for allocations that
3939 * cannot be reclaimed or migrated.
3941 static int __init
cmdline_parse_kernelcore(char *p
)
3943 return cmdline_parse_core(p
, &required_kernelcore
);
3947 * movablecore=size sets the amount of memory for use for allocations that
3948 * can be reclaimed or migrated.
3950 static int __init
cmdline_parse_movablecore(char *p
)
3952 return cmdline_parse_core(p
, &required_movablecore
);
3955 early_param("kernelcore", cmdline_parse_kernelcore
);
3956 early_param("movablecore", cmdline_parse_movablecore
);
3958 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3961 * set_dma_reserve - set the specified number of pages reserved in the first zone
3962 * @new_dma_reserve: The number of pages to mark reserved
3964 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3965 * In the DMA zone, a significant percentage may be consumed by kernel image
3966 * and other unfreeable allocations which can skew the watermarks badly. This
3967 * function may optionally be used to account for unfreeable pages in the
3968 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3969 * smaller per-cpu batchsize.
3971 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
3973 dma_reserve
= new_dma_reserve
;
3976 #ifndef CONFIG_NEED_MULTIPLE_NODES
3977 static bootmem_data_t contig_bootmem_data
;
3978 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
3980 EXPORT_SYMBOL(contig_page_data
);
3983 void __init
free_area_init(unsigned long *zones_size
)
3985 free_area_init_node(0, NODE_DATA(0), zones_size
,
3986 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
3989 static int page_alloc_cpu_notify(struct notifier_block
*self
,
3990 unsigned long action
, void *hcpu
)
3992 int cpu
= (unsigned long)hcpu
;
3994 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3998 * Spill the event counters of the dead processor
3999 * into the current processors event counters.
4000 * This artificially elevates the count of the current
4003 vm_events_fold_cpu(cpu
);
4006 * Zero the differential counters of the dead processor
4007 * so that the vm statistics are consistent.
4009 * This is only okay since the processor is dead and cannot
4010 * race with what we are doing.
4012 refresh_cpu_vm_stats(cpu
);
4017 void __init
page_alloc_init(void)
4019 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4023 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4024 * or min_free_kbytes changes.
4026 static void calculate_totalreserve_pages(void)
4028 struct pglist_data
*pgdat
;
4029 unsigned long reserve_pages
= 0;
4030 enum zone_type i
, j
;
4032 for_each_online_pgdat(pgdat
) {
4033 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4034 struct zone
*zone
= pgdat
->node_zones
+ i
;
4035 unsigned long max
= 0;
4037 /* Find valid and maximum lowmem_reserve in the zone */
4038 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4039 if (zone
->lowmem_reserve
[j
] > max
)
4040 max
= zone
->lowmem_reserve
[j
];
4043 /* we treat pages_high as reserved pages. */
4044 max
+= zone
->pages_high
;
4046 if (max
> zone
->present_pages
)
4047 max
= zone
->present_pages
;
4048 reserve_pages
+= max
;
4051 totalreserve_pages
= reserve_pages
;
4055 * setup_per_zone_lowmem_reserve - called whenever
4056 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4057 * has a correct pages reserved value, so an adequate number of
4058 * pages are left in the zone after a successful __alloc_pages().
4060 static void setup_per_zone_lowmem_reserve(void)
4062 struct pglist_data
*pgdat
;
4063 enum zone_type j
, idx
;
4065 for_each_online_pgdat(pgdat
) {
4066 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4067 struct zone
*zone
= pgdat
->node_zones
+ j
;
4068 unsigned long present_pages
= zone
->present_pages
;
4070 zone
->lowmem_reserve
[j
] = 0;
4074 struct zone
*lower_zone
;
4078 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4079 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4081 lower_zone
= pgdat
->node_zones
+ idx
;
4082 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4083 sysctl_lowmem_reserve_ratio
[idx
];
4084 present_pages
+= lower_zone
->present_pages
;
4089 /* update totalreserve_pages */
4090 calculate_totalreserve_pages();
4094 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4096 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4097 * with respect to min_free_kbytes.
4099 void setup_per_zone_pages_min(void)
4101 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4102 unsigned long lowmem_pages
= 0;
4104 unsigned long flags
;
4106 /* Calculate total number of !ZONE_HIGHMEM pages */
4107 for_each_zone(zone
) {
4108 if (!is_highmem(zone
))
4109 lowmem_pages
+= zone
->present_pages
;
4112 for_each_zone(zone
) {
4115 spin_lock_irqsave(&zone
->lru_lock
, flags
);
4116 tmp
= (u64
)pages_min
* zone
->present_pages
;
4117 do_div(tmp
, lowmem_pages
);
4118 if (is_highmem(zone
)) {
4120 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4121 * need highmem pages, so cap pages_min to a small
4124 * The (pages_high-pages_low) and (pages_low-pages_min)
4125 * deltas controls asynch page reclaim, and so should
4126 * not be capped for highmem.
4130 min_pages
= zone
->present_pages
/ 1024;
4131 if (min_pages
< SWAP_CLUSTER_MAX
)
4132 min_pages
= SWAP_CLUSTER_MAX
;
4133 if (min_pages
> 128)
4135 zone
->pages_min
= min_pages
;
4138 * If it's a lowmem zone, reserve a number of pages
4139 * proportionate to the zone's size.
4141 zone
->pages_min
= tmp
;
4144 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4145 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4146 setup_zone_migrate_reserve(zone
);
4147 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4150 /* update totalreserve_pages */
4151 calculate_totalreserve_pages();
4155 * Initialise min_free_kbytes.
4157 * For small machines we want it small (128k min). For large machines
4158 * we want it large (64MB max). But it is not linear, because network
4159 * bandwidth does not increase linearly with machine size. We use
4161 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4162 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4178 static int __init
init_per_zone_pages_min(void)
4180 unsigned long lowmem_kbytes
;
4182 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4184 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4185 if (min_free_kbytes
< 128)
4186 min_free_kbytes
= 128;
4187 if (min_free_kbytes
> 65536)
4188 min_free_kbytes
= 65536;
4189 setup_per_zone_pages_min();
4190 setup_per_zone_lowmem_reserve();
4193 module_init(init_per_zone_pages_min
)
4196 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4197 * that we can call two helper functions whenever min_free_kbytes
4200 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4201 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4203 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4205 setup_per_zone_pages_min();
4210 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4211 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4216 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4221 zone
->min_unmapped_pages
= (zone
->present_pages
*
4222 sysctl_min_unmapped_ratio
) / 100;
4226 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4227 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4232 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4237 zone
->min_slab_pages
= (zone
->present_pages
*
4238 sysctl_min_slab_ratio
) / 100;
4244 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4245 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4246 * whenever sysctl_lowmem_reserve_ratio changes.
4248 * The reserve ratio obviously has absolutely no relation with the
4249 * pages_min watermarks. The lowmem reserve ratio can only make sense
4250 * if in function of the boot time zone sizes.
4252 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4253 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4255 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4256 setup_per_zone_lowmem_reserve();
4261 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4262 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4263 * can have before it gets flushed back to buddy allocator.
4266 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4267 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4273 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4274 if (!write
|| (ret
== -EINVAL
))
4276 for_each_zone(zone
) {
4277 for_each_online_cpu(cpu
) {
4279 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4280 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4286 int hashdist
= HASHDIST_DEFAULT
;
4289 static int __init
set_hashdist(char *str
)
4293 hashdist
= simple_strtoul(str
, &str
, 0);
4296 __setup("hashdist=", set_hashdist
);
4300 * allocate a large system hash table from bootmem
4301 * - it is assumed that the hash table must contain an exact power-of-2
4302 * quantity of entries
4303 * - limit is the number of hash buckets, not the total allocation size
4305 void *__init
alloc_large_system_hash(const char *tablename
,
4306 unsigned long bucketsize
,
4307 unsigned long numentries
,
4310 unsigned int *_hash_shift
,
4311 unsigned int *_hash_mask
,
4312 unsigned long limit
)
4314 unsigned long long max
= limit
;
4315 unsigned long log2qty
, size
;
4318 /* allow the kernel cmdline to have a say */
4320 /* round applicable memory size up to nearest megabyte */
4321 numentries
= nr_kernel_pages
;
4322 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4323 numentries
>>= 20 - PAGE_SHIFT
;
4324 numentries
<<= 20 - PAGE_SHIFT
;
4326 /* limit to 1 bucket per 2^scale bytes of low memory */
4327 if (scale
> PAGE_SHIFT
)
4328 numentries
>>= (scale
- PAGE_SHIFT
);
4330 numentries
<<= (PAGE_SHIFT
- scale
);
4332 /* Make sure we've got at least a 0-order allocation.. */
4333 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4334 numentries
= PAGE_SIZE
/ bucketsize
;
4336 numentries
= roundup_pow_of_two(numentries
);
4338 /* limit allocation size to 1/16 total memory by default */
4340 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4341 do_div(max
, bucketsize
);
4344 if (numentries
> max
)
4347 log2qty
= ilog2(numentries
);
4350 size
= bucketsize
<< log2qty
;
4351 if (flags
& HASH_EARLY
)
4352 table
= alloc_bootmem(size
);
4354 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4356 unsigned long order
= get_order(size
);
4357 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4359 * If bucketsize is not a power-of-two, we may free
4360 * some pages at the end of hash table.
4363 unsigned long alloc_end
= (unsigned long)table
+
4364 (PAGE_SIZE
<< order
);
4365 unsigned long used
= (unsigned long)table
+
4367 split_page(virt_to_page(table
), order
);
4368 while (used
< alloc_end
) {
4374 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4377 panic("Failed to allocate %s hash table\n", tablename
);
4379 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4382 ilog2(size
) - PAGE_SHIFT
,
4386 *_hash_shift
= log2qty
;
4388 *_hash_mask
= (1 << log2qty
) - 1;
4393 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4394 struct page
*pfn_to_page(unsigned long pfn
)
4396 return __pfn_to_page(pfn
);
4398 unsigned long page_to_pfn(struct page
*page
)
4400 return __page_to_pfn(page
);
4402 EXPORT_SYMBOL(pfn_to_page
);
4403 EXPORT_SYMBOL(page_to_pfn
);
4404 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4406 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4407 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4410 #ifdef CONFIG_SPARSEMEM
4411 return __pfn_to_section(pfn
)->pageblock_flags
;
4413 return zone
->pageblock_flags
;
4414 #endif /* CONFIG_SPARSEMEM */
4417 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4419 #ifdef CONFIG_SPARSEMEM
4420 pfn
&= (PAGES_PER_SECTION
-1);
4421 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4423 pfn
= pfn
- zone
->zone_start_pfn
;
4424 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4425 #endif /* CONFIG_SPARSEMEM */
4429 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4430 * @page: The page within the block of interest
4431 * @start_bitidx: The first bit of interest to retrieve
4432 * @end_bitidx: The last bit of interest
4433 * returns pageblock_bits flags
4435 unsigned long get_pageblock_flags_group(struct page
*page
,
4436 int start_bitidx
, int end_bitidx
)
4439 unsigned long *bitmap
;
4440 unsigned long pfn
, bitidx
;
4441 unsigned long flags
= 0;
4442 unsigned long value
= 1;
4444 zone
= page_zone(page
);
4445 pfn
= page_to_pfn(page
);
4446 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4447 bitidx
= pfn_to_bitidx(zone
, pfn
);
4449 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4450 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4457 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4458 * @page: The page within the block of interest
4459 * @start_bitidx: The first bit of interest
4460 * @end_bitidx: The last bit of interest
4461 * @flags: The flags to set
4463 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4464 int start_bitidx
, int end_bitidx
)
4467 unsigned long *bitmap
;
4468 unsigned long pfn
, bitidx
;
4469 unsigned long value
= 1;
4471 zone
= page_zone(page
);
4472 pfn
= page_to_pfn(page
);
4473 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4474 bitidx
= pfn_to_bitidx(zone
, pfn
);
4475 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4476 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4478 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4480 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4482 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4486 * This is designed as sub function...plz see page_isolation.c also.
4487 * set/clear page block's type to be ISOLATE.
4488 * page allocater never alloc memory from ISOLATE block.
4491 int set_migratetype_isolate(struct page
*page
)
4494 unsigned long flags
;
4497 zone
= page_zone(page
);
4498 spin_lock_irqsave(&zone
->lock
, flags
);
4500 * In future, more migrate types will be able to be isolation target.
4502 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4504 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4505 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4508 spin_unlock_irqrestore(&zone
->lock
, flags
);
4514 void unset_migratetype_isolate(struct page
*page
)
4517 unsigned long flags
;
4518 zone
= page_zone(page
);
4519 spin_lock_irqsave(&zone
->lock
, flags
);
4520 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4522 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4523 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4525 spin_unlock_irqrestore(&zone
->lock
, flags
);
4528 #ifdef CONFIG_MEMORY_HOTREMOVE
4530 * All pages in the range must be isolated before calling this.
4533 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4539 unsigned long flags
;
4540 /* find the first valid pfn */
4541 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4546 zone
= page_zone(pfn_to_page(pfn
));
4547 spin_lock_irqsave(&zone
->lock
, flags
);
4549 while (pfn
< end_pfn
) {
4550 if (!pfn_valid(pfn
)) {
4554 page
= pfn_to_page(pfn
);
4555 BUG_ON(page_count(page
));
4556 BUG_ON(!PageBuddy(page
));
4557 order
= page_order(page
);
4558 #ifdef CONFIG_DEBUG_VM
4559 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4560 pfn
, 1 << order
, end_pfn
);
4562 list_del(&page
->lru
);
4563 rmv_page_order(page
);
4564 zone
->free_area
[order
].nr_free
--;
4565 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4567 for (i
= 0; i
< (1 << order
); i
++)
4568 SetPageReserved((page
+i
));
4569 pfn
+= (1 << order
);
4571 spin_unlock_irqrestore(&zone
->lock
, flags
);