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 if (!pfn_valid_within(page_to_pfn(page
))) {
701 if (!PageBuddy(page
)) {
706 order
= page_order(page
);
707 list_del(&page
->lru
);
709 &zone
->free_area
[order
].free_list
[migratetype
]);
711 pages_moved
+= 1 << order
;
717 int move_freepages_block(struct zone
*zone
, struct page
*page
, int migratetype
)
719 unsigned long start_pfn
, end_pfn
;
720 struct page
*start_page
, *end_page
;
722 start_pfn
= page_to_pfn(page
);
723 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
724 start_page
= pfn_to_page(start_pfn
);
725 end_page
= start_page
+ pageblock_nr_pages
- 1;
726 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
728 /* Do not cross zone boundaries */
729 if (start_pfn
< zone
->zone_start_pfn
)
731 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
734 return move_freepages(zone
, start_page
, end_page
, migratetype
);
737 /* Remove an element from the buddy allocator from the fallback list */
738 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
739 int start_migratetype
)
741 struct free_area
* area
;
746 /* Find the largest possible block of pages in the other list */
747 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
749 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
750 migratetype
= fallbacks
[start_migratetype
][i
];
752 /* MIGRATE_RESERVE handled later if necessary */
753 if (migratetype
== MIGRATE_RESERVE
)
756 area
= &(zone
->free_area
[current_order
]);
757 if (list_empty(&area
->free_list
[migratetype
]))
760 page
= list_entry(area
->free_list
[migratetype
].next
,
765 * If breaking a large block of pages, move all free
766 * pages to the preferred allocation list. If falling
767 * back for a reclaimable kernel allocation, be more
768 * agressive about taking ownership of free pages
770 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
771 start_migratetype
== MIGRATE_RECLAIMABLE
) {
773 pages
= move_freepages_block(zone
, page
,
776 /* Claim the whole block if over half of it is free */
777 if (pages
>= (1 << (pageblock_order
-1)))
778 set_pageblock_migratetype(page
,
781 migratetype
= start_migratetype
;
784 /* Remove the page from the freelists */
785 list_del(&page
->lru
);
786 rmv_page_order(page
);
787 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
790 if (current_order
== pageblock_order
)
791 set_pageblock_migratetype(page
,
794 expand(zone
, page
, order
, current_order
, area
, migratetype
);
799 /* Use MIGRATE_RESERVE rather than fail an allocation */
800 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
804 * Do the hard work of removing an element from the buddy allocator.
805 * Call me with the zone->lock already held.
807 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
812 page
= __rmqueue_smallest(zone
, order
, migratetype
);
815 page
= __rmqueue_fallback(zone
, order
, migratetype
);
821 * Obtain a specified number of elements from the buddy allocator, all under
822 * a single hold of the lock, for efficiency. Add them to the supplied list.
823 * Returns the number of new pages which were placed at *list.
825 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
826 unsigned long count
, struct list_head
*list
,
831 spin_lock(&zone
->lock
);
832 for (i
= 0; i
< count
; ++i
) {
833 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
834 if (unlikely(page
== NULL
))
838 * Split buddy pages returned by expand() are received here
839 * in physical page order. The page is added to the callers and
840 * list and the list head then moves forward. From the callers
841 * perspective, the linked list is ordered by page number in
842 * some conditions. This is useful for IO devices that can
843 * merge IO requests if the physical pages are ordered
846 list_add(&page
->lru
, list
);
847 set_page_private(page
, migratetype
);
850 spin_unlock(&zone
->lock
);
856 * Called from the vmstat counter updater to drain pagesets of this
857 * currently executing processor on remote nodes after they have
860 * Note that this function must be called with the thread pinned to
861 * a single processor.
863 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
868 local_irq_save(flags
);
869 if (pcp
->count
>= pcp
->batch
)
870 to_drain
= pcp
->batch
;
872 to_drain
= pcp
->count
;
873 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
874 pcp
->count
-= to_drain
;
875 local_irq_restore(flags
);
880 * Drain pages of the indicated processor.
882 * The processor must either be the current processor and the
883 * thread pinned to the current processor or a processor that
886 static void drain_pages(unsigned int cpu
)
891 for_each_zone(zone
) {
892 struct per_cpu_pageset
*pset
;
893 struct per_cpu_pages
*pcp
;
895 if (!populated_zone(zone
))
898 pset
= zone_pcp(zone
, cpu
);
901 local_irq_save(flags
);
902 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
904 local_irq_restore(flags
);
909 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
911 void drain_local_pages(void *arg
)
913 drain_pages(smp_processor_id());
917 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
919 void drain_all_pages(void)
921 on_each_cpu(drain_local_pages
, NULL
, 1);
924 #ifdef CONFIG_HIBERNATION
926 void mark_free_pages(struct zone
*zone
)
928 unsigned long pfn
, max_zone_pfn
;
931 struct list_head
*curr
;
933 if (!zone
->spanned_pages
)
936 spin_lock_irqsave(&zone
->lock
, flags
);
938 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
939 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
940 if (pfn_valid(pfn
)) {
941 struct page
*page
= pfn_to_page(pfn
);
943 if (!swsusp_page_is_forbidden(page
))
944 swsusp_unset_page_free(page
);
947 for_each_migratetype_order(order
, t
) {
948 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
951 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
952 for (i
= 0; i
< (1UL << order
); i
++)
953 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
956 spin_unlock_irqrestore(&zone
->lock
, flags
);
958 #endif /* CONFIG_PM */
961 * Free a 0-order page
963 static void free_hot_cold_page(struct page
*page
, int cold
)
965 struct zone
*zone
= page_zone(page
);
966 struct per_cpu_pages
*pcp
;
970 page
->mapping
= NULL
;
971 if (free_pages_check(page
))
974 if (!PageHighMem(page
)) {
975 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
976 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
978 arch_free_page(page
, 0);
979 kernel_map_pages(page
, 1, 0);
981 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
982 local_irq_save(flags
);
983 __count_vm_event(PGFREE
);
985 list_add_tail(&page
->lru
, &pcp
->list
);
987 list_add(&page
->lru
, &pcp
->list
);
988 set_page_private(page
, get_pageblock_migratetype(page
));
990 if (pcp
->count
>= pcp
->high
) {
991 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
992 pcp
->count
-= pcp
->batch
;
994 local_irq_restore(flags
);
998 void free_hot_page(struct page
*page
)
1000 free_hot_cold_page(page
, 0);
1003 void free_cold_page(struct page
*page
)
1005 free_hot_cold_page(page
, 1);
1009 * split_page takes a non-compound higher-order page, and splits it into
1010 * n (1<<order) sub-pages: page[0..n]
1011 * Each sub-page must be freed individually.
1013 * Note: this is probably too low level an operation for use in drivers.
1014 * Please consult with lkml before using this in your driver.
1016 void split_page(struct page
*page
, unsigned int order
)
1020 VM_BUG_ON(PageCompound(page
));
1021 VM_BUG_ON(!page_count(page
));
1022 for (i
= 1; i
< (1 << order
); i
++)
1023 set_page_refcounted(page
+ i
);
1027 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1028 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1031 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1032 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1034 unsigned long flags
;
1036 int cold
= !!(gfp_flags
& __GFP_COLD
);
1038 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1042 if (likely(order
== 0)) {
1043 struct per_cpu_pages
*pcp
;
1045 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1046 local_irq_save(flags
);
1048 pcp
->count
= rmqueue_bulk(zone
, 0,
1049 pcp
->batch
, &pcp
->list
, migratetype
);
1050 if (unlikely(!pcp
->count
))
1054 /* Find a page of the appropriate migrate type */
1056 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1057 if (page_private(page
) == migratetype
)
1060 list_for_each_entry(page
, &pcp
->list
, lru
)
1061 if (page_private(page
) == migratetype
)
1065 /* Allocate more to the pcp list if necessary */
1066 if (unlikely(&page
->lru
== &pcp
->list
)) {
1067 pcp
->count
+= rmqueue_bulk(zone
, 0,
1068 pcp
->batch
, &pcp
->list
, migratetype
);
1069 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1072 list_del(&page
->lru
);
1075 spin_lock_irqsave(&zone
->lock
, flags
);
1076 page
= __rmqueue(zone
, order
, migratetype
);
1077 spin_unlock(&zone
->lock
);
1082 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1083 zone_statistics(preferred_zone
, zone
);
1084 local_irq_restore(flags
);
1087 VM_BUG_ON(bad_range(zone
, page
));
1088 if (prep_new_page(page
, order
, gfp_flags
))
1093 local_irq_restore(flags
);
1098 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1099 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1100 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1101 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1102 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1103 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1104 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1106 #ifdef CONFIG_FAIL_PAGE_ALLOC
1108 static struct fail_page_alloc_attr
{
1109 struct fault_attr attr
;
1111 u32 ignore_gfp_highmem
;
1112 u32 ignore_gfp_wait
;
1115 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1117 struct dentry
*ignore_gfp_highmem_file
;
1118 struct dentry
*ignore_gfp_wait_file
;
1119 struct dentry
*min_order_file
;
1121 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1123 } fail_page_alloc
= {
1124 .attr
= FAULT_ATTR_INITIALIZER
,
1125 .ignore_gfp_wait
= 1,
1126 .ignore_gfp_highmem
= 1,
1130 static int __init
setup_fail_page_alloc(char *str
)
1132 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1134 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1136 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1138 if (order
< fail_page_alloc
.min_order
)
1140 if (gfp_mask
& __GFP_NOFAIL
)
1142 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1144 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1147 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1150 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1152 static int __init
fail_page_alloc_debugfs(void)
1154 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1158 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1162 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1164 fail_page_alloc
.ignore_gfp_wait_file
=
1165 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1166 &fail_page_alloc
.ignore_gfp_wait
);
1168 fail_page_alloc
.ignore_gfp_highmem_file
=
1169 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1170 &fail_page_alloc
.ignore_gfp_highmem
);
1171 fail_page_alloc
.min_order_file
=
1172 debugfs_create_u32("min-order", mode
, dir
,
1173 &fail_page_alloc
.min_order
);
1175 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1176 !fail_page_alloc
.ignore_gfp_highmem_file
||
1177 !fail_page_alloc
.min_order_file
) {
1179 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1180 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1181 debugfs_remove(fail_page_alloc
.min_order_file
);
1182 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1188 late_initcall(fail_page_alloc_debugfs
);
1190 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1192 #else /* CONFIG_FAIL_PAGE_ALLOC */
1194 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1199 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1202 * Return 1 if free pages are above 'mark'. This takes into account the order
1203 * of the allocation.
1205 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1206 int classzone_idx
, int alloc_flags
)
1208 /* free_pages my go negative - that's OK */
1210 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1213 if (alloc_flags
& ALLOC_HIGH
)
1215 if (alloc_flags
& ALLOC_HARDER
)
1218 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1220 for (o
= 0; o
< order
; o
++) {
1221 /* At the next order, this order's pages become unavailable */
1222 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1224 /* Require fewer higher order pages to be free */
1227 if (free_pages
<= min
)
1235 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1236 * skip over zones that are not allowed by the cpuset, or that have
1237 * been recently (in last second) found to be nearly full. See further
1238 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1239 * that have to skip over a lot of full or unallowed zones.
1241 * If the zonelist cache is present in the passed in zonelist, then
1242 * returns a pointer to the allowed node mask (either the current
1243 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1245 * If the zonelist cache is not available for this zonelist, does
1246 * nothing and returns NULL.
1248 * If the fullzones BITMAP in the zonelist cache is stale (more than
1249 * a second since last zap'd) then we zap it out (clear its bits.)
1251 * We hold off even calling zlc_setup, until after we've checked the
1252 * first zone in the zonelist, on the theory that most allocations will
1253 * be satisfied from that first zone, so best to examine that zone as
1254 * quickly as we can.
1256 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1258 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1259 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1261 zlc
= zonelist
->zlcache_ptr
;
1265 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1266 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1267 zlc
->last_full_zap
= jiffies
;
1270 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1271 &cpuset_current_mems_allowed
:
1272 &node_states
[N_HIGH_MEMORY
];
1273 return allowednodes
;
1277 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1278 * if it is worth looking at further for free memory:
1279 * 1) Check that the zone isn't thought to be full (doesn't have its
1280 * bit set in the zonelist_cache fullzones BITMAP).
1281 * 2) Check that the zones node (obtained from the zonelist_cache
1282 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1283 * Return true (non-zero) if zone is worth looking at further, or
1284 * else return false (zero) if it is not.
1286 * This check -ignores- the distinction between various watermarks,
1287 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1288 * found to be full for any variation of these watermarks, it will
1289 * be considered full for up to one second by all requests, unless
1290 * we are so low on memory on all allowed nodes that we are forced
1291 * into the second scan of the zonelist.
1293 * In the second scan we ignore this zonelist cache and exactly
1294 * apply the watermarks to all zones, even it is slower to do so.
1295 * We are low on memory in the second scan, and should leave no stone
1296 * unturned looking for a free page.
1298 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1299 nodemask_t
*allowednodes
)
1301 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1302 int i
; /* index of *z in zonelist zones */
1303 int n
; /* node that zone *z is on */
1305 zlc
= zonelist
->zlcache_ptr
;
1309 i
= z
- zonelist
->_zonerefs
;
1312 /* This zone is worth trying if it is allowed but not full */
1313 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1317 * Given 'z' scanning a zonelist, set the corresponding bit in
1318 * zlc->fullzones, so that subsequent attempts to allocate a page
1319 * from that zone don't waste time re-examining it.
1321 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1323 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1324 int i
; /* index of *z in zonelist zones */
1326 zlc
= zonelist
->zlcache_ptr
;
1330 i
= z
- zonelist
->_zonerefs
;
1332 set_bit(i
, zlc
->fullzones
);
1335 #else /* CONFIG_NUMA */
1337 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1342 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1343 nodemask_t
*allowednodes
)
1348 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1351 #endif /* CONFIG_NUMA */
1354 * get_page_from_freelist goes through the zonelist trying to allocate
1357 static struct page
*
1358 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1359 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1362 struct page
*page
= NULL
;
1364 struct zone
*zone
, *preferred_zone
;
1365 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1366 int zlc_active
= 0; /* set if using zonelist_cache */
1367 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1369 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1371 if (!preferred_zone
)
1374 classzone_idx
= zone_idx(preferred_zone
);
1378 * Scan zonelist, looking for a zone with enough free.
1379 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1381 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1382 high_zoneidx
, nodemask
) {
1383 if (NUMA_BUILD
&& zlc_active
&&
1384 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1386 if ((alloc_flags
& ALLOC_CPUSET
) &&
1387 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1390 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1392 if (alloc_flags
& ALLOC_WMARK_MIN
)
1393 mark
= zone
->pages_min
;
1394 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1395 mark
= zone
->pages_low
;
1397 mark
= zone
->pages_high
;
1398 if (!zone_watermark_ok(zone
, order
, mark
,
1399 classzone_idx
, alloc_flags
)) {
1400 if (!zone_reclaim_mode
||
1401 !zone_reclaim(zone
, gfp_mask
, order
))
1402 goto this_zone_full
;
1406 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1411 zlc_mark_zone_full(zonelist
, z
);
1413 if (NUMA_BUILD
&& !did_zlc_setup
) {
1414 /* we do zlc_setup after the first zone is tried */
1415 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1421 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1422 /* Disable zlc cache for second zonelist scan */
1430 * This is the 'heart' of the zoned buddy allocator.
1433 __alloc_pages_internal(gfp_t gfp_mask
, unsigned int order
,
1434 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1436 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1437 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1441 struct reclaim_state reclaim_state
;
1442 struct task_struct
*p
= current
;
1445 unsigned long did_some_progress
;
1446 unsigned long pages_reclaimed
= 0;
1448 might_sleep_if(wait
);
1450 if (should_fail_alloc_page(gfp_mask
, order
))
1454 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1456 if (unlikely(!z
->zone
)) {
1458 * Happens if we have an empty zonelist as a result of
1459 * GFP_THISNODE being used on a memoryless node
1464 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1465 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1470 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1471 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1472 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1473 * using a larger set of nodes after it has established that the
1474 * allowed per node queues are empty and that nodes are
1477 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1480 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1481 wakeup_kswapd(zone
, order
);
1484 * OK, we're below the kswapd watermark and have kicked background
1485 * reclaim. Now things get more complex, so set up alloc_flags according
1486 * to how we want to proceed.
1488 * The caller may dip into page reserves a bit more if the caller
1489 * cannot run direct reclaim, or if the caller has realtime scheduling
1490 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1491 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1493 alloc_flags
= ALLOC_WMARK_MIN
;
1494 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1495 alloc_flags
|= ALLOC_HARDER
;
1496 if (gfp_mask
& __GFP_HIGH
)
1497 alloc_flags
|= ALLOC_HIGH
;
1499 alloc_flags
|= ALLOC_CPUSET
;
1502 * Go through the zonelist again. Let __GFP_HIGH and allocations
1503 * coming from realtime tasks go deeper into reserves.
1505 * This is the last chance, in general, before the goto nopage.
1506 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1507 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1509 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1510 high_zoneidx
, alloc_flags
);
1514 /* This allocation should allow future memory freeing. */
1517 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1518 && !in_interrupt()) {
1519 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1521 /* go through the zonelist yet again, ignoring mins */
1522 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1523 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1526 if (gfp_mask
& __GFP_NOFAIL
) {
1527 congestion_wait(WRITE
, HZ
/50);
1534 /* Atomic allocations - we can't balance anything */
1540 /* We now go into synchronous reclaim */
1541 cpuset_memory_pressure_bump();
1542 p
->flags
|= PF_MEMALLOC
;
1543 reclaim_state
.reclaimed_slab
= 0;
1544 p
->reclaim_state
= &reclaim_state
;
1546 did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
);
1548 p
->reclaim_state
= NULL
;
1549 p
->flags
&= ~PF_MEMALLOC
;
1556 if (likely(did_some_progress
)) {
1557 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1558 zonelist
, high_zoneidx
, alloc_flags
);
1561 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1562 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1563 schedule_timeout_uninterruptible(1);
1568 * Go through the zonelist yet one more time, keep
1569 * very high watermark here, this is only to catch
1570 * a parallel oom killing, we must fail if we're still
1571 * under heavy pressure.
1573 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1574 order
, zonelist
, high_zoneidx
,
1575 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1577 clear_zonelist_oom(zonelist
, gfp_mask
);
1581 /* The OOM killer will not help higher order allocs so fail */
1582 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1583 clear_zonelist_oom(zonelist
, gfp_mask
);
1587 out_of_memory(zonelist
, gfp_mask
, order
);
1588 clear_zonelist_oom(zonelist
, gfp_mask
);
1593 * Don't let big-order allocations loop unless the caller explicitly
1594 * requests that. Wait for some write requests to complete then retry.
1596 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1597 * means __GFP_NOFAIL, but that may not be true in other
1600 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1601 * specified, then we retry until we no longer reclaim any pages
1602 * (above), or we've reclaimed an order of pages at least as
1603 * large as the allocation's order. In both cases, if the
1604 * allocation still fails, we stop retrying.
1606 pages_reclaimed
+= did_some_progress
;
1608 if (!(gfp_mask
& __GFP_NORETRY
)) {
1609 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1612 if (gfp_mask
& __GFP_REPEAT
&&
1613 pages_reclaimed
< (1 << order
))
1616 if (gfp_mask
& __GFP_NOFAIL
)
1620 congestion_wait(WRITE
, HZ
/50);
1625 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1626 printk(KERN_WARNING
"%s: page allocation failure."
1627 " order:%d, mode:0x%x\n",
1628 p
->comm
, order
, gfp_mask
);
1635 EXPORT_SYMBOL(__alloc_pages_internal
);
1638 * Common helper functions.
1640 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1643 page
= alloc_pages(gfp_mask
, order
);
1646 return (unsigned long) page_address(page
);
1649 EXPORT_SYMBOL(__get_free_pages
);
1651 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1656 * get_zeroed_page() returns a 32-bit address, which cannot represent
1659 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1661 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1663 return (unsigned long) page_address(page
);
1667 EXPORT_SYMBOL(get_zeroed_page
);
1669 void __pagevec_free(struct pagevec
*pvec
)
1671 int i
= pagevec_count(pvec
);
1674 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1677 void __free_pages(struct page
*page
, unsigned int order
)
1679 if (put_page_testzero(page
)) {
1681 free_hot_page(page
);
1683 __free_pages_ok(page
, order
);
1687 EXPORT_SYMBOL(__free_pages
);
1689 void free_pages(unsigned long addr
, unsigned int order
)
1692 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1693 __free_pages(virt_to_page((void *)addr
), order
);
1697 EXPORT_SYMBOL(free_pages
);
1699 static unsigned int nr_free_zone_pages(int offset
)
1704 /* Just pick one node, since fallback list is circular */
1705 unsigned int sum
= 0;
1707 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1709 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1710 unsigned long size
= zone
->present_pages
;
1711 unsigned long high
= zone
->pages_high
;
1720 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1722 unsigned int nr_free_buffer_pages(void)
1724 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1726 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1729 * Amount of free RAM allocatable within all zones
1731 unsigned int nr_free_pagecache_pages(void)
1733 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1736 static inline void show_node(struct zone
*zone
)
1739 printk("Node %d ", zone_to_nid(zone
));
1742 void si_meminfo(struct sysinfo
*val
)
1744 val
->totalram
= totalram_pages
;
1746 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1747 val
->bufferram
= nr_blockdev_pages();
1748 val
->totalhigh
= totalhigh_pages
;
1749 val
->freehigh
= nr_free_highpages();
1750 val
->mem_unit
= PAGE_SIZE
;
1753 EXPORT_SYMBOL(si_meminfo
);
1756 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1758 pg_data_t
*pgdat
= NODE_DATA(nid
);
1760 val
->totalram
= pgdat
->node_present_pages
;
1761 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1762 #ifdef CONFIG_HIGHMEM
1763 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1764 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1770 val
->mem_unit
= PAGE_SIZE
;
1774 #define K(x) ((x) << (PAGE_SHIFT-10))
1777 * Show free area list (used inside shift_scroll-lock stuff)
1778 * We also calculate the percentage fragmentation. We do this by counting the
1779 * memory on each free list with the exception of the first item on the list.
1781 void show_free_areas(void)
1786 for_each_zone(zone
) {
1787 if (!populated_zone(zone
))
1791 printk("%s per-cpu:\n", zone
->name
);
1793 for_each_online_cpu(cpu
) {
1794 struct per_cpu_pageset
*pageset
;
1796 pageset
= zone_pcp(zone
, cpu
);
1798 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1799 cpu
, pageset
->pcp
.high
,
1800 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1804 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1805 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1806 global_page_state(NR_ACTIVE
),
1807 global_page_state(NR_INACTIVE
),
1808 global_page_state(NR_FILE_DIRTY
),
1809 global_page_state(NR_WRITEBACK
),
1810 global_page_state(NR_UNSTABLE_NFS
),
1811 global_page_state(NR_FREE_PAGES
),
1812 global_page_state(NR_SLAB_RECLAIMABLE
) +
1813 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1814 global_page_state(NR_FILE_MAPPED
),
1815 global_page_state(NR_PAGETABLE
),
1816 global_page_state(NR_BOUNCE
));
1818 for_each_zone(zone
) {
1821 if (!populated_zone(zone
))
1833 " pages_scanned:%lu"
1834 " all_unreclaimable? %s"
1837 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1840 K(zone
->pages_high
),
1841 K(zone_page_state(zone
, NR_ACTIVE
)),
1842 K(zone_page_state(zone
, NR_INACTIVE
)),
1843 K(zone
->present_pages
),
1844 zone
->pages_scanned
,
1845 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1847 printk("lowmem_reserve[]:");
1848 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1849 printk(" %lu", zone
->lowmem_reserve
[i
]);
1853 for_each_zone(zone
) {
1854 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1856 if (!populated_zone(zone
))
1860 printk("%s: ", zone
->name
);
1862 spin_lock_irqsave(&zone
->lock
, flags
);
1863 for (order
= 0; order
< MAX_ORDER
; order
++) {
1864 nr
[order
] = zone
->free_area
[order
].nr_free
;
1865 total
+= nr
[order
] << order
;
1867 spin_unlock_irqrestore(&zone
->lock
, flags
);
1868 for (order
= 0; order
< MAX_ORDER
; order
++)
1869 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1870 printk("= %lukB\n", K(total
));
1873 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
1875 show_swap_cache_info();
1878 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
1880 zoneref
->zone
= zone
;
1881 zoneref
->zone_idx
= zone_idx(zone
);
1885 * Builds allocation fallback zone lists.
1887 * Add all populated zones of a node to the zonelist.
1889 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1890 int nr_zones
, enum zone_type zone_type
)
1894 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1899 zone
= pgdat
->node_zones
+ zone_type
;
1900 if (populated_zone(zone
)) {
1901 zoneref_set_zone(zone
,
1902 &zonelist
->_zonerefs
[nr_zones
++]);
1903 check_highest_zone(zone_type
);
1906 } while (zone_type
);
1913 * 0 = automatic detection of better ordering.
1914 * 1 = order by ([node] distance, -zonetype)
1915 * 2 = order by (-zonetype, [node] distance)
1917 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1918 * the same zonelist. So only NUMA can configure this param.
1920 #define ZONELIST_ORDER_DEFAULT 0
1921 #define ZONELIST_ORDER_NODE 1
1922 #define ZONELIST_ORDER_ZONE 2
1924 /* zonelist order in the kernel.
1925 * set_zonelist_order() will set this to NODE or ZONE.
1927 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1928 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
1932 /* The value user specified ....changed by config */
1933 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1934 /* string for sysctl */
1935 #define NUMA_ZONELIST_ORDER_LEN 16
1936 char numa_zonelist_order
[16] = "default";
1939 * interface for configure zonelist ordering.
1940 * command line option "numa_zonelist_order"
1941 * = "[dD]efault - default, automatic configuration.
1942 * = "[nN]ode - order by node locality, then by zone within node
1943 * = "[zZ]one - order by zone, then by locality within zone
1946 static int __parse_numa_zonelist_order(char *s
)
1948 if (*s
== 'd' || *s
== 'D') {
1949 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1950 } else if (*s
== 'n' || *s
== 'N') {
1951 user_zonelist_order
= ZONELIST_ORDER_NODE
;
1952 } else if (*s
== 'z' || *s
== 'Z') {
1953 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
1956 "Ignoring invalid numa_zonelist_order value: "
1963 static __init
int setup_numa_zonelist_order(char *s
)
1966 return __parse_numa_zonelist_order(s
);
1969 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
1972 * sysctl handler for numa_zonelist_order
1974 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
1975 struct file
*file
, void __user
*buffer
, size_t *length
,
1978 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
1982 strncpy(saved_string
, (char*)table
->data
,
1983 NUMA_ZONELIST_ORDER_LEN
);
1984 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
1988 int oldval
= user_zonelist_order
;
1989 if (__parse_numa_zonelist_order((char*)table
->data
)) {
1991 * bogus value. restore saved string
1993 strncpy((char*)table
->data
, saved_string
,
1994 NUMA_ZONELIST_ORDER_LEN
);
1995 user_zonelist_order
= oldval
;
1996 } else if (oldval
!= user_zonelist_order
)
1997 build_all_zonelists();
2003 #define MAX_NODE_LOAD (num_online_nodes())
2004 static int node_load
[MAX_NUMNODES
];
2007 * find_next_best_node - find the next node that should appear in a given node's fallback list
2008 * @node: node whose fallback list we're appending
2009 * @used_node_mask: nodemask_t of already used nodes
2011 * We use a number of factors to determine which is the next node that should
2012 * appear on a given node's fallback list. The node should not have appeared
2013 * already in @node's fallback list, and it should be the next closest node
2014 * according to the distance array (which contains arbitrary distance values
2015 * from each node to each node in the system), and should also prefer nodes
2016 * with no CPUs, since presumably they'll have very little allocation pressure
2017 * on them otherwise.
2018 * It returns -1 if no node is found.
2020 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2023 int min_val
= INT_MAX
;
2025 node_to_cpumask_ptr(tmp
, 0);
2027 /* Use the local node if we haven't already */
2028 if (!node_isset(node
, *used_node_mask
)) {
2029 node_set(node
, *used_node_mask
);
2033 for_each_node_state(n
, N_HIGH_MEMORY
) {
2035 /* Don't want a node to appear more than once */
2036 if (node_isset(n
, *used_node_mask
))
2039 /* Use the distance array to find the distance */
2040 val
= node_distance(node
, n
);
2042 /* Penalize nodes under us ("prefer the next node") */
2045 /* Give preference to headless and unused nodes */
2046 node_to_cpumask_ptr_next(tmp
, n
);
2047 if (!cpus_empty(*tmp
))
2048 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2050 /* Slight preference for less loaded node */
2051 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2052 val
+= node_load
[n
];
2054 if (val
< min_val
) {
2061 node_set(best_node
, *used_node_mask
);
2068 * Build zonelists ordered by node and zones within node.
2069 * This results in maximum locality--normal zone overflows into local
2070 * DMA zone, if any--but risks exhausting DMA zone.
2072 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2075 struct zonelist
*zonelist
;
2077 zonelist
= &pgdat
->node_zonelists
[0];
2078 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2080 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2082 zonelist
->_zonerefs
[j
].zone
= NULL
;
2083 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2087 * Build gfp_thisnode zonelists
2089 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2092 struct zonelist
*zonelist
;
2094 zonelist
= &pgdat
->node_zonelists
[1];
2095 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2096 zonelist
->_zonerefs
[j
].zone
= NULL
;
2097 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2101 * Build zonelists ordered by zone and nodes within zones.
2102 * This results in conserving DMA zone[s] until all Normal memory is
2103 * exhausted, but results in overflowing to remote node while memory
2104 * may still exist in local DMA zone.
2106 static int node_order
[MAX_NUMNODES
];
2108 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2111 int zone_type
; /* needs to be signed */
2113 struct zonelist
*zonelist
;
2115 zonelist
= &pgdat
->node_zonelists
[0];
2117 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2118 for (j
= 0; j
< nr_nodes
; j
++) {
2119 node
= node_order
[j
];
2120 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2121 if (populated_zone(z
)) {
2123 &zonelist
->_zonerefs
[pos
++]);
2124 check_highest_zone(zone_type
);
2128 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2129 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2132 static int default_zonelist_order(void)
2135 unsigned long low_kmem_size
,total_size
;
2139 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2140 * If they are really small and used heavily, the system can fall
2141 * into OOM very easily.
2142 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2144 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2147 for_each_online_node(nid
) {
2148 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2149 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2150 if (populated_zone(z
)) {
2151 if (zone_type
< ZONE_NORMAL
)
2152 low_kmem_size
+= z
->present_pages
;
2153 total_size
+= z
->present_pages
;
2157 if (!low_kmem_size
|| /* there are no DMA area. */
2158 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2159 return ZONELIST_ORDER_NODE
;
2161 * look into each node's config.
2162 * If there is a node whose DMA/DMA32 memory is very big area on
2163 * local memory, NODE_ORDER may be suitable.
2165 average_size
= total_size
/
2166 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2167 for_each_online_node(nid
) {
2170 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2171 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2172 if (populated_zone(z
)) {
2173 if (zone_type
< ZONE_NORMAL
)
2174 low_kmem_size
+= z
->present_pages
;
2175 total_size
+= z
->present_pages
;
2178 if (low_kmem_size
&&
2179 total_size
> average_size
&& /* ignore small node */
2180 low_kmem_size
> total_size
* 70/100)
2181 return ZONELIST_ORDER_NODE
;
2183 return ZONELIST_ORDER_ZONE
;
2186 static void set_zonelist_order(void)
2188 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2189 current_zonelist_order
= default_zonelist_order();
2191 current_zonelist_order
= user_zonelist_order
;
2194 static void build_zonelists(pg_data_t
*pgdat
)
2198 nodemask_t used_mask
;
2199 int local_node
, prev_node
;
2200 struct zonelist
*zonelist
;
2201 int order
= current_zonelist_order
;
2203 /* initialize zonelists */
2204 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2205 zonelist
= pgdat
->node_zonelists
+ i
;
2206 zonelist
->_zonerefs
[0].zone
= NULL
;
2207 zonelist
->_zonerefs
[0].zone_idx
= 0;
2210 /* NUMA-aware ordering of nodes */
2211 local_node
= pgdat
->node_id
;
2212 load
= num_online_nodes();
2213 prev_node
= local_node
;
2214 nodes_clear(used_mask
);
2216 memset(node_load
, 0, sizeof(node_load
));
2217 memset(node_order
, 0, sizeof(node_order
));
2220 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2221 int distance
= node_distance(local_node
, node
);
2224 * If another node is sufficiently far away then it is better
2225 * to reclaim pages in a zone before going off node.
2227 if (distance
> RECLAIM_DISTANCE
)
2228 zone_reclaim_mode
= 1;
2231 * We don't want to pressure a particular node.
2232 * So adding penalty to the first node in same
2233 * distance group to make it round-robin.
2235 if (distance
!= node_distance(local_node
, prev_node
))
2236 node_load
[node
] = load
;
2240 if (order
== ZONELIST_ORDER_NODE
)
2241 build_zonelists_in_node_order(pgdat
, node
);
2243 node_order
[j
++] = node
; /* remember order */
2246 if (order
== ZONELIST_ORDER_ZONE
) {
2247 /* calculate node order -- i.e., DMA last! */
2248 build_zonelists_in_zone_order(pgdat
, j
);
2251 build_thisnode_zonelists(pgdat
);
2254 /* Construct the zonelist performance cache - see further mmzone.h */
2255 static void build_zonelist_cache(pg_data_t
*pgdat
)
2257 struct zonelist
*zonelist
;
2258 struct zonelist_cache
*zlc
;
2261 zonelist
= &pgdat
->node_zonelists
[0];
2262 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2263 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2264 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2265 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2269 #else /* CONFIG_NUMA */
2271 static void set_zonelist_order(void)
2273 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2276 static void build_zonelists(pg_data_t
*pgdat
)
2278 int node
, local_node
;
2280 struct zonelist
*zonelist
;
2282 local_node
= pgdat
->node_id
;
2284 zonelist
= &pgdat
->node_zonelists
[0];
2285 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2288 * Now we build the zonelist so that it contains the zones
2289 * of all the other nodes.
2290 * We don't want to pressure a particular node, so when
2291 * building the zones for node N, we make sure that the
2292 * zones coming right after the local ones are those from
2293 * node N+1 (modulo N)
2295 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2296 if (!node_online(node
))
2298 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2301 for (node
= 0; node
< local_node
; node
++) {
2302 if (!node_online(node
))
2304 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2308 zonelist
->_zonerefs
[j
].zone
= NULL
;
2309 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2312 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2313 static void build_zonelist_cache(pg_data_t
*pgdat
)
2315 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2318 #endif /* CONFIG_NUMA */
2320 /* return values int ....just for stop_machine_run() */
2321 static int __build_all_zonelists(void *dummy
)
2325 for_each_online_node(nid
) {
2326 pg_data_t
*pgdat
= NODE_DATA(nid
);
2328 build_zonelists(pgdat
);
2329 build_zonelist_cache(pgdat
);
2334 void build_all_zonelists(void)
2336 set_zonelist_order();
2338 if (system_state
== SYSTEM_BOOTING
) {
2339 __build_all_zonelists(NULL
);
2340 mminit_verify_zonelist();
2341 cpuset_init_current_mems_allowed();
2343 /* we have to stop all cpus to guarantee there is no user
2345 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
2346 /* cpuset refresh routine should be here */
2348 vm_total_pages
= nr_free_pagecache_pages();
2350 * Disable grouping by mobility if the number of pages in the
2351 * system is too low to allow the mechanism to work. It would be
2352 * more accurate, but expensive to check per-zone. This check is
2353 * made on memory-hotadd so a system can start with mobility
2354 * disabled and enable it later
2356 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2357 page_group_by_mobility_disabled
= 1;
2359 page_group_by_mobility_disabled
= 0;
2361 printk("Built %i zonelists in %s order, mobility grouping %s. "
2362 "Total pages: %ld\n",
2364 zonelist_order_name
[current_zonelist_order
],
2365 page_group_by_mobility_disabled
? "off" : "on",
2368 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2373 * Helper functions to size the waitqueue hash table.
2374 * Essentially these want to choose hash table sizes sufficiently
2375 * large so that collisions trying to wait on pages are rare.
2376 * But in fact, the number of active page waitqueues on typical
2377 * systems is ridiculously low, less than 200. So this is even
2378 * conservative, even though it seems large.
2380 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2381 * waitqueues, i.e. the size of the waitq table given the number of pages.
2383 #define PAGES_PER_WAITQUEUE 256
2385 #ifndef CONFIG_MEMORY_HOTPLUG
2386 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2388 unsigned long size
= 1;
2390 pages
/= PAGES_PER_WAITQUEUE
;
2392 while (size
< pages
)
2396 * Once we have dozens or even hundreds of threads sleeping
2397 * on IO we've got bigger problems than wait queue collision.
2398 * Limit the size of the wait table to a reasonable size.
2400 size
= min(size
, 4096UL);
2402 return max(size
, 4UL);
2406 * A zone's size might be changed by hot-add, so it is not possible to determine
2407 * a suitable size for its wait_table. So we use the maximum size now.
2409 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2411 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2412 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2413 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2415 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2416 * or more by the traditional way. (See above). It equals:
2418 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2419 * ia64(16K page size) : = ( 8G + 4M)byte.
2420 * powerpc (64K page size) : = (32G +16M)byte.
2422 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2429 * This is an integer logarithm so that shifts can be used later
2430 * to extract the more random high bits from the multiplicative
2431 * hash function before the remainder is taken.
2433 static inline unsigned long wait_table_bits(unsigned long size
)
2438 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2441 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2442 * of blocks reserved is based on zone->pages_min. The memory within the
2443 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2444 * higher will lead to a bigger reserve which will get freed as contiguous
2445 * blocks as reclaim kicks in
2447 static void setup_zone_migrate_reserve(struct zone
*zone
)
2449 unsigned long start_pfn
, pfn
, end_pfn
;
2451 unsigned long reserve
, block_migratetype
;
2453 /* Get the start pfn, end pfn and the number of blocks to reserve */
2454 start_pfn
= zone
->zone_start_pfn
;
2455 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2456 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2459 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2460 if (!pfn_valid(pfn
))
2462 page
= pfn_to_page(pfn
);
2464 /* Blocks with reserved pages will never free, skip them. */
2465 if (PageReserved(page
))
2468 block_migratetype
= get_pageblock_migratetype(page
);
2470 /* If this block is reserved, account for it */
2471 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2476 /* Suitable for reserving if this block is movable */
2477 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2478 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2479 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2485 * If the reserve is met and this is a previous reserved block,
2488 if (block_migratetype
== MIGRATE_RESERVE
) {
2489 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2490 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2496 * Initially all pages are reserved - free ones are freed
2497 * up by free_all_bootmem() once the early boot process is
2498 * done. Non-atomic initialization, single-pass.
2500 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2501 unsigned long start_pfn
, enum memmap_context context
)
2504 unsigned long end_pfn
= start_pfn
+ size
;
2508 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2509 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2511 * There can be holes in boot-time mem_map[]s
2512 * handed to this function. They do not
2513 * exist on hotplugged memory.
2515 if (context
== MEMMAP_EARLY
) {
2516 if (!early_pfn_valid(pfn
))
2518 if (!early_pfn_in_nid(pfn
, nid
))
2521 page
= pfn_to_page(pfn
);
2522 set_page_links(page
, zone
, nid
, pfn
);
2523 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2524 init_page_count(page
);
2525 reset_page_mapcount(page
);
2526 SetPageReserved(page
);
2528 * Mark the block movable so that blocks are reserved for
2529 * movable at startup. This will force kernel allocations
2530 * to reserve their blocks rather than leaking throughout
2531 * the address space during boot when many long-lived
2532 * kernel allocations are made. Later some blocks near
2533 * the start are marked MIGRATE_RESERVE by
2534 * setup_zone_migrate_reserve()
2536 * bitmap is created for zone's valid pfn range. but memmap
2537 * can be created for invalid pages (for alignment)
2538 * check here not to call set_pageblock_migratetype() against
2541 if ((z
->zone_start_pfn
<= pfn
)
2542 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2543 && !(pfn
& (pageblock_nr_pages
- 1)))
2544 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2546 INIT_LIST_HEAD(&page
->lru
);
2547 #ifdef WANT_PAGE_VIRTUAL
2548 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2549 if (!is_highmem_idx(zone
))
2550 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2555 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2558 for_each_migratetype_order(order
, t
) {
2559 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2560 zone
->free_area
[order
].nr_free
= 0;
2564 #ifndef __HAVE_ARCH_MEMMAP_INIT
2565 #define memmap_init(size, nid, zone, start_pfn) \
2566 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2569 static int zone_batchsize(struct zone
*zone
)
2574 * The per-cpu-pages pools are set to around 1000th of the
2575 * size of the zone. But no more than 1/2 of a meg.
2577 * OK, so we don't know how big the cache is. So guess.
2579 batch
= zone
->present_pages
/ 1024;
2580 if (batch
* PAGE_SIZE
> 512 * 1024)
2581 batch
= (512 * 1024) / PAGE_SIZE
;
2582 batch
/= 4; /* We effectively *= 4 below */
2587 * Clamp the batch to a 2^n - 1 value. Having a power
2588 * of 2 value was found to be more likely to have
2589 * suboptimal cache aliasing properties in some cases.
2591 * For example if 2 tasks are alternately allocating
2592 * batches of pages, one task can end up with a lot
2593 * of pages of one half of the possible page colors
2594 * and the other with pages of the other colors.
2596 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2601 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2603 struct per_cpu_pages
*pcp
;
2605 memset(p
, 0, sizeof(*p
));
2609 pcp
->high
= 6 * batch
;
2610 pcp
->batch
= max(1UL, 1 * batch
);
2611 INIT_LIST_HEAD(&pcp
->list
);
2615 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2616 * to the value high for the pageset p.
2619 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2622 struct per_cpu_pages
*pcp
;
2626 pcp
->batch
= max(1UL, high
/4);
2627 if ((high
/4) > (PAGE_SHIFT
* 8))
2628 pcp
->batch
= PAGE_SHIFT
* 8;
2634 * Boot pageset table. One per cpu which is going to be used for all
2635 * zones and all nodes. The parameters will be set in such a way
2636 * that an item put on a list will immediately be handed over to
2637 * the buddy list. This is safe since pageset manipulation is done
2638 * with interrupts disabled.
2640 * Some NUMA counter updates may also be caught by the boot pagesets.
2642 * The boot_pagesets must be kept even after bootup is complete for
2643 * unused processors and/or zones. They do play a role for bootstrapping
2644 * hotplugged processors.
2646 * zoneinfo_show() and maybe other functions do
2647 * not check if the processor is online before following the pageset pointer.
2648 * Other parts of the kernel may not check if the zone is available.
2650 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2653 * Dynamically allocate memory for the
2654 * per cpu pageset array in struct zone.
2656 static int __cpuinit
process_zones(int cpu
)
2658 struct zone
*zone
, *dzone
;
2659 int node
= cpu_to_node(cpu
);
2661 node_set_state(node
, N_CPU
); /* this node has a cpu */
2663 for_each_zone(zone
) {
2665 if (!populated_zone(zone
))
2668 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2670 if (!zone_pcp(zone
, cpu
))
2673 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2675 if (percpu_pagelist_fraction
)
2676 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2677 (zone
->present_pages
/ percpu_pagelist_fraction
));
2682 for_each_zone(dzone
) {
2683 if (!populated_zone(dzone
))
2687 kfree(zone_pcp(dzone
, cpu
));
2688 zone_pcp(dzone
, cpu
) = NULL
;
2693 static inline void free_zone_pagesets(int cpu
)
2697 for_each_zone(zone
) {
2698 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2700 /* Free per_cpu_pageset if it is slab allocated */
2701 if (pset
!= &boot_pageset
[cpu
])
2703 zone_pcp(zone
, cpu
) = NULL
;
2707 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2708 unsigned long action
,
2711 int cpu
= (long)hcpu
;
2712 int ret
= NOTIFY_OK
;
2715 case CPU_UP_PREPARE
:
2716 case CPU_UP_PREPARE_FROZEN
:
2717 if (process_zones(cpu
))
2720 case CPU_UP_CANCELED
:
2721 case CPU_UP_CANCELED_FROZEN
:
2723 case CPU_DEAD_FROZEN
:
2724 free_zone_pagesets(cpu
);
2732 static struct notifier_block __cpuinitdata pageset_notifier
=
2733 { &pageset_cpuup_callback
, NULL
, 0 };
2735 void __init
setup_per_cpu_pageset(void)
2739 /* Initialize per_cpu_pageset for cpu 0.
2740 * A cpuup callback will do this for every cpu
2741 * as it comes online
2743 err
= process_zones(smp_processor_id());
2745 register_cpu_notifier(&pageset_notifier
);
2750 static noinline __init_refok
2751 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2754 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2758 * The per-page waitqueue mechanism uses hashed waitqueues
2761 zone
->wait_table_hash_nr_entries
=
2762 wait_table_hash_nr_entries(zone_size_pages
);
2763 zone
->wait_table_bits
=
2764 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2765 alloc_size
= zone
->wait_table_hash_nr_entries
2766 * sizeof(wait_queue_head_t
);
2768 if (!slab_is_available()) {
2769 zone
->wait_table
= (wait_queue_head_t
*)
2770 alloc_bootmem_node(pgdat
, alloc_size
);
2773 * This case means that a zone whose size was 0 gets new memory
2774 * via memory hot-add.
2775 * But it may be the case that a new node was hot-added. In
2776 * this case vmalloc() will not be able to use this new node's
2777 * memory - this wait_table must be initialized to use this new
2778 * node itself as well.
2779 * To use this new node's memory, further consideration will be
2782 zone
->wait_table
= vmalloc(alloc_size
);
2784 if (!zone
->wait_table
)
2787 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2788 init_waitqueue_head(zone
->wait_table
+ i
);
2793 static __meminit
void zone_pcp_init(struct zone
*zone
)
2796 unsigned long batch
= zone_batchsize(zone
);
2798 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2800 /* Early boot. Slab allocator not functional yet */
2801 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2802 setup_pageset(&boot_pageset
[cpu
],0);
2804 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2807 if (zone
->present_pages
)
2808 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2809 zone
->name
, zone
->present_pages
, batch
);
2812 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2813 unsigned long zone_start_pfn
,
2815 enum memmap_context context
)
2817 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2819 ret
= zone_wait_table_init(zone
, size
);
2822 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2824 zone
->zone_start_pfn
= zone_start_pfn
;
2826 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
2827 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2829 (unsigned long)zone_idx(zone
),
2830 zone_start_pfn
, (zone_start_pfn
+ size
));
2832 zone_init_free_lists(zone
);
2837 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2839 * Basic iterator support. Return the first range of PFNs for a node
2840 * Note: nid == MAX_NUMNODES returns first region regardless of node
2842 static int __meminit
first_active_region_index_in_nid(int nid
)
2846 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2847 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2854 * Basic iterator support. Return the next active range of PFNs for a node
2855 * Note: nid == MAX_NUMNODES returns next region regardless of node
2857 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2859 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2860 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2866 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2868 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2869 * Architectures may implement their own version but if add_active_range()
2870 * was used and there are no special requirements, this is a convenient
2873 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2877 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2878 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2879 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2881 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2882 return early_node_map
[i
].nid
;
2887 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2889 /* Basic iterator support to walk early_node_map[] */
2890 #define for_each_active_range_index_in_nid(i, nid) \
2891 for (i = first_active_region_index_in_nid(nid); i != -1; \
2892 i = next_active_region_index_in_nid(i, nid))
2895 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2896 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2897 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2899 * If an architecture guarantees that all ranges registered with
2900 * add_active_ranges() contain no holes and may be freed, this
2901 * this function may be used instead of calling free_bootmem() manually.
2903 void __init
free_bootmem_with_active_regions(int nid
,
2904 unsigned long max_low_pfn
)
2908 for_each_active_range_index_in_nid(i
, nid
) {
2909 unsigned long size_pages
= 0;
2910 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2912 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2915 if (end_pfn
> max_low_pfn
)
2916 end_pfn
= max_low_pfn
;
2918 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2919 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2920 PFN_PHYS(early_node_map
[i
].start_pfn
),
2921 size_pages
<< PAGE_SHIFT
);
2925 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
2930 for_each_active_range_index_in_nid(i
, nid
) {
2931 ret
= work_fn(early_node_map
[i
].start_pfn
,
2932 early_node_map
[i
].end_pfn
, data
);
2938 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2939 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2941 * If an architecture guarantees that all ranges registered with
2942 * add_active_ranges() contain no holes and may be freed, this
2943 * function may be used instead of calling memory_present() manually.
2945 void __init
sparse_memory_present_with_active_regions(int nid
)
2949 for_each_active_range_index_in_nid(i
, nid
)
2950 memory_present(early_node_map
[i
].nid
,
2951 early_node_map
[i
].start_pfn
,
2952 early_node_map
[i
].end_pfn
);
2956 * push_node_boundaries - Push node boundaries to at least the requested boundary
2957 * @nid: The nid of the node to push the boundary for
2958 * @start_pfn: The start pfn of the node
2959 * @end_pfn: The end pfn of the node
2961 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2962 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2963 * be hotplugged even though no physical memory exists. This function allows
2964 * an arch to push out the node boundaries so mem_map is allocated that can
2967 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2968 void __init
push_node_boundaries(unsigned int nid
,
2969 unsigned long start_pfn
, unsigned long end_pfn
)
2971 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
2972 "Entering push_node_boundaries(%u, %lu, %lu)\n",
2973 nid
, start_pfn
, end_pfn
);
2975 /* Initialise the boundary for this node if necessary */
2976 if (node_boundary_end_pfn
[nid
] == 0)
2977 node_boundary_start_pfn
[nid
] = -1UL;
2979 /* Update the boundaries */
2980 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2981 node_boundary_start_pfn
[nid
] = start_pfn
;
2982 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2983 node_boundary_end_pfn
[nid
] = end_pfn
;
2986 /* If necessary, push the node boundary out for reserve hotadd */
2987 static void __meminit
account_node_boundary(unsigned int nid
,
2988 unsigned long *start_pfn
, unsigned long *end_pfn
)
2990 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
2991 "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
;
3366 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
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 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3378 "%s zone: %lu pages reserved\n",
3379 zone_names
[0], dma_reserve
);
3382 if (!is_highmem_idx(j
))
3383 nr_kernel_pages
+= realsize
;
3384 nr_all_pages
+= realsize
;
3386 zone
->spanned_pages
= size
;
3387 zone
->present_pages
= realsize
;
3390 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3392 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3394 zone
->name
= zone_names
[j
];
3395 spin_lock_init(&zone
->lock
);
3396 spin_lock_init(&zone
->lru_lock
);
3397 zone_seqlock_init(zone
);
3398 zone
->zone_pgdat
= pgdat
;
3400 zone
->prev_priority
= DEF_PRIORITY
;
3402 zone_pcp_init(zone
);
3403 INIT_LIST_HEAD(&zone
->active_list
);
3404 INIT_LIST_HEAD(&zone
->inactive_list
);
3405 zone
->nr_scan_active
= 0;
3406 zone
->nr_scan_inactive
= 0;
3407 zap_zone_vm_stats(zone
);
3412 set_pageblock_order(pageblock_default_order());
3413 setup_usemap(pgdat
, zone
, size
);
3414 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3415 size
, MEMMAP_EARLY
);
3417 memmap_init(size
, nid
, j
, zone_start_pfn
);
3418 zone_start_pfn
+= size
;
3422 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3424 /* Skip empty nodes */
3425 if (!pgdat
->node_spanned_pages
)
3428 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3429 /* ia64 gets its own node_mem_map, before this, without bootmem */
3430 if (!pgdat
->node_mem_map
) {
3431 unsigned long size
, start
, end
;
3435 * The zone's endpoints aren't required to be MAX_ORDER
3436 * aligned but the node_mem_map endpoints must be in order
3437 * for the buddy allocator to function correctly.
3439 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3440 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3441 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3442 size
= (end
- start
) * sizeof(struct page
);
3443 map
= alloc_remap(pgdat
->node_id
, size
);
3445 map
= alloc_bootmem_node(pgdat
, size
);
3446 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3448 #ifndef CONFIG_NEED_MULTIPLE_NODES
3450 * With no DISCONTIG, the global mem_map is just set as node 0's
3452 if (pgdat
== NODE_DATA(0)) {
3453 mem_map
= NODE_DATA(0)->node_mem_map
;
3454 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3455 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3456 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3457 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3460 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3463 void __paginginit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
3464 unsigned long *zones_size
, unsigned long node_start_pfn
,
3465 unsigned long *zholes_size
)
3467 pgdat
->node_id
= nid
;
3468 pgdat
->node_start_pfn
= node_start_pfn
;
3469 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3471 alloc_node_mem_map(pgdat
);
3472 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3473 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3474 nid
, (unsigned long)pgdat
,
3475 (unsigned long)pgdat
->node_mem_map
);
3478 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3481 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3483 #if MAX_NUMNODES > 1
3485 * Figure out the number of possible node ids.
3487 static void __init
setup_nr_node_ids(void)
3490 unsigned int highest
= 0;
3492 for_each_node_mask(node
, node_possible_map
)
3494 nr_node_ids
= highest
+ 1;
3497 static inline void setup_nr_node_ids(void)
3503 * add_active_range - Register a range of PFNs backed by physical memory
3504 * @nid: The node ID the range resides on
3505 * @start_pfn: The start PFN of the available physical memory
3506 * @end_pfn: The end PFN of the available physical memory
3508 * These ranges are stored in an early_node_map[] and later used by
3509 * free_area_init_nodes() to calculate zone sizes and holes. If the
3510 * range spans a memory hole, it is up to the architecture to ensure
3511 * the memory is not freed by the bootmem allocator. If possible
3512 * the range being registered will be merged with existing ranges.
3514 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3515 unsigned long end_pfn
)
3519 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3520 "Entering add_active_range(%d, %#lx, %#lx) "
3521 "%d entries of %d used\n",
3522 nid
, start_pfn
, end_pfn
,
3523 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3525 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3527 /* Merge with existing active regions if possible */
3528 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3529 if (early_node_map
[i
].nid
!= nid
)
3532 /* Skip if an existing region covers this new one */
3533 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3534 end_pfn
<= early_node_map
[i
].end_pfn
)
3537 /* Merge forward if suitable */
3538 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3539 end_pfn
> early_node_map
[i
].end_pfn
) {
3540 early_node_map
[i
].end_pfn
= end_pfn
;
3544 /* Merge backward if suitable */
3545 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3546 end_pfn
>= early_node_map
[i
].start_pfn
) {
3547 early_node_map
[i
].start_pfn
= start_pfn
;
3552 /* Check that early_node_map is large enough */
3553 if (i
>= MAX_ACTIVE_REGIONS
) {
3554 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3555 MAX_ACTIVE_REGIONS
);
3559 early_node_map
[i
].nid
= nid
;
3560 early_node_map
[i
].start_pfn
= start_pfn
;
3561 early_node_map
[i
].end_pfn
= end_pfn
;
3562 nr_nodemap_entries
= i
+ 1;
3566 * remove_active_range - Shrink an existing registered range of PFNs
3567 * @nid: The node id the range is on that should be shrunk
3568 * @start_pfn: The new PFN of the range
3569 * @end_pfn: The new PFN of the range
3571 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3572 * The map is kept near the end physical page range that has already been
3573 * registered. This function allows an arch to shrink an existing registered
3576 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3577 unsigned long end_pfn
)
3582 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3583 nid
, start_pfn
, end_pfn
);
3585 /* Find the old active region end and shrink */
3586 for_each_active_range_index_in_nid(i
, nid
) {
3587 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3588 early_node_map
[i
].end_pfn
<= end_pfn
) {
3590 early_node_map
[i
].start_pfn
= 0;
3591 early_node_map
[i
].end_pfn
= 0;
3595 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3596 early_node_map
[i
].end_pfn
> start_pfn
) {
3597 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3598 early_node_map
[i
].end_pfn
= start_pfn
;
3599 if (temp_end_pfn
> end_pfn
)
3600 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3603 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3604 early_node_map
[i
].end_pfn
> end_pfn
&&
3605 early_node_map
[i
].start_pfn
< end_pfn
) {
3606 early_node_map
[i
].start_pfn
= end_pfn
;
3614 /* remove the blank ones */
3615 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3616 if (early_node_map
[i
].nid
!= nid
)
3618 if (early_node_map
[i
].end_pfn
)
3620 /* we found it, get rid of it */
3621 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3622 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3623 sizeof(early_node_map
[j
]));
3624 j
= nr_nodemap_entries
- 1;
3625 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3626 nr_nodemap_entries
--;
3631 * remove_all_active_ranges - Remove all currently registered regions
3633 * During discovery, it may be found that a table like SRAT is invalid
3634 * and an alternative discovery method must be used. This function removes
3635 * all currently registered regions.
3637 void __init
remove_all_active_ranges(void)
3639 memset(early_node_map
, 0, sizeof(early_node_map
));
3640 nr_nodemap_entries
= 0;
3641 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3642 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3643 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3644 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3647 /* Compare two active node_active_regions */
3648 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3650 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3651 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3653 /* Done this way to avoid overflows */
3654 if (arange
->start_pfn
> brange
->start_pfn
)
3656 if (arange
->start_pfn
< brange
->start_pfn
)
3662 /* sort the node_map by start_pfn */
3663 static void __init
sort_node_map(void)
3665 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3666 sizeof(struct node_active_region
),
3667 cmp_node_active_region
, NULL
);
3670 /* Find the lowest pfn for a node */
3671 unsigned long __init
find_min_pfn_for_node(int nid
)
3674 unsigned long min_pfn
= ULONG_MAX
;
3676 /* Assuming a sorted map, the first range found has the starting pfn */
3677 for_each_active_range_index_in_nid(i
, nid
)
3678 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3680 if (min_pfn
== ULONG_MAX
) {
3682 "Could not find start_pfn for node %d\n", nid
);
3690 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3692 * It returns the minimum PFN based on information provided via
3693 * add_active_range().
3695 unsigned long __init
find_min_pfn_with_active_regions(void)
3697 return find_min_pfn_for_node(MAX_NUMNODES
);
3701 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3703 * It returns the maximum PFN based on information provided via
3704 * add_active_range().
3706 unsigned long __init
find_max_pfn_with_active_regions(void)
3709 unsigned long max_pfn
= 0;
3711 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3712 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
3718 * early_calculate_totalpages()
3719 * Sum pages in active regions for movable zone.
3720 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3722 static unsigned long __init
early_calculate_totalpages(void)
3725 unsigned long totalpages
= 0;
3727 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3728 unsigned long pages
= early_node_map
[i
].end_pfn
-
3729 early_node_map
[i
].start_pfn
;
3730 totalpages
+= pages
;
3732 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3738 * Find the PFN the Movable zone begins in each node. Kernel memory
3739 * is spread evenly between nodes as long as the nodes have enough
3740 * memory. When they don't, some nodes will have more kernelcore than
3743 void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3746 unsigned long usable_startpfn
;
3747 unsigned long kernelcore_node
, kernelcore_remaining
;
3748 unsigned long totalpages
= early_calculate_totalpages();
3749 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3752 * If movablecore was specified, calculate what size of
3753 * kernelcore that corresponds so that memory usable for
3754 * any allocation type is evenly spread. If both kernelcore
3755 * and movablecore are specified, then the value of kernelcore
3756 * will be used for required_kernelcore if it's greater than
3757 * what movablecore would have allowed.
3759 if (required_movablecore
) {
3760 unsigned long corepages
;
3763 * Round-up so that ZONE_MOVABLE is at least as large as what
3764 * was requested by the user
3766 required_movablecore
=
3767 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3768 corepages
= totalpages
- required_movablecore
;
3770 required_kernelcore
= max(required_kernelcore
, corepages
);
3773 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3774 if (!required_kernelcore
)
3777 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3778 find_usable_zone_for_movable();
3779 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3782 /* Spread kernelcore memory as evenly as possible throughout nodes */
3783 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3784 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3786 * Recalculate kernelcore_node if the division per node
3787 * now exceeds what is necessary to satisfy the requested
3788 * amount of memory for the kernel
3790 if (required_kernelcore
< kernelcore_node
)
3791 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3794 * As the map is walked, we track how much memory is usable
3795 * by the kernel using kernelcore_remaining. When it is
3796 * 0, the rest of the node is usable by ZONE_MOVABLE
3798 kernelcore_remaining
= kernelcore_node
;
3800 /* Go through each range of PFNs within this node */
3801 for_each_active_range_index_in_nid(i
, nid
) {
3802 unsigned long start_pfn
, end_pfn
;
3803 unsigned long size_pages
;
3805 start_pfn
= max(early_node_map
[i
].start_pfn
,
3806 zone_movable_pfn
[nid
]);
3807 end_pfn
= early_node_map
[i
].end_pfn
;
3808 if (start_pfn
>= end_pfn
)
3811 /* Account for what is only usable for kernelcore */
3812 if (start_pfn
< usable_startpfn
) {
3813 unsigned long kernel_pages
;
3814 kernel_pages
= min(end_pfn
, usable_startpfn
)
3817 kernelcore_remaining
-= min(kernel_pages
,
3818 kernelcore_remaining
);
3819 required_kernelcore
-= min(kernel_pages
,
3820 required_kernelcore
);
3822 /* Continue if range is now fully accounted */
3823 if (end_pfn
<= usable_startpfn
) {
3826 * Push zone_movable_pfn to the end so
3827 * that if we have to rebalance
3828 * kernelcore across nodes, we will
3829 * not double account here
3831 zone_movable_pfn
[nid
] = end_pfn
;
3834 start_pfn
= usable_startpfn
;
3838 * The usable PFN range for ZONE_MOVABLE is from
3839 * start_pfn->end_pfn. Calculate size_pages as the
3840 * number of pages used as kernelcore
3842 size_pages
= end_pfn
- start_pfn
;
3843 if (size_pages
> kernelcore_remaining
)
3844 size_pages
= kernelcore_remaining
;
3845 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3848 * Some kernelcore has been met, update counts and
3849 * break if the kernelcore for this node has been
3852 required_kernelcore
-= min(required_kernelcore
,
3854 kernelcore_remaining
-= size_pages
;
3855 if (!kernelcore_remaining
)
3861 * If there is still required_kernelcore, we do another pass with one
3862 * less node in the count. This will push zone_movable_pfn[nid] further
3863 * along on the nodes that still have memory until kernelcore is
3867 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3870 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3871 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3872 zone_movable_pfn
[nid
] =
3873 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3876 /* Any regular memory on that node ? */
3877 static void check_for_regular_memory(pg_data_t
*pgdat
)
3879 #ifdef CONFIG_HIGHMEM
3880 enum zone_type zone_type
;
3882 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3883 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3884 if (zone
->present_pages
)
3885 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3891 * free_area_init_nodes - Initialise all pg_data_t and zone data
3892 * @max_zone_pfn: an array of max PFNs for each zone
3894 * This will call free_area_init_node() for each active node in the system.
3895 * Using the page ranges provided by add_active_range(), the size of each
3896 * zone in each node and their holes is calculated. If the maximum PFN
3897 * between two adjacent zones match, it is assumed that the zone is empty.
3898 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3899 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3900 * starts where the previous one ended. For example, ZONE_DMA32 starts
3901 * at arch_max_dma_pfn.
3903 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3908 /* Sort early_node_map as initialisation assumes it is sorted */
3911 /* Record where the zone boundaries are */
3912 memset(arch_zone_lowest_possible_pfn
, 0,
3913 sizeof(arch_zone_lowest_possible_pfn
));
3914 memset(arch_zone_highest_possible_pfn
, 0,
3915 sizeof(arch_zone_highest_possible_pfn
));
3916 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3917 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3918 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3919 if (i
== ZONE_MOVABLE
)
3921 arch_zone_lowest_possible_pfn
[i
] =
3922 arch_zone_highest_possible_pfn
[i
-1];
3923 arch_zone_highest_possible_pfn
[i
] =
3924 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3926 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3927 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3929 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3930 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3931 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3933 /* Print out the zone ranges */
3934 printk("Zone PFN ranges:\n");
3935 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3936 if (i
== ZONE_MOVABLE
)
3938 printk(" %-8s %0#10lx -> %0#10lx\n",
3940 arch_zone_lowest_possible_pfn
[i
],
3941 arch_zone_highest_possible_pfn
[i
]);
3944 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3945 printk("Movable zone start PFN for each node\n");
3946 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
3947 if (zone_movable_pfn
[i
])
3948 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
3951 /* Print out the early_node_map[] */
3952 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
3953 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3954 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
3955 early_node_map
[i
].start_pfn
,
3956 early_node_map
[i
].end_pfn
);
3958 /* Initialise every node */
3959 mminit_verify_pageflags_layout();
3960 setup_nr_node_ids();
3961 for_each_online_node(nid
) {
3962 pg_data_t
*pgdat
= NODE_DATA(nid
);
3963 free_area_init_node(nid
, pgdat
, NULL
,
3964 find_min_pfn_for_node(nid
), NULL
);
3966 /* Any memory on that node */
3967 if (pgdat
->node_present_pages
)
3968 node_set_state(nid
, N_HIGH_MEMORY
);
3969 check_for_regular_memory(pgdat
);
3973 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
3975 unsigned long long coremem
;
3979 coremem
= memparse(p
, &p
);
3980 *core
= coremem
>> PAGE_SHIFT
;
3982 /* Paranoid check that UL is enough for the coremem value */
3983 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
3989 * kernelcore=size sets the amount of memory for use for allocations that
3990 * cannot be reclaimed or migrated.
3992 static int __init
cmdline_parse_kernelcore(char *p
)
3994 return cmdline_parse_core(p
, &required_kernelcore
);
3998 * movablecore=size sets the amount of memory for use for allocations that
3999 * can be reclaimed or migrated.
4001 static int __init
cmdline_parse_movablecore(char *p
)
4003 return cmdline_parse_core(p
, &required_movablecore
);
4006 early_param("kernelcore", cmdline_parse_kernelcore
);
4007 early_param("movablecore", cmdline_parse_movablecore
);
4009 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4012 * set_dma_reserve - set the specified number of pages reserved in the first zone
4013 * @new_dma_reserve: The number of pages to mark reserved
4015 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4016 * In the DMA zone, a significant percentage may be consumed by kernel image
4017 * and other unfreeable allocations which can skew the watermarks badly. This
4018 * function may optionally be used to account for unfreeable pages in the
4019 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4020 * smaller per-cpu batchsize.
4022 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4024 dma_reserve
= new_dma_reserve
;
4027 #ifndef CONFIG_NEED_MULTIPLE_NODES
4028 struct pglist_data contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4029 EXPORT_SYMBOL(contig_page_data
);
4032 void __init
free_area_init(unsigned long *zones_size
)
4034 free_area_init_node(0, NODE_DATA(0), zones_size
,
4035 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4038 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4039 unsigned long action
, void *hcpu
)
4041 int cpu
= (unsigned long)hcpu
;
4043 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4047 * Spill the event counters of the dead processor
4048 * into the current processors event counters.
4049 * This artificially elevates the count of the current
4052 vm_events_fold_cpu(cpu
);
4055 * Zero the differential counters of the dead processor
4056 * so that the vm statistics are consistent.
4058 * This is only okay since the processor is dead and cannot
4059 * race with what we are doing.
4061 refresh_cpu_vm_stats(cpu
);
4066 void __init
page_alloc_init(void)
4068 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4072 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4073 * or min_free_kbytes changes.
4075 static void calculate_totalreserve_pages(void)
4077 struct pglist_data
*pgdat
;
4078 unsigned long reserve_pages
= 0;
4079 enum zone_type i
, j
;
4081 for_each_online_pgdat(pgdat
) {
4082 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4083 struct zone
*zone
= pgdat
->node_zones
+ i
;
4084 unsigned long max
= 0;
4086 /* Find valid and maximum lowmem_reserve in the zone */
4087 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4088 if (zone
->lowmem_reserve
[j
] > max
)
4089 max
= zone
->lowmem_reserve
[j
];
4092 /* we treat pages_high as reserved pages. */
4093 max
+= zone
->pages_high
;
4095 if (max
> zone
->present_pages
)
4096 max
= zone
->present_pages
;
4097 reserve_pages
+= max
;
4100 totalreserve_pages
= reserve_pages
;
4104 * setup_per_zone_lowmem_reserve - called whenever
4105 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4106 * has a correct pages reserved value, so an adequate number of
4107 * pages are left in the zone after a successful __alloc_pages().
4109 static void setup_per_zone_lowmem_reserve(void)
4111 struct pglist_data
*pgdat
;
4112 enum zone_type j
, idx
;
4114 for_each_online_pgdat(pgdat
) {
4115 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4116 struct zone
*zone
= pgdat
->node_zones
+ j
;
4117 unsigned long present_pages
= zone
->present_pages
;
4119 zone
->lowmem_reserve
[j
] = 0;
4123 struct zone
*lower_zone
;
4127 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4128 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4130 lower_zone
= pgdat
->node_zones
+ idx
;
4131 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4132 sysctl_lowmem_reserve_ratio
[idx
];
4133 present_pages
+= lower_zone
->present_pages
;
4138 /* update totalreserve_pages */
4139 calculate_totalreserve_pages();
4143 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4145 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4146 * with respect to min_free_kbytes.
4148 void setup_per_zone_pages_min(void)
4150 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4151 unsigned long lowmem_pages
= 0;
4153 unsigned long flags
;
4155 /* Calculate total number of !ZONE_HIGHMEM pages */
4156 for_each_zone(zone
) {
4157 if (!is_highmem(zone
))
4158 lowmem_pages
+= zone
->present_pages
;
4161 for_each_zone(zone
) {
4164 spin_lock_irqsave(&zone
->lru_lock
, flags
);
4165 tmp
= (u64
)pages_min
* zone
->present_pages
;
4166 do_div(tmp
, lowmem_pages
);
4167 if (is_highmem(zone
)) {
4169 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4170 * need highmem pages, so cap pages_min to a small
4173 * The (pages_high-pages_low) and (pages_low-pages_min)
4174 * deltas controls asynch page reclaim, and so should
4175 * not be capped for highmem.
4179 min_pages
= zone
->present_pages
/ 1024;
4180 if (min_pages
< SWAP_CLUSTER_MAX
)
4181 min_pages
= SWAP_CLUSTER_MAX
;
4182 if (min_pages
> 128)
4184 zone
->pages_min
= min_pages
;
4187 * If it's a lowmem zone, reserve a number of pages
4188 * proportionate to the zone's size.
4190 zone
->pages_min
= tmp
;
4193 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4194 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4195 setup_zone_migrate_reserve(zone
);
4196 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4199 /* update totalreserve_pages */
4200 calculate_totalreserve_pages();
4204 * Initialise min_free_kbytes.
4206 * For small machines we want it small (128k min). For large machines
4207 * we want it large (64MB max). But it is not linear, because network
4208 * bandwidth does not increase linearly with machine size. We use
4210 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4211 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4227 static int __init
init_per_zone_pages_min(void)
4229 unsigned long lowmem_kbytes
;
4231 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4233 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4234 if (min_free_kbytes
< 128)
4235 min_free_kbytes
= 128;
4236 if (min_free_kbytes
> 65536)
4237 min_free_kbytes
= 65536;
4238 setup_per_zone_pages_min();
4239 setup_per_zone_lowmem_reserve();
4242 module_init(init_per_zone_pages_min
)
4245 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4246 * that we can call two helper functions whenever min_free_kbytes
4249 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4250 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4252 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4254 setup_per_zone_pages_min();
4259 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4260 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4265 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4270 zone
->min_unmapped_pages
= (zone
->present_pages
*
4271 sysctl_min_unmapped_ratio
) / 100;
4275 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4276 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4281 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4286 zone
->min_slab_pages
= (zone
->present_pages
*
4287 sysctl_min_slab_ratio
) / 100;
4293 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4294 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4295 * whenever sysctl_lowmem_reserve_ratio changes.
4297 * The reserve ratio obviously has absolutely no relation with the
4298 * pages_min watermarks. The lowmem reserve ratio can only make sense
4299 * if in function of the boot time zone sizes.
4301 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4302 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4304 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4305 setup_per_zone_lowmem_reserve();
4310 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4311 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4312 * can have before it gets flushed back to buddy allocator.
4315 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4316 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4322 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4323 if (!write
|| (ret
== -EINVAL
))
4325 for_each_zone(zone
) {
4326 for_each_online_cpu(cpu
) {
4328 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4329 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4335 int hashdist
= HASHDIST_DEFAULT
;
4338 static int __init
set_hashdist(char *str
)
4342 hashdist
= simple_strtoul(str
, &str
, 0);
4345 __setup("hashdist=", set_hashdist
);
4349 * allocate a large system hash table from bootmem
4350 * - it is assumed that the hash table must contain an exact power-of-2
4351 * quantity of entries
4352 * - limit is the number of hash buckets, not the total allocation size
4354 void *__init
alloc_large_system_hash(const char *tablename
,
4355 unsigned long bucketsize
,
4356 unsigned long numentries
,
4359 unsigned int *_hash_shift
,
4360 unsigned int *_hash_mask
,
4361 unsigned long limit
)
4363 unsigned long long max
= limit
;
4364 unsigned long log2qty
, size
;
4367 /* allow the kernel cmdline to have a say */
4369 /* round applicable memory size up to nearest megabyte */
4370 numentries
= nr_kernel_pages
;
4371 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4372 numentries
>>= 20 - PAGE_SHIFT
;
4373 numentries
<<= 20 - PAGE_SHIFT
;
4375 /* limit to 1 bucket per 2^scale bytes of low memory */
4376 if (scale
> PAGE_SHIFT
)
4377 numentries
>>= (scale
- PAGE_SHIFT
);
4379 numentries
<<= (PAGE_SHIFT
- scale
);
4381 /* Make sure we've got at least a 0-order allocation.. */
4382 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4383 numentries
= PAGE_SIZE
/ bucketsize
;
4385 numentries
= roundup_pow_of_two(numentries
);
4387 /* limit allocation size to 1/16 total memory by default */
4389 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4390 do_div(max
, bucketsize
);
4393 if (numentries
> max
)
4396 log2qty
= ilog2(numentries
);
4399 size
= bucketsize
<< log2qty
;
4400 if (flags
& HASH_EARLY
)
4401 table
= alloc_bootmem(size
);
4403 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4405 unsigned long order
= get_order(size
);
4406 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4408 * If bucketsize is not a power-of-two, we may free
4409 * some pages at the end of hash table.
4412 unsigned long alloc_end
= (unsigned long)table
+
4413 (PAGE_SIZE
<< order
);
4414 unsigned long used
= (unsigned long)table
+
4416 split_page(virt_to_page(table
), order
);
4417 while (used
< alloc_end
) {
4423 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4426 panic("Failed to allocate %s hash table\n", tablename
);
4428 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4431 ilog2(size
) - PAGE_SHIFT
,
4435 *_hash_shift
= log2qty
;
4437 *_hash_mask
= (1 << log2qty
) - 1;
4442 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4443 struct page
*pfn_to_page(unsigned long pfn
)
4445 return __pfn_to_page(pfn
);
4447 unsigned long page_to_pfn(struct page
*page
)
4449 return __page_to_pfn(page
);
4451 EXPORT_SYMBOL(pfn_to_page
);
4452 EXPORT_SYMBOL(page_to_pfn
);
4453 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4455 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4456 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4459 #ifdef CONFIG_SPARSEMEM
4460 return __pfn_to_section(pfn
)->pageblock_flags
;
4462 return zone
->pageblock_flags
;
4463 #endif /* CONFIG_SPARSEMEM */
4466 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4468 #ifdef CONFIG_SPARSEMEM
4469 pfn
&= (PAGES_PER_SECTION
-1);
4470 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4472 pfn
= pfn
- zone
->zone_start_pfn
;
4473 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4474 #endif /* CONFIG_SPARSEMEM */
4478 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4479 * @page: The page within the block of interest
4480 * @start_bitidx: The first bit of interest to retrieve
4481 * @end_bitidx: The last bit of interest
4482 * returns pageblock_bits flags
4484 unsigned long get_pageblock_flags_group(struct page
*page
,
4485 int start_bitidx
, int end_bitidx
)
4488 unsigned long *bitmap
;
4489 unsigned long pfn
, bitidx
;
4490 unsigned long flags
= 0;
4491 unsigned long value
= 1;
4493 zone
= page_zone(page
);
4494 pfn
= page_to_pfn(page
);
4495 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4496 bitidx
= pfn_to_bitidx(zone
, pfn
);
4498 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4499 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4506 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4507 * @page: The page within the block of interest
4508 * @start_bitidx: The first bit of interest
4509 * @end_bitidx: The last bit of interest
4510 * @flags: The flags to set
4512 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4513 int start_bitidx
, int end_bitidx
)
4516 unsigned long *bitmap
;
4517 unsigned long pfn
, bitidx
;
4518 unsigned long value
= 1;
4520 zone
= page_zone(page
);
4521 pfn
= page_to_pfn(page
);
4522 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4523 bitidx
= pfn_to_bitidx(zone
, pfn
);
4524 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4525 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4527 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4529 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4531 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4535 * This is designed as sub function...plz see page_isolation.c also.
4536 * set/clear page block's type to be ISOLATE.
4537 * page allocater never alloc memory from ISOLATE block.
4540 int set_migratetype_isolate(struct page
*page
)
4543 unsigned long flags
;
4546 zone
= page_zone(page
);
4547 spin_lock_irqsave(&zone
->lock
, flags
);
4549 * In future, more migrate types will be able to be isolation target.
4551 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4553 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4554 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4557 spin_unlock_irqrestore(&zone
->lock
, flags
);
4563 void unset_migratetype_isolate(struct page
*page
)
4566 unsigned long flags
;
4567 zone
= page_zone(page
);
4568 spin_lock_irqsave(&zone
->lock
, flags
);
4569 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4571 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4572 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4574 spin_unlock_irqrestore(&zone
->lock
, flags
);
4577 #ifdef CONFIG_MEMORY_HOTREMOVE
4579 * All pages in the range must be isolated before calling this.
4582 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4588 unsigned long flags
;
4589 /* find the first valid pfn */
4590 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4595 zone
= page_zone(pfn_to_page(pfn
));
4596 spin_lock_irqsave(&zone
->lock
, flags
);
4598 while (pfn
< end_pfn
) {
4599 if (!pfn_valid(pfn
)) {
4603 page
= pfn_to_page(pfn
);
4604 BUG_ON(page_count(page
));
4605 BUG_ON(!PageBuddy(page
));
4606 order
= page_order(page
);
4607 #ifdef CONFIG_DEBUG_VM
4608 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4609 pfn
, 1 << order
, end_pfn
);
4611 list_del(&page
->lru
);
4612 rmv_page_order(page
);
4613 zone
->free_area
[order
].nr_free
--;
4614 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4616 for (i
= 0; i
< (1 << order
); i
++)
4617 SetPageReserved((page
+i
));
4618 pfn
+= (1 << order
);
4620 spin_unlock_irqrestore(&zone
->lock
, flags
);