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 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
))
437 /* Our buddy is free, merge with it and move up one order. */
438 list_del(&buddy
->lru
);
439 zone
->free_area
[order
].nr_free
--;
440 rmv_page_order(buddy
);
441 combined_idx
= __find_combined_index(page_idx
, order
);
442 page
= page
+ (combined_idx
- page_idx
);
443 page_idx
= combined_idx
;
446 set_page_order(page
, order
);
448 &zone
->free_area
[order
].free_list
[migratetype
]);
449 zone
->free_area
[order
].nr_free
++;
452 static inline int free_pages_check(struct page
*page
)
454 if (unlikely(page_mapcount(page
) |
455 (page
->mapping
!= NULL
) |
456 (page_get_page_cgroup(page
) != NULL
) |
457 (page_count(page
) != 0) |
458 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)))
461 __ClearPageDirty(page
);
463 * For now, we report if PG_reserved was found set, but do not
464 * clear it, and do not free the page. But we shall soon need
465 * to do more, for when the ZERO_PAGE count wraps negative.
467 return PageReserved(page
);
471 * Frees a list of pages.
472 * Assumes all pages on list are in same zone, and of same order.
473 * count is the number of pages to free.
475 * If the zone was previously in an "all pages pinned" state then look to
476 * see if this freeing clears that state.
478 * And clear the zone's pages_scanned counter, to hold off the "all pages are
479 * pinned" detection logic.
481 static void free_pages_bulk(struct zone
*zone
, int count
,
482 struct list_head
*list
, int order
)
484 spin_lock(&zone
->lock
);
485 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
486 zone
->pages_scanned
= 0;
490 VM_BUG_ON(list_empty(list
));
491 page
= list_entry(list
->prev
, struct page
, lru
);
492 /* have to delete it as __free_one_page list manipulates */
493 list_del(&page
->lru
);
494 __free_one_page(page
, zone
, order
);
496 spin_unlock(&zone
->lock
);
499 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
501 spin_lock(&zone
->lock
);
502 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
503 zone
->pages_scanned
= 0;
504 __free_one_page(page
, zone
, order
);
505 spin_unlock(&zone
->lock
);
508 static void __free_pages_ok(struct page
*page
, unsigned int order
)
514 for (i
= 0 ; i
< (1 << order
) ; ++i
)
515 reserved
+= free_pages_check(page
+ i
);
519 if (!PageHighMem(page
)) {
520 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
521 debug_check_no_obj_freed(page_address(page
),
524 arch_free_page(page
, order
);
525 kernel_map_pages(page
, 1 << order
, 0);
527 local_irq_save(flags
);
528 __count_vm_events(PGFREE
, 1 << order
);
529 free_one_page(page_zone(page
), page
, order
);
530 local_irq_restore(flags
);
534 * permit the bootmem allocator to evade page validation on high-order frees
536 void __free_pages_bootmem(struct page
*page
, unsigned int order
)
539 __ClearPageReserved(page
);
540 set_page_count(page
, 0);
541 set_page_refcounted(page
);
547 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
548 struct page
*p
= &page
[loop
];
550 if (loop
+ 1 < BITS_PER_LONG
)
552 __ClearPageReserved(p
);
553 set_page_count(p
, 0);
556 set_page_refcounted(page
);
557 __free_pages(page
, order
);
563 * The order of subdivision here is critical for the IO subsystem.
564 * Please do not alter this order without good reasons and regression
565 * testing. Specifically, as large blocks of memory are subdivided,
566 * the order in which smaller blocks are delivered depends on the order
567 * they're subdivided in this function. This is the primary factor
568 * influencing the order in which pages are delivered to the IO
569 * subsystem according to empirical testing, and this is also justified
570 * by considering the behavior of a buddy system containing a single
571 * large block of memory acted on by a series of small allocations.
572 * This behavior is a critical factor in sglist merging's success.
576 static inline void expand(struct zone
*zone
, struct page
*page
,
577 int low
, int high
, struct free_area
*area
,
580 unsigned long size
= 1 << high
;
586 VM_BUG_ON(bad_range(zone
, &page
[size
]));
587 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
589 set_page_order(&page
[size
], high
);
594 * This page is about to be returned from the page allocator
596 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
598 if (unlikely(page_mapcount(page
) |
599 (page
->mapping
!= NULL
) |
600 (page_get_page_cgroup(page
) != NULL
) |
601 (page_count(page
) != 0) |
602 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)))
606 * For now, we report if PG_reserved was found set, but do not
607 * clear it, and do not allocate the page: as a safety net.
609 if (PageReserved(page
))
612 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_reclaim
|
613 1 << PG_referenced
| 1 << PG_arch_1
|
614 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
);
615 set_page_private(page
, 0);
616 set_page_refcounted(page
);
618 arch_alloc_page(page
, order
);
619 kernel_map_pages(page
, 1 << order
, 1);
621 if (gfp_flags
& __GFP_ZERO
)
622 prep_zero_page(page
, order
, gfp_flags
);
624 if (order
&& (gfp_flags
& __GFP_COMP
))
625 prep_compound_page(page
, order
);
631 * Go through the free lists for the given migratetype and remove
632 * the smallest available page from the freelists
634 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
637 unsigned int current_order
;
638 struct free_area
* area
;
641 /* Find a page of the appropriate size in the preferred list */
642 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
643 area
= &(zone
->free_area
[current_order
]);
644 if (list_empty(&area
->free_list
[migratetype
]))
647 page
= list_entry(area
->free_list
[migratetype
].next
,
649 list_del(&page
->lru
);
650 rmv_page_order(page
);
652 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
653 expand(zone
, page
, order
, current_order
, area
, migratetype
);
662 * This array describes the order lists are fallen back to when
663 * the free lists for the desirable migrate type are depleted
665 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
666 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
667 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
668 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
669 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
673 * Move the free pages in a range to the free lists of the requested type.
674 * Note that start_page and end_pages are not aligned on a pageblock
675 * boundary. If alignment is required, use move_freepages_block()
677 int move_freepages(struct zone
*zone
,
678 struct page
*start_page
, struct page
*end_page
,
685 #ifndef CONFIG_HOLES_IN_ZONE
687 * page_zone is not safe to call in this context when
688 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
689 * anyway as we check zone boundaries in move_freepages_block().
690 * Remove at a later date when no bug reports exist related to
691 * grouping pages by mobility
693 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
696 for (page
= start_page
; page
<= end_page
;) {
697 if (!pfn_valid_within(page_to_pfn(page
))) {
702 if (!PageBuddy(page
)) {
707 order
= page_order(page
);
708 list_del(&page
->lru
);
710 &zone
->free_area
[order
].free_list
[migratetype
]);
712 pages_moved
+= 1 << order
;
718 int move_freepages_block(struct zone
*zone
, struct page
*page
, int migratetype
)
720 unsigned long start_pfn
, end_pfn
;
721 struct page
*start_page
, *end_page
;
723 start_pfn
= page_to_pfn(page
);
724 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
725 start_page
= pfn_to_page(start_pfn
);
726 end_page
= start_page
+ pageblock_nr_pages
- 1;
727 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
729 /* Do not cross zone boundaries */
730 if (start_pfn
< zone
->zone_start_pfn
)
732 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
735 return move_freepages(zone
, start_page
, end_page
, migratetype
);
738 /* Remove an element from the buddy allocator from the fallback list */
739 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
740 int start_migratetype
)
742 struct free_area
* area
;
747 /* Find the largest possible block of pages in the other list */
748 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
750 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
751 migratetype
= fallbacks
[start_migratetype
][i
];
753 /* MIGRATE_RESERVE handled later if necessary */
754 if (migratetype
== MIGRATE_RESERVE
)
757 area
= &(zone
->free_area
[current_order
]);
758 if (list_empty(&area
->free_list
[migratetype
]))
761 page
= list_entry(area
->free_list
[migratetype
].next
,
766 * If breaking a large block of pages, move all free
767 * pages to the preferred allocation list. If falling
768 * back for a reclaimable kernel allocation, be more
769 * agressive about taking ownership of free pages
771 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
772 start_migratetype
== MIGRATE_RECLAIMABLE
) {
774 pages
= move_freepages_block(zone
, page
,
777 /* Claim the whole block if over half of it is free */
778 if (pages
>= (1 << (pageblock_order
-1)))
779 set_pageblock_migratetype(page
,
782 migratetype
= start_migratetype
;
785 /* Remove the page from the freelists */
786 list_del(&page
->lru
);
787 rmv_page_order(page
);
788 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
791 if (current_order
== pageblock_order
)
792 set_pageblock_migratetype(page
,
795 expand(zone
, page
, order
, current_order
, area
, migratetype
);
800 /* Use MIGRATE_RESERVE rather than fail an allocation */
801 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
805 * Do the hard work of removing an element from the buddy allocator.
806 * Call me with the zone->lock already held.
808 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
813 page
= __rmqueue_smallest(zone
, order
, migratetype
);
816 page
= __rmqueue_fallback(zone
, order
, migratetype
);
822 * Obtain a specified number of elements from the buddy allocator, all under
823 * a single hold of the lock, for efficiency. Add them to the supplied list.
824 * Returns the number of new pages which were placed at *list.
826 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
827 unsigned long count
, struct list_head
*list
,
832 spin_lock(&zone
->lock
);
833 for (i
= 0; i
< count
; ++i
) {
834 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
835 if (unlikely(page
== NULL
))
839 * Split buddy pages returned by expand() are received here
840 * in physical page order. The page is added to the callers and
841 * list and the list head then moves forward. From the callers
842 * perspective, the linked list is ordered by page number in
843 * some conditions. This is useful for IO devices that can
844 * merge IO requests if the physical pages are ordered
847 list_add(&page
->lru
, list
);
848 set_page_private(page
, migratetype
);
851 spin_unlock(&zone
->lock
);
857 * Called from the vmstat counter updater to drain pagesets of this
858 * currently executing processor on remote nodes after they have
861 * Note that this function must be called with the thread pinned to
862 * a single processor.
864 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
869 local_irq_save(flags
);
870 if (pcp
->count
>= pcp
->batch
)
871 to_drain
= pcp
->batch
;
873 to_drain
= pcp
->count
;
874 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
875 pcp
->count
-= to_drain
;
876 local_irq_restore(flags
);
881 * Drain pages of the indicated processor.
883 * The processor must either be the current processor and the
884 * thread pinned to the current processor or a processor that
887 static void drain_pages(unsigned int cpu
)
892 for_each_zone(zone
) {
893 struct per_cpu_pageset
*pset
;
894 struct per_cpu_pages
*pcp
;
896 if (!populated_zone(zone
))
899 pset
= zone_pcp(zone
, cpu
);
902 local_irq_save(flags
);
903 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
905 local_irq_restore(flags
);
910 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
912 void drain_local_pages(void *arg
)
914 drain_pages(smp_processor_id());
918 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
920 void drain_all_pages(void)
922 on_each_cpu(drain_local_pages
, NULL
, 1);
925 #ifdef CONFIG_HIBERNATION
927 void mark_free_pages(struct zone
*zone
)
929 unsigned long pfn
, max_zone_pfn
;
932 struct list_head
*curr
;
934 if (!zone
->spanned_pages
)
937 spin_lock_irqsave(&zone
->lock
, flags
);
939 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
940 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
941 if (pfn_valid(pfn
)) {
942 struct page
*page
= pfn_to_page(pfn
);
944 if (!swsusp_page_is_forbidden(page
))
945 swsusp_unset_page_free(page
);
948 for_each_migratetype_order(order
, t
) {
949 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
952 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
953 for (i
= 0; i
< (1UL << order
); i
++)
954 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
957 spin_unlock_irqrestore(&zone
->lock
, flags
);
959 #endif /* CONFIG_PM */
962 * Free a 0-order page
964 static void free_hot_cold_page(struct page
*page
, int cold
)
966 struct zone
*zone
= page_zone(page
);
967 struct per_cpu_pages
*pcp
;
971 page
->mapping
= NULL
;
972 if (free_pages_check(page
))
975 if (!PageHighMem(page
)) {
976 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
977 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
979 arch_free_page(page
, 0);
980 kernel_map_pages(page
, 1, 0);
982 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
983 local_irq_save(flags
);
984 __count_vm_event(PGFREE
);
986 list_add_tail(&page
->lru
, &pcp
->list
);
988 list_add(&page
->lru
, &pcp
->list
);
989 set_page_private(page
, get_pageblock_migratetype(page
));
991 if (pcp
->count
>= pcp
->high
) {
992 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
993 pcp
->count
-= pcp
->batch
;
995 local_irq_restore(flags
);
999 void free_hot_page(struct page
*page
)
1001 free_hot_cold_page(page
, 0);
1004 void free_cold_page(struct page
*page
)
1006 free_hot_cold_page(page
, 1);
1010 * split_page takes a non-compound higher-order page, and splits it into
1011 * n (1<<order) sub-pages: page[0..n]
1012 * Each sub-page must be freed individually.
1014 * Note: this is probably too low level an operation for use in drivers.
1015 * Please consult with lkml before using this in your driver.
1017 void split_page(struct page
*page
, unsigned int order
)
1021 VM_BUG_ON(PageCompound(page
));
1022 VM_BUG_ON(!page_count(page
));
1023 for (i
= 1; i
< (1 << order
); i
++)
1024 set_page_refcounted(page
+ i
);
1028 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1029 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1032 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1033 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1035 unsigned long flags
;
1037 int cold
= !!(gfp_flags
& __GFP_COLD
);
1039 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1043 if (likely(order
== 0)) {
1044 struct per_cpu_pages
*pcp
;
1046 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1047 local_irq_save(flags
);
1049 pcp
->count
= rmqueue_bulk(zone
, 0,
1050 pcp
->batch
, &pcp
->list
, migratetype
);
1051 if (unlikely(!pcp
->count
))
1055 /* Find a page of the appropriate migrate type */
1057 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1058 if (page_private(page
) == migratetype
)
1061 list_for_each_entry(page
, &pcp
->list
, lru
)
1062 if (page_private(page
) == migratetype
)
1066 /* Allocate more to the pcp list if necessary */
1067 if (unlikely(&page
->lru
== &pcp
->list
)) {
1068 pcp
->count
+= rmqueue_bulk(zone
, 0,
1069 pcp
->batch
, &pcp
->list
, migratetype
);
1070 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1073 list_del(&page
->lru
);
1076 spin_lock_irqsave(&zone
->lock
, flags
);
1077 page
= __rmqueue(zone
, order
, migratetype
);
1078 spin_unlock(&zone
->lock
);
1083 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1084 zone_statistics(preferred_zone
, zone
);
1085 local_irq_restore(flags
);
1088 VM_BUG_ON(bad_range(zone
, page
));
1089 if (prep_new_page(page
, order
, gfp_flags
))
1094 local_irq_restore(flags
);
1099 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1100 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1101 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1102 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1103 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1104 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1105 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1107 #ifdef CONFIG_FAIL_PAGE_ALLOC
1109 static struct fail_page_alloc_attr
{
1110 struct fault_attr attr
;
1112 u32 ignore_gfp_highmem
;
1113 u32 ignore_gfp_wait
;
1116 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1118 struct dentry
*ignore_gfp_highmem_file
;
1119 struct dentry
*ignore_gfp_wait_file
;
1120 struct dentry
*min_order_file
;
1122 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1124 } fail_page_alloc
= {
1125 .attr
= FAULT_ATTR_INITIALIZER
,
1126 .ignore_gfp_wait
= 1,
1127 .ignore_gfp_highmem
= 1,
1131 static int __init
setup_fail_page_alloc(char *str
)
1133 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1135 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1137 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1139 if (order
< fail_page_alloc
.min_order
)
1141 if (gfp_mask
& __GFP_NOFAIL
)
1143 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1145 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1148 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1151 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1153 static int __init
fail_page_alloc_debugfs(void)
1155 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1159 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1163 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1165 fail_page_alloc
.ignore_gfp_wait_file
=
1166 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1167 &fail_page_alloc
.ignore_gfp_wait
);
1169 fail_page_alloc
.ignore_gfp_highmem_file
=
1170 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1171 &fail_page_alloc
.ignore_gfp_highmem
);
1172 fail_page_alloc
.min_order_file
=
1173 debugfs_create_u32("min-order", mode
, dir
,
1174 &fail_page_alloc
.min_order
);
1176 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1177 !fail_page_alloc
.ignore_gfp_highmem_file
||
1178 !fail_page_alloc
.min_order_file
) {
1180 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1181 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1182 debugfs_remove(fail_page_alloc
.min_order_file
);
1183 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1189 late_initcall(fail_page_alloc_debugfs
);
1191 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1193 #else /* CONFIG_FAIL_PAGE_ALLOC */
1195 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1200 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1203 * Return 1 if free pages are above 'mark'. This takes into account the order
1204 * of the allocation.
1206 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1207 int classzone_idx
, int alloc_flags
)
1209 /* free_pages my go negative - that's OK */
1211 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1214 if (alloc_flags
& ALLOC_HIGH
)
1216 if (alloc_flags
& ALLOC_HARDER
)
1219 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1221 for (o
= 0; o
< order
; o
++) {
1222 /* At the next order, this order's pages become unavailable */
1223 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1225 /* Require fewer higher order pages to be free */
1228 if (free_pages
<= min
)
1236 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1237 * skip over zones that are not allowed by the cpuset, or that have
1238 * been recently (in last second) found to be nearly full. See further
1239 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1240 * that have to skip over a lot of full or unallowed zones.
1242 * If the zonelist cache is present in the passed in zonelist, then
1243 * returns a pointer to the allowed node mask (either the current
1244 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1246 * If the zonelist cache is not available for this zonelist, does
1247 * nothing and returns NULL.
1249 * If the fullzones BITMAP in the zonelist cache is stale (more than
1250 * a second since last zap'd) then we zap it out (clear its bits.)
1252 * We hold off even calling zlc_setup, until after we've checked the
1253 * first zone in the zonelist, on the theory that most allocations will
1254 * be satisfied from that first zone, so best to examine that zone as
1255 * quickly as we can.
1257 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1259 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1260 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1262 zlc
= zonelist
->zlcache_ptr
;
1266 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1267 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1268 zlc
->last_full_zap
= jiffies
;
1271 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1272 &cpuset_current_mems_allowed
:
1273 &node_states
[N_HIGH_MEMORY
];
1274 return allowednodes
;
1278 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1279 * if it is worth looking at further for free memory:
1280 * 1) Check that the zone isn't thought to be full (doesn't have its
1281 * bit set in the zonelist_cache fullzones BITMAP).
1282 * 2) Check that the zones node (obtained from the zonelist_cache
1283 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1284 * Return true (non-zero) if zone is worth looking at further, or
1285 * else return false (zero) if it is not.
1287 * This check -ignores- the distinction between various watermarks,
1288 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1289 * found to be full for any variation of these watermarks, it will
1290 * be considered full for up to one second by all requests, unless
1291 * we are so low on memory on all allowed nodes that we are forced
1292 * into the second scan of the zonelist.
1294 * In the second scan we ignore this zonelist cache and exactly
1295 * apply the watermarks to all zones, even it is slower to do so.
1296 * We are low on memory in the second scan, and should leave no stone
1297 * unturned looking for a free page.
1299 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1300 nodemask_t
*allowednodes
)
1302 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1303 int i
; /* index of *z in zonelist zones */
1304 int n
; /* node that zone *z is on */
1306 zlc
= zonelist
->zlcache_ptr
;
1310 i
= z
- zonelist
->_zonerefs
;
1313 /* This zone is worth trying if it is allowed but not full */
1314 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1318 * Given 'z' scanning a zonelist, set the corresponding bit in
1319 * zlc->fullzones, so that subsequent attempts to allocate a page
1320 * from that zone don't waste time re-examining it.
1322 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1324 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1325 int i
; /* index of *z in zonelist zones */
1327 zlc
= zonelist
->zlcache_ptr
;
1331 i
= z
- zonelist
->_zonerefs
;
1333 set_bit(i
, zlc
->fullzones
);
1336 #else /* CONFIG_NUMA */
1338 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1343 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1344 nodemask_t
*allowednodes
)
1349 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1352 #endif /* CONFIG_NUMA */
1355 * get_page_from_freelist goes through the zonelist trying to allocate
1358 static struct page
*
1359 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1360 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1363 struct page
*page
= NULL
;
1365 struct zone
*zone
, *preferred_zone
;
1366 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1367 int zlc_active
= 0; /* set if using zonelist_cache */
1368 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1370 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1372 if (!preferred_zone
)
1375 classzone_idx
= zone_idx(preferred_zone
);
1379 * Scan zonelist, looking for a zone with enough free.
1380 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1382 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1383 high_zoneidx
, nodemask
) {
1384 if (NUMA_BUILD
&& zlc_active
&&
1385 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1387 if ((alloc_flags
& ALLOC_CPUSET
) &&
1388 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1391 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1393 if (alloc_flags
& ALLOC_WMARK_MIN
)
1394 mark
= zone
->pages_min
;
1395 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1396 mark
= zone
->pages_low
;
1398 mark
= zone
->pages_high
;
1399 if (!zone_watermark_ok(zone
, order
, mark
,
1400 classzone_idx
, alloc_flags
)) {
1401 if (!zone_reclaim_mode
||
1402 !zone_reclaim(zone
, gfp_mask
, order
))
1403 goto this_zone_full
;
1407 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1412 zlc_mark_zone_full(zonelist
, z
);
1414 if (NUMA_BUILD
&& !did_zlc_setup
) {
1415 /* we do zlc_setup after the first zone is tried */
1416 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1422 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1423 /* Disable zlc cache for second zonelist scan */
1431 * This is the 'heart' of the zoned buddy allocator.
1434 __alloc_pages_internal(gfp_t gfp_mask
, unsigned int order
,
1435 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1437 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1438 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1442 struct reclaim_state reclaim_state
;
1443 struct task_struct
*p
= current
;
1446 unsigned long did_some_progress
;
1447 unsigned long pages_reclaimed
= 0;
1449 might_sleep_if(wait
);
1451 if (should_fail_alloc_page(gfp_mask
, order
))
1455 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1457 if (unlikely(!z
->zone
)) {
1459 * Happens if we have an empty zonelist as a result of
1460 * GFP_THISNODE being used on a memoryless node
1465 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1466 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1471 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1472 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1473 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1474 * using a larger set of nodes after it has established that the
1475 * allowed per node queues are empty and that nodes are
1478 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1481 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1482 wakeup_kswapd(zone
, order
);
1485 * OK, we're below the kswapd watermark and have kicked background
1486 * reclaim. Now things get more complex, so set up alloc_flags according
1487 * to how we want to proceed.
1489 * The caller may dip into page reserves a bit more if the caller
1490 * cannot run direct reclaim, or if the caller has realtime scheduling
1491 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1492 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1494 alloc_flags
= ALLOC_WMARK_MIN
;
1495 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1496 alloc_flags
|= ALLOC_HARDER
;
1497 if (gfp_mask
& __GFP_HIGH
)
1498 alloc_flags
|= ALLOC_HIGH
;
1500 alloc_flags
|= ALLOC_CPUSET
;
1503 * Go through the zonelist again. Let __GFP_HIGH and allocations
1504 * coming from realtime tasks go deeper into reserves.
1506 * This is the last chance, in general, before the goto nopage.
1507 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1508 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1510 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1511 high_zoneidx
, alloc_flags
);
1515 /* This allocation should allow future memory freeing. */
1518 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1519 && !in_interrupt()) {
1520 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1522 /* go through the zonelist yet again, ignoring mins */
1523 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1524 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1527 if (gfp_mask
& __GFP_NOFAIL
) {
1528 congestion_wait(WRITE
, HZ
/50);
1535 /* Atomic allocations - we can't balance anything */
1541 /* We now go into synchronous reclaim */
1542 cpuset_memory_pressure_bump();
1543 p
->flags
|= PF_MEMALLOC
;
1544 reclaim_state
.reclaimed_slab
= 0;
1545 p
->reclaim_state
= &reclaim_state
;
1547 did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
);
1549 p
->reclaim_state
= NULL
;
1550 p
->flags
&= ~PF_MEMALLOC
;
1557 if (likely(did_some_progress
)) {
1558 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1559 zonelist
, high_zoneidx
, alloc_flags
);
1562 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1563 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1564 schedule_timeout_uninterruptible(1);
1569 * Go through the zonelist yet one more time, keep
1570 * very high watermark here, this is only to catch
1571 * a parallel oom killing, we must fail if we're still
1572 * under heavy pressure.
1574 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1575 order
, zonelist
, high_zoneidx
,
1576 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1578 clear_zonelist_oom(zonelist
, gfp_mask
);
1582 /* The OOM killer will not help higher order allocs so fail */
1583 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1584 clear_zonelist_oom(zonelist
, gfp_mask
);
1588 out_of_memory(zonelist
, gfp_mask
, order
);
1589 clear_zonelist_oom(zonelist
, gfp_mask
);
1594 * Don't let big-order allocations loop unless the caller explicitly
1595 * requests that. Wait for some write requests to complete then retry.
1597 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1598 * means __GFP_NOFAIL, but that may not be true in other
1601 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1602 * specified, then we retry until we no longer reclaim any pages
1603 * (above), or we've reclaimed an order of pages at least as
1604 * large as the allocation's order. In both cases, if the
1605 * allocation still fails, we stop retrying.
1607 pages_reclaimed
+= did_some_progress
;
1609 if (!(gfp_mask
& __GFP_NORETRY
)) {
1610 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1613 if (gfp_mask
& __GFP_REPEAT
&&
1614 pages_reclaimed
< (1 << order
))
1617 if (gfp_mask
& __GFP_NOFAIL
)
1621 congestion_wait(WRITE
, HZ
/50);
1626 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1627 printk(KERN_WARNING
"%s: page allocation failure."
1628 " order:%d, mode:0x%x\n",
1629 p
->comm
, order
, gfp_mask
);
1636 EXPORT_SYMBOL(__alloc_pages_internal
);
1639 * Common helper functions.
1641 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1644 page
= alloc_pages(gfp_mask
, order
);
1647 return (unsigned long) page_address(page
);
1650 EXPORT_SYMBOL(__get_free_pages
);
1652 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1657 * get_zeroed_page() returns a 32-bit address, which cannot represent
1660 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1662 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1664 return (unsigned long) page_address(page
);
1668 EXPORT_SYMBOL(get_zeroed_page
);
1670 void __pagevec_free(struct pagevec
*pvec
)
1672 int i
= pagevec_count(pvec
);
1675 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1678 void __free_pages(struct page
*page
, unsigned int order
)
1680 if (put_page_testzero(page
)) {
1682 free_hot_page(page
);
1684 __free_pages_ok(page
, order
);
1688 EXPORT_SYMBOL(__free_pages
);
1690 void free_pages(unsigned long addr
, unsigned int order
)
1693 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1694 __free_pages(virt_to_page((void *)addr
), order
);
1698 EXPORT_SYMBOL(free_pages
);
1701 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1702 * @size: the number of bytes to allocate
1703 * @gfp_mask: GFP flags for the allocation
1705 * This function is similar to alloc_pages(), except that it allocates the
1706 * minimum number of pages to satisfy the request. alloc_pages() can only
1707 * allocate memory in power-of-two pages.
1709 * This function is also limited by MAX_ORDER.
1711 * Memory allocated by this function must be released by free_pages_exact().
1713 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1715 unsigned int order
= get_order(size
);
1718 addr
= __get_free_pages(gfp_mask
, order
);
1720 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1721 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1723 split_page(virt_to_page(addr
), order
);
1724 while (used
< alloc_end
) {
1730 return (void *)addr
;
1732 EXPORT_SYMBOL(alloc_pages_exact
);
1735 * free_pages_exact - release memory allocated via alloc_pages_exact()
1736 * @virt: the value returned by alloc_pages_exact.
1737 * @size: size of allocation, same value as passed to alloc_pages_exact().
1739 * Release the memory allocated by a previous call to alloc_pages_exact.
1741 void free_pages_exact(void *virt
, size_t size
)
1743 unsigned long addr
= (unsigned long)virt
;
1744 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1746 while (addr
< end
) {
1751 EXPORT_SYMBOL(free_pages_exact
);
1753 static unsigned int nr_free_zone_pages(int offset
)
1758 /* Just pick one node, since fallback list is circular */
1759 unsigned int sum
= 0;
1761 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1763 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1764 unsigned long size
= zone
->present_pages
;
1765 unsigned long high
= zone
->pages_high
;
1774 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1776 unsigned int nr_free_buffer_pages(void)
1778 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1780 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1783 * Amount of free RAM allocatable within all zones
1785 unsigned int nr_free_pagecache_pages(void)
1787 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1790 static inline void show_node(struct zone
*zone
)
1793 printk("Node %d ", zone_to_nid(zone
));
1796 void si_meminfo(struct sysinfo
*val
)
1798 val
->totalram
= totalram_pages
;
1800 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1801 val
->bufferram
= nr_blockdev_pages();
1802 val
->totalhigh
= totalhigh_pages
;
1803 val
->freehigh
= nr_free_highpages();
1804 val
->mem_unit
= PAGE_SIZE
;
1807 EXPORT_SYMBOL(si_meminfo
);
1810 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1812 pg_data_t
*pgdat
= NODE_DATA(nid
);
1814 val
->totalram
= pgdat
->node_present_pages
;
1815 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1816 #ifdef CONFIG_HIGHMEM
1817 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1818 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1824 val
->mem_unit
= PAGE_SIZE
;
1828 #define K(x) ((x) << (PAGE_SHIFT-10))
1831 * Show free area list (used inside shift_scroll-lock stuff)
1832 * We also calculate the percentage fragmentation. We do this by counting the
1833 * memory on each free list with the exception of the first item on the list.
1835 void show_free_areas(void)
1840 for_each_zone(zone
) {
1841 if (!populated_zone(zone
))
1845 printk("%s per-cpu:\n", zone
->name
);
1847 for_each_online_cpu(cpu
) {
1848 struct per_cpu_pageset
*pageset
;
1850 pageset
= zone_pcp(zone
, cpu
);
1852 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1853 cpu
, pageset
->pcp
.high
,
1854 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1858 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1859 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1860 global_page_state(NR_ACTIVE
),
1861 global_page_state(NR_INACTIVE
),
1862 global_page_state(NR_FILE_DIRTY
),
1863 global_page_state(NR_WRITEBACK
),
1864 global_page_state(NR_UNSTABLE_NFS
),
1865 global_page_state(NR_FREE_PAGES
),
1866 global_page_state(NR_SLAB_RECLAIMABLE
) +
1867 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1868 global_page_state(NR_FILE_MAPPED
),
1869 global_page_state(NR_PAGETABLE
),
1870 global_page_state(NR_BOUNCE
));
1872 for_each_zone(zone
) {
1875 if (!populated_zone(zone
))
1887 " pages_scanned:%lu"
1888 " all_unreclaimable? %s"
1891 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1894 K(zone
->pages_high
),
1895 K(zone_page_state(zone
, NR_ACTIVE
)),
1896 K(zone_page_state(zone
, NR_INACTIVE
)),
1897 K(zone
->present_pages
),
1898 zone
->pages_scanned
,
1899 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1901 printk("lowmem_reserve[]:");
1902 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1903 printk(" %lu", zone
->lowmem_reserve
[i
]);
1907 for_each_zone(zone
) {
1908 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1910 if (!populated_zone(zone
))
1914 printk("%s: ", zone
->name
);
1916 spin_lock_irqsave(&zone
->lock
, flags
);
1917 for (order
= 0; order
< MAX_ORDER
; order
++) {
1918 nr
[order
] = zone
->free_area
[order
].nr_free
;
1919 total
+= nr
[order
] << order
;
1921 spin_unlock_irqrestore(&zone
->lock
, flags
);
1922 for (order
= 0; order
< MAX_ORDER
; order
++)
1923 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1924 printk("= %lukB\n", K(total
));
1927 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
1929 show_swap_cache_info();
1932 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
1934 zoneref
->zone
= zone
;
1935 zoneref
->zone_idx
= zone_idx(zone
);
1939 * Builds allocation fallback zone lists.
1941 * Add all populated zones of a node to the zonelist.
1943 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1944 int nr_zones
, enum zone_type zone_type
)
1948 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1953 zone
= pgdat
->node_zones
+ zone_type
;
1954 if (populated_zone(zone
)) {
1955 zoneref_set_zone(zone
,
1956 &zonelist
->_zonerefs
[nr_zones
++]);
1957 check_highest_zone(zone_type
);
1960 } while (zone_type
);
1967 * 0 = automatic detection of better ordering.
1968 * 1 = order by ([node] distance, -zonetype)
1969 * 2 = order by (-zonetype, [node] distance)
1971 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1972 * the same zonelist. So only NUMA can configure this param.
1974 #define ZONELIST_ORDER_DEFAULT 0
1975 #define ZONELIST_ORDER_NODE 1
1976 #define ZONELIST_ORDER_ZONE 2
1978 /* zonelist order in the kernel.
1979 * set_zonelist_order() will set this to NODE or ZONE.
1981 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1982 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
1986 /* The value user specified ....changed by config */
1987 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1988 /* string for sysctl */
1989 #define NUMA_ZONELIST_ORDER_LEN 16
1990 char numa_zonelist_order
[16] = "default";
1993 * interface for configure zonelist ordering.
1994 * command line option "numa_zonelist_order"
1995 * = "[dD]efault - default, automatic configuration.
1996 * = "[nN]ode - order by node locality, then by zone within node
1997 * = "[zZ]one - order by zone, then by locality within zone
2000 static int __parse_numa_zonelist_order(char *s
)
2002 if (*s
== 'd' || *s
== 'D') {
2003 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2004 } else if (*s
== 'n' || *s
== 'N') {
2005 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2006 } else if (*s
== 'z' || *s
== 'Z') {
2007 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2010 "Ignoring invalid numa_zonelist_order value: "
2017 static __init
int setup_numa_zonelist_order(char *s
)
2020 return __parse_numa_zonelist_order(s
);
2023 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2026 * sysctl handler for numa_zonelist_order
2028 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2029 struct file
*file
, void __user
*buffer
, size_t *length
,
2032 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2036 strncpy(saved_string
, (char*)table
->data
,
2037 NUMA_ZONELIST_ORDER_LEN
);
2038 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2042 int oldval
= user_zonelist_order
;
2043 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2045 * bogus value. restore saved string
2047 strncpy((char*)table
->data
, saved_string
,
2048 NUMA_ZONELIST_ORDER_LEN
);
2049 user_zonelist_order
= oldval
;
2050 } else if (oldval
!= user_zonelist_order
)
2051 build_all_zonelists();
2057 #define MAX_NODE_LOAD (num_online_nodes())
2058 static int node_load
[MAX_NUMNODES
];
2061 * find_next_best_node - find the next node that should appear in a given node's fallback list
2062 * @node: node whose fallback list we're appending
2063 * @used_node_mask: nodemask_t of already used nodes
2065 * We use a number of factors to determine which is the next node that should
2066 * appear on a given node's fallback list. The node should not have appeared
2067 * already in @node's fallback list, and it should be the next closest node
2068 * according to the distance array (which contains arbitrary distance values
2069 * from each node to each node in the system), and should also prefer nodes
2070 * with no CPUs, since presumably they'll have very little allocation pressure
2071 * on them otherwise.
2072 * It returns -1 if no node is found.
2074 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2077 int min_val
= INT_MAX
;
2079 node_to_cpumask_ptr(tmp
, 0);
2081 /* Use the local node if we haven't already */
2082 if (!node_isset(node
, *used_node_mask
)) {
2083 node_set(node
, *used_node_mask
);
2087 for_each_node_state(n
, N_HIGH_MEMORY
) {
2089 /* Don't want a node to appear more than once */
2090 if (node_isset(n
, *used_node_mask
))
2093 /* Use the distance array to find the distance */
2094 val
= node_distance(node
, n
);
2096 /* Penalize nodes under us ("prefer the next node") */
2099 /* Give preference to headless and unused nodes */
2100 node_to_cpumask_ptr_next(tmp
, n
);
2101 if (!cpus_empty(*tmp
))
2102 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2104 /* Slight preference for less loaded node */
2105 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2106 val
+= node_load
[n
];
2108 if (val
< min_val
) {
2115 node_set(best_node
, *used_node_mask
);
2122 * Build zonelists ordered by node and zones within node.
2123 * This results in maximum locality--normal zone overflows into local
2124 * DMA zone, if any--but risks exhausting DMA zone.
2126 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2129 struct zonelist
*zonelist
;
2131 zonelist
= &pgdat
->node_zonelists
[0];
2132 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2134 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2136 zonelist
->_zonerefs
[j
].zone
= NULL
;
2137 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2141 * Build gfp_thisnode zonelists
2143 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2146 struct zonelist
*zonelist
;
2148 zonelist
= &pgdat
->node_zonelists
[1];
2149 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2150 zonelist
->_zonerefs
[j
].zone
= NULL
;
2151 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2155 * Build zonelists ordered by zone and nodes within zones.
2156 * This results in conserving DMA zone[s] until all Normal memory is
2157 * exhausted, but results in overflowing to remote node while memory
2158 * may still exist in local DMA zone.
2160 static int node_order
[MAX_NUMNODES
];
2162 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2165 int zone_type
; /* needs to be signed */
2167 struct zonelist
*zonelist
;
2169 zonelist
= &pgdat
->node_zonelists
[0];
2171 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2172 for (j
= 0; j
< nr_nodes
; j
++) {
2173 node
= node_order
[j
];
2174 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2175 if (populated_zone(z
)) {
2177 &zonelist
->_zonerefs
[pos
++]);
2178 check_highest_zone(zone_type
);
2182 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2183 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2186 static int default_zonelist_order(void)
2189 unsigned long low_kmem_size
,total_size
;
2193 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2194 * If they are really small and used heavily, the system can fall
2195 * into OOM very easily.
2196 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2198 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2201 for_each_online_node(nid
) {
2202 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2203 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2204 if (populated_zone(z
)) {
2205 if (zone_type
< ZONE_NORMAL
)
2206 low_kmem_size
+= z
->present_pages
;
2207 total_size
+= z
->present_pages
;
2211 if (!low_kmem_size
|| /* there are no DMA area. */
2212 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2213 return ZONELIST_ORDER_NODE
;
2215 * look into each node's config.
2216 * If there is a node whose DMA/DMA32 memory is very big area on
2217 * local memory, NODE_ORDER may be suitable.
2219 average_size
= total_size
/
2220 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2221 for_each_online_node(nid
) {
2224 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2225 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2226 if (populated_zone(z
)) {
2227 if (zone_type
< ZONE_NORMAL
)
2228 low_kmem_size
+= z
->present_pages
;
2229 total_size
+= z
->present_pages
;
2232 if (low_kmem_size
&&
2233 total_size
> average_size
&& /* ignore small node */
2234 low_kmem_size
> total_size
* 70/100)
2235 return ZONELIST_ORDER_NODE
;
2237 return ZONELIST_ORDER_ZONE
;
2240 static void set_zonelist_order(void)
2242 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2243 current_zonelist_order
= default_zonelist_order();
2245 current_zonelist_order
= user_zonelist_order
;
2248 static void build_zonelists(pg_data_t
*pgdat
)
2252 nodemask_t used_mask
;
2253 int local_node
, prev_node
;
2254 struct zonelist
*zonelist
;
2255 int order
= current_zonelist_order
;
2257 /* initialize zonelists */
2258 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2259 zonelist
= pgdat
->node_zonelists
+ i
;
2260 zonelist
->_zonerefs
[0].zone
= NULL
;
2261 zonelist
->_zonerefs
[0].zone_idx
= 0;
2264 /* NUMA-aware ordering of nodes */
2265 local_node
= pgdat
->node_id
;
2266 load
= num_online_nodes();
2267 prev_node
= local_node
;
2268 nodes_clear(used_mask
);
2270 memset(node_load
, 0, sizeof(node_load
));
2271 memset(node_order
, 0, sizeof(node_order
));
2274 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2275 int distance
= node_distance(local_node
, node
);
2278 * If another node is sufficiently far away then it is better
2279 * to reclaim pages in a zone before going off node.
2281 if (distance
> RECLAIM_DISTANCE
)
2282 zone_reclaim_mode
= 1;
2285 * We don't want to pressure a particular node.
2286 * So adding penalty to the first node in same
2287 * distance group to make it round-robin.
2289 if (distance
!= node_distance(local_node
, prev_node
))
2290 node_load
[node
] = load
;
2294 if (order
== ZONELIST_ORDER_NODE
)
2295 build_zonelists_in_node_order(pgdat
, node
);
2297 node_order
[j
++] = node
; /* remember order */
2300 if (order
== ZONELIST_ORDER_ZONE
) {
2301 /* calculate node order -- i.e., DMA last! */
2302 build_zonelists_in_zone_order(pgdat
, j
);
2305 build_thisnode_zonelists(pgdat
);
2308 /* Construct the zonelist performance cache - see further mmzone.h */
2309 static void build_zonelist_cache(pg_data_t
*pgdat
)
2311 struct zonelist
*zonelist
;
2312 struct zonelist_cache
*zlc
;
2315 zonelist
= &pgdat
->node_zonelists
[0];
2316 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2317 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2318 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2319 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2323 #else /* CONFIG_NUMA */
2325 static void set_zonelist_order(void)
2327 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2330 static void build_zonelists(pg_data_t
*pgdat
)
2332 int node
, local_node
;
2334 struct zonelist
*zonelist
;
2336 local_node
= pgdat
->node_id
;
2338 zonelist
= &pgdat
->node_zonelists
[0];
2339 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2342 * Now we build the zonelist so that it contains the zones
2343 * of all the other nodes.
2344 * We don't want to pressure a particular node, so when
2345 * building the zones for node N, we make sure that the
2346 * zones coming right after the local ones are those from
2347 * node N+1 (modulo N)
2349 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2350 if (!node_online(node
))
2352 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2355 for (node
= 0; node
< local_node
; node
++) {
2356 if (!node_online(node
))
2358 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2362 zonelist
->_zonerefs
[j
].zone
= NULL
;
2363 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2366 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2367 static void build_zonelist_cache(pg_data_t
*pgdat
)
2369 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2372 #endif /* CONFIG_NUMA */
2374 /* return values int ....just for stop_machine_run() */
2375 static int __build_all_zonelists(void *dummy
)
2379 for_each_online_node(nid
) {
2380 pg_data_t
*pgdat
= NODE_DATA(nid
);
2382 build_zonelists(pgdat
);
2383 build_zonelist_cache(pgdat
);
2388 void build_all_zonelists(void)
2390 set_zonelist_order();
2392 if (system_state
== SYSTEM_BOOTING
) {
2393 __build_all_zonelists(NULL
);
2394 mminit_verify_zonelist();
2395 cpuset_init_current_mems_allowed();
2397 /* we have to stop all cpus to guarantee there is no user
2399 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
2400 /* cpuset refresh routine should be here */
2402 vm_total_pages
= nr_free_pagecache_pages();
2404 * Disable grouping by mobility if the number of pages in the
2405 * system is too low to allow the mechanism to work. It would be
2406 * more accurate, but expensive to check per-zone. This check is
2407 * made on memory-hotadd so a system can start with mobility
2408 * disabled and enable it later
2410 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2411 page_group_by_mobility_disabled
= 1;
2413 page_group_by_mobility_disabled
= 0;
2415 printk("Built %i zonelists in %s order, mobility grouping %s. "
2416 "Total pages: %ld\n",
2418 zonelist_order_name
[current_zonelist_order
],
2419 page_group_by_mobility_disabled
? "off" : "on",
2422 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2427 * Helper functions to size the waitqueue hash table.
2428 * Essentially these want to choose hash table sizes sufficiently
2429 * large so that collisions trying to wait on pages are rare.
2430 * But in fact, the number of active page waitqueues on typical
2431 * systems is ridiculously low, less than 200. So this is even
2432 * conservative, even though it seems large.
2434 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2435 * waitqueues, i.e. the size of the waitq table given the number of pages.
2437 #define PAGES_PER_WAITQUEUE 256
2439 #ifndef CONFIG_MEMORY_HOTPLUG
2440 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2442 unsigned long size
= 1;
2444 pages
/= PAGES_PER_WAITQUEUE
;
2446 while (size
< pages
)
2450 * Once we have dozens or even hundreds of threads sleeping
2451 * on IO we've got bigger problems than wait queue collision.
2452 * Limit the size of the wait table to a reasonable size.
2454 size
= min(size
, 4096UL);
2456 return max(size
, 4UL);
2460 * A zone's size might be changed by hot-add, so it is not possible to determine
2461 * a suitable size for its wait_table. So we use the maximum size now.
2463 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2465 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2466 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2467 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2469 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2470 * or more by the traditional way. (See above). It equals:
2472 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2473 * ia64(16K page size) : = ( 8G + 4M)byte.
2474 * powerpc (64K page size) : = (32G +16M)byte.
2476 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2483 * This is an integer logarithm so that shifts can be used later
2484 * to extract the more random high bits from the multiplicative
2485 * hash function before the remainder is taken.
2487 static inline unsigned long wait_table_bits(unsigned long size
)
2492 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2495 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2496 * of blocks reserved is based on zone->pages_min. The memory within the
2497 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2498 * higher will lead to a bigger reserve which will get freed as contiguous
2499 * blocks as reclaim kicks in
2501 static void setup_zone_migrate_reserve(struct zone
*zone
)
2503 unsigned long start_pfn
, pfn
, end_pfn
;
2505 unsigned long reserve
, block_migratetype
;
2507 /* Get the start pfn, end pfn and the number of blocks to reserve */
2508 start_pfn
= zone
->zone_start_pfn
;
2509 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2510 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2513 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2514 if (!pfn_valid(pfn
))
2516 page
= pfn_to_page(pfn
);
2518 /* Blocks with reserved pages will never free, skip them. */
2519 if (PageReserved(page
))
2522 block_migratetype
= get_pageblock_migratetype(page
);
2524 /* If this block is reserved, account for it */
2525 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2530 /* Suitable for reserving if this block is movable */
2531 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2532 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2533 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2539 * If the reserve is met and this is a previous reserved block,
2542 if (block_migratetype
== MIGRATE_RESERVE
) {
2543 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2544 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2550 * Initially all pages are reserved - free ones are freed
2551 * up by free_all_bootmem() once the early boot process is
2552 * done. Non-atomic initialization, single-pass.
2554 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2555 unsigned long start_pfn
, enum memmap_context context
)
2558 unsigned long end_pfn
= start_pfn
+ size
;
2562 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2563 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2565 * There can be holes in boot-time mem_map[]s
2566 * handed to this function. They do not
2567 * exist on hotplugged memory.
2569 if (context
== MEMMAP_EARLY
) {
2570 if (!early_pfn_valid(pfn
))
2572 if (!early_pfn_in_nid(pfn
, nid
))
2575 page
= pfn_to_page(pfn
);
2576 set_page_links(page
, zone
, nid
, pfn
);
2577 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2578 init_page_count(page
);
2579 reset_page_mapcount(page
);
2580 SetPageReserved(page
);
2582 * Mark the block movable so that blocks are reserved for
2583 * movable at startup. This will force kernel allocations
2584 * to reserve their blocks rather than leaking throughout
2585 * the address space during boot when many long-lived
2586 * kernel allocations are made. Later some blocks near
2587 * the start are marked MIGRATE_RESERVE by
2588 * setup_zone_migrate_reserve()
2590 * bitmap is created for zone's valid pfn range. but memmap
2591 * can be created for invalid pages (for alignment)
2592 * check here not to call set_pageblock_migratetype() against
2595 if ((z
->zone_start_pfn
<= pfn
)
2596 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2597 && !(pfn
& (pageblock_nr_pages
- 1)))
2598 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2600 INIT_LIST_HEAD(&page
->lru
);
2601 #ifdef WANT_PAGE_VIRTUAL
2602 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2603 if (!is_highmem_idx(zone
))
2604 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2609 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2612 for_each_migratetype_order(order
, t
) {
2613 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2614 zone
->free_area
[order
].nr_free
= 0;
2618 #ifndef __HAVE_ARCH_MEMMAP_INIT
2619 #define memmap_init(size, nid, zone, start_pfn) \
2620 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2623 static int zone_batchsize(struct zone
*zone
)
2628 * The per-cpu-pages pools are set to around 1000th of the
2629 * size of the zone. But no more than 1/2 of a meg.
2631 * OK, so we don't know how big the cache is. So guess.
2633 batch
= zone
->present_pages
/ 1024;
2634 if (batch
* PAGE_SIZE
> 512 * 1024)
2635 batch
= (512 * 1024) / PAGE_SIZE
;
2636 batch
/= 4; /* We effectively *= 4 below */
2641 * Clamp the batch to a 2^n - 1 value. Having a power
2642 * of 2 value was found to be more likely to have
2643 * suboptimal cache aliasing properties in some cases.
2645 * For example if 2 tasks are alternately allocating
2646 * batches of pages, one task can end up with a lot
2647 * of pages of one half of the possible page colors
2648 * and the other with pages of the other colors.
2650 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2655 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2657 struct per_cpu_pages
*pcp
;
2659 memset(p
, 0, sizeof(*p
));
2663 pcp
->high
= 6 * batch
;
2664 pcp
->batch
= max(1UL, 1 * batch
);
2665 INIT_LIST_HEAD(&pcp
->list
);
2669 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2670 * to the value high for the pageset p.
2673 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2676 struct per_cpu_pages
*pcp
;
2680 pcp
->batch
= max(1UL, high
/4);
2681 if ((high
/4) > (PAGE_SHIFT
* 8))
2682 pcp
->batch
= PAGE_SHIFT
* 8;
2688 * Boot pageset table. One per cpu which is going to be used for all
2689 * zones and all nodes. The parameters will be set in such a way
2690 * that an item put on a list will immediately be handed over to
2691 * the buddy list. This is safe since pageset manipulation is done
2692 * with interrupts disabled.
2694 * Some NUMA counter updates may also be caught by the boot pagesets.
2696 * The boot_pagesets must be kept even after bootup is complete for
2697 * unused processors and/or zones. They do play a role for bootstrapping
2698 * hotplugged processors.
2700 * zoneinfo_show() and maybe other functions do
2701 * not check if the processor is online before following the pageset pointer.
2702 * Other parts of the kernel may not check if the zone is available.
2704 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2707 * Dynamically allocate memory for the
2708 * per cpu pageset array in struct zone.
2710 static int __cpuinit
process_zones(int cpu
)
2712 struct zone
*zone
, *dzone
;
2713 int node
= cpu_to_node(cpu
);
2715 node_set_state(node
, N_CPU
); /* this node has a cpu */
2717 for_each_zone(zone
) {
2719 if (!populated_zone(zone
))
2722 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2724 if (!zone_pcp(zone
, cpu
))
2727 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2729 if (percpu_pagelist_fraction
)
2730 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2731 (zone
->present_pages
/ percpu_pagelist_fraction
));
2736 for_each_zone(dzone
) {
2737 if (!populated_zone(dzone
))
2741 kfree(zone_pcp(dzone
, cpu
));
2742 zone_pcp(dzone
, cpu
) = NULL
;
2747 static inline void free_zone_pagesets(int cpu
)
2751 for_each_zone(zone
) {
2752 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2754 /* Free per_cpu_pageset if it is slab allocated */
2755 if (pset
!= &boot_pageset
[cpu
])
2757 zone_pcp(zone
, cpu
) = NULL
;
2761 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2762 unsigned long action
,
2765 int cpu
= (long)hcpu
;
2766 int ret
= NOTIFY_OK
;
2769 case CPU_UP_PREPARE
:
2770 case CPU_UP_PREPARE_FROZEN
:
2771 if (process_zones(cpu
))
2774 case CPU_UP_CANCELED
:
2775 case CPU_UP_CANCELED_FROZEN
:
2777 case CPU_DEAD_FROZEN
:
2778 free_zone_pagesets(cpu
);
2786 static struct notifier_block __cpuinitdata pageset_notifier
=
2787 { &pageset_cpuup_callback
, NULL
, 0 };
2789 void __init
setup_per_cpu_pageset(void)
2793 /* Initialize per_cpu_pageset for cpu 0.
2794 * A cpuup callback will do this for every cpu
2795 * as it comes online
2797 err
= process_zones(smp_processor_id());
2799 register_cpu_notifier(&pageset_notifier
);
2804 static noinline __init_refok
2805 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2808 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2812 * The per-page waitqueue mechanism uses hashed waitqueues
2815 zone
->wait_table_hash_nr_entries
=
2816 wait_table_hash_nr_entries(zone_size_pages
);
2817 zone
->wait_table_bits
=
2818 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2819 alloc_size
= zone
->wait_table_hash_nr_entries
2820 * sizeof(wait_queue_head_t
);
2822 if (!slab_is_available()) {
2823 zone
->wait_table
= (wait_queue_head_t
*)
2824 alloc_bootmem_node(pgdat
, alloc_size
);
2827 * This case means that a zone whose size was 0 gets new memory
2828 * via memory hot-add.
2829 * But it may be the case that a new node was hot-added. In
2830 * this case vmalloc() will not be able to use this new node's
2831 * memory - this wait_table must be initialized to use this new
2832 * node itself as well.
2833 * To use this new node's memory, further consideration will be
2836 zone
->wait_table
= vmalloc(alloc_size
);
2838 if (!zone
->wait_table
)
2841 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2842 init_waitqueue_head(zone
->wait_table
+ i
);
2847 static __meminit
void zone_pcp_init(struct zone
*zone
)
2850 unsigned long batch
= zone_batchsize(zone
);
2852 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2854 /* Early boot. Slab allocator not functional yet */
2855 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2856 setup_pageset(&boot_pageset
[cpu
],0);
2858 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2861 if (zone
->present_pages
)
2862 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2863 zone
->name
, zone
->present_pages
, batch
);
2866 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2867 unsigned long zone_start_pfn
,
2869 enum memmap_context context
)
2871 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2873 ret
= zone_wait_table_init(zone
, size
);
2876 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2878 zone
->zone_start_pfn
= zone_start_pfn
;
2880 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
2881 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2883 (unsigned long)zone_idx(zone
),
2884 zone_start_pfn
, (zone_start_pfn
+ size
));
2886 zone_init_free_lists(zone
);
2891 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2893 * Basic iterator support. Return the first range of PFNs for a node
2894 * Note: nid == MAX_NUMNODES returns first region regardless of node
2896 static int __meminit
first_active_region_index_in_nid(int nid
)
2900 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2901 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2908 * Basic iterator support. Return the next active range of PFNs for a node
2909 * Note: nid == MAX_NUMNODES returns next region regardless of node
2911 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2913 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2914 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2920 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2922 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2923 * Architectures may implement their own version but if add_active_range()
2924 * was used and there are no special requirements, this is a convenient
2927 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2931 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2932 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2933 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2935 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2936 return early_node_map
[i
].nid
;
2941 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2943 /* Basic iterator support to walk early_node_map[] */
2944 #define for_each_active_range_index_in_nid(i, nid) \
2945 for (i = first_active_region_index_in_nid(nid); i != -1; \
2946 i = next_active_region_index_in_nid(i, nid))
2949 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2950 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2951 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2953 * If an architecture guarantees that all ranges registered with
2954 * add_active_ranges() contain no holes and may be freed, this
2955 * this function may be used instead of calling free_bootmem() manually.
2957 void __init
free_bootmem_with_active_regions(int nid
,
2958 unsigned long max_low_pfn
)
2962 for_each_active_range_index_in_nid(i
, nid
) {
2963 unsigned long size_pages
= 0;
2964 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2966 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2969 if (end_pfn
> max_low_pfn
)
2970 end_pfn
= max_low_pfn
;
2972 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2973 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2974 PFN_PHYS(early_node_map
[i
].start_pfn
),
2975 size_pages
<< PAGE_SHIFT
);
2979 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
2984 for_each_active_range_index_in_nid(i
, nid
) {
2985 ret
= work_fn(early_node_map
[i
].start_pfn
,
2986 early_node_map
[i
].end_pfn
, data
);
2992 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2993 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2995 * If an architecture guarantees that all ranges registered with
2996 * add_active_ranges() contain no holes and may be freed, this
2997 * function may be used instead of calling memory_present() manually.
2999 void __init
sparse_memory_present_with_active_regions(int nid
)
3003 for_each_active_range_index_in_nid(i
, nid
)
3004 memory_present(early_node_map
[i
].nid
,
3005 early_node_map
[i
].start_pfn
,
3006 early_node_map
[i
].end_pfn
);
3010 * push_node_boundaries - Push node boundaries to at least the requested boundary
3011 * @nid: The nid of the node to push the boundary for
3012 * @start_pfn: The start pfn of the node
3013 * @end_pfn: The end pfn of the node
3015 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3016 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3017 * be hotplugged even though no physical memory exists. This function allows
3018 * an arch to push out the node boundaries so mem_map is allocated that can
3021 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3022 void __init
push_node_boundaries(unsigned int nid
,
3023 unsigned long start_pfn
, unsigned long end_pfn
)
3025 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3026 "Entering push_node_boundaries(%u, %lu, %lu)\n",
3027 nid
, start_pfn
, end_pfn
);
3029 /* Initialise the boundary for this node if necessary */
3030 if (node_boundary_end_pfn
[nid
] == 0)
3031 node_boundary_start_pfn
[nid
] = -1UL;
3033 /* Update the boundaries */
3034 if (node_boundary_start_pfn
[nid
] > start_pfn
)
3035 node_boundary_start_pfn
[nid
] = start_pfn
;
3036 if (node_boundary_end_pfn
[nid
] < end_pfn
)
3037 node_boundary_end_pfn
[nid
] = end_pfn
;
3040 /* If necessary, push the node boundary out for reserve hotadd */
3041 static void __meminit
account_node_boundary(unsigned int nid
,
3042 unsigned long *start_pfn
, unsigned long *end_pfn
)
3044 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3045 "Entering account_node_boundary(%u, %lu, %lu)\n",
3046 nid
, *start_pfn
, *end_pfn
);
3048 /* Return if boundary information has not been provided */
3049 if (node_boundary_end_pfn
[nid
] == 0)
3052 /* Check the boundaries and update if necessary */
3053 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
3054 *start_pfn
= node_boundary_start_pfn
[nid
];
3055 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
3056 *end_pfn
= node_boundary_end_pfn
[nid
];
3059 void __init
push_node_boundaries(unsigned int nid
,
3060 unsigned long start_pfn
, unsigned long end_pfn
) {}
3062 static void __meminit
account_node_boundary(unsigned int nid
,
3063 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3068 * get_pfn_range_for_nid - Return the start and end page frames for a node
3069 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3070 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3071 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3073 * It returns the start and end page frame of a node based on information
3074 * provided by an arch calling add_active_range(). If called for a node
3075 * with no available memory, a warning is printed and the start and end
3078 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3079 unsigned long *start_pfn
, unsigned long *end_pfn
)
3085 for_each_active_range_index_in_nid(i
, nid
) {
3086 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3087 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3090 if (*start_pfn
== -1UL)
3093 /* Push the node boundaries out if requested */
3094 account_node_boundary(nid
, start_pfn
, end_pfn
);
3098 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3099 * assumption is made that zones within a node are ordered in monotonic
3100 * increasing memory addresses so that the "highest" populated zone is used
3102 void __init
find_usable_zone_for_movable(void)
3105 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3106 if (zone_index
== ZONE_MOVABLE
)
3109 if (arch_zone_highest_possible_pfn
[zone_index
] >
3110 arch_zone_lowest_possible_pfn
[zone_index
])
3114 VM_BUG_ON(zone_index
== -1);
3115 movable_zone
= zone_index
;
3119 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3120 * because it is sized independant of architecture. Unlike the other zones,
3121 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3122 * in each node depending on the size of each node and how evenly kernelcore
3123 * is distributed. This helper function adjusts the zone ranges
3124 * provided by the architecture for a given node by using the end of the
3125 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3126 * zones within a node are in order of monotonic increases memory addresses
3128 void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3129 unsigned long zone_type
,
3130 unsigned long node_start_pfn
,
3131 unsigned long node_end_pfn
,
3132 unsigned long *zone_start_pfn
,
3133 unsigned long *zone_end_pfn
)
3135 /* Only adjust if ZONE_MOVABLE is on this node */
3136 if (zone_movable_pfn
[nid
]) {
3137 /* Size ZONE_MOVABLE */
3138 if (zone_type
== ZONE_MOVABLE
) {
3139 *zone_start_pfn
= zone_movable_pfn
[nid
];
3140 *zone_end_pfn
= min(node_end_pfn
,
3141 arch_zone_highest_possible_pfn
[movable_zone
]);
3143 /* Adjust for ZONE_MOVABLE starting within this range */
3144 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3145 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3146 *zone_end_pfn
= zone_movable_pfn
[nid
];
3148 /* Check if this whole range is within ZONE_MOVABLE */
3149 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3150 *zone_start_pfn
= *zone_end_pfn
;
3155 * Return the number of pages a zone spans in a node, including holes
3156 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3158 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3159 unsigned long zone_type
,
3160 unsigned long *ignored
)
3162 unsigned long node_start_pfn
, node_end_pfn
;
3163 unsigned long zone_start_pfn
, zone_end_pfn
;
3165 /* Get the start and end of the node and zone */
3166 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3167 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3168 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3169 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3170 node_start_pfn
, node_end_pfn
,
3171 &zone_start_pfn
, &zone_end_pfn
);
3173 /* Check that this node has pages within the zone's required range */
3174 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3177 /* Move the zone boundaries inside the node if necessary */
3178 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3179 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3181 /* Return the spanned pages */
3182 return zone_end_pfn
- zone_start_pfn
;
3186 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3187 * then all holes in the requested range will be accounted for.
3189 unsigned long __meminit
__absent_pages_in_range(int nid
,
3190 unsigned long range_start_pfn
,
3191 unsigned long range_end_pfn
)
3194 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3195 unsigned long start_pfn
;
3197 /* Find the end_pfn of the first active range of pfns in the node */
3198 i
= first_active_region_index_in_nid(nid
);
3202 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3204 /* Account for ranges before physical memory on this node */
3205 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3206 hole_pages
= prev_end_pfn
- range_start_pfn
;
3208 /* Find all holes for the zone within the node */
3209 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3211 /* No need to continue if prev_end_pfn is outside the zone */
3212 if (prev_end_pfn
>= range_end_pfn
)
3215 /* Make sure the end of the zone is not within the hole */
3216 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3217 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3219 /* Update the hole size cound and move on */
3220 if (start_pfn
> range_start_pfn
) {
3221 BUG_ON(prev_end_pfn
> start_pfn
);
3222 hole_pages
+= start_pfn
- prev_end_pfn
;
3224 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3227 /* Account for ranges past physical memory on this node */
3228 if (range_end_pfn
> prev_end_pfn
)
3229 hole_pages
+= range_end_pfn
-
3230 max(range_start_pfn
, prev_end_pfn
);
3236 * absent_pages_in_range - Return number of page frames in holes within a range
3237 * @start_pfn: The start PFN to start searching for holes
3238 * @end_pfn: The end PFN to stop searching for holes
3240 * It returns the number of pages frames in memory holes within a range.
3242 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3243 unsigned long end_pfn
)
3245 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3248 /* Return the number of page frames in holes in a zone on a node */
3249 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3250 unsigned long zone_type
,
3251 unsigned long *ignored
)
3253 unsigned long node_start_pfn
, node_end_pfn
;
3254 unsigned long zone_start_pfn
, zone_end_pfn
;
3256 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3257 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3259 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3262 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3263 node_start_pfn
, node_end_pfn
,
3264 &zone_start_pfn
, &zone_end_pfn
);
3265 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3269 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3270 unsigned long zone_type
,
3271 unsigned long *zones_size
)
3273 return zones_size
[zone_type
];
3276 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3277 unsigned long zone_type
,
3278 unsigned long *zholes_size
)
3283 return zholes_size
[zone_type
];
3288 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3289 unsigned long *zones_size
, unsigned long *zholes_size
)
3291 unsigned long realtotalpages
, totalpages
= 0;
3294 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3295 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3297 pgdat
->node_spanned_pages
= totalpages
;
3299 realtotalpages
= totalpages
;
3300 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3302 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3304 pgdat
->node_present_pages
= realtotalpages
;
3305 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3309 #ifndef CONFIG_SPARSEMEM
3311 * Calculate the size of the zone->blockflags rounded to an unsigned long
3312 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3313 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3314 * round what is now in bits to nearest long in bits, then return it in
3317 static unsigned long __init
usemap_size(unsigned long zonesize
)
3319 unsigned long usemapsize
;
3321 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3322 usemapsize
= usemapsize
>> pageblock_order
;
3323 usemapsize
*= NR_PAGEBLOCK_BITS
;
3324 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3326 return usemapsize
/ 8;
3329 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3330 struct zone
*zone
, unsigned long zonesize
)
3332 unsigned long usemapsize
= usemap_size(zonesize
);
3333 zone
->pageblock_flags
= NULL
;
3335 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3336 memset(zone
->pageblock_flags
, 0, usemapsize
);
3340 static void inline setup_usemap(struct pglist_data
*pgdat
,
3341 struct zone
*zone
, unsigned long zonesize
) {}
3342 #endif /* CONFIG_SPARSEMEM */
3344 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3346 /* Return a sensible default order for the pageblock size. */
3347 static inline int pageblock_default_order(void)
3349 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3350 return HUGETLB_PAGE_ORDER
;
3355 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3356 static inline void __init
set_pageblock_order(unsigned int order
)
3358 /* Check that pageblock_nr_pages has not already been setup */
3359 if (pageblock_order
)
3363 * Assume the largest contiguous order of interest is a huge page.
3364 * This value may be variable depending on boot parameters on IA64
3366 pageblock_order
= order
;
3368 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3371 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3372 * and pageblock_default_order() are unused as pageblock_order is set
3373 * at compile-time. See include/linux/pageblock-flags.h for the values of
3374 * pageblock_order based on the kernel config
3376 static inline int pageblock_default_order(unsigned int order
)
3380 #define set_pageblock_order(x) do {} while (0)
3382 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3385 * Set up the zone data structures:
3386 * - mark all pages reserved
3387 * - mark all memory queues empty
3388 * - clear the memory bitmaps
3390 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3391 unsigned long *zones_size
, unsigned long *zholes_size
)
3394 int nid
= pgdat
->node_id
;
3395 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3398 pgdat_resize_init(pgdat
);
3399 pgdat
->nr_zones
= 0;
3400 init_waitqueue_head(&pgdat
->kswapd_wait
);
3401 pgdat
->kswapd_max_order
= 0;
3403 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3404 struct zone
*zone
= pgdat
->node_zones
+ j
;
3405 unsigned long size
, realsize
, memmap_pages
;
3407 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3408 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3412 * Adjust realsize so that it accounts for how much memory
3413 * is used by this zone for memmap. This affects the watermark
3414 * and per-cpu initialisations
3417 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3418 if (realsize
>= memmap_pages
) {
3419 realsize
-= memmap_pages
;
3420 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3421 "%s zone: %lu pages used for memmap\n",
3422 zone_names
[j
], memmap_pages
);
3425 " %s zone: %lu pages exceeds realsize %lu\n",
3426 zone_names
[j
], memmap_pages
, realsize
);
3428 /* Account for reserved pages */
3429 if (j
== 0 && realsize
> dma_reserve
) {
3430 realsize
-= dma_reserve
;
3431 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3432 "%s zone: %lu pages reserved\n",
3433 zone_names
[0], dma_reserve
);
3436 if (!is_highmem_idx(j
))
3437 nr_kernel_pages
+= realsize
;
3438 nr_all_pages
+= realsize
;
3440 zone
->spanned_pages
= size
;
3441 zone
->present_pages
= realsize
;
3444 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3446 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3448 zone
->name
= zone_names
[j
];
3449 spin_lock_init(&zone
->lock
);
3450 spin_lock_init(&zone
->lru_lock
);
3451 zone_seqlock_init(zone
);
3452 zone
->zone_pgdat
= pgdat
;
3454 zone
->prev_priority
= DEF_PRIORITY
;
3456 zone_pcp_init(zone
);
3457 INIT_LIST_HEAD(&zone
->active_list
);
3458 INIT_LIST_HEAD(&zone
->inactive_list
);
3459 zone
->nr_scan_active
= 0;
3460 zone
->nr_scan_inactive
= 0;
3461 zap_zone_vm_stats(zone
);
3466 set_pageblock_order(pageblock_default_order());
3467 setup_usemap(pgdat
, zone
, size
);
3468 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3469 size
, MEMMAP_EARLY
);
3471 memmap_init(size
, nid
, j
, zone_start_pfn
);
3472 zone_start_pfn
+= size
;
3476 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3478 /* Skip empty nodes */
3479 if (!pgdat
->node_spanned_pages
)
3482 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3483 /* ia64 gets its own node_mem_map, before this, without bootmem */
3484 if (!pgdat
->node_mem_map
) {
3485 unsigned long size
, start
, end
;
3489 * The zone's endpoints aren't required to be MAX_ORDER
3490 * aligned but the node_mem_map endpoints must be in order
3491 * for the buddy allocator to function correctly.
3493 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3494 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3495 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3496 size
= (end
- start
) * sizeof(struct page
);
3497 map
= alloc_remap(pgdat
->node_id
, size
);
3499 map
= alloc_bootmem_node(pgdat
, size
);
3500 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3502 #ifndef CONFIG_NEED_MULTIPLE_NODES
3504 * With no DISCONTIG, the global mem_map is just set as node 0's
3506 if (pgdat
== NODE_DATA(0)) {
3507 mem_map
= NODE_DATA(0)->node_mem_map
;
3508 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3509 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3510 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3511 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3514 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3517 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3518 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3520 pg_data_t
*pgdat
= NODE_DATA(nid
);
3522 pgdat
->node_id
= nid
;
3523 pgdat
->node_start_pfn
= node_start_pfn
;
3524 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3526 alloc_node_mem_map(pgdat
);
3527 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3528 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3529 nid
, (unsigned long)pgdat
,
3530 (unsigned long)pgdat
->node_mem_map
);
3533 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3536 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3538 #if MAX_NUMNODES > 1
3540 * Figure out the number of possible node ids.
3542 static void __init
setup_nr_node_ids(void)
3545 unsigned int highest
= 0;
3547 for_each_node_mask(node
, node_possible_map
)
3549 nr_node_ids
= highest
+ 1;
3552 static inline void setup_nr_node_ids(void)
3558 * add_active_range - Register a range of PFNs backed by physical memory
3559 * @nid: The node ID the range resides on
3560 * @start_pfn: The start PFN of the available physical memory
3561 * @end_pfn: The end PFN of the available physical memory
3563 * These ranges are stored in an early_node_map[] and later used by
3564 * free_area_init_nodes() to calculate zone sizes and holes. If the
3565 * range spans a memory hole, it is up to the architecture to ensure
3566 * the memory is not freed by the bootmem allocator. If possible
3567 * the range being registered will be merged with existing ranges.
3569 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3570 unsigned long end_pfn
)
3574 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3575 "Entering add_active_range(%d, %#lx, %#lx) "
3576 "%d entries of %d used\n",
3577 nid
, start_pfn
, end_pfn
,
3578 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3580 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3582 /* Merge with existing active regions if possible */
3583 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3584 if (early_node_map
[i
].nid
!= nid
)
3587 /* Skip if an existing region covers this new one */
3588 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3589 end_pfn
<= early_node_map
[i
].end_pfn
)
3592 /* Merge forward if suitable */
3593 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3594 end_pfn
> early_node_map
[i
].end_pfn
) {
3595 early_node_map
[i
].end_pfn
= end_pfn
;
3599 /* Merge backward if suitable */
3600 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3601 end_pfn
>= early_node_map
[i
].start_pfn
) {
3602 early_node_map
[i
].start_pfn
= start_pfn
;
3607 /* Check that early_node_map is large enough */
3608 if (i
>= MAX_ACTIVE_REGIONS
) {
3609 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3610 MAX_ACTIVE_REGIONS
);
3614 early_node_map
[i
].nid
= nid
;
3615 early_node_map
[i
].start_pfn
= start_pfn
;
3616 early_node_map
[i
].end_pfn
= end_pfn
;
3617 nr_nodemap_entries
= i
+ 1;
3621 * remove_active_range - Shrink an existing registered range of PFNs
3622 * @nid: The node id the range is on that should be shrunk
3623 * @start_pfn: The new PFN of the range
3624 * @end_pfn: The new PFN of the range
3626 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3627 * The map is kept near the end physical page range that has already been
3628 * registered. This function allows an arch to shrink an existing registered
3631 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3632 unsigned long end_pfn
)
3637 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3638 nid
, start_pfn
, end_pfn
);
3640 /* Find the old active region end and shrink */
3641 for_each_active_range_index_in_nid(i
, nid
) {
3642 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3643 early_node_map
[i
].end_pfn
<= end_pfn
) {
3645 early_node_map
[i
].start_pfn
= 0;
3646 early_node_map
[i
].end_pfn
= 0;
3650 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3651 early_node_map
[i
].end_pfn
> start_pfn
) {
3652 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3653 early_node_map
[i
].end_pfn
= start_pfn
;
3654 if (temp_end_pfn
> end_pfn
)
3655 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3658 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3659 early_node_map
[i
].end_pfn
> end_pfn
&&
3660 early_node_map
[i
].start_pfn
< end_pfn
) {
3661 early_node_map
[i
].start_pfn
= end_pfn
;
3669 /* remove the blank ones */
3670 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3671 if (early_node_map
[i
].nid
!= nid
)
3673 if (early_node_map
[i
].end_pfn
)
3675 /* we found it, get rid of it */
3676 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3677 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3678 sizeof(early_node_map
[j
]));
3679 j
= nr_nodemap_entries
- 1;
3680 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3681 nr_nodemap_entries
--;
3686 * remove_all_active_ranges - Remove all currently registered regions
3688 * During discovery, it may be found that a table like SRAT is invalid
3689 * and an alternative discovery method must be used. This function removes
3690 * all currently registered regions.
3692 void __init
remove_all_active_ranges(void)
3694 memset(early_node_map
, 0, sizeof(early_node_map
));
3695 nr_nodemap_entries
= 0;
3696 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3697 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3698 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3699 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3702 /* Compare two active node_active_regions */
3703 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3705 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3706 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3708 /* Done this way to avoid overflows */
3709 if (arange
->start_pfn
> brange
->start_pfn
)
3711 if (arange
->start_pfn
< brange
->start_pfn
)
3717 /* sort the node_map by start_pfn */
3718 static void __init
sort_node_map(void)
3720 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3721 sizeof(struct node_active_region
),
3722 cmp_node_active_region
, NULL
);
3725 /* Find the lowest pfn for a node */
3726 unsigned long __init
find_min_pfn_for_node(int nid
)
3729 unsigned long min_pfn
= ULONG_MAX
;
3731 /* Assuming a sorted map, the first range found has the starting pfn */
3732 for_each_active_range_index_in_nid(i
, nid
)
3733 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3735 if (min_pfn
== ULONG_MAX
) {
3737 "Could not find start_pfn for node %d\n", nid
);
3745 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3747 * It returns the minimum PFN based on information provided via
3748 * add_active_range().
3750 unsigned long __init
find_min_pfn_with_active_regions(void)
3752 return find_min_pfn_for_node(MAX_NUMNODES
);
3756 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3758 * It returns the maximum PFN based on information provided via
3759 * add_active_range().
3761 unsigned long __init
find_max_pfn_with_active_regions(void)
3764 unsigned long max_pfn
= 0;
3766 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3767 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
3773 * early_calculate_totalpages()
3774 * Sum pages in active regions for movable zone.
3775 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3777 static unsigned long __init
early_calculate_totalpages(void)
3780 unsigned long totalpages
= 0;
3782 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3783 unsigned long pages
= early_node_map
[i
].end_pfn
-
3784 early_node_map
[i
].start_pfn
;
3785 totalpages
+= pages
;
3787 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3793 * Find the PFN the Movable zone begins in each node. Kernel memory
3794 * is spread evenly between nodes as long as the nodes have enough
3795 * memory. When they don't, some nodes will have more kernelcore than
3798 void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3801 unsigned long usable_startpfn
;
3802 unsigned long kernelcore_node
, kernelcore_remaining
;
3803 unsigned long totalpages
= early_calculate_totalpages();
3804 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3807 * If movablecore was specified, calculate what size of
3808 * kernelcore that corresponds so that memory usable for
3809 * any allocation type is evenly spread. If both kernelcore
3810 * and movablecore are specified, then the value of kernelcore
3811 * will be used for required_kernelcore if it's greater than
3812 * what movablecore would have allowed.
3814 if (required_movablecore
) {
3815 unsigned long corepages
;
3818 * Round-up so that ZONE_MOVABLE is at least as large as what
3819 * was requested by the user
3821 required_movablecore
=
3822 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3823 corepages
= totalpages
- required_movablecore
;
3825 required_kernelcore
= max(required_kernelcore
, corepages
);
3828 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3829 if (!required_kernelcore
)
3832 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3833 find_usable_zone_for_movable();
3834 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3837 /* Spread kernelcore memory as evenly as possible throughout nodes */
3838 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3839 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3841 * Recalculate kernelcore_node if the division per node
3842 * now exceeds what is necessary to satisfy the requested
3843 * amount of memory for the kernel
3845 if (required_kernelcore
< kernelcore_node
)
3846 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3849 * As the map is walked, we track how much memory is usable
3850 * by the kernel using kernelcore_remaining. When it is
3851 * 0, the rest of the node is usable by ZONE_MOVABLE
3853 kernelcore_remaining
= kernelcore_node
;
3855 /* Go through each range of PFNs within this node */
3856 for_each_active_range_index_in_nid(i
, nid
) {
3857 unsigned long start_pfn
, end_pfn
;
3858 unsigned long size_pages
;
3860 start_pfn
= max(early_node_map
[i
].start_pfn
,
3861 zone_movable_pfn
[nid
]);
3862 end_pfn
= early_node_map
[i
].end_pfn
;
3863 if (start_pfn
>= end_pfn
)
3866 /* Account for what is only usable for kernelcore */
3867 if (start_pfn
< usable_startpfn
) {
3868 unsigned long kernel_pages
;
3869 kernel_pages
= min(end_pfn
, usable_startpfn
)
3872 kernelcore_remaining
-= min(kernel_pages
,
3873 kernelcore_remaining
);
3874 required_kernelcore
-= min(kernel_pages
,
3875 required_kernelcore
);
3877 /* Continue if range is now fully accounted */
3878 if (end_pfn
<= usable_startpfn
) {
3881 * Push zone_movable_pfn to the end so
3882 * that if we have to rebalance
3883 * kernelcore across nodes, we will
3884 * not double account here
3886 zone_movable_pfn
[nid
] = end_pfn
;
3889 start_pfn
= usable_startpfn
;
3893 * The usable PFN range for ZONE_MOVABLE is from
3894 * start_pfn->end_pfn. Calculate size_pages as the
3895 * number of pages used as kernelcore
3897 size_pages
= end_pfn
- start_pfn
;
3898 if (size_pages
> kernelcore_remaining
)
3899 size_pages
= kernelcore_remaining
;
3900 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3903 * Some kernelcore has been met, update counts and
3904 * break if the kernelcore for this node has been
3907 required_kernelcore
-= min(required_kernelcore
,
3909 kernelcore_remaining
-= size_pages
;
3910 if (!kernelcore_remaining
)
3916 * If there is still required_kernelcore, we do another pass with one
3917 * less node in the count. This will push zone_movable_pfn[nid] further
3918 * along on the nodes that still have memory until kernelcore is
3922 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3925 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3926 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3927 zone_movable_pfn
[nid
] =
3928 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3931 /* Any regular memory on that node ? */
3932 static void check_for_regular_memory(pg_data_t
*pgdat
)
3934 #ifdef CONFIG_HIGHMEM
3935 enum zone_type zone_type
;
3937 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3938 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3939 if (zone
->present_pages
)
3940 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3946 * free_area_init_nodes - Initialise all pg_data_t and zone data
3947 * @max_zone_pfn: an array of max PFNs for each zone
3949 * This will call free_area_init_node() for each active node in the system.
3950 * Using the page ranges provided by add_active_range(), the size of each
3951 * zone in each node and their holes is calculated. If the maximum PFN
3952 * between two adjacent zones match, it is assumed that the zone is empty.
3953 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3954 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3955 * starts where the previous one ended. For example, ZONE_DMA32 starts
3956 * at arch_max_dma_pfn.
3958 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3963 /* Sort early_node_map as initialisation assumes it is sorted */
3966 /* Record where the zone boundaries are */
3967 memset(arch_zone_lowest_possible_pfn
, 0,
3968 sizeof(arch_zone_lowest_possible_pfn
));
3969 memset(arch_zone_highest_possible_pfn
, 0,
3970 sizeof(arch_zone_highest_possible_pfn
));
3971 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3972 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3973 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3974 if (i
== ZONE_MOVABLE
)
3976 arch_zone_lowest_possible_pfn
[i
] =
3977 arch_zone_highest_possible_pfn
[i
-1];
3978 arch_zone_highest_possible_pfn
[i
] =
3979 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3981 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3982 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3984 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3985 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3986 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3988 /* Print out the zone ranges */
3989 printk("Zone PFN ranges:\n");
3990 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3991 if (i
== ZONE_MOVABLE
)
3993 printk(" %-8s %0#10lx -> %0#10lx\n",
3995 arch_zone_lowest_possible_pfn
[i
],
3996 arch_zone_highest_possible_pfn
[i
]);
3999 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4000 printk("Movable zone start PFN for each node\n");
4001 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4002 if (zone_movable_pfn
[i
])
4003 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4006 /* Print out the early_node_map[] */
4007 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4008 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4009 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4010 early_node_map
[i
].start_pfn
,
4011 early_node_map
[i
].end_pfn
);
4013 /* Initialise every node */
4014 mminit_verify_pageflags_layout();
4015 setup_nr_node_ids();
4016 for_each_online_node(nid
) {
4017 pg_data_t
*pgdat
= NODE_DATA(nid
);
4018 free_area_init_node(nid
, NULL
,
4019 find_min_pfn_for_node(nid
), NULL
);
4021 /* Any memory on that node */
4022 if (pgdat
->node_present_pages
)
4023 node_set_state(nid
, N_HIGH_MEMORY
);
4024 check_for_regular_memory(pgdat
);
4028 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4030 unsigned long long coremem
;
4034 coremem
= memparse(p
, &p
);
4035 *core
= coremem
>> PAGE_SHIFT
;
4037 /* Paranoid check that UL is enough for the coremem value */
4038 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4044 * kernelcore=size sets the amount of memory for use for allocations that
4045 * cannot be reclaimed or migrated.
4047 static int __init
cmdline_parse_kernelcore(char *p
)
4049 return cmdline_parse_core(p
, &required_kernelcore
);
4053 * movablecore=size sets the amount of memory for use for allocations that
4054 * can be reclaimed or migrated.
4056 static int __init
cmdline_parse_movablecore(char *p
)
4058 return cmdline_parse_core(p
, &required_movablecore
);
4061 early_param("kernelcore", cmdline_parse_kernelcore
);
4062 early_param("movablecore", cmdline_parse_movablecore
);
4064 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4067 * set_dma_reserve - set the specified number of pages reserved in the first zone
4068 * @new_dma_reserve: The number of pages to mark reserved
4070 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4071 * In the DMA zone, a significant percentage may be consumed by kernel image
4072 * and other unfreeable allocations which can skew the watermarks badly. This
4073 * function may optionally be used to account for unfreeable pages in the
4074 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4075 * smaller per-cpu batchsize.
4077 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4079 dma_reserve
= new_dma_reserve
;
4082 #ifndef CONFIG_NEED_MULTIPLE_NODES
4083 struct pglist_data contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4084 EXPORT_SYMBOL(contig_page_data
);
4087 void __init
free_area_init(unsigned long *zones_size
)
4089 free_area_init_node(0, zones_size
,
4090 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4093 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4094 unsigned long action
, void *hcpu
)
4096 int cpu
= (unsigned long)hcpu
;
4098 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4102 * Spill the event counters of the dead processor
4103 * into the current processors event counters.
4104 * This artificially elevates the count of the current
4107 vm_events_fold_cpu(cpu
);
4110 * Zero the differential counters of the dead processor
4111 * so that the vm statistics are consistent.
4113 * This is only okay since the processor is dead and cannot
4114 * race with what we are doing.
4116 refresh_cpu_vm_stats(cpu
);
4121 void __init
page_alloc_init(void)
4123 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4127 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4128 * or min_free_kbytes changes.
4130 static void calculate_totalreserve_pages(void)
4132 struct pglist_data
*pgdat
;
4133 unsigned long reserve_pages
= 0;
4134 enum zone_type i
, j
;
4136 for_each_online_pgdat(pgdat
) {
4137 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4138 struct zone
*zone
= pgdat
->node_zones
+ i
;
4139 unsigned long max
= 0;
4141 /* Find valid and maximum lowmem_reserve in the zone */
4142 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4143 if (zone
->lowmem_reserve
[j
] > max
)
4144 max
= zone
->lowmem_reserve
[j
];
4147 /* we treat pages_high as reserved pages. */
4148 max
+= zone
->pages_high
;
4150 if (max
> zone
->present_pages
)
4151 max
= zone
->present_pages
;
4152 reserve_pages
+= max
;
4155 totalreserve_pages
= reserve_pages
;
4159 * setup_per_zone_lowmem_reserve - called whenever
4160 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4161 * has a correct pages reserved value, so an adequate number of
4162 * pages are left in the zone after a successful __alloc_pages().
4164 static void setup_per_zone_lowmem_reserve(void)
4166 struct pglist_data
*pgdat
;
4167 enum zone_type j
, idx
;
4169 for_each_online_pgdat(pgdat
) {
4170 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4171 struct zone
*zone
= pgdat
->node_zones
+ j
;
4172 unsigned long present_pages
= zone
->present_pages
;
4174 zone
->lowmem_reserve
[j
] = 0;
4178 struct zone
*lower_zone
;
4182 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4183 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4185 lower_zone
= pgdat
->node_zones
+ idx
;
4186 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4187 sysctl_lowmem_reserve_ratio
[idx
];
4188 present_pages
+= lower_zone
->present_pages
;
4193 /* update totalreserve_pages */
4194 calculate_totalreserve_pages();
4198 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4200 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4201 * with respect to min_free_kbytes.
4203 void setup_per_zone_pages_min(void)
4205 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4206 unsigned long lowmem_pages
= 0;
4208 unsigned long flags
;
4210 /* Calculate total number of !ZONE_HIGHMEM pages */
4211 for_each_zone(zone
) {
4212 if (!is_highmem(zone
))
4213 lowmem_pages
+= zone
->present_pages
;
4216 for_each_zone(zone
) {
4219 spin_lock_irqsave(&zone
->lru_lock
, flags
);
4220 tmp
= (u64
)pages_min
* zone
->present_pages
;
4221 do_div(tmp
, lowmem_pages
);
4222 if (is_highmem(zone
)) {
4224 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4225 * need highmem pages, so cap pages_min to a small
4228 * The (pages_high-pages_low) and (pages_low-pages_min)
4229 * deltas controls asynch page reclaim, and so should
4230 * not be capped for highmem.
4234 min_pages
= zone
->present_pages
/ 1024;
4235 if (min_pages
< SWAP_CLUSTER_MAX
)
4236 min_pages
= SWAP_CLUSTER_MAX
;
4237 if (min_pages
> 128)
4239 zone
->pages_min
= min_pages
;
4242 * If it's a lowmem zone, reserve a number of pages
4243 * proportionate to the zone's size.
4245 zone
->pages_min
= tmp
;
4248 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4249 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4250 setup_zone_migrate_reserve(zone
);
4251 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4254 /* update totalreserve_pages */
4255 calculate_totalreserve_pages();
4259 * Initialise min_free_kbytes.
4261 * For small machines we want it small (128k min). For large machines
4262 * we want it large (64MB max). But it is not linear, because network
4263 * bandwidth does not increase linearly with machine size. We use
4265 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4266 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4282 static int __init
init_per_zone_pages_min(void)
4284 unsigned long lowmem_kbytes
;
4286 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4288 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4289 if (min_free_kbytes
< 128)
4290 min_free_kbytes
= 128;
4291 if (min_free_kbytes
> 65536)
4292 min_free_kbytes
= 65536;
4293 setup_per_zone_pages_min();
4294 setup_per_zone_lowmem_reserve();
4297 module_init(init_per_zone_pages_min
)
4300 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4301 * that we can call two helper functions whenever min_free_kbytes
4304 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4305 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4307 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4309 setup_per_zone_pages_min();
4314 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4315 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4320 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4325 zone
->min_unmapped_pages
= (zone
->present_pages
*
4326 sysctl_min_unmapped_ratio
) / 100;
4330 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4331 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4336 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4341 zone
->min_slab_pages
= (zone
->present_pages
*
4342 sysctl_min_slab_ratio
) / 100;
4348 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4349 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4350 * whenever sysctl_lowmem_reserve_ratio changes.
4352 * The reserve ratio obviously has absolutely no relation with the
4353 * pages_min watermarks. The lowmem reserve ratio can only make sense
4354 * if in function of the boot time zone sizes.
4356 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4357 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4359 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4360 setup_per_zone_lowmem_reserve();
4365 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4366 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4367 * can have before it gets flushed back to buddy allocator.
4370 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4371 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4377 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4378 if (!write
|| (ret
== -EINVAL
))
4380 for_each_zone(zone
) {
4381 for_each_online_cpu(cpu
) {
4383 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4384 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4390 int hashdist
= HASHDIST_DEFAULT
;
4393 static int __init
set_hashdist(char *str
)
4397 hashdist
= simple_strtoul(str
, &str
, 0);
4400 __setup("hashdist=", set_hashdist
);
4404 * allocate a large system hash table from bootmem
4405 * - it is assumed that the hash table must contain an exact power-of-2
4406 * quantity of entries
4407 * - limit is the number of hash buckets, not the total allocation size
4409 void *__init
alloc_large_system_hash(const char *tablename
,
4410 unsigned long bucketsize
,
4411 unsigned long numentries
,
4414 unsigned int *_hash_shift
,
4415 unsigned int *_hash_mask
,
4416 unsigned long limit
)
4418 unsigned long long max
= limit
;
4419 unsigned long log2qty
, size
;
4422 /* allow the kernel cmdline to have a say */
4424 /* round applicable memory size up to nearest megabyte */
4425 numentries
= nr_kernel_pages
;
4426 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4427 numentries
>>= 20 - PAGE_SHIFT
;
4428 numentries
<<= 20 - PAGE_SHIFT
;
4430 /* limit to 1 bucket per 2^scale bytes of low memory */
4431 if (scale
> PAGE_SHIFT
)
4432 numentries
>>= (scale
- PAGE_SHIFT
);
4434 numentries
<<= (PAGE_SHIFT
- scale
);
4436 /* Make sure we've got at least a 0-order allocation.. */
4437 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4438 numentries
= PAGE_SIZE
/ bucketsize
;
4440 numentries
= roundup_pow_of_two(numentries
);
4442 /* limit allocation size to 1/16 total memory by default */
4444 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4445 do_div(max
, bucketsize
);
4448 if (numentries
> max
)
4451 log2qty
= ilog2(numentries
);
4454 size
= bucketsize
<< log2qty
;
4455 if (flags
& HASH_EARLY
)
4456 table
= alloc_bootmem(size
);
4458 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4460 unsigned long order
= get_order(size
);
4461 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4463 * If bucketsize is not a power-of-two, we may free
4464 * some pages at the end of hash table.
4467 unsigned long alloc_end
= (unsigned long)table
+
4468 (PAGE_SIZE
<< order
);
4469 unsigned long used
= (unsigned long)table
+
4471 split_page(virt_to_page(table
), order
);
4472 while (used
< alloc_end
) {
4478 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4481 panic("Failed to allocate %s hash table\n", tablename
);
4483 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4486 ilog2(size
) - PAGE_SHIFT
,
4490 *_hash_shift
= log2qty
;
4492 *_hash_mask
= (1 << log2qty
) - 1;
4497 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4498 struct page
*pfn_to_page(unsigned long pfn
)
4500 return __pfn_to_page(pfn
);
4502 unsigned long page_to_pfn(struct page
*page
)
4504 return __page_to_pfn(page
);
4506 EXPORT_SYMBOL(pfn_to_page
);
4507 EXPORT_SYMBOL(page_to_pfn
);
4508 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4510 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4511 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4514 #ifdef CONFIG_SPARSEMEM
4515 return __pfn_to_section(pfn
)->pageblock_flags
;
4517 return zone
->pageblock_flags
;
4518 #endif /* CONFIG_SPARSEMEM */
4521 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4523 #ifdef CONFIG_SPARSEMEM
4524 pfn
&= (PAGES_PER_SECTION
-1);
4525 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4527 pfn
= pfn
- zone
->zone_start_pfn
;
4528 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4529 #endif /* CONFIG_SPARSEMEM */
4533 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4534 * @page: The page within the block of interest
4535 * @start_bitidx: The first bit of interest to retrieve
4536 * @end_bitidx: The last bit of interest
4537 * returns pageblock_bits flags
4539 unsigned long get_pageblock_flags_group(struct page
*page
,
4540 int start_bitidx
, int end_bitidx
)
4543 unsigned long *bitmap
;
4544 unsigned long pfn
, bitidx
;
4545 unsigned long flags
= 0;
4546 unsigned long value
= 1;
4548 zone
= page_zone(page
);
4549 pfn
= page_to_pfn(page
);
4550 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4551 bitidx
= pfn_to_bitidx(zone
, pfn
);
4553 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4554 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4561 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4562 * @page: The page within the block of interest
4563 * @start_bitidx: The first bit of interest
4564 * @end_bitidx: The last bit of interest
4565 * @flags: The flags to set
4567 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4568 int start_bitidx
, int end_bitidx
)
4571 unsigned long *bitmap
;
4572 unsigned long pfn
, bitidx
;
4573 unsigned long value
= 1;
4575 zone
= page_zone(page
);
4576 pfn
= page_to_pfn(page
);
4577 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4578 bitidx
= pfn_to_bitidx(zone
, pfn
);
4579 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4580 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4582 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4584 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4586 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4590 * This is designed as sub function...plz see page_isolation.c also.
4591 * set/clear page block's type to be ISOLATE.
4592 * page allocater never alloc memory from ISOLATE block.
4595 int set_migratetype_isolate(struct page
*page
)
4598 unsigned long flags
;
4601 zone
= page_zone(page
);
4602 spin_lock_irqsave(&zone
->lock
, flags
);
4604 * In future, more migrate types will be able to be isolation target.
4606 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4608 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4609 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4612 spin_unlock_irqrestore(&zone
->lock
, flags
);
4618 void unset_migratetype_isolate(struct page
*page
)
4621 unsigned long flags
;
4622 zone
= page_zone(page
);
4623 spin_lock_irqsave(&zone
->lock
, flags
);
4624 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4626 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4627 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4629 spin_unlock_irqrestore(&zone
->lock
, flags
);
4632 #ifdef CONFIG_MEMORY_HOTREMOVE
4634 * All pages in the range must be isolated before calling this.
4637 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4643 unsigned long flags
;
4644 /* find the first valid pfn */
4645 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4650 zone
= page_zone(pfn_to_page(pfn
));
4651 spin_lock_irqsave(&zone
->lock
, flags
);
4653 while (pfn
< end_pfn
) {
4654 if (!pfn_valid(pfn
)) {
4658 page
= pfn_to_page(pfn
);
4659 BUG_ON(page_count(page
));
4660 BUG_ON(!PageBuddy(page
));
4661 order
= page_order(page
);
4662 #ifdef CONFIG_DEBUG_VM
4663 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4664 pfn
, 1 << order
, end_pfn
);
4666 list_del(&page
->lru
);
4667 rmv_page_order(page
);
4668 zone
->free_area
[order
].nr_free
--;
4669 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4671 for (i
= 0; i
< (1 << order
); i
++)
4672 SetPageReserved((page
+i
));
4673 pfn
+= (1 << order
);
4675 spin_unlock_irqrestore(&zone
->lock
, flags
);