2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/oom.h>
32 #include <linux/notifier.h>
33 #include <linux/topology.h>
34 #include <linux/sysctl.h>
35 #include <linux/cpu.h>
36 #include <linux/cpuset.h>
37 #include <linux/memory_hotplug.h>
38 #include <linux/nodemask.h>
39 #include <linux/vmalloc.h>
40 #include <linux/mempolicy.h>
41 #include <linux/stop_machine.h>
42 #include <linux/sort.h>
43 #include <linux/pfn.h>
44 #include <linux/backing-dev.h>
45 #include <linux/fault-inject.h>
46 #include <linux/page-isolation.h>
47 #include <linux/page_cgroup.h>
48 #include <linux/debugobjects.h>
50 #include <asm/tlbflush.h>
51 #include <asm/div64.h>
55 * Array of node states.
57 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
58 [N_POSSIBLE
] = NODE_MASK_ALL
,
59 [N_ONLINE
] = { { [0] = 1UL } },
61 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
63 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
65 [N_CPU
] = { { [0] = 1UL } },
68 EXPORT_SYMBOL(node_states
);
70 unsigned long totalram_pages __read_mostly
;
71 unsigned long totalreserve_pages __read_mostly
;
73 int percpu_pagelist_fraction
;
75 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
76 int pageblock_order __read_mostly
;
79 static void __free_pages_ok(struct page
*page
, unsigned int order
);
82 * results with 256, 32 in the lowmem_reserve sysctl:
83 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
84 * 1G machine -> (16M dma, 784M normal, 224M high)
85 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
86 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
87 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
89 * TBD: should special case ZONE_DMA32 machines here - in those we normally
90 * don't need any ZONE_NORMAL reservation
92 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
93 #ifdef CONFIG_ZONE_DMA
96 #ifdef CONFIG_ZONE_DMA32
105 EXPORT_SYMBOL(totalram_pages
);
107 static char * const zone_names
[MAX_NR_ZONES
] = {
108 #ifdef CONFIG_ZONE_DMA
111 #ifdef CONFIG_ZONE_DMA32
115 #ifdef CONFIG_HIGHMEM
121 int min_free_kbytes
= 1024;
123 unsigned long __meminitdata nr_kernel_pages
;
124 unsigned long __meminitdata nr_all_pages
;
125 static unsigned long __meminitdata dma_reserve
;
127 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
129 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
130 * ranges of memory (RAM) that may be registered with add_active_range().
131 * Ranges passed to add_active_range() will be merged if possible
132 * so the number of times add_active_range() can be called is
133 * related to the number of nodes and the number of holes
135 #ifdef CONFIG_MAX_ACTIVE_REGIONS
136 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
137 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
139 #if MAX_NUMNODES >= 32
140 /* If there can be many nodes, allow up to 50 holes per node */
141 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
143 /* By default, allow up to 256 distinct regions */
144 #define MAX_ACTIVE_REGIONS 256
148 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
149 static int __meminitdata nr_nodemap_entries
;
150 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
151 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
152 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
153 static unsigned long __meminitdata node_boundary_start_pfn
[MAX_NUMNODES
];
154 static unsigned long __meminitdata node_boundary_end_pfn
[MAX_NUMNODES
];
155 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
156 static unsigned long __initdata required_kernelcore
;
157 static unsigned long __initdata required_movablecore
;
158 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
160 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
162 EXPORT_SYMBOL(movable_zone
);
163 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
166 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
167 EXPORT_SYMBOL(nr_node_ids
);
170 int page_group_by_mobility_disabled __read_mostly
;
172 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
174 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
175 PB_migrate
, PB_migrate_end
);
178 #ifdef CONFIG_DEBUG_VM
179 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
183 unsigned long pfn
= page_to_pfn(page
);
186 seq
= zone_span_seqbegin(zone
);
187 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
189 else if (pfn
< zone
->zone_start_pfn
)
191 } while (zone_span_seqretry(zone
, seq
));
196 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
198 if (!pfn_valid_within(page_to_pfn(page
)))
200 if (zone
!= page_zone(page
))
206 * Temporary debugging check for pages not lying within a given zone.
208 static int bad_range(struct zone
*zone
, struct page
*page
)
210 if (page_outside_zone_boundaries(zone
, page
))
212 if (!page_is_consistent(zone
, page
))
218 static inline int bad_range(struct zone
*zone
, struct page
*page
)
224 static void bad_page(struct page
*page
)
226 printk(KERN_EMERG
"Bad page state in process '%s'\n" KERN_EMERG
227 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
228 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
229 (unsigned long)page
->flags
, page
->mapping
,
230 page_mapcount(page
), page_count(page
));
232 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
233 KERN_EMERG
"Backtrace:\n");
235 page
->flags
&= ~PAGE_FLAGS_CLEAR_WHEN_BAD
;
236 set_page_count(page
, 0);
237 reset_page_mapcount(page
);
238 page
->mapping
= NULL
;
239 add_taint(TAINT_BAD_PAGE
);
243 * Higher-order pages are called "compound pages". They are structured thusly:
245 * The first PAGE_SIZE page is called the "head page".
247 * The remaining PAGE_SIZE pages are called "tail pages".
249 * All pages have PG_compound set. All pages have their ->private pointing at
250 * the head page (even the head page has this).
252 * The first tail page's ->lru.next holds the address of the compound page's
253 * put_page() function. Its ->lru.prev holds the order of allocation.
254 * This usage means that zero-order pages may not be compound.
257 static void free_compound_page(struct page
*page
)
259 __free_pages_ok(page
, compound_order(page
));
262 void prep_compound_page(struct page
*page
, unsigned long order
)
265 int nr_pages
= 1 << order
;
267 set_compound_page_dtor(page
, free_compound_page
);
268 set_compound_order(page
, order
);
270 for (i
= 1; i
< nr_pages
; i
++) {
271 struct page
*p
= page
+ i
;
274 p
->first_page
= page
;
278 #ifdef CONFIG_HUGETLBFS
279 void prep_compound_gigantic_page(struct page
*page
, unsigned long order
)
282 int nr_pages
= 1 << order
;
283 struct page
*p
= page
+ 1;
285 set_compound_page_dtor(page
, free_compound_page
);
286 set_compound_order(page
, order
);
288 for (i
= 1; i
< nr_pages
; i
++, p
= mem_map_next(p
, page
, i
)) {
290 p
->first_page
= page
;
295 static void destroy_compound_page(struct page
*page
, unsigned long order
)
298 int nr_pages
= 1 << order
;
300 if (unlikely(compound_order(page
) != order
))
303 if (unlikely(!PageHead(page
)))
305 __ClearPageHead(page
);
306 for (i
= 1; i
< nr_pages
; i
++) {
307 struct page
*p
= page
+ i
;
309 if (unlikely(!PageTail(p
) |
310 (p
->first_page
!= page
)))
316 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
321 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
322 * and __GFP_HIGHMEM from hard or soft interrupt context.
324 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
325 for (i
= 0; i
< (1 << order
); i
++)
326 clear_highpage(page
+ i
);
329 static inline void set_page_order(struct page
*page
, int order
)
331 set_page_private(page
, order
);
332 __SetPageBuddy(page
);
335 static inline void rmv_page_order(struct page
*page
)
337 __ClearPageBuddy(page
);
338 set_page_private(page
, 0);
342 * Locate the struct page for both the matching buddy in our
343 * pair (buddy1) and the combined O(n+1) page they form (page).
345 * 1) Any buddy B1 will have an order O twin B2 which satisfies
346 * the following equation:
348 * For example, if the starting buddy (buddy2) is #8 its order
350 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
352 * 2) Any buddy B will have an order O+1 parent P which
353 * satisfies the following equation:
356 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
358 static inline struct page
*
359 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
361 unsigned long buddy_idx
= page_idx
^ (1 << order
);
363 return page
+ (buddy_idx
- page_idx
);
366 static inline unsigned long
367 __find_combined_index(unsigned long page_idx
, unsigned int order
)
369 return (page_idx
& ~(1 << order
));
373 * This function checks whether a page is free && is the buddy
374 * we can do coalesce a page and its buddy if
375 * (a) the buddy is not in a hole &&
376 * (b) the buddy is in the buddy system &&
377 * (c) a page and its buddy have the same order &&
378 * (d) a page and its buddy are in the same zone.
380 * For recording whether a page is in the buddy system, we use PG_buddy.
381 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
383 * For recording page's order, we use page_private(page).
385 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
388 if (!pfn_valid_within(page_to_pfn(buddy
)))
391 if (page_zone_id(page
) != page_zone_id(buddy
))
394 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
395 BUG_ON(page_count(buddy
) != 0);
402 * Freeing function for a buddy system allocator.
404 * The concept of a buddy system is to maintain direct-mapped table
405 * (containing bit values) for memory blocks of various "orders".
406 * The bottom level table contains the map for the smallest allocatable
407 * units of memory (here, pages), and each level above it describes
408 * pairs of units from the levels below, hence, "buddies".
409 * At a high level, all that happens here is marking the table entry
410 * at the bottom level available, and propagating the changes upward
411 * as necessary, plus some accounting needed to play nicely with other
412 * parts of the VM system.
413 * At each level, we keep a list of pages, which are heads of continuous
414 * free pages of length of (1 << order) and marked with PG_buddy. Page's
415 * order is recorded in page_private(page) field.
416 * So when we are allocating or freeing one, we can derive the state of the
417 * other. That is, if we allocate a small block, and both were
418 * free, the remainder of the region must be split into blocks.
419 * If a block is freed, and its buddy is also free, then this
420 * triggers coalescing into a block of larger size.
425 static inline void __free_one_page(struct page
*page
,
426 struct zone
*zone
, unsigned int order
)
428 unsigned long page_idx
;
429 int order_size
= 1 << order
;
430 int migratetype
= get_pageblock_migratetype(page
);
432 if (unlikely(PageCompound(page
)))
433 destroy_compound_page(page
, order
);
435 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
437 VM_BUG_ON(page_idx
& (order_size
- 1));
438 VM_BUG_ON(bad_range(zone
, page
));
440 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
441 while (order
< MAX_ORDER
-1) {
442 unsigned long combined_idx
;
445 buddy
= __page_find_buddy(page
, page_idx
, order
);
446 if (!page_is_buddy(page
, buddy
, order
))
449 /* Our buddy is free, merge with it and move up one order. */
450 list_del(&buddy
->lru
);
451 zone
->free_area
[order
].nr_free
--;
452 rmv_page_order(buddy
);
453 combined_idx
= __find_combined_index(page_idx
, order
);
454 page
= page
+ (combined_idx
- page_idx
);
455 page_idx
= combined_idx
;
458 set_page_order(page
, order
);
460 &zone
->free_area
[order
].free_list
[migratetype
]);
461 zone
->free_area
[order
].nr_free
++;
464 static inline int free_pages_check(struct page
*page
)
466 free_page_mlock(page
);
467 if (unlikely(page_mapcount(page
) |
468 (page
->mapping
!= NULL
) |
469 (page_count(page
) != 0) |
470 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)))
473 __ClearPageDirty(page
);
474 if (PageSwapBacked(page
))
475 __ClearPageSwapBacked(page
);
477 * For now, we report if PG_reserved was found set, but do not
478 * clear it, and do not free the page. But we shall soon need
479 * to do more, for when the ZERO_PAGE count wraps negative.
481 return PageReserved(page
);
485 * Frees a list of pages.
486 * Assumes all pages on list are in same zone, and of same order.
487 * count is the number of pages to free.
489 * If the zone was previously in an "all pages pinned" state then look to
490 * see if this freeing clears that state.
492 * And clear the zone's pages_scanned counter, to hold off the "all pages are
493 * pinned" detection logic.
495 static void free_pages_bulk(struct zone
*zone
, int count
,
496 struct list_head
*list
, int order
)
498 spin_lock(&zone
->lock
);
499 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
500 zone
->pages_scanned
= 0;
504 VM_BUG_ON(list_empty(list
));
505 page
= list_entry(list
->prev
, struct page
, lru
);
506 /* have to delete it as __free_one_page list manipulates */
507 list_del(&page
->lru
);
508 __free_one_page(page
, zone
, order
);
510 spin_unlock(&zone
->lock
);
513 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
515 spin_lock(&zone
->lock
);
516 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
517 zone
->pages_scanned
= 0;
518 __free_one_page(page
, zone
, order
);
519 spin_unlock(&zone
->lock
);
522 static void __free_pages_ok(struct page
*page
, unsigned int order
)
528 for (i
= 0 ; i
< (1 << order
) ; ++i
)
529 reserved
+= free_pages_check(page
+ i
);
533 if (!PageHighMem(page
)) {
534 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
535 debug_check_no_obj_freed(page_address(page
),
538 arch_free_page(page
, order
);
539 kernel_map_pages(page
, 1 << order
, 0);
541 local_irq_save(flags
);
542 __count_vm_events(PGFREE
, 1 << order
);
543 free_one_page(page_zone(page
), page
, order
);
544 local_irq_restore(flags
);
548 * permit the bootmem allocator to evade page validation on high-order frees
550 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
553 __ClearPageReserved(page
);
554 set_page_count(page
, 0);
555 set_page_refcounted(page
);
561 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
562 struct page
*p
= &page
[loop
];
564 if (loop
+ 1 < BITS_PER_LONG
)
566 __ClearPageReserved(p
);
567 set_page_count(p
, 0);
570 set_page_refcounted(page
);
571 __free_pages(page
, order
);
577 * The order of subdivision here is critical for the IO subsystem.
578 * Please do not alter this order without good reasons and regression
579 * testing. Specifically, as large blocks of memory are subdivided,
580 * the order in which smaller blocks are delivered depends on the order
581 * they're subdivided in this function. This is the primary factor
582 * influencing the order in which pages are delivered to the IO
583 * subsystem according to empirical testing, and this is also justified
584 * by considering the behavior of a buddy system containing a single
585 * large block of memory acted on by a series of small allocations.
586 * This behavior is a critical factor in sglist merging's success.
590 static inline void expand(struct zone
*zone
, struct page
*page
,
591 int low
, int high
, struct free_area
*area
,
594 unsigned long size
= 1 << high
;
600 VM_BUG_ON(bad_range(zone
, &page
[size
]));
601 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
603 set_page_order(&page
[size
], high
);
608 * This page is about to be returned from the page allocator
610 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
612 if (unlikely(page_mapcount(page
) |
613 (page
->mapping
!= NULL
) |
614 (page_count(page
) != 0) |
615 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)))
619 * For now, we report if PG_reserved was found set, but do not
620 * clear it, and do not allocate the page: as a safety net.
622 if (PageReserved(page
))
625 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_reclaim
|
626 1 << PG_referenced
| 1 << PG_arch_1
|
627 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
628 #ifdef CONFIG_UNEVICTABLE_LRU
632 set_page_private(page
, 0);
633 set_page_refcounted(page
);
635 arch_alloc_page(page
, order
);
636 kernel_map_pages(page
, 1 << order
, 1);
638 if (gfp_flags
& __GFP_ZERO
)
639 prep_zero_page(page
, order
, gfp_flags
);
641 if (order
&& (gfp_flags
& __GFP_COMP
))
642 prep_compound_page(page
, order
);
648 * Go through the free lists for the given migratetype and remove
649 * the smallest available page from the freelists
651 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
654 unsigned int current_order
;
655 struct free_area
* area
;
658 /* Find a page of the appropriate size in the preferred list */
659 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
660 area
= &(zone
->free_area
[current_order
]);
661 if (list_empty(&area
->free_list
[migratetype
]))
664 page
= list_entry(area
->free_list
[migratetype
].next
,
666 list_del(&page
->lru
);
667 rmv_page_order(page
);
669 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
670 expand(zone
, page
, order
, current_order
, area
, migratetype
);
679 * This array describes the order lists are fallen back to when
680 * the free lists for the desirable migrate type are depleted
682 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
683 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
684 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
685 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
686 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
690 * Move the free pages in a range to the free lists of the requested type.
691 * Note that start_page and end_pages are not aligned on a pageblock
692 * boundary. If alignment is required, use move_freepages_block()
694 static int move_freepages(struct zone
*zone
,
695 struct page
*start_page
, struct page
*end_page
,
702 #ifndef CONFIG_HOLES_IN_ZONE
704 * page_zone is not safe to call in this context when
705 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
706 * anyway as we check zone boundaries in move_freepages_block().
707 * Remove at a later date when no bug reports exist related to
708 * grouping pages by mobility
710 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
713 for (page
= start_page
; page
<= end_page
;) {
714 /* Make sure we are not inadvertently changing nodes */
715 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
717 if (!pfn_valid_within(page_to_pfn(page
))) {
722 if (!PageBuddy(page
)) {
727 order
= page_order(page
);
728 list_del(&page
->lru
);
730 &zone
->free_area
[order
].free_list
[migratetype
]);
732 pages_moved
+= 1 << order
;
738 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
741 unsigned long start_pfn
, end_pfn
;
742 struct page
*start_page
, *end_page
;
744 start_pfn
= page_to_pfn(page
);
745 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
746 start_page
= pfn_to_page(start_pfn
);
747 end_page
= start_page
+ pageblock_nr_pages
- 1;
748 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
750 /* Do not cross zone boundaries */
751 if (start_pfn
< zone
->zone_start_pfn
)
753 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
756 return move_freepages(zone
, start_page
, end_page
, migratetype
);
759 /* Remove an element from the buddy allocator from the fallback list */
760 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
761 int start_migratetype
)
763 struct free_area
* area
;
768 /* Find the largest possible block of pages in the other list */
769 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
771 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
772 migratetype
= fallbacks
[start_migratetype
][i
];
774 /* MIGRATE_RESERVE handled later if necessary */
775 if (migratetype
== MIGRATE_RESERVE
)
778 area
= &(zone
->free_area
[current_order
]);
779 if (list_empty(&area
->free_list
[migratetype
]))
782 page
= list_entry(area
->free_list
[migratetype
].next
,
787 * If breaking a large block of pages, move all free
788 * pages to the preferred allocation list. If falling
789 * back for a reclaimable kernel allocation, be more
790 * agressive about taking ownership of free pages
792 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
793 start_migratetype
== MIGRATE_RECLAIMABLE
) {
795 pages
= move_freepages_block(zone
, page
,
798 /* Claim the whole block if over half of it is free */
799 if (pages
>= (1 << (pageblock_order
-1)))
800 set_pageblock_migratetype(page
,
803 migratetype
= start_migratetype
;
806 /* Remove the page from the freelists */
807 list_del(&page
->lru
);
808 rmv_page_order(page
);
809 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
812 if (current_order
== pageblock_order
)
813 set_pageblock_migratetype(page
,
816 expand(zone
, page
, order
, current_order
, area
, migratetype
);
821 /* Use MIGRATE_RESERVE rather than fail an allocation */
822 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
826 * Do the hard work of removing an element from the buddy allocator.
827 * Call me with the zone->lock already held.
829 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
834 page
= __rmqueue_smallest(zone
, order
, migratetype
);
837 page
= __rmqueue_fallback(zone
, order
, migratetype
);
843 * Obtain a specified number of elements from the buddy allocator, all under
844 * a single hold of the lock, for efficiency. Add them to the supplied list.
845 * Returns the number of new pages which were placed at *list.
847 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
848 unsigned long count
, struct list_head
*list
,
853 spin_lock(&zone
->lock
);
854 for (i
= 0; i
< count
; ++i
) {
855 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
856 if (unlikely(page
== NULL
))
860 * Split buddy pages returned by expand() are received here
861 * in physical page order. The page is added to the callers and
862 * list and the list head then moves forward. From the callers
863 * perspective, the linked list is ordered by page number in
864 * some conditions. This is useful for IO devices that can
865 * merge IO requests if the physical pages are ordered
868 list_add(&page
->lru
, list
);
869 set_page_private(page
, migratetype
);
872 spin_unlock(&zone
->lock
);
878 * Called from the vmstat counter updater to drain pagesets of this
879 * currently executing processor on remote nodes after they have
882 * Note that this function must be called with the thread pinned to
883 * a single processor.
885 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
890 local_irq_save(flags
);
891 if (pcp
->count
>= pcp
->batch
)
892 to_drain
= pcp
->batch
;
894 to_drain
= pcp
->count
;
895 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
896 pcp
->count
-= to_drain
;
897 local_irq_restore(flags
);
902 * Drain pages of the indicated processor.
904 * The processor must either be the current processor and the
905 * thread pinned to the current processor or a processor that
908 static void drain_pages(unsigned int cpu
)
913 for_each_zone(zone
) {
914 struct per_cpu_pageset
*pset
;
915 struct per_cpu_pages
*pcp
;
917 if (!populated_zone(zone
))
920 pset
= zone_pcp(zone
, cpu
);
923 local_irq_save(flags
);
924 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
926 local_irq_restore(flags
);
931 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
933 void drain_local_pages(void *arg
)
935 drain_pages(smp_processor_id());
939 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
941 void drain_all_pages(void)
943 on_each_cpu(drain_local_pages
, NULL
, 1);
946 #ifdef CONFIG_HIBERNATION
948 void mark_free_pages(struct zone
*zone
)
950 unsigned long pfn
, max_zone_pfn
;
953 struct list_head
*curr
;
955 if (!zone
->spanned_pages
)
958 spin_lock_irqsave(&zone
->lock
, flags
);
960 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
961 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
962 if (pfn_valid(pfn
)) {
963 struct page
*page
= pfn_to_page(pfn
);
965 if (!swsusp_page_is_forbidden(page
))
966 swsusp_unset_page_free(page
);
969 for_each_migratetype_order(order
, t
) {
970 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
973 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
974 for (i
= 0; i
< (1UL << order
); i
++)
975 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
978 spin_unlock_irqrestore(&zone
->lock
, flags
);
980 #endif /* CONFIG_PM */
983 * Free a 0-order page
985 static void free_hot_cold_page(struct page
*page
, int cold
)
987 struct zone
*zone
= page_zone(page
);
988 struct per_cpu_pages
*pcp
;
992 page
->mapping
= NULL
;
993 if (free_pages_check(page
))
996 if (!PageHighMem(page
)) {
997 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
998 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1000 arch_free_page(page
, 0);
1001 kernel_map_pages(page
, 1, 0);
1003 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1004 local_irq_save(flags
);
1005 __count_vm_event(PGFREE
);
1007 list_add_tail(&page
->lru
, &pcp
->list
);
1009 list_add(&page
->lru
, &pcp
->list
);
1010 set_page_private(page
, get_pageblock_migratetype(page
));
1012 if (pcp
->count
>= pcp
->high
) {
1013 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1014 pcp
->count
-= pcp
->batch
;
1016 local_irq_restore(flags
);
1020 void free_hot_page(struct page
*page
)
1022 free_hot_cold_page(page
, 0);
1025 void free_cold_page(struct page
*page
)
1027 free_hot_cold_page(page
, 1);
1031 * split_page takes a non-compound higher-order page, and splits it into
1032 * n (1<<order) sub-pages: page[0..n]
1033 * Each sub-page must be freed individually.
1035 * Note: this is probably too low level an operation for use in drivers.
1036 * Please consult with lkml before using this in your driver.
1038 void split_page(struct page
*page
, unsigned int order
)
1042 VM_BUG_ON(PageCompound(page
));
1043 VM_BUG_ON(!page_count(page
));
1044 for (i
= 1; i
< (1 << order
); i
++)
1045 set_page_refcounted(page
+ i
);
1049 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1050 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1053 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1054 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1056 unsigned long flags
;
1058 int cold
= !!(gfp_flags
& __GFP_COLD
);
1060 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1064 if (likely(order
== 0)) {
1065 struct per_cpu_pages
*pcp
;
1067 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1068 local_irq_save(flags
);
1070 pcp
->count
= rmqueue_bulk(zone
, 0,
1071 pcp
->batch
, &pcp
->list
, migratetype
);
1072 if (unlikely(!pcp
->count
))
1076 /* Find a page of the appropriate migrate type */
1078 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1079 if (page_private(page
) == migratetype
)
1082 list_for_each_entry(page
, &pcp
->list
, lru
)
1083 if (page_private(page
) == migratetype
)
1087 /* Allocate more to the pcp list if necessary */
1088 if (unlikely(&page
->lru
== &pcp
->list
)) {
1089 pcp
->count
+= rmqueue_bulk(zone
, 0,
1090 pcp
->batch
, &pcp
->list
, migratetype
);
1091 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1094 list_del(&page
->lru
);
1097 spin_lock_irqsave(&zone
->lock
, flags
);
1098 page
= __rmqueue(zone
, order
, migratetype
);
1099 spin_unlock(&zone
->lock
);
1104 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1105 zone_statistics(preferred_zone
, zone
);
1106 local_irq_restore(flags
);
1109 VM_BUG_ON(bad_range(zone
, page
));
1110 if (prep_new_page(page
, order
, gfp_flags
))
1115 local_irq_restore(flags
);
1120 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1121 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1122 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1123 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1124 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1125 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1126 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1128 #ifdef CONFIG_FAIL_PAGE_ALLOC
1130 static struct fail_page_alloc_attr
{
1131 struct fault_attr attr
;
1133 u32 ignore_gfp_highmem
;
1134 u32 ignore_gfp_wait
;
1137 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1139 struct dentry
*ignore_gfp_highmem_file
;
1140 struct dentry
*ignore_gfp_wait_file
;
1141 struct dentry
*min_order_file
;
1143 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1145 } fail_page_alloc
= {
1146 .attr
= FAULT_ATTR_INITIALIZER
,
1147 .ignore_gfp_wait
= 1,
1148 .ignore_gfp_highmem
= 1,
1152 static int __init
setup_fail_page_alloc(char *str
)
1154 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1156 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1158 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1160 if (order
< fail_page_alloc
.min_order
)
1162 if (gfp_mask
& __GFP_NOFAIL
)
1164 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1166 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1169 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1172 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1174 static int __init
fail_page_alloc_debugfs(void)
1176 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1180 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1184 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1186 fail_page_alloc
.ignore_gfp_wait_file
=
1187 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1188 &fail_page_alloc
.ignore_gfp_wait
);
1190 fail_page_alloc
.ignore_gfp_highmem_file
=
1191 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1192 &fail_page_alloc
.ignore_gfp_highmem
);
1193 fail_page_alloc
.min_order_file
=
1194 debugfs_create_u32("min-order", mode
, dir
,
1195 &fail_page_alloc
.min_order
);
1197 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1198 !fail_page_alloc
.ignore_gfp_highmem_file
||
1199 !fail_page_alloc
.min_order_file
) {
1201 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1202 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1203 debugfs_remove(fail_page_alloc
.min_order_file
);
1204 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1210 late_initcall(fail_page_alloc_debugfs
);
1212 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1214 #else /* CONFIG_FAIL_PAGE_ALLOC */
1216 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1221 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1224 * Return 1 if free pages are above 'mark'. This takes into account the order
1225 * of the allocation.
1227 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1228 int classzone_idx
, int alloc_flags
)
1230 /* free_pages my go negative - that's OK */
1232 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1235 if (alloc_flags
& ALLOC_HIGH
)
1237 if (alloc_flags
& ALLOC_HARDER
)
1240 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1242 for (o
= 0; o
< order
; o
++) {
1243 /* At the next order, this order's pages become unavailable */
1244 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1246 /* Require fewer higher order pages to be free */
1249 if (free_pages
<= min
)
1257 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1258 * skip over zones that are not allowed by the cpuset, or that have
1259 * been recently (in last second) found to be nearly full. See further
1260 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1261 * that have to skip over a lot of full or unallowed zones.
1263 * If the zonelist cache is present in the passed in zonelist, then
1264 * returns a pointer to the allowed node mask (either the current
1265 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1267 * If the zonelist cache is not available for this zonelist, does
1268 * nothing and returns NULL.
1270 * If the fullzones BITMAP in the zonelist cache is stale (more than
1271 * a second since last zap'd) then we zap it out (clear its bits.)
1273 * We hold off even calling zlc_setup, until after we've checked the
1274 * first zone in the zonelist, on the theory that most allocations will
1275 * be satisfied from that first zone, so best to examine that zone as
1276 * quickly as we can.
1278 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1280 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1281 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1283 zlc
= zonelist
->zlcache_ptr
;
1287 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1288 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1289 zlc
->last_full_zap
= jiffies
;
1292 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1293 &cpuset_current_mems_allowed
:
1294 &node_states
[N_HIGH_MEMORY
];
1295 return allowednodes
;
1299 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1300 * if it is worth looking at further for free memory:
1301 * 1) Check that the zone isn't thought to be full (doesn't have its
1302 * bit set in the zonelist_cache fullzones BITMAP).
1303 * 2) Check that the zones node (obtained from the zonelist_cache
1304 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1305 * Return true (non-zero) if zone is worth looking at further, or
1306 * else return false (zero) if it is not.
1308 * This check -ignores- the distinction between various watermarks,
1309 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1310 * found to be full for any variation of these watermarks, it will
1311 * be considered full for up to one second by all requests, unless
1312 * we are so low on memory on all allowed nodes that we are forced
1313 * into the second scan of the zonelist.
1315 * In the second scan we ignore this zonelist cache and exactly
1316 * apply the watermarks to all zones, even it is slower to do so.
1317 * We are low on memory in the second scan, and should leave no stone
1318 * unturned looking for a free page.
1320 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1321 nodemask_t
*allowednodes
)
1323 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1324 int i
; /* index of *z in zonelist zones */
1325 int n
; /* node that zone *z is on */
1327 zlc
= zonelist
->zlcache_ptr
;
1331 i
= z
- zonelist
->_zonerefs
;
1334 /* This zone is worth trying if it is allowed but not full */
1335 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1339 * Given 'z' scanning a zonelist, set the corresponding bit in
1340 * zlc->fullzones, so that subsequent attempts to allocate a page
1341 * from that zone don't waste time re-examining it.
1343 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1345 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1346 int i
; /* index of *z in zonelist zones */
1348 zlc
= zonelist
->zlcache_ptr
;
1352 i
= z
- zonelist
->_zonerefs
;
1354 set_bit(i
, zlc
->fullzones
);
1357 #else /* CONFIG_NUMA */
1359 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1364 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1365 nodemask_t
*allowednodes
)
1370 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1373 #endif /* CONFIG_NUMA */
1376 * get_page_from_freelist goes through the zonelist trying to allocate
1379 static struct page
*
1380 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1381 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1384 struct page
*page
= NULL
;
1386 struct zone
*zone
, *preferred_zone
;
1387 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1388 int zlc_active
= 0; /* set if using zonelist_cache */
1389 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1391 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1393 if (!preferred_zone
)
1396 classzone_idx
= zone_idx(preferred_zone
);
1400 * Scan zonelist, looking for a zone with enough free.
1401 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1403 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1404 high_zoneidx
, nodemask
) {
1405 if (NUMA_BUILD
&& zlc_active
&&
1406 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1408 if ((alloc_flags
& ALLOC_CPUSET
) &&
1409 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1412 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1414 if (alloc_flags
& ALLOC_WMARK_MIN
)
1415 mark
= zone
->pages_min
;
1416 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1417 mark
= zone
->pages_low
;
1419 mark
= zone
->pages_high
;
1420 if (!zone_watermark_ok(zone
, order
, mark
,
1421 classzone_idx
, alloc_flags
)) {
1422 if (!zone_reclaim_mode
||
1423 !zone_reclaim(zone
, gfp_mask
, order
))
1424 goto this_zone_full
;
1428 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1433 zlc_mark_zone_full(zonelist
, z
);
1435 if (NUMA_BUILD
&& !did_zlc_setup
) {
1436 /* we do zlc_setup after the first zone is tried */
1437 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1443 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1444 /* Disable zlc cache for second zonelist scan */
1452 * This is the 'heart' of the zoned buddy allocator.
1455 __alloc_pages_internal(gfp_t gfp_mask
, unsigned int order
,
1456 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1458 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1459 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1463 struct reclaim_state reclaim_state
;
1464 struct task_struct
*p
= current
;
1467 unsigned long did_some_progress
;
1468 unsigned long pages_reclaimed
= 0;
1470 might_sleep_if(wait
);
1472 if (should_fail_alloc_page(gfp_mask
, order
))
1476 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1478 if (unlikely(!z
->zone
)) {
1480 * Happens if we have an empty zonelist as a result of
1481 * GFP_THISNODE being used on a memoryless node
1486 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1487 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1492 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1493 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1494 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1495 * using a larger set of nodes after it has established that the
1496 * allowed per node queues are empty and that nodes are
1499 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1502 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1503 wakeup_kswapd(zone
, order
);
1506 * OK, we're below the kswapd watermark and have kicked background
1507 * reclaim. Now things get more complex, so set up alloc_flags according
1508 * to how we want to proceed.
1510 * The caller may dip into page reserves a bit more if the caller
1511 * cannot run direct reclaim, or if the caller has realtime scheduling
1512 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1513 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1515 alloc_flags
= ALLOC_WMARK_MIN
;
1516 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1517 alloc_flags
|= ALLOC_HARDER
;
1518 if (gfp_mask
& __GFP_HIGH
)
1519 alloc_flags
|= ALLOC_HIGH
;
1521 alloc_flags
|= ALLOC_CPUSET
;
1524 * Go through the zonelist again. Let __GFP_HIGH and allocations
1525 * coming from realtime tasks go deeper into reserves.
1527 * This is the last chance, in general, before the goto nopage.
1528 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1529 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1531 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1532 high_zoneidx
, alloc_flags
);
1536 /* This allocation should allow future memory freeing. */
1539 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1540 && !in_interrupt()) {
1541 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1543 /* go through the zonelist yet again, ignoring mins */
1544 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1545 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1548 if (gfp_mask
& __GFP_NOFAIL
) {
1549 congestion_wait(WRITE
, HZ
/50);
1556 /* Atomic allocations - we can't balance anything */
1562 /* We now go into synchronous reclaim */
1563 cpuset_memory_pressure_bump();
1564 p
->flags
|= PF_MEMALLOC
;
1565 reclaim_state
.reclaimed_slab
= 0;
1566 p
->reclaim_state
= &reclaim_state
;
1568 did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
);
1570 p
->reclaim_state
= NULL
;
1571 p
->flags
&= ~PF_MEMALLOC
;
1578 if (likely(did_some_progress
)) {
1579 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1580 zonelist
, high_zoneidx
, alloc_flags
);
1583 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1584 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1585 schedule_timeout_uninterruptible(1);
1590 * Go through the zonelist yet one more time, keep
1591 * very high watermark here, this is only to catch
1592 * a parallel oom killing, we must fail if we're still
1593 * under heavy pressure.
1595 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1596 order
, zonelist
, high_zoneidx
,
1597 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1599 clear_zonelist_oom(zonelist
, gfp_mask
);
1603 /* The OOM killer will not help higher order allocs so fail */
1604 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1605 clear_zonelist_oom(zonelist
, gfp_mask
);
1609 out_of_memory(zonelist
, gfp_mask
, order
);
1610 clear_zonelist_oom(zonelist
, gfp_mask
);
1615 * Don't let big-order allocations loop unless the caller explicitly
1616 * requests that. Wait for some write requests to complete then retry.
1618 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1619 * means __GFP_NOFAIL, but that may not be true in other
1622 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1623 * specified, then we retry until we no longer reclaim any pages
1624 * (above), or we've reclaimed an order of pages at least as
1625 * large as the allocation's order. In both cases, if the
1626 * allocation still fails, we stop retrying.
1628 pages_reclaimed
+= did_some_progress
;
1630 if (!(gfp_mask
& __GFP_NORETRY
)) {
1631 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1634 if (gfp_mask
& __GFP_REPEAT
&&
1635 pages_reclaimed
< (1 << order
))
1638 if (gfp_mask
& __GFP_NOFAIL
)
1642 congestion_wait(WRITE
, HZ
/50);
1647 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1648 printk(KERN_WARNING
"%s: page allocation failure."
1649 " order:%d, mode:0x%x\n",
1650 p
->comm
, order
, gfp_mask
);
1657 EXPORT_SYMBOL(__alloc_pages_internal
);
1660 * Common helper functions.
1662 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1665 page
= alloc_pages(gfp_mask
, order
);
1668 return (unsigned long) page_address(page
);
1671 EXPORT_SYMBOL(__get_free_pages
);
1673 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1678 * get_zeroed_page() returns a 32-bit address, which cannot represent
1681 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1683 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1685 return (unsigned long) page_address(page
);
1689 EXPORT_SYMBOL(get_zeroed_page
);
1691 void __pagevec_free(struct pagevec
*pvec
)
1693 int i
= pagevec_count(pvec
);
1696 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1699 void __free_pages(struct page
*page
, unsigned int order
)
1701 if (put_page_testzero(page
)) {
1703 free_hot_page(page
);
1705 __free_pages_ok(page
, order
);
1709 EXPORT_SYMBOL(__free_pages
);
1711 void free_pages(unsigned long addr
, unsigned int order
)
1714 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1715 __free_pages(virt_to_page((void *)addr
), order
);
1719 EXPORT_SYMBOL(free_pages
);
1722 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1723 * @size: the number of bytes to allocate
1724 * @gfp_mask: GFP flags for the allocation
1726 * This function is similar to alloc_pages(), except that it allocates the
1727 * minimum number of pages to satisfy the request. alloc_pages() can only
1728 * allocate memory in power-of-two pages.
1730 * This function is also limited by MAX_ORDER.
1732 * Memory allocated by this function must be released by free_pages_exact().
1734 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1736 unsigned int order
= get_order(size
);
1739 addr
= __get_free_pages(gfp_mask
, order
);
1741 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1742 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1744 split_page(virt_to_page(addr
), order
);
1745 while (used
< alloc_end
) {
1751 return (void *)addr
;
1753 EXPORT_SYMBOL(alloc_pages_exact
);
1756 * free_pages_exact - release memory allocated via alloc_pages_exact()
1757 * @virt: the value returned by alloc_pages_exact.
1758 * @size: size of allocation, same value as passed to alloc_pages_exact().
1760 * Release the memory allocated by a previous call to alloc_pages_exact.
1762 void free_pages_exact(void *virt
, size_t size
)
1764 unsigned long addr
= (unsigned long)virt
;
1765 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1767 while (addr
< end
) {
1772 EXPORT_SYMBOL(free_pages_exact
);
1774 static unsigned int nr_free_zone_pages(int offset
)
1779 /* Just pick one node, since fallback list is circular */
1780 unsigned int sum
= 0;
1782 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1784 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1785 unsigned long size
= zone
->present_pages
;
1786 unsigned long high
= zone
->pages_high
;
1795 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1797 unsigned int nr_free_buffer_pages(void)
1799 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1801 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1804 * Amount of free RAM allocatable within all zones
1806 unsigned int nr_free_pagecache_pages(void)
1808 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1811 static inline void show_node(struct zone
*zone
)
1814 printk("Node %d ", zone_to_nid(zone
));
1817 void si_meminfo(struct sysinfo
*val
)
1819 val
->totalram
= totalram_pages
;
1821 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1822 val
->bufferram
= nr_blockdev_pages();
1823 val
->totalhigh
= totalhigh_pages
;
1824 val
->freehigh
= nr_free_highpages();
1825 val
->mem_unit
= PAGE_SIZE
;
1828 EXPORT_SYMBOL(si_meminfo
);
1831 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1833 pg_data_t
*pgdat
= NODE_DATA(nid
);
1835 val
->totalram
= pgdat
->node_present_pages
;
1836 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1837 #ifdef CONFIG_HIGHMEM
1838 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1839 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1845 val
->mem_unit
= PAGE_SIZE
;
1849 #define K(x) ((x) << (PAGE_SHIFT-10))
1852 * Show free area list (used inside shift_scroll-lock stuff)
1853 * We also calculate the percentage fragmentation. We do this by counting the
1854 * memory on each free list with the exception of the first item on the list.
1856 void show_free_areas(void)
1861 for_each_zone(zone
) {
1862 if (!populated_zone(zone
))
1866 printk("%s per-cpu:\n", zone
->name
);
1868 for_each_online_cpu(cpu
) {
1869 struct per_cpu_pageset
*pageset
;
1871 pageset
= zone_pcp(zone
, cpu
);
1873 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1874 cpu
, pageset
->pcp
.high
,
1875 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1879 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1880 " inactive_file:%lu"
1881 //TODO: check/adjust line lengths
1882 #ifdef CONFIG_UNEVICTABLE_LRU
1885 " dirty:%lu writeback:%lu unstable:%lu\n"
1886 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1887 global_page_state(NR_ACTIVE_ANON
),
1888 global_page_state(NR_ACTIVE_FILE
),
1889 global_page_state(NR_INACTIVE_ANON
),
1890 global_page_state(NR_INACTIVE_FILE
),
1891 #ifdef CONFIG_UNEVICTABLE_LRU
1892 global_page_state(NR_UNEVICTABLE
),
1894 global_page_state(NR_FILE_DIRTY
),
1895 global_page_state(NR_WRITEBACK
),
1896 global_page_state(NR_UNSTABLE_NFS
),
1897 global_page_state(NR_FREE_PAGES
),
1898 global_page_state(NR_SLAB_RECLAIMABLE
) +
1899 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1900 global_page_state(NR_FILE_MAPPED
),
1901 global_page_state(NR_PAGETABLE
),
1902 global_page_state(NR_BOUNCE
));
1904 for_each_zone(zone
) {
1907 if (!populated_zone(zone
))
1916 " active_anon:%lukB"
1917 " inactive_anon:%lukB"
1918 " active_file:%lukB"
1919 " inactive_file:%lukB"
1920 #ifdef CONFIG_UNEVICTABLE_LRU
1921 " unevictable:%lukB"
1924 " pages_scanned:%lu"
1925 " all_unreclaimable? %s"
1928 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1931 K(zone
->pages_high
),
1932 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
1933 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
1934 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
1935 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
1936 #ifdef CONFIG_UNEVICTABLE_LRU
1937 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
1939 K(zone
->present_pages
),
1940 zone
->pages_scanned
,
1941 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1943 printk("lowmem_reserve[]:");
1944 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1945 printk(" %lu", zone
->lowmem_reserve
[i
]);
1949 for_each_zone(zone
) {
1950 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1952 if (!populated_zone(zone
))
1956 printk("%s: ", zone
->name
);
1958 spin_lock_irqsave(&zone
->lock
, flags
);
1959 for (order
= 0; order
< MAX_ORDER
; order
++) {
1960 nr
[order
] = zone
->free_area
[order
].nr_free
;
1961 total
+= nr
[order
] << order
;
1963 spin_unlock_irqrestore(&zone
->lock
, flags
);
1964 for (order
= 0; order
< MAX_ORDER
; order
++)
1965 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1966 printk("= %lukB\n", K(total
));
1969 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
1971 show_swap_cache_info();
1974 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
1976 zoneref
->zone
= zone
;
1977 zoneref
->zone_idx
= zone_idx(zone
);
1981 * Builds allocation fallback zone lists.
1983 * Add all populated zones of a node to the zonelist.
1985 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1986 int nr_zones
, enum zone_type zone_type
)
1990 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1995 zone
= pgdat
->node_zones
+ zone_type
;
1996 if (populated_zone(zone
)) {
1997 zoneref_set_zone(zone
,
1998 &zonelist
->_zonerefs
[nr_zones
++]);
1999 check_highest_zone(zone_type
);
2002 } while (zone_type
);
2009 * 0 = automatic detection of better ordering.
2010 * 1 = order by ([node] distance, -zonetype)
2011 * 2 = order by (-zonetype, [node] distance)
2013 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2014 * the same zonelist. So only NUMA can configure this param.
2016 #define ZONELIST_ORDER_DEFAULT 0
2017 #define ZONELIST_ORDER_NODE 1
2018 #define ZONELIST_ORDER_ZONE 2
2020 /* zonelist order in the kernel.
2021 * set_zonelist_order() will set this to NODE or ZONE.
2023 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2024 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2028 /* The value user specified ....changed by config */
2029 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2030 /* string for sysctl */
2031 #define NUMA_ZONELIST_ORDER_LEN 16
2032 char numa_zonelist_order
[16] = "default";
2035 * interface for configure zonelist ordering.
2036 * command line option "numa_zonelist_order"
2037 * = "[dD]efault - default, automatic configuration.
2038 * = "[nN]ode - order by node locality, then by zone within node
2039 * = "[zZ]one - order by zone, then by locality within zone
2042 static int __parse_numa_zonelist_order(char *s
)
2044 if (*s
== 'd' || *s
== 'D') {
2045 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2046 } else if (*s
== 'n' || *s
== 'N') {
2047 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2048 } else if (*s
== 'z' || *s
== 'Z') {
2049 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2052 "Ignoring invalid numa_zonelist_order value: "
2059 static __init
int setup_numa_zonelist_order(char *s
)
2062 return __parse_numa_zonelist_order(s
);
2065 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2068 * sysctl handler for numa_zonelist_order
2070 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2071 struct file
*file
, void __user
*buffer
, size_t *length
,
2074 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2078 strncpy(saved_string
, (char*)table
->data
,
2079 NUMA_ZONELIST_ORDER_LEN
);
2080 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2084 int oldval
= user_zonelist_order
;
2085 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2087 * bogus value. restore saved string
2089 strncpy((char*)table
->data
, saved_string
,
2090 NUMA_ZONELIST_ORDER_LEN
);
2091 user_zonelist_order
= oldval
;
2092 } else if (oldval
!= user_zonelist_order
)
2093 build_all_zonelists();
2099 #define MAX_NODE_LOAD (num_online_nodes())
2100 static int node_load
[MAX_NUMNODES
];
2103 * find_next_best_node - find the next node that should appear in a given node's fallback list
2104 * @node: node whose fallback list we're appending
2105 * @used_node_mask: nodemask_t of already used nodes
2107 * We use a number of factors to determine which is the next node that should
2108 * appear on a given node's fallback list. The node should not have appeared
2109 * already in @node's fallback list, and it should be the next closest node
2110 * according to the distance array (which contains arbitrary distance values
2111 * from each node to each node in the system), and should also prefer nodes
2112 * with no CPUs, since presumably they'll have very little allocation pressure
2113 * on them otherwise.
2114 * It returns -1 if no node is found.
2116 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2119 int min_val
= INT_MAX
;
2121 node_to_cpumask_ptr(tmp
, 0);
2123 /* Use the local node if we haven't already */
2124 if (!node_isset(node
, *used_node_mask
)) {
2125 node_set(node
, *used_node_mask
);
2129 for_each_node_state(n
, N_HIGH_MEMORY
) {
2131 /* Don't want a node to appear more than once */
2132 if (node_isset(n
, *used_node_mask
))
2135 /* Use the distance array to find the distance */
2136 val
= node_distance(node
, n
);
2138 /* Penalize nodes under us ("prefer the next node") */
2141 /* Give preference to headless and unused nodes */
2142 node_to_cpumask_ptr_next(tmp
, n
);
2143 if (!cpus_empty(*tmp
))
2144 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2146 /* Slight preference for less loaded node */
2147 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2148 val
+= node_load
[n
];
2150 if (val
< min_val
) {
2157 node_set(best_node
, *used_node_mask
);
2164 * Build zonelists ordered by node and zones within node.
2165 * This results in maximum locality--normal zone overflows into local
2166 * DMA zone, if any--but risks exhausting DMA zone.
2168 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2171 struct zonelist
*zonelist
;
2173 zonelist
= &pgdat
->node_zonelists
[0];
2174 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2176 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2178 zonelist
->_zonerefs
[j
].zone
= NULL
;
2179 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2183 * Build gfp_thisnode zonelists
2185 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2188 struct zonelist
*zonelist
;
2190 zonelist
= &pgdat
->node_zonelists
[1];
2191 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2192 zonelist
->_zonerefs
[j
].zone
= NULL
;
2193 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2197 * Build zonelists ordered by zone and nodes within zones.
2198 * This results in conserving DMA zone[s] until all Normal memory is
2199 * exhausted, but results in overflowing to remote node while memory
2200 * may still exist in local DMA zone.
2202 static int node_order
[MAX_NUMNODES
];
2204 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2207 int zone_type
; /* needs to be signed */
2209 struct zonelist
*zonelist
;
2211 zonelist
= &pgdat
->node_zonelists
[0];
2213 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2214 for (j
= 0; j
< nr_nodes
; j
++) {
2215 node
= node_order
[j
];
2216 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2217 if (populated_zone(z
)) {
2219 &zonelist
->_zonerefs
[pos
++]);
2220 check_highest_zone(zone_type
);
2224 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2225 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2228 static int default_zonelist_order(void)
2231 unsigned long low_kmem_size
,total_size
;
2235 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2236 * If they are really small and used heavily, the system can fall
2237 * into OOM very easily.
2238 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2240 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2243 for_each_online_node(nid
) {
2244 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2245 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2246 if (populated_zone(z
)) {
2247 if (zone_type
< ZONE_NORMAL
)
2248 low_kmem_size
+= z
->present_pages
;
2249 total_size
+= z
->present_pages
;
2253 if (!low_kmem_size
|| /* there are no DMA area. */
2254 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2255 return ZONELIST_ORDER_NODE
;
2257 * look into each node's config.
2258 * If there is a node whose DMA/DMA32 memory is very big area on
2259 * local memory, NODE_ORDER may be suitable.
2261 average_size
= total_size
/
2262 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2263 for_each_online_node(nid
) {
2266 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2267 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2268 if (populated_zone(z
)) {
2269 if (zone_type
< ZONE_NORMAL
)
2270 low_kmem_size
+= z
->present_pages
;
2271 total_size
+= z
->present_pages
;
2274 if (low_kmem_size
&&
2275 total_size
> average_size
&& /* ignore small node */
2276 low_kmem_size
> total_size
* 70/100)
2277 return ZONELIST_ORDER_NODE
;
2279 return ZONELIST_ORDER_ZONE
;
2282 static void set_zonelist_order(void)
2284 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2285 current_zonelist_order
= default_zonelist_order();
2287 current_zonelist_order
= user_zonelist_order
;
2290 static void build_zonelists(pg_data_t
*pgdat
)
2294 nodemask_t used_mask
;
2295 int local_node
, prev_node
;
2296 struct zonelist
*zonelist
;
2297 int order
= current_zonelist_order
;
2299 /* initialize zonelists */
2300 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2301 zonelist
= pgdat
->node_zonelists
+ i
;
2302 zonelist
->_zonerefs
[0].zone
= NULL
;
2303 zonelist
->_zonerefs
[0].zone_idx
= 0;
2306 /* NUMA-aware ordering of nodes */
2307 local_node
= pgdat
->node_id
;
2308 load
= num_online_nodes();
2309 prev_node
= local_node
;
2310 nodes_clear(used_mask
);
2312 memset(node_load
, 0, sizeof(node_load
));
2313 memset(node_order
, 0, sizeof(node_order
));
2316 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2317 int distance
= node_distance(local_node
, node
);
2320 * If another node is sufficiently far away then it is better
2321 * to reclaim pages in a zone before going off node.
2323 if (distance
> RECLAIM_DISTANCE
)
2324 zone_reclaim_mode
= 1;
2327 * We don't want to pressure a particular node.
2328 * So adding penalty to the first node in same
2329 * distance group to make it round-robin.
2331 if (distance
!= node_distance(local_node
, prev_node
))
2332 node_load
[node
] = load
;
2336 if (order
== ZONELIST_ORDER_NODE
)
2337 build_zonelists_in_node_order(pgdat
, node
);
2339 node_order
[j
++] = node
; /* remember order */
2342 if (order
== ZONELIST_ORDER_ZONE
) {
2343 /* calculate node order -- i.e., DMA last! */
2344 build_zonelists_in_zone_order(pgdat
, j
);
2347 build_thisnode_zonelists(pgdat
);
2350 /* Construct the zonelist performance cache - see further mmzone.h */
2351 static void build_zonelist_cache(pg_data_t
*pgdat
)
2353 struct zonelist
*zonelist
;
2354 struct zonelist_cache
*zlc
;
2357 zonelist
= &pgdat
->node_zonelists
[0];
2358 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2359 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2360 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2361 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2365 #else /* CONFIG_NUMA */
2367 static void set_zonelist_order(void)
2369 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2372 static void build_zonelists(pg_data_t
*pgdat
)
2374 int node
, local_node
;
2376 struct zonelist
*zonelist
;
2378 local_node
= pgdat
->node_id
;
2380 zonelist
= &pgdat
->node_zonelists
[0];
2381 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2384 * Now we build the zonelist so that it contains the zones
2385 * of all the other nodes.
2386 * We don't want to pressure a particular node, so when
2387 * building the zones for node N, we make sure that the
2388 * zones coming right after the local ones are those from
2389 * node N+1 (modulo N)
2391 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2392 if (!node_online(node
))
2394 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2397 for (node
= 0; node
< local_node
; node
++) {
2398 if (!node_online(node
))
2400 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2404 zonelist
->_zonerefs
[j
].zone
= NULL
;
2405 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2408 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2409 static void build_zonelist_cache(pg_data_t
*pgdat
)
2411 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2414 #endif /* CONFIG_NUMA */
2416 /* return values int ....just for stop_machine() */
2417 static int __build_all_zonelists(void *dummy
)
2421 for_each_online_node(nid
) {
2422 pg_data_t
*pgdat
= NODE_DATA(nid
);
2424 build_zonelists(pgdat
);
2425 build_zonelist_cache(pgdat
);
2430 void build_all_zonelists(void)
2432 set_zonelist_order();
2434 if (system_state
== SYSTEM_BOOTING
) {
2435 __build_all_zonelists(NULL
);
2436 mminit_verify_zonelist();
2437 cpuset_init_current_mems_allowed();
2439 /* we have to stop all cpus to guarantee there is no user
2441 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2442 /* cpuset refresh routine should be here */
2444 vm_total_pages
= nr_free_pagecache_pages();
2446 * Disable grouping by mobility if the number of pages in the
2447 * system is too low to allow the mechanism to work. It would be
2448 * more accurate, but expensive to check per-zone. This check is
2449 * made on memory-hotadd so a system can start with mobility
2450 * disabled and enable it later
2452 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2453 page_group_by_mobility_disabled
= 1;
2455 page_group_by_mobility_disabled
= 0;
2457 printk("Built %i zonelists in %s order, mobility grouping %s. "
2458 "Total pages: %ld\n",
2460 zonelist_order_name
[current_zonelist_order
],
2461 page_group_by_mobility_disabled
? "off" : "on",
2464 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2469 * Helper functions to size the waitqueue hash table.
2470 * Essentially these want to choose hash table sizes sufficiently
2471 * large so that collisions trying to wait on pages are rare.
2472 * But in fact, the number of active page waitqueues on typical
2473 * systems is ridiculously low, less than 200. So this is even
2474 * conservative, even though it seems large.
2476 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2477 * waitqueues, i.e. the size of the waitq table given the number of pages.
2479 #define PAGES_PER_WAITQUEUE 256
2481 #ifndef CONFIG_MEMORY_HOTPLUG
2482 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2484 unsigned long size
= 1;
2486 pages
/= PAGES_PER_WAITQUEUE
;
2488 while (size
< pages
)
2492 * Once we have dozens or even hundreds of threads sleeping
2493 * on IO we've got bigger problems than wait queue collision.
2494 * Limit the size of the wait table to a reasonable size.
2496 size
= min(size
, 4096UL);
2498 return max(size
, 4UL);
2502 * A zone's size might be changed by hot-add, so it is not possible to determine
2503 * a suitable size for its wait_table. So we use the maximum size now.
2505 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2507 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2508 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2509 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2511 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2512 * or more by the traditional way. (See above). It equals:
2514 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2515 * ia64(16K page size) : = ( 8G + 4M)byte.
2516 * powerpc (64K page size) : = (32G +16M)byte.
2518 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2525 * This is an integer logarithm so that shifts can be used later
2526 * to extract the more random high bits from the multiplicative
2527 * hash function before the remainder is taken.
2529 static inline unsigned long wait_table_bits(unsigned long size
)
2534 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2537 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2538 * of blocks reserved is based on zone->pages_min. The memory within the
2539 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2540 * higher will lead to a bigger reserve which will get freed as contiguous
2541 * blocks as reclaim kicks in
2543 static void setup_zone_migrate_reserve(struct zone
*zone
)
2545 unsigned long start_pfn
, pfn
, end_pfn
;
2547 unsigned long reserve
, block_migratetype
;
2549 /* Get the start pfn, end pfn and the number of blocks to reserve */
2550 start_pfn
= zone
->zone_start_pfn
;
2551 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2552 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2555 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2556 if (!pfn_valid(pfn
))
2558 page
= pfn_to_page(pfn
);
2560 /* Watch out for overlapping nodes */
2561 if (page_to_nid(page
) != zone_to_nid(zone
))
2564 /* Blocks with reserved pages will never free, skip them. */
2565 if (PageReserved(page
))
2568 block_migratetype
= get_pageblock_migratetype(page
);
2570 /* If this block is reserved, account for it */
2571 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2576 /* Suitable for reserving if this block is movable */
2577 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2578 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2579 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2585 * If the reserve is met and this is a previous reserved block,
2588 if (block_migratetype
== MIGRATE_RESERVE
) {
2589 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2590 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2596 * Initially all pages are reserved - free ones are freed
2597 * up by free_all_bootmem() once the early boot process is
2598 * done. Non-atomic initialization, single-pass.
2600 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2601 unsigned long start_pfn
, enum memmap_context context
)
2604 unsigned long end_pfn
= start_pfn
+ size
;
2608 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2609 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2611 * There can be holes in boot-time mem_map[]s
2612 * handed to this function. They do not
2613 * exist on hotplugged memory.
2615 if (context
== MEMMAP_EARLY
) {
2616 if (!early_pfn_valid(pfn
))
2618 if (!early_pfn_in_nid(pfn
, nid
))
2621 page
= pfn_to_page(pfn
);
2622 set_page_links(page
, zone
, nid
, pfn
);
2623 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2624 init_page_count(page
);
2625 reset_page_mapcount(page
);
2626 SetPageReserved(page
);
2628 * Mark the block movable so that blocks are reserved for
2629 * movable at startup. This will force kernel allocations
2630 * to reserve their blocks rather than leaking throughout
2631 * the address space during boot when many long-lived
2632 * kernel allocations are made. Later some blocks near
2633 * the start are marked MIGRATE_RESERVE by
2634 * setup_zone_migrate_reserve()
2636 * bitmap is created for zone's valid pfn range. but memmap
2637 * can be created for invalid pages (for alignment)
2638 * check here not to call set_pageblock_migratetype() against
2641 if ((z
->zone_start_pfn
<= pfn
)
2642 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2643 && !(pfn
& (pageblock_nr_pages
- 1)))
2644 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2646 INIT_LIST_HEAD(&page
->lru
);
2647 #ifdef WANT_PAGE_VIRTUAL
2648 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2649 if (!is_highmem_idx(zone
))
2650 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2655 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2658 for_each_migratetype_order(order
, t
) {
2659 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2660 zone
->free_area
[order
].nr_free
= 0;
2664 #ifndef __HAVE_ARCH_MEMMAP_INIT
2665 #define memmap_init(size, nid, zone, start_pfn) \
2666 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2669 static int zone_batchsize(struct zone
*zone
)
2674 * The per-cpu-pages pools are set to around 1000th of the
2675 * size of the zone. But no more than 1/2 of a meg.
2677 * OK, so we don't know how big the cache is. So guess.
2679 batch
= zone
->present_pages
/ 1024;
2680 if (batch
* PAGE_SIZE
> 512 * 1024)
2681 batch
= (512 * 1024) / PAGE_SIZE
;
2682 batch
/= 4; /* We effectively *= 4 below */
2687 * Clamp the batch to a 2^n - 1 value. Having a power
2688 * of 2 value was found to be more likely to have
2689 * suboptimal cache aliasing properties in some cases.
2691 * For example if 2 tasks are alternately allocating
2692 * batches of pages, one task can end up with a lot
2693 * of pages of one half of the possible page colors
2694 * and the other with pages of the other colors.
2696 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2701 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2703 struct per_cpu_pages
*pcp
;
2705 memset(p
, 0, sizeof(*p
));
2709 pcp
->high
= 6 * batch
;
2710 pcp
->batch
= max(1UL, 1 * batch
);
2711 INIT_LIST_HEAD(&pcp
->list
);
2715 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2716 * to the value high for the pageset p.
2719 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2722 struct per_cpu_pages
*pcp
;
2726 pcp
->batch
= max(1UL, high
/4);
2727 if ((high
/4) > (PAGE_SHIFT
* 8))
2728 pcp
->batch
= PAGE_SHIFT
* 8;
2734 * Boot pageset table. One per cpu which is going to be used for all
2735 * zones and all nodes. The parameters will be set in such a way
2736 * that an item put on a list will immediately be handed over to
2737 * the buddy list. This is safe since pageset manipulation is done
2738 * with interrupts disabled.
2740 * Some NUMA counter updates may also be caught by the boot pagesets.
2742 * The boot_pagesets must be kept even after bootup is complete for
2743 * unused processors and/or zones. They do play a role for bootstrapping
2744 * hotplugged processors.
2746 * zoneinfo_show() and maybe other functions do
2747 * not check if the processor is online before following the pageset pointer.
2748 * Other parts of the kernel may not check if the zone is available.
2750 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2753 * Dynamically allocate memory for the
2754 * per cpu pageset array in struct zone.
2756 static int __cpuinit
process_zones(int cpu
)
2758 struct zone
*zone
, *dzone
;
2759 int node
= cpu_to_node(cpu
);
2761 node_set_state(node
, N_CPU
); /* this node has a cpu */
2763 for_each_zone(zone
) {
2765 if (!populated_zone(zone
))
2768 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2770 if (!zone_pcp(zone
, cpu
))
2773 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2775 if (percpu_pagelist_fraction
)
2776 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2777 (zone
->present_pages
/ percpu_pagelist_fraction
));
2782 for_each_zone(dzone
) {
2783 if (!populated_zone(dzone
))
2787 kfree(zone_pcp(dzone
, cpu
));
2788 zone_pcp(dzone
, cpu
) = NULL
;
2793 static inline void free_zone_pagesets(int cpu
)
2797 for_each_zone(zone
) {
2798 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2800 /* Free per_cpu_pageset if it is slab allocated */
2801 if (pset
!= &boot_pageset
[cpu
])
2803 zone_pcp(zone
, cpu
) = NULL
;
2807 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2808 unsigned long action
,
2811 int cpu
= (long)hcpu
;
2812 int ret
= NOTIFY_OK
;
2815 case CPU_UP_PREPARE
:
2816 case CPU_UP_PREPARE_FROZEN
:
2817 if (process_zones(cpu
))
2820 case CPU_UP_CANCELED
:
2821 case CPU_UP_CANCELED_FROZEN
:
2823 case CPU_DEAD_FROZEN
:
2824 free_zone_pagesets(cpu
);
2832 static struct notifier_block __cpuinitdata pageset_notifier
=
2833 { &pageset_cpuup_callback
, NULL
, 0 };
2835 void __init
setup_per_cpu_pageset(void)
2839 /* Initialize per_cpu_pageset for cpu 0.
2840 * A cpuup callback will do this for every cpu
2841 * as it comes online
2843 err
= process_zones(smp_processor_id());
2845 register_cpu_notifier(&pageset_notifier
);
2850 static noinline __init_refok
2851 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2854 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2858 * The per-page waitqueue mechanism uses hashed waitqueues
2861 zone
->wait_table_hash_nr_entries
=
2862 wait_table_hash_nr_entries(zone_size_pages
);
2863 zone
->wait_table_bits
=
2864 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2865 alloc_size
= zone
->wait_table_hash_nr_entries
2866 * sizeof(wait_queue_head_t
);
2868 if (!slab_is_available()) {
2869 zone
->wait_table
= (wait_queue_head_t
*)
2870 alloc_bootmem_node(pgdat
, alloc_size
);
2873 * This case means that a zone whose size was 0 gets new memory
2874 * via memory hot-add.
2875 * But it may be the case that a new node was hot-added. In
2876 * this case vmalloc() will not be able to use this new node's
2877 * memory - this wait_table must be initialized to use this new
2878 * node itself as well.
2879 * To use this new node's memory, further consideration will be
2882 zone
->wait_table
= vmalloc(alloc_size
);
2884 if (!zone
->wait_table
)
2887 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2888 init_waitqueue_head(zone
->wait_table
+ i
);
2893 static __meminit
void zone_pcp_init(struct zone
*zone
)
2896 unsigned long batch
= zone_batchsize(zone
);
2898 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2900 /* Early boot. Slab allocator not functional yet */
2901 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2902 setup_pageset(&boot_pageset
[cpu
],0);
2904 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2907 if (zone
->present_pages
)
2908 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2909 zone
->name
, zone
->present_pages
, batch
);
2912 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2913 unsigned long zone_start_pfn
,
2915 enum memmap_context context
)
2917 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2919 ret
= zone_wait_table_init(zone
, size
);
2922 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2924 zone
->zone_start_pfn
= zone_start_pfn
;
2926 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
2927 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2929 (unsigned long)zone_idx(zone
),
2930 zone_start_pfn
, (zone_start_pfn
+ size
));
2932 zone_init_free_lists(zone
);
2937 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2939 * Basic iterator support. Return the first range of PFNs for a node
2940 * Note: nid == MAX_NUMNODES returns first region regardless of node
2942 static int __meminit
first_active_region_index_in_nid(int nid
)
2946 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2947 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2954 * Basic iterator support. Return the next active range of PFNs for a node
2955 * Note: nid == MAX_NUMNODES returns next region regardless of node
2957 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2959 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2960 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2966 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2968 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2969 * Architectures may implement their own version but if add_active_range()
2970 * was used and there are no special requirements, this is a convenient
2973 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2977 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2978 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2979 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2981 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2982 return early_node_map
[i
].nid
;
2987 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2989 /* Basic iterator support to walk early_node_map[] */
2990 #define for_each_active_range_index_in_nid(i, nid) \
2991 for (i = first_active_region_index_in_nid(nid); i != -1; \
2992 i = next_active_region_index_in_nid(i, nid))
2995 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2996 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2997 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2999 * If an architecture guarantees that all ranges registered with
3000 * add_active_ranges() contain no holes and may be freed, this
3001 * this function may be used instead of calling free_bootmem() manually.
3003 void __init
free_bootmem_with_active_regions(int nid
,
3004 unsigned long max_low_pfn
)
3008 for_each_active_range_index_in_nid(i
, nid
) {
3009 unsigned long size_pages
= 0;
3010 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3012 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3015 if (end_pfn
> max_low_pfn
)
3016 end_pfn
= max_low_pfn
;
3018 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3019 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3020 PFN_PHYS(early_node_map
[i
].start_pfn
),
3021 size_pages
<< PAGE_SHIFT
);
3025 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3030 for_each_active_range_index_in_nid(i
, nid
) {
3031 ret
= work_fn(early_node_map
[i
].start_pfn
,
3032 early_node_map
[i
].end_pfn
, data
);
3038 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3039 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3041 * If an architecture guarantees that all ranges registered with
3042 * add_active_ranges() contain no holes and may be freed, this
3043 * function may be used instead of calling memory_present() manually.
3045 void __init
sparse_memory_present_with_active_regions(int nid
)
3049 for_each_active_range_index_in_nid(i
, nid
)
3050 memory_present(early_node_map
[i
].nid
,
3051 early_node_map
[i
].start_pfn
,
3052 early_node_map
[i
].end_pfn
);
3056 * push_node_boundaries - Push node boundaries to at least the requested boundary
3057 * @nid: The nid of the node to push the boundary for
3058 * @start_pfn: The start pfn of the node
3059 * @end_pfn: The end pfn of the node
3061 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3062 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3063 * be hotplugged even though no physical memory exists. This function allows
3064 * an arch to push out the node boundaries so mem_map is allocated that can
3067 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3068 void __init
push_node_boundaries(unsigned int nid
,
3069 unsigned long start_pfn
, unsigned long end_pfn
)
3071 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3072 "Entering push_node_boundaries(%u, %lu, %lu)\n",
3073 nid
, start_pfn
, end_pfn
);
3075 /* Initialise the boundary for this node if necessary */
3076 if (node_boundary_end_pfn
[nid
] == 0)
3077 node_boundary_start_pfn
[nid
] = -1UL;
3079 /* Update the boundaries */
3080 if (node_boundary_start_pfn
[nid
] > start_pfn
)
3081 node_boundary_start_pfn
[nid
] = start_pfn
;
3082 if (node_boundary_end_pfn
[nid
] < end_pfn
)
3083 node_boundary_end_pfn
[nid
] = end_pfn
;
3086 /* If necessary, push the node boundary out for reserve hotadd */
3087 static void __meminit
account_node_boundary(unsigned int nid
,
3088 unsigned long *start_pfn
, unsigned long *end_pfn
)
3090 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3091 "Entering account_node_boundary(%u, %lu, %lu)\n",
3092 nid
, *start_pfn
, *end_pfn
);
3094 /* Return if boundary information has not been provided */
3095 if (node_boundary_end_pfn
[nid
] == 0)
3098 /* Check the boundaries and update if necessary */
3099 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
3100 *start_pfn
= node_boundary_start_pfn
[nid
];
3101 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
3102 *end_pfn
= node_boundary_end_pfn
[nid
];
3105 void __init
push_node_boundaries(unsigned int nid
,
3106 unsigned long start_pfn
, unsigned long end_pfn
) {}
3108 static void __meminit
account_node_boundary(unsigned int nid
,
3109 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3114 * get_pfn_range_for_nid - Return the start and end page frames for a node
3115 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3116 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3117 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3119 * It returns the start and end page frame of a node based on information
3120 * provided by an arch calling add_active_range(). If called for a node
3121 * with no available memory, a warning is printed and the start and end
3124 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3125 unsigned long *start_pfn
, unsigned long *end_pfn
)
3131 for_each_active_range_index_in_nid(i
, nid
) {
3132 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3133 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3136 if (*start_pfn
== -1UL)
3139 /* Push the node boundaries out if requested */
3140 account_node_boundary(nid
, start_pfn
, end_pfn
);
3144 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3145 * assumption is made that zones within a node are ordered in monotonic
3146 * increasing memory addresses so that the "highest" populated zone is used
3148 static void __init
find_usable_zone_for_movable(void)
3151 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3152 if (zone_index
== ZONE_MOVABLE
)
3155 if (arch_zone_highest_possible_pfn
[zone_index
] >
3156 arch_zone_lowest_possible_pfn
[zone_index
])
3160 VM_BUG_ON(zone_index
== -1);
3161 movable_zone
= zone_index
;
3165 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3166 * because it is sized independant of architecture. Unlike the other zones,
3167 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3168 * in each node depending on the size of each node and how evenly kernelcore
3169 * is distributed. This helper function adjusts the zone ranges
3170 * provided by the architecture for a given node by using the end of the
3171 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3172 * zones within a node are in order of monotonic increases memory addresses
3174 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3175 unsigned long zone_type
,
3176 unsigned long node_start_pfn
,
3177 unsigned long node_end_pfn
,
3178 unsigned long *zone_start_pfn
,
3179 unsigned long *zone_end_pfn
)
3181 /* Only adjust if ZONE_MOVABLE is on this node */
3182 if (zone_movable_pfn
[nid
]) {
3183 /* Size ZONE_MOVABLE */
3184 if (zone_type
== ZONE_MOVABLE
) {
3185 *zone_start_pfn
= zone_movable_pfn
[nid
];
3186 *zone_end_pfn
= min(node_end_pfn
,
3187 arch_zone_highest_possible_pfn
[movable_zone
]);
3189 /* Adjust for ZONE_MOVABLE starting within this range */
3190 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3191 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3192 *zone_end_pfn
= zone_movable_pfn
[nid
];
3194 /* Check if this whole range is within ZONE_MOVABLE */
3195 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3196 *zone_start_pfn
= *zone_end_pfn
;
3201 * Return the number of pages a zone spans in a node, including holes
3202 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3204 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3205 unsigned long zone_type
,
3206 unsigned long *ignored
)
3208 unsigned long node_start_pfn
, node_end_pfn
;
3209 unsigned long zone_start_pfn
, zone_end_pfn
;
3211 /* Get the start and end of the node and zone */
3212 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3213 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3214 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3215 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3216 node_start_pfn
, node_end_pfn
,
3217 &zone_start_pfn
, &zone_end_pfn
);
3219 /* Check that this node has pages within the zone's required range */
3220 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3223 /* Move the zone boundaries inside the node if necessary */
3224 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3225 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3227 /* Return the spanned pages */
3228 return zone_end_pfn
- zone_start_pfn
;
3232 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3233 * then all holes in the requested range will be accounted for.
3235 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3236 unsigned long range_start_pfn
,
3237 unsigned long range_end_pfn
)
3240 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3241 unsigned long start_pfn
;
3243 /* Find the end_pfn of the first active range of pfns in the node */
3244 i
= first_active_region_index_in_nid(nid
);
3248 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3250 /* Account for ranges before physical memory on this node */
3251 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3252 hole_pages
= prev_end_pfn
- range_start_pfn
;
3254 /* Find all holes for the zone within the node */
3255 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3257 /* No need to continue if prev_end_pfn is outside the zone */
3258 if (prev_end_pfn
>= range_end_pfn
)
3261 /* Make sure the end of the zone is not within the hole */
3262 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3263 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3265 /* Update the hole size cound and move on */
3266 if (start_pfn
> range_start_pfn
) {
3267 BUG_ON(prev_end_pfn
> start_pfn
);
3268 hole_pages
+= start_pfn
- prev_end_pfn
;
3270 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3273 /* Account for ranges past physical memory on this node */
3274 if (range_end_pfn
> prev_end_pfn
)
3275 hole_pages
+= range_end_pfn
-
3276 max(range_start_pfn
, prev_end_pfn
);
3282 * absent_pages_in_range - Return number of page frames in holes within a range
3283 * @start_pfn: The start PFN to start searching for holes
3284 * @end_pfn: The end PFN to stop searching for holes
3286 * It returns the number of pages frames in memory holes within a range.
3288 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3289 unsigned long end_pfn
)
3291 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3294 /* Return the number of page frames in holes in a zone on a node */
3295 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3296 unsigned long zone_type
,
3297 unsigned long *ignored
)
3299 unsigned long node_start_pfn
, node_end_pfn
;
3300 unsigned long zone_start_pfn
, zone_end_pfn
;
3302 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3303 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3305 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3308 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3309 node_start_pfn
, node_end_pfn
,
3310 &zone_start_pfn
, &zone_end_pfn
);
3311 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3315 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3316 unsigned long zone_type
,
3317 unsigned long *zones_size
)
3319 return zones_size
[zone_type
];
3322 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3323 unsigned long zone_type
,
3324 unsigned long *zholes_size
)
3329 return zholes_size
[zone_type
];
3334 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3335 unsigned long *zones_size
, unsigned long *zholes_size
)
3337 unsigned long realtotalpages
, totalpages
= 0;
3340 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3341 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3343 pgdat
->node_spanned_pages
= totalpages
;
3345 realtotalpages
= totalpages
;
3346 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3348 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3350 pgdat
->node_present_pages
= realtotalpages
;
3351 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3355 #ifndef CONFIG_SPARSEMEM
3357 * Calculate the size of the zone->blockflags rounded to an unsigned long
3358 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3359 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3360 * round what is now in bits to nearest long in bits, then return it in
3363 static unsigned long __init
usemap_size(unsigned long zonesize
)
3365 unsigned long usemapsize
;
3367 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3368 usemapsize
= usemapsize
>> pageblock_order
;
3369 usemapsize
*= NR_PAGEBLOCK_BITS
;
3370 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3372 return usemapsize
/ 8;
3375 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3376 struct zone
*zone
, unsigned long zonesize
)
3378 unsigned long usemapsize
= usemap_size(zonesize
);
3379 zone
->pageblock_flags
= NULL
;
3381 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3382 memset(zone
->pageblock_flags
, 0, usemapsize
);
3386 static void inline setup_usemap(struct pglist_data
*pgdat
,
3387 struct zone
*zone
, unsigned long zonesize
) {}
3388 #endif /* CONFIG_SPARSEMEM */
3390 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3392 /* Return a sensible default order for the pageblock size. */
3393 static inline int pageblock_default_order(void)
3395 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3396 return HUGETLB_PAGE_ORDER
;
3401 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3402 static inline void __init
set_pageblock_order(unsigned int order
)
3404 /* Check that pageblock_nr_pages has not already been setup */
3405 if (pageblock_order
)
3409 * Assume the largest contiguous order of interest is a huge page.
3410 * This value may be variable depending on boot parameters on IA64
3412 pageblock_order
= order
;
3414 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3417 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3418 * and pageblock_default_order() are unused as pageblock_order is set
3419 * at compile-time. See include/linux/pageblock-flags.h for the values of
3420 * pageblock_order based on the kernel config
3422 static inline int pageblock_default_order(unsigned int order
)
3426 #define set_pageblock_order(x) do {} while (0)
3428 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3431 * Set up the zone data structures:
3432 * - mark all pages reserved
3433 * - mark all memory queues empty
3434 * - clear the memory bitmaps
3436 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3437 unsigned long *zones_size
, unsigned long *zholes_size
)
3440 int nid
= pgdat
->node_id
;
3441 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3444 pgdat_resize_init(pgdat
);
3445 pgdat
->nr_zones
= 0;
3446 init_waitqueue_head(&pgdat
->kswapd_wait
);
3447 pgdat
->kswapd_max_order
= 0;
3448 pgdat_page_cgroup_init(pgdat
);
3450 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3451 struct zone
*zone
= pgdat
->node_zones
+ j
;
3452 unsigned long size
, realsize
, memmap_pages
;
3455 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3456 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3460 * Adjust realsize so that it accounts for how much memory
3461 * is used by this zone for memmap. This affects the watermark
3462 * and per-cpu initialisations
3465 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3466 if (realsize
>= memmap_pages
) {
3467 realsize
-= memmap_pages
;
3469 " %s zone: %lu pages used for memmap\n",
3470 zone_names
[j
], memmap_pages
);
3473 " %s zone: %lu pages exceeds realsize %lu\n",
3474 zone_names
[j
], memmap_pages
, realsize
);
3476 /* Account for reserved pages */
3477 if (j
== 0 && realsize
> dma_reserve
) {
3478 realsize
-= dma_reserve
;
3479 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3480 zone_names
[0], dma_reserve
);
3483 if (!is_highmem_idx(j
))
3484 nr_kernel_pages
+= realsize
;
3485 nr_all_pages
+= realsize
;
3487 zone
->spanned_pages
= size
;
3488 zone
->present_pages
= realsize
;
3491 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3493 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3495 zone
->name
= zone_names
[j
];
3496 spin_lock_init(&zone
->lock
);
3497 spin_lock_init(&zone
->lru_lock
);
3498 zone_seqlock_init(zone
);
3499 zone
->zone_pgdat
= pgdat
;
3501 zone
->prev_priority
= DEF_PRIORITY
;
3503 zone_pcp_init(zone
);
3505 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3506 zone
->lru
[l
].nr_scan
= 0;
3508 zone
->recent_rotated
[0] = 0;
3509 zone
->recent_rotated
[1] = 0;
3510 zone
->recent_scanned
[0] = 0;
3511 zone
->recent_scanned
[1] = 0;
3512 zap_zone_vm_stats(zone
);
3517 set_pageblock_order(pageblock_default_order());
3518 setup_usemap(pgdat
, zone
, size
);
3519 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3520 size
, MEMMAP_EARLY
);
3522 memmap_init(size
, nid
, j
, zone_start_pfn
);
3523 zone_start_pfn
+= size
;
3527 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3529 /* Skip empty nodes */
3530 if (!pgdat
->node_spanned_pages
)
3533 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3534 /* ia64 gets its own node_mem_map, before this, without bootmem */
3535 if (!pgdat
->node_mem_map
) {
3536 unsigned long size
, start
, end
;
3540 * The zone's endpoints aren't required to be MAX_ORDER
3541 * aligned but the node_mem_map endpoints must be in order
3542 * for the buddy allocator to function correctly.
3544 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3545 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3546 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3547 size
= (end
- start
) * sizeof(struct page
);
3548 map
= alloc_remap(pgdat
->node_id
, size
);
3550 map
= alloc_bootmem_node(pgdat
, size
);
3551 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3553 #ifndef CONFIG_NEED_MULTIPLE_NODES
3555 * With no DISCONTIG, the global mem_map is just set as node 0's
3557 if (pgdat
== NODE_DATA(0)) {
3558 mem_map
= NODE_DATA(0)->node_mem_map
;
3559 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3560 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3561 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3562 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3565 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3568 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3569 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3571 pg_data_t
*pgdat
= NODE_DATA(nid
);
3573 pgdat
->node_id
= nid
;
3574 pgdat
->node_start_pfn
= node_start_pfn
;
3575 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3577 alloc_node_mem_map(pgdat
);
3578 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3579 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3580 nid
, (unsigned long)pgdat
,
3581 (unsigned long)pgdat
->node_mem_map
);
3584 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3587 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3589 #if MAX_NUMNODES > 1
3591 * Figure out the number of possible node ids.
3593 static void __init
setup_nr_node_ids(void)
3596 unsigned int highest
= 0;
3598 for_each_node_mask(node
, node_possible_map
)
3600 nr_node_ids
= highest
+ 1;
3603 static inline void setup_nr_node_ids(void)
3609 * add_active_range - Register a range of PFNs backed by physical memory
3610 * @nid: The node ID the range resides on
3611 * @start_pfn: The start PFN of the available physical memory
3612 * @end_pfn: The end PFN of the available physical memory
3614 * These ranges are stored in an early_node_map[] and later used by
3615 * free_area_init_nodes() to calculate zone sizes and holes. If the
3616 * range spans a memory hole, it is up to the architecture to ensure
3617 * the memory is not freed by the bootmem allocator. If possible
3618 * the range being registered will be merged with existing ranges.
3620 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3621 unsigned long end_pfn
)
3625 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3626 "Entering add_active_range(%d, %#lx, %#lx) "
3627 "%d entries of %d used\n",
3628 nid
, start_pfn
, end_pfn
,
3629 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3631 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3633 /* Merge with existing active regions if possible */
3634 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3635 if (early_node_map
[i
].nid
!= nid
)
3638 /* Skip if an existing region covers this new one */
3639 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3640 end_pfn
<= early_node_map
[i
].end_pfn
)
3643 /* Merge forward if suitable */
3644 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3645 end_pfn
> early_node_map
[i
].end_pfn
) {
3646 early_node_map
[i
].end_pfn
= end_pfn
;
3650 /* Merge backward if suitable */
3651 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3652 end_pfn
>= early_node_map
[i
].start_pfn
) {
3653 early_node_map
[i
].start_pfn
= start_pfn
;
3658 /* Check that early_node_map is large enough */
3659 if (i
>= MAX_ACTIVE_REGIONS
) {
3660 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3661 MAX_ACTIVE_REGIONS
);
3665 early_node_map
[i
].nid
= nid
;
3666 early_node_map
[i
].start_pfn
= start_pfn
;
3667 early_node_map
[i
].end_pfn
= end_pfn
;
3668 nr_nodemap_entries
= i
+ 1;
3672 * remove_active_range - Shrink an existing registered range of PFNs
3673 * @nid: The node id the range is on that should be shrunk
3674 * @start_pfn: The new PFN of the range
3675 * @end_pfn: The new PFN of the range
3677 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3678 * The map is kept near the end physical page range that has already been
3679 * registered. This function allows an arch to shrink an existing registered
3682 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3683 unsigned long end_pfn
)
3688 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3689 nid
, start_pfn
, end_pfn
);
3691 /* Find the old active region end and shrink */
3692 for_each_active_range_index_in_nid(i
, nid
) {
3693 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3694 early_node_map
[i
].end_pfn
<= end_pfn
) {
3696 early_node_map
[i
].start_pfn
= 0;
3697 early_node_map
[i
].end_pfn
= 0;
3701 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3702 early_node_map
[i
].end_pfn
> start_pfn
) {
3703 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3704 early_node_map
[i
].end_pfn
= start_pfn
;
3705 if (temp_end_pfn
> end_pfn
)
3706 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3709 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3710 early_node_map
[i
].end_pfn
> end_pfn
&&
3711 early_node_map
[i
].start_pfn
< end_pfn
) {
3712 early_node_map
[i
].start_pfn
= end_pfn
;
3720 /* remove the blank ones */
3721 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3722 if (early_node_map
[i
].nid
!= nid
)
3724 if (early_node_map
[i
].end_pfn
)
3726 /* we found it, get rid of it */
3727 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3728 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3729 sizeof(early_node_map
[j
]));
3730 j
= nr_nodemap_entries
- 1;
3731 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3732 nr_nodemap_entries
--;
3737 * remove_all_active_ranges - Remove all currently registered regions
3739 * During discovery, it may be found that a table like SRAT is invalid
3740 * and an alternative discovery method must be used. This function removes
3741 * all currently registered regions.
3743 void __init
remove_all_active_ranges(void)
3745 memset(early_node_map
, 0, sizeof(early_node_map
));
3746 nr_nodemap_entries
= 0;
3747 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3748 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3749 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3750 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3753 /* Compare two active node_active_regions */
3754 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3756 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3757 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3759 /* Done this way to avoid overflows */
3760 if (arange
->start_pfn
> brange
->start_pfn
)
3762 if (arange
->start_pfn
< brange
->start_pfn
)
3768 /* sort the node_map by start_pfn */
3769 static void __init
sort_node_map(void)
3771 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3772 sizeof(struct node_active_region
),
3773 cmp_node_active_region
, NULL
);
3776 /* Find the lowest pfn for a node */
3777 static unsigned long __init
find_min_pfn_for_node(int nid
)
3780 unsigned long min_pfn
= ULONG_MAX
;
3782 /* Assuming a sorted map, the first range found has the starting pfn */
3783 for_each_active_range_index_in_nid(i
, nid
)
3784 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3786 if (min_pfn
== ULONG_MAX
) {
3788 "Could not find start_pfn for node %d\n", nid
);
3796 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3798 * It returns the minimum PFN based on information provided via
3799 * add_active_range().
3801 unsigned long __init
find_min_pfn_with_active_regions(void)
3803 return find_min_pfn_for_node(MAX_NUMNODES
);
3807 * early_calculate_totalpages()
3808 * Sum pages in active regions for movable zone.
3809 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3811 static unsigned long __init
early_calculate_totalpages(void)
3814 unsigned long totalpages
= 0;
3816 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3817 unsigned long pages
= early_node_map
[i
].end_pfn
-
3818 early_node_map
[i
].start_pfn
;
3819 totalpages
+= pages
;
3821 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3827 * Find the PFN the Movable zone begins in each node. Kernel memory
3828 * is spread evenly between nodes as long as the nodes have enough
3829 * memory. When they don't, some nodes will have more kernelcore than
3832 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3835 unsigned long usable_startpfn
;
3836 unsigned long kernelcore_node
, kernelcore_remaining
;
3837 unsigned long totalpages
= early_calculate_totalpages();
3838 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3841 * If movablecore was specified, calculate what size of
3842 * kernelcore that corresponds so that memory usable for
3843 * any allocation type is evenly spread. If both kernelcore
3844 * and movablecore are specified, then the value of kernelcore
3845 * will be used for required_kernelcore if it's greater than
3846 * what movablecore would have allowed.
3848 if (required_movablecore
) {
3849 unsigned long corepages
;
3852 * Round-up so that ZONE_MOVABLE is at least as large as what
3853 * was requested by the user
3855 required_movablecore
=
3856 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3857 corepages
= totalpages
- required_movablecore
;
3859 required_kernelcore
= max(required_kernelcore
, corepages
);
3862 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3863 if (!required_kernelcore
)
3866 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3867 find_usable_zone_for_movable();
3868 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3871 /* Spread kernelcore memory as evenly as possible throughout nodes */
3872 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3873 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3875 * Recalculate kernelcore_node if the division per node
3876 * now exceeds what is necessary to satisfy the requested
3877 * amount of memory for the kernel
3879 if (required_kernelcore
< kernelcore_node
)
3880 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3883 * As the map is walked, we track how much memory is usable
3884 * by the kernel using kernelcore_remaining. When it is
3885 * 0, the rest of the node is usable by ZONE_MOVABLE
3887 kernelcore_remaining
= kernelcore_node
;
3889 /* Go through each range of PFNs within this node */
3890 for_each_active_range_index_in_nid(i
, nid
) {
3891 unsigned long start_pfn
, end_pfn
;
3892 unsigned long size_pages
;
3894 start_pfn
= max(early_node_map
[i
].start_pfn
,
3895 zone_movable_pfn
[nid
]);
3896 end_pfn
= early_node_map
[i
].end_pfn
;
3897 if (start_pfn
>= end_pfn
)
3900 /* Account for what is only usable for kernelcore */
3901 if (start_pfn
< usable_startpfn
) {
3902 unsigned long kernel_pages
;
3903 kernel_pages
= min(end_pfn
, usable_startpfn
)
3906 kernelcore_remaining
-= min(kernel_pages
,
3907 kernelcore_remaining
);
3908 required_kernelcore
-= min(kernel_pages
,
3909 required_kernelcore
);
3911 /* Continue if range is now fully accounted */
3912 if (end_pfn
<= usable_startpfn
) {
3915 * Push zone_movable_pfn to the end so
3916 * that if we have to rebalance
3917 * kernelcore across nodes, we will
3918 * not double account here
3920 zone_movable_pfn
[nid
] = end_pfn
;
3923 start_pfn
= usable_startpfn
;
3927 * The usable PFN range for ZONE_MOVABLE is from
3928 * start_pfn->end_pfn. Calculate size_pages as the
3929 * number of pages used as kernelcore
3931 size_pages
= end_pfn
- start_pfn
;
3932 if (size_pages
> kernelcore_remaining
)
3933 size_pages
= kernelcore_remaining
;
3934 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3937 * Some kernelcore has been met, update counts and
3938 * break if the kernelcore for this node has been
3941 required_kernelcore
-= min(required_kernelcore
,
3943 kernelcore_remaining
-= size_pages
;
3944 if (!kernelcore_remaining
)
3950 * If there is still required_kernelcore, we do another pass with one
3951 * less node in the count. This will push zone_movable_pfn[nid] further
3952 * along on the nodes that still have memory until kernelcore is
3956 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3959 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3960 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3961 zone_movable_pfn
[nid
] =
3962 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3965 /* Any regular memory on that node ? */
3966 static void check_for_regular_memory(pg_data_t
*pgdat
)
3968 #ifdef CONFIG_HIGHMEM
3969 enum zone_type zone_type
;
3971 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3972 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3973 if (zone
->present_pages
)
3974 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3980 * free_area_init_nodes - Initialise all pg_data_t and zone data
3981 * @max_zone_pfn: an array of max PFNs for each zone
3983 * This will call free_area_init_node() for each active node in the system.
3984 * Using the page ranges provided by add_active_range(), the size of each
3985 * zone in each node and their holes is calculated. If the maximum PFN
3986 * between two adjacent zones match, it is assumed that the zone is empty.
3987 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3988 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3989 * starts where the previous one ended. For example, ZONE_DMA32 starts
3990 * at arch_max_dma_pfn.
3992 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3997 /* Sort early_node_map as initialisation assumes it is sorted */
4000 /* Record where the zone boundaries are */
4001 memset(arch_zone_lowest_possible_pfn
, 0,
4002 sizeof(arch_zone_lowest_possible_pfn
));
4003 memset(arch_zone_highest_possible_pfn
, 0,
4004 sizeof(arch_zone_highest_possible_pfn
));
4005 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4006 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4007 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4008 if (i
== ZONE_MOVABLE
)
4010 arch_zone_lowest_possible_pfn
[i
] =
4011 arch_zone_highest_possible_pfn
[i
-1];
4012 arch_zone_highest_possible_pfn
[i
] =
4013 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4015 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4016 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4018 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4019 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4020 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4022 /* Print out the zone ranges */
4023 printk("Zone PFN ranges:\n");
4024 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4025 if (i
== ZONE_MOVABLE
)
4027 printk(" %-8s %0#10lx -> %0#10lx\n",
4029 arch_zone_lowest_possible_pfn
[i
],
4030 arch_zone_highest_possible_pfn
[i
]);
4033 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4034 printk("Movable zone start PFN for each node\n");
4035 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4036 if (zone_movable_pfn
[i
])
4037 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4040 /* Print out the early_node_map[] */
4041 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4042 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4043 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4044 early_node_map
[i
].start_pfn
,
4045 early_node_map
[i
].end_pfn
);
4047 /* Initialise every node */
4048 mminit_verify_pageflags_layout();
4049 setup_nr_node_ids();
4050 for_each_online_node(nid
) {
4051 pg_data_t
*pgdat
= NODE_DATA(nid
);
4052 free_area_init_node(nid
, NULL
,
4053 find_min_pfn_for_node(nid
), NULL
);
4055 /* Any memory on that node */
4056 if (pgdat
->node_present_pages
)
4057 node_set_state(nid
, N_HIGH_MEMORY
);
4058 check_for_regular_memory(pgdat
);
4062 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4064 unsigned long long coremem
;
4068 coremem
= memparse(p
, &p
);
4069 *core
= coremem
>> PAGE_SHIFT
;
4071 /* Paranoid check that UL is enough for the coremem value */
4072 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4078 * kernelcore=size sets the amount of memory for use for allocations that
4079 * cannot be reclaimed or migrated.
4081 static int __init
cmdline_parse_kernelcore(char *p
)
4083 return cmdline_parse_core(p
, &required_kernelcore
);
4087 * movablecore=size sets the amount of memory for use for allocations that
4088 * can be reclaimed or migrated.
4090 static int __init
cmdline_parse_movablecore(char *p
)
4092 return cmdline_parse_core(p
, &required_movablecore
);
4095 early_param("kernelcore", cmdline_parse_kernelcore
);
4096 early_param("movablecore", cmdline_parse_movablecore
);
4098 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4101 * set_dma_reserve - set the specified number of pages reserved in the first zone
4102 * @new_dma_reserve: The number of pages to mark reserved
4104 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4105 * In the DMA zone, a significant percentage may be consumed by kernel image
4106 * and other unfreeable allocations which can skew the watermarks badly. This
4107 * function may optionally be used to account for unfreeable pages in the
4108 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4109 * smaller per-cpu batchsize.
4111 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4113 dma_reserve
= new_dma_reserve
;
4116 #ifndef CONFIG_NEED_MULTIPLE_NODES
4117 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4118 EXPORT_SYMBOL(contig_page_data
);
4121 void __init
free_area_init(unsigned long *zones_size
)
4123 free_area_init_node(0, zones_size
,
4124 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4127 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4128 unsigned long action
, void *hcpu
)
4130 int cpu
= (unsigned long)hcpu
;
4132 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4136 * Spill the event counters of the dead processor
4137 * into the current processors event counters.
4138 * This artificially elevates the count of the current
4141 vm_events_fold_cpu(cpu
);
4144 * Zero the differential counters of the dead processor
4145 * so that the vm statistics are consistent.
4147 * This is only okay since the processor is dead and cannot
4148 * race with what we are doing.
4150 refresh_cpu_vm_stats(cpu
);
4155 void __init
page_alloc_init(void)
4157 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4161 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4162 * or min_free_kbytes changes.
4164 static void calculate_totalreserve_pages(void)
4166 struct pglist_data
*pgdat
;
4167 unsigned long reserve_pages
= 0;
4168 enum zone_type i
, j
;
4170 for_each_online_pgdat(pgdat
) {
4171 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4172 struct zone
*zone
= pgdat
->node_zones
+ i
;
4173 unsigned long max
= 0;
4175 /* Find valid and maximum lowmem_reserve in the zone */
4176 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4177 if (zone
->lowmem_reserve
[j
] > max
)
4178 max
= zone
->lowmem_reserve
[j
];
4181 /* we treat pages_high as reserved pages. */
4182 max
+= zone
->pages_high
;
4184 if (max
> zone
->present_pages
)
4185 max
= zone
->present_pages
;
4186 reserve_pages
+= max
;
4189 totalreserve_pages
= reserve_pages
;
4193 * setup_per_zone_lowmem_reserve - called whenever
4194 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4195 * has a correct pages reserved value, so an adequate number of
4196 * pages are left in the zone after a successful __alloc_pages().
4198 static void setup_per_zone_lowmem_reserve(void)
4200 struct pglist_data
*pgdat
;
4201 enum zone_type j
, idx
;
4203 for_each_online_pgdat(pgdat
) {
4204 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4205 struct zone
*zone
= pgdat
->node_zones
+ j
;
4206 unsigned long present_pages
= zone
->present_pages
;
4208 zone
->lowmem_reserve
[j
] = 0;
4212 struct zone
*lower_zone
;
4216 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4217 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4219 lower_zone
= pgdat
->node_zones
+ idx
;
4220 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4221 sysctl_lowmem_reserve_ratio
[idx
];
4222 present_pages
+= lower_zone
->present_pages
;
4227 /* update totalreserve_pages */
4228 calculate_totalreserve_pages();
4232 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4234 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4235 * with respect to min_free_kbytes.
4237 void setup_per_zone_pages_min(void)
4239 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4240 unsigned long lowmem_pages
= 0;
4242 unsigned long flags
;
4244 /* Calculate total number of !ZONE_HIGHMEM pages */
4245 for_each_zone(zone
) {
4246 if (!is_highmem(zone
))
4247 lowmem_pages
+= zone
->present_pages
;
4250 for_each_zone(zone
) {
4253 spin_lock_irqsave(&zone
->lock
, flags
);
4254 tmp
= (u64
)pages_min
* zone
->present_pages
;
4255 do_div(tmp
, lowmem_pages
);
4256 if (is_highmem(zone
)) {
4258 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4259 * need highmem pages, so cap pages_min to a small
4262 * The (pages_high-pages_low) and (pages_low-pages_min)
4263 * deltas controls asynch page reclaim, and so should
4264 * not be capped for highmem.
4268 min_pages
= zone
->present_pages
/ 1024;
4269 if (min_pages
< SWAP_CLUSTER_MAX
)
4270 min_pages
= SWAP_CLUSTER_MAX
;
4271 if (min_pages
> 128)
4273 zone
->pages_min
= min_pages
;
4276 * If it's a lowmem zone, reserve a number of pages
4277 * proportionate to the zone's size.
4279 zone
->pages_min
= tmp
;
4282 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4283 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4284 setup_zone_migrate_reserve(zone
);
4285 spin_unlock_irqrestore(&zone
->lock
, flags
);
4288 /* update totalreserve_pages */
4289 calculate_totalreserve_pages();
4293 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4295 * The inactive anon list should be small enough that the VM never has to
4296 * do too much work, but large enough that each inactive page has a chance
4297 * to be referenced again before it is swapped out.
4299 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4300 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4301 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4302 * the anonymous pages are kept on the inactive list.
4305 * memory ratio inactive anon
4306 * -------------------------------------
4315 void setup_per_zone_inactive_ratio(void)
4319 for_each_zone(zone
) {
4320 unsigned int gb
, ratio
;
4322 /* Zone size in gigabytes */
4323 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4324 ratio
= int_sqrt(10 * gb
);
4328 zone
->inactive_ratio
= ratio
;
4333 * Initialise min_free_kbytes.
4335 * For small machines we want it small (128k min). For large machines
4336 * we want it large (64MB max). But it is not linear, because network
4337 * bandwidth does not increase linearly with machine size. We use
4339 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4340 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4356 static int __init
init_per_zone_pages_min(void)
4358 unsigned long lowmem_kbytes
;
4360 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4362 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4363 if (min_free_kbytes
< 128)
4364 min_free_kbytes
= 128;
4365 if (min_free_kbytes
> 65536)
4366 min_free_kbytes
= 65536;
4367 setup_per_zone_pages_min();
4368 setup_per_zone_lowmem_reserve();
4369 setup_per_zone_inactive_ratio();
4372 module_init(init_per_zone_pages_min
)
4375 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4376 * that we can call two helper functions whenever min_free_kbytes
4379 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4380 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4382 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4384 setup_per_zone_pages_min();
4389 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4390 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4395 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4400 zone
->min_unmapped_pages
= (zone
->present_pages
*
4401 sysctl_min_unmapped_ratio
) / 100;
4405 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4406 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4411 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4416 zone
->min_slab_pages
= (zone
->present_pages
*
4417 sysctl_min_slab_ratio
) / 100;
4423 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4424 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4425 * whenever sysctl_lowmem_reserve_ratio changes.
4427 * The reserve ratio obviously has absolutely no relation with the
4428 * pages_min watermarks. The lowmem reserve ratio can only make sense
4429 * if in function of the boot time zone sizes.
4431 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4432 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4434 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4435 setup_per_zone_lowmem_reserve();
4440 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4441 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4442 * can have before it gets flushed back to buddy allocator.
4445 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4446 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4452 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4453 if (!write
|| (ret
== -EINVAL
))
4455 for_each_zone(zone
) {
4456 for_each_online_cpu(cpu
) {
4458 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4459 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4465 int hashdist
= HASHDIST_DEFAULT
;
4468 static int __init
set_hashdist(char *str
)
4472 hashdist
= simple_strtoul(str
, &str
, 0);
4475 __setup("hashdist=", set_hashdist
);
4479 * allocate a large system hash table from bootmem
4480 * - it is assumed that the hash table must contain an exact power-of-2
4481 * quantity of entries
4482 * - limit is the number of hash buckets, not the total allocation size
4484 void *__init
alloc_large_system_hash(const char *tablename
,
4485 unsigned long bucketsize
,
4486 unsigned long numentries
,
4489 unsigned int *_hash_shift
,
4490 unsigned int *_hash_mask
,
4491 unsigned long limit
)
4493 unsigned long long max
= limit
;
4494 unsigned long log2qty
, size
;
4497 /* allow the kernel cmdline to have a say */
4499 /* round applicable memory size up to nearest megabyte */
4500 numentries
= nr_kernel_pages
;
4501 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4502 numentries
>>= 20 - PAGE_SHIFT
;
4503 numentries
<<= 20 - PAGE_SHIFT
;
4505 /* limit to 1 bucket per 2^scale bytes of low memory */
4506 if (scale
> PAGE_SHIFT
)
4507 numentries
>>= (scale
- PAGE_SHIFT
);
4509 numentries
<<= (PAGE_SHIFT
- scale
);
4511 /* Make sure we've got at least a 0-order allocation.. */
4512 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4513 numentries
= PAGE_SIZE
/ bucketsize
;
4515 numentries
= roundup_pow_of_two(numentries
);
4517 /* limit allocation size to 1/16 total memory by default */
4519 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4520 do_div(max
, bucketsize
);
4523 if (numentries
> max
)
4526 log2qty
= ilog2(numentries
);
4529 size
= bucketsize
<< log2qty
;
4530 if (flags
& HASH_EARLY
)
4531 table
= alloc_bootmem_nopanic(size
);
4533 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4535 unsigned long order
= get_order(size
);
4536 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4538 * If bucketsize is not a power-of-two, we may free
4539 * some pages at the end of hash table.
4542 unsigned long alloc_end
= (unsigned long)table
+
4543 (PAGE_SIZE
<< order
);
4544 unsigned long used
= (unsigned long)table
+
4546 split_page(virt_to_page(table
), order
);
4547 while (used
< alloc_end
) {
4553 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4556 panic("Failed to allocate %s hash table\n", tablename
);
4558 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4561 ilog2(size
) - PAGE_SHIFT
,
4565 *_hash_shift
= log2qty
;
4567 *_hash_mask
= (1 << log2qty
) - 1;
4572 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4573 struct page
*pfn_to_page(unsigned long pfn
)
4575 return __pfn_to_page(pfn
);
4577 unsigned long page_to_pfn(struct page
*page
)
4579 return __page_to_pfn(page
);
4581 EXPORT_SYMBOL(pfn_to_page
);
4582 EXPORT_SYMBOL(page_to_pfn
);
4583 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4585 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4586 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4589 #ifdef CONFIG_SPARSEMEM
4590 return __pfn_to_section(pfn
)->pageblock_flags
;
4592 return zone
->pageblock_flags
;
4593 #endif /* CONFIG_SPARSEMEM */
4596 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4598 #ifdef CONFIG_SPARSEMEM
4599 pfn
&= (PAGES_PER_SECTION
-1);
4600 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4602 pfn
= pfn
- zone
->zone_start_pfn
;
4603 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4604 #endif /* CONFIG_SPARSEMEM */
4608 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4609 * @page: The page within the block of interest
4610 * @start_bitidx: The first bit of interest to retrieve
4611 * @end_bitidx: The last bit of interest
4612 * returns pageblock_bits flags
4614 unsigned long get_pageblock_flags_group(struct page
*page
,
4615 int start_bitidx
, int end_bitidx
)
4618 unsigned long *bitmap
;
4619 unsigned long pfn
, bitidx
;
4620 unsigned long flags
= 0;
4621 unsigned long value
= 1;
4623 zone
= page_zone(page
);
4624 pfn
= page_to_pfn(page
);
4625 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4626 bitidx
= pfn_to_bitidx(zone
, pfn
);
4628 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4629 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4636 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4637 * @page: The page within the block of interest
4638 * @start_bitidx: The first bit of interest
4639 * @end_bitidx: The last bit of interest
4640 * @flags: The flags to set
4642 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4643 int start_bitidx
, int end_bitidx
)
4646 unsigned long *bitmap
;
4647 unsigned long pfn
, bitidx
;
4648 unsigned long value
= 1;
4650 zone
= page_zone(page
);
4651 pfn
= page_to_pfn(page
);
4652 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4653 bitidx
= pfn_to_bitidx(zone
, pfn
);
4654 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4655 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4657 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4659 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4661 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4665 * This is designed as sub function...plz see page_isolation.c also.
4666 * set/clear page block's type to be ISOLATE.
4667 * page allocater never alloc memory from ISOLATE block.
4670 int set_migratetype_isolate(struct page
*page
)
4673 unsigned long flags
;
4676 zone
= page_zone(page
);
4677 spin_lock_irqsave(&zone
->lock
, flags
);
4679 * In future, more migrate types will be able to be isolation target.
4681 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4683 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4684 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4687 spin_unlock_irqrestore(&zone
->lock
, flags
);
4693 void unset_migratetype_isolate(struct page
*page
)
4696 unsigned long flags
;
4697 zone
= page_zone(page
);
4698 spin_lock_irqsave(&zone
->lock
, flags
);
4699 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4701 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4702 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4704 spin_unlock_irqrestore(&zone
->lock
, flags
);
4707 #ifdef CONFIG_MEMORY_HOTREMOVE
4709 * All pages in the range must be isolated before calling this.
4712 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4718 unsigned long flags
;
4719 /* find the first valid pfn */
4720 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4725 zone
= page_zone(pfn_to_page(pfn
));
4726 spin_lock_irqsave(&zone
->lock
, flags
);
4728 while (pfn
< end_pfn
) {
4729 if (!pfn_valid(pfn
)) {
4733 page
= pfn_to_page(pfn
);
4734 BUG_ON(page_count(page
));
4735 BUG_ON(!PageBuddy(page
));
4736 order
= page_order(page
);
4737 #ifdef CONFIG_DEBUG_VM
4738 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4739 pfn
, 1 << order
, end_pfn
);
4741 list_del(&page
->lru
);
4742 rmv_page_order(page
);
4743 zone
->free_area
[order
].nr_free
--;
4744 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4746 for (i
= 0; i
< (1 << order
); i
++)
4747 SetPageReserved((page
+i
));
4748 pfn
+= (1 << order
);
4750 spin_unlock_irqrestore(&zone
->lock
, flags
);