2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/oom.h>
32 #include <linux/notifier.h>
33 #include <linux/topology.h>
34 #include <linux/sysctl.h>
35 #include <linux/cpu.h>
36 #include <linux/cpuset.h>
37 #include <linux/memory_hotplug.h>
38 #include <linux/nodemask.h>
39 #include <linux/vmalloc.h>
40 #include <linux/mempolicy.h>
41 #include <linux/stop_machine.h>
42 #include <linux/sort.h>
43 #include <linux/pfn.h>
44 #include <linux/backing-dev.h>
45 #include <linux/fault-inject.h>
46 #include <linux/page-isolation.h>
47 #include <linux/memcontrol.h>
48 #include <linux/debugobjects.h>
50 #include <asm/tlbflush.h>
51 #include <asm/div64.h>
55 * Array of node states.
57 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
58 [N_POSSIBLE
] = NODE_MASK_ALL
,
59 [N_ONLINE
] = { { [0] = 1UL } },
61 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
63 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
65 [N_CPU
] = { { [0] = 1UL } },
68 EXPORT_SYMBOL(node_states
);
70 unsigned long totalram_pages __read_mostly
;
71 unsigned long totalreserve_pages __read_mostly
;
73 int percpu_pagelist_fraction
;
75 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
76 int pageblock_order __read_mostly
;
79 static void __free_pages_ok(struct page
*page
, unsigned int order
);
82 * results with 256, 32 in the lowmem_reserve sysctl:
83 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
84 * 1G machine -> (16M dma, 784M normal, 224M high)
85 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
86 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
87 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
89 * TBD: should special case ZONE_DMA32 machines here - in those we normally
90 * don't need any ZONE_NORMAL reservation
92 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
93 #ifdef CONFIG_ZONE_DMA
96 #ifdef CONFIG_ZONE_DMA32
105 EXPORT_SYMBOL(totalram_pages
);
107 static char * const zone_names
[MAX_NR_ZONES
] = {
108 #ifdef CONFIG_ZONE_DMA
111 #ifdef CONFIG_ZONE_DMA32
115 #ifdef CONFIG_HIGHMEM
121 int min_free_kbytes
= 1024;
123 unsigned long __meminitdata nr_kernel_pages
;
124 unsigned long __meminitdata nr_all_pages
;
125 static unsigned long __meminitdata dma_reserve
;
127 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
129 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
130 * ranges of memory (RAM) that may be registered with add_active_range().
131 * Ranges passed to add_active_range() will be merged if possible
132 * so the number of times add_active_range() can be called is
133 * related to the number of nodes and the number of holes
135 #ifdef CONFIG_MAX_ACTIVE_REGIONS
136 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
137 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
139 #if MAX_NUMNODES >= 32
140 /* If there can be many nodes, allow up to 50 holes per node */
141 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
143 /* By default, allow up to 256 distinct regions */
144 #define MAX_ACTIVE_REGIONS 256
148 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
149 static int __meminitdata nr_nodemap_entries
;
150 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
151 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
152 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
153 static unsigned long __meminitdata node_boundary_start_pfn
[MAX_NUMNODES
];
154 static unsigned long __meminitdata node_boundary_end_pfn
[MAX_NUMNODES
];
155 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
156 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 void *pc
= page_get_page_cgroup(page
);
228 printk(KERN_EMERG
"Bad page state in process '%s'\n" KERN_EMERG
229 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
230 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
231 (unsigned long)page
->flags
, page
->mapping
,
232 page_mapcount(page
), page_count(page
));
234 printk(KERN_EMERG
"cgroup:%p\n", pc
);
235 page_reset_bad_cgroup(page
);
237 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
238 KERN_EMERG
"Backtrace:\n");
240 page
->flags
&= ~PAGE_FLAGS_CLEAR_WHEN_BAD
;
241 set_page_count(page
, 0);
242 reset_page_mapcount(page
);
243 page
->mapping
= NULL
;
244 add_taint(TAINT_BAD_PAGE
);
248 * Higher-order pages are called "compound pages". They are structured thusly:
250 * The first PAGE_SIZE page is called the "head page".
252 * The remaining PAGE_SIZE pages are called "tail pages".
254 * All pages have PG_compound set. All pages have their ->private pointing at
255 * the head page (even the head page has this).
257 * The first tail page's ->lru.next holds the address of the compound page's
258 * put_page() function. Its ->lru.prev holds the order of allocation.
259 * This usage means that zero-order pages may not be compound.
262 static void free_compound_page(struct page
*page
)
264 __free_pages_ok(page
, compound_order(page
));
267 void prep_compound_page(struct page
*page
, unsigned long order
)
270 int nr_pages
= 1 << order
;
271 struct page
*p
= page
+ 1;
273 set_compound_page_dtor(page
, free_compound_page
);
274 set_compound_order(page
, order
);
276 for (i
= 1; i
< nr_pages
; i
++, p
++) {
277 if (unlikely((i
& (MAX_ORDER_NR_PAGES
- 1)) == 0))
278 p
= pfn_to_page(page_to_pfn(page
) + i
);
280 p
->first_page
= page
;
284 static void destroy_compound_page(struct page
*page
, unsigned long order
)
287 int nr_pages
= 1 << order
;
288 struct page
*p
= page
+ 1;
290 if (unlikely(compound_order(page
) != order
))
293 if (unlikely(!PageHead(page
)))
295 __ClearPageHead(page
);
296 for (i
= 1; i
< nr_pages
; i
++, p
++) {
297 if (unlikely((i
& (MAX_ORDER_NR_PAGES
- 1)) == 0))
298 p
= pfn_to_page(page_to_pfn(page
) + i
);
300 if (unlikely(!PageTail(p
) |
301 (p
->first_page
!= page
)))
307 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
312 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
313 * and __GFP_HIGHMEM from hard or soft interrupt context.
315 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
316 for (i
= 0; i
< (1 << order
); i
++)
317 clear_highpage(page
+ i
);
320 static inline void set_page_order(struct page
*page
, int order
)
322 set_page_private(page
, order
);
323 __SetPageBuddy(page
);
326 static inline void rmv_page_order(struct page
*page
)
328 __ClearPageBuddy(page
);
329 set_page_private(page
, 0);
333 * Locate the struct page for both the matching buddy in our
334 * pair (buddy1) and the combined O(n+1) page they form (page).
336 * 1) Any buddy B1 will have an order O twin B2 which satisfies
337 * the following equation:
339 * For example, if the starting buddy (buddy2) is #8 its order
341 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
343 * 2) Any buddy B will have an order O+1 parent P which
344 * satisfies the following equation:
347 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
349 static inline struct page
*
350 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
352 unsigned long buddy_idx
= page_idx
^ (1 << order
);
354 return page
+ (buddy_idx
- page_idx
);
357 static inline unsigned long
358 __find_combined_index(unsigned long page_idx
, unsigned int order
)
360 return (page_idx
& ~(1 << order
));
364 * This function checks whether a page is free && is the buddy
365 * we can do coalesce a page and its buddy if
366 * (a) the buddy is not in a hole &&
367 * (b) the buddy is in the buddy system &&
368 * (c) a page and its buddy have the same order &&
369 * (d) a page and its buddy are in the same zone.
371 * For recording whether a page is in the buddy system, we use PG_buddy.
372 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
374 * For recording page's order, we use page_private(page).
376 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
379 if (!pfn_valid_within(page_to_pfn(buddy
)))
382 if (page_zone_id(page
) != page_zone_id(buddy
))
385 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
386 BUG_ON(page_count(buddy
) != 0);
393 * Freeing function for a buddy system allocator.
395 * The concept of a buddy system is to maintain direct-mapped table
396 * (containing bit values) for memory blocks of various "orders".
397 * The bottom level table contains the map for the smallest allocatable
398 * units of memory (here, pages), and each level above it describes
399 * pairs of units from the levels below, hence, "buddies".
400 * At a high level, all that happens here is marking the table entry
401 * at the bottom level available, and propagating the changes upward
402 * as necessary, plus some accounting needed to play nicely with other
403 * parts of the VM system.
404 * At each level, we keep a list of pages, which are heads of continuous
405 * free pages of length of (1 << order) and marked with PG_buddy. Page's
406 * order is recorded in page_private(page) field.
407 * So when we are allocating or freeing one, we can derive the state of the
408 * other. That is, if we allocate a small block, and both were
409 * free, the remainder of the region must be split into blocks.
410 * If a block is freed, and its buddy is also free, then this
411 * triggers coalescing into a block of larger size.
416 static inline void __free_one_page(struct page
*page
,
417 struct zone
*zone
, unsigned int order
)
419 unsigned long page_idx
;
420 int order_size
= 1 << order
;
421 int migratetype
= get_pageblock_migratetype(page
);
423 if (unlikely(PageCompound(page
)))
424 destroy_compound_page(page
, order
);
426 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
428 VM_BUG_ON(page_idx
& (order_size
- 1));
429 VM_BUG_ON(bad_range(zone
, page
));
431 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
432 while (order
< MAX_ORDER
-1) {
433 unsigned long combined_idx
;
436 buddy
= __page_find_buddy(page
, page_idx
, order
);
437 if (!page_is_buddy(page
, buddy
, order
))
440 /* Our buddy is free, merge with it and move up one order. */
441 list_del(&buddy
->lru
);
442 zone
->free_area
[order
].nr_free
--;
443 rmv_page_order(buddy
);
444 combined_idx
= __find_combined_index(page_idx
, order
);
445 page
= page
+ (combined_idx
- page_idx
);
446 page_idx
= combined_idx
;
449 set_page_order(page
, order
);
451 &zone
->free_area
[order
].free_list
[migratetype
]);
452 zone
->free_area
[order
].nr_free
++;
455 static inline int free_pages_check(struct page
*page
)
457 free_page_mlock(page
);
458 if (unlikely(page_mapcount(page
) |
459 (page
->mapping
!= NULL
) |
460 (page_get_page_cgroup(page
) != NULL
) |
461 (page_count(page
) != 0) |
462 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)))
465 __ClearPageDirty(page
);
466 if (PageSwapBacked(page
))
467 __ClearPageSwapBacked(page
);
469 * For now, we report if PG_reserved was found set, but do not
470 * clear it, and do not free the page. But we shall soon need
471 * to do more, for when the ZERO_PAGE count wraps negative.
473 return PageReserved(page
);
477 * Frees a list of pages.
478 * Assumes all pages on list are in same zone, and of same order.
479 * count is the number of pages to free.
481 * If the zone was previously in an "all pages pinned" state then look to
482 * see if this freeing clears that state.
484 * And clear the zone's pages_scanned counter, to hold off the "all pages are
485 * pinned" detection logic.
487 static void free_pages_bulk(struct zone
*zone
, int count
,
488 struct list_head
*list
, int order
)
490 spin_lock(&zone
->lock
);
491 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
492 zone
->pages_scanned
= 0;
496 VM_BUG_ON(list_empty(list
));
497 page
= list_entry(list
->prev
, struct page
, lru
);
498 /* have to delete it as __free_one_page list manipulates */
499 list_del(&page
->lru
);
500 __free_one_page(page
, zone
, order
);
502 spin_unlock(&zone
->lock
);
505 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
507 spin_lock(&zone
->lock
);
508 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
509 zone
->pages_scanned
= 0;
510 __free_one_page(page
, zone
, order
);
511 spin_unlock(&zone
->lock
);
514 static void __free_pages_ok(struct page
*page
, unsigned int order
)
520 for (i
= 0 ; i
< (1 << order
) ; ++i
)
521 reserved
+= free_pages_check(page
+ i
);
525 if (!PageHighMem(page
)) {
526 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
527 debug_check_no_obj_freed(page_address(page
),
530 arch_free_page(page
, order
);
531 kernel_map_pages(page
, 1 << order
, 0);
533 local_irq_save(flags
);
534 __count_vm_events(PGFREE
, 1 << order
);
535 free_one_page(page_zone(page
), page
, order
);
536 local_irq_restore(flags
);
540 * permit the bootmem allocator to evade page validation on high-order frees
542 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
545 __ClearPageReserved(page
);
546 set_page_count(page
, 0);
547 set_page_refcounted(page
);
553 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
554 struct page
*p
= &page
[loop
];
556 if (loop
+ 1 < BITS_PER_LONG
)
558 __ClearPageReserved(p
);
559 set_page_count(p
, 0);
562 set_page_refcounted(page
);
563 __free_pages(page
, order
);
569 * The order of subdivision here is critical for the IO subsystem.
570 * Please do not alter this order without good reasons and regression
571 * testing. Specifically, as large blocks of memory are subdivided,
572 * the order in which smaller blocks are delivered depends on the order
573 * they're subdivided in this function. This is the primary factor
574 * influencing the order in which pages are delivered to the IO
575 * subsystem according to empirical testing, and this is also justified
576 * by considering the behavior of a buddy system containing a single
577 * large block of memory acted on by a series of small allocations.
578 * This behavior is a critical factor in sglist merging's success.
582 static inline void expand(struct zone
*zone
, struct page
*page
,
583 int low
, int high
, struct free_area
*area
,
586 unsigned long size
= 1 << high
;
592 VM_BUG_ON(bad_range(zone
, &page
[size
]));
593 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
595 set_page_order(&page
[size
], high
);
600 * This page is about to be returned from the page allocator
602 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
604 if (unlikely(page_mapcount(page
) |
605 (page
->mapping
!= NULL
) |
606 (page_get_page_cgroup(page
) != NULL
) |
607 (page_count(page
) != 0) |
608 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)))
612 * For now, we report if PG_reserved was found set, but do not
613 * clear it, and do not allocate the page: as a safety net.
615 if (PageReserved(page
))
618 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_reclaim
|
619 1 << PG_referenced
| 1 << PG_arch_1
|
620 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
621 #ifdef CONFIG_UNEVICTABLE_LRU
625 set_page_private(page
, 0);
626 set_page_refcounted(page
);
628 arch_alloc_page(page
, order
);
629 kernel_map_pages(page
, 1 << order
, 1);
631 if (gfp_flags
& __GFP_ZERO
)
632 prep_zero_page(page
, order
, gfp_flags
);
634 if (order
&& (gfp_flags
& __GFP_COMP
))
635 prep_compound_page(page
, order
);
641 * Go through the free lists for the given migratetype and remove
642 * the smallest available page from the freelists
644 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
647 unsigned int current_order
;
648 struct free_area
* area
;
651 /* Find a page of the appropriate size in the preferred list */
652 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
653 area
= &(zone
->free_area
[current_order
]);
654 if (list_empty(&area
->free_list
[migratetype
]))
657 page
= list_entry(area
->free_list
[migratetype
].next
,
659 list_del(&page
->lru
);
660 rmv_page_order(page
);
662 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
663 expand(zone
, page
, order
, current_order
, area
, migratetype
);
672 * This array describes the order lists are fallen back to when
673 * the free lists for the desirable migrate type are depleted
675 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
676 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
677 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
678 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
679 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
683 * Move the free pages in a range to the free lists of the requested type.
684 * Note that start_page and end_pages are not aligned on a pageblock
685 * boundary. If alignment is required, use move_freepages_block()
687 static int move_freepages(struct zone
*zone
,
688 struct page
*start_page
, struct page
*end_page
,
695 #ifndef CONFIG_HOLES_IN_ZONE
697 * page_zone is not safe to call in this context when
698 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
699 * anyway as we check zone boundaries in move_freepages_block().
700 * Remove at a later date when no bug reports exist related to
701 * grouping pages by mobility
703 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
706 for (page
= start_page
; page
<= end_page
;) {
707 /* Make sure we are not inadvertently changing nodes */
708 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
710 if (!pfn_valid_within(page_to_pfn(page
))) {
715 if (!PageBuddy(page
)) {
720 order
= page_order(page
);
721 list_del(&page
->lru
);
723 &zone
->free_area
[order
].free_list
[migratetype
]);
725 pages_moved
+= 1 << order
;
731 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
734 unsigned long start_pfn
, end_pfn
;
735 struct page
*start_page
, *end_page
;
737 start_pfn
= page_to_pfn(page
);
738 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
739 start_page
= pfn_to_page(start_pfn
);
740 end_page
= start_page
+ pageblock_nr_pages
- 1;
741 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
743 /* Do not cross zone boundaries */
744 if (start_pfn
< zone
->zone_start_pfn
)
746 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
749 return move_freepages(zone
, start_page
, end_page
, migratetype
);
752 /* Remove an element from the buddy allocator from the fallback list */
753 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
754 int start_migratetype
)
756 struct free_area
* area
;
761 /* Find the largest possible block of pages in the other list */
762 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
764 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
765 migratetype
= fallbacks
[start_migratetype
][i
];
767 /* MIGRATE_RESERVE handled later if necessary */
768 if (migratetype
== MIGRATE_RESERVE
)
771 area
= &(zone
->free_area
[current_order
]);
772 if (list_empty(&area
->free_list
[migratetype
]))
775 page
= list_entry(area
->free_list
[migratetype
].next
,
780 * If breaking a large block of pages, move all free
781 * pages to the preferred allocation list. If falling
782 * back for a reclaimable kernel allocation, be more
783 * agressive about taking ownership of free pages
785 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
786 start_migratetype
== MIGRATE_RECLAIMABLE
) {
788 pages
= move_freepages_block(zone
, page
,
791 /* Claim the whole block if over half of it is free */
792 if (pages
>= (1 << (pageblock_order
-1)))
793 set_pageblock_migratetype(page
,
796 migratetype
= start_migratetype
;
799 /* Remove the page from the freelists */
800 list_del(&page
->lru
);
801 rmv_page_order(page
);
802 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
805 if (current_order
== pageblock_order
)
806 set_pageblock_migratetype(page
,
809 expand(zone
, page
, order
, current_order
, area
, migratetype
);
814 /* Use MIGRATE_RESERVE rather than fail an allocation */
815 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
819 * Do the hard work of removing an element from the buddy allocator.
820 * Call me with the zone->lock already held.
822 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
827 page
= __rmqueue_smallest(zone
, order
, migratetype
);
830 page
= __rmqueue_fallback(zone
, order
, migratetype
);
836 * Obtain a specified number of elements from the buddy allocator, all under
837 * a single hold of the lock, for efficiency. Add them to the supplied list.
838 * Returns the number of new pages which were placed at *list.
840 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
841 unsigned long count
, struct list_head
*list
,
846 spin_lock(&zone
->lock
);
847 for (i
= 0; i
< count
; ++i
) {
848 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
849 if (unlikely(page
== NULL
))
853 * Split buddy pages returned by expand() are received here
854 * in physical page order. The page is added to the callers and
855 * list and the list head then moves forward. From the callers
856 * perspective, the linked list is ordered by page number in
857 * some conditions. This is useful for IO devices that can
858 * merge IO requests if the physical pages are ordered
861 list_add(&page
->lru
, list
);
862 set_page_private(page
, migratetype
);
865 spin_unlock(&zone
->lock
);
871 * Called from the vmstat counter updater to drain pagesets of this
872 * currently executing processor on remote nodes after they have
875 * Note that this function must be called with the thread pinned to
876 * a single processor.
878 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
883 local_irq_save(flags
);
884 if (pcp
->count
>= pcp
->batch
)
885 to_drain
= pcp
->batch
;
887 to_drain
= pcp
->count
;
888 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
889 pcp
->count
-= to_drain
;
890 local_irq_restore(flags
);
895 * Drain pages of the indicated processor.
897 * The processor must either be the current processor and the
898 * thread pinned to the current processor or a processor that
901 static void drain_pages(unsigned int cpu
)
906 for_each_zone(zone
) {
907 struct per_cpu_pageset
*pset
;
908 struct per_cpu_pages
*pcp
;
910 if (!populated_zone(zone
))
913 pset
= zone_pcp(zone
, cpu
);
916 local_irq_save(flags
);
917 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
919 local_irq_restore(flags
);
924 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
926 void drain_local_pages(void *arg
)
928 drain_pages(smp_processor_id());
932 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
934 void drain_all_pages(void)
936 on_each_cpu(drain_local_pages
, NULL
, 1);
939 #ifdef CONFIG_HIBERNATION
941 void mark_free_pages(struct zone
*zone
)
943 unsigned long pfn
, max_zone_pfn
;
946 struct list_head
*curr
;
948 if (!zone
->spanned_pages
)
951 spin_lock_irqsave(&zone
->lock
, flags
);
953 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
954 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
955 if (pfn_valid(pfn
)) {
956 struct page
*page
= pfn_to_page(pfn
);
958 if (!swsusp_page_is_forbidden(page
))
959 swsusp_unset_page_free(page
);
962 for_each_migratetype_order(order
, t
) {
963 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
966 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
967 for (i
= 0; i
< (1UL << order
); i
++)
968 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
971 spin_unlock_irqrestore(&zone
->lock
, flags
);
973 #endif /* CONFIG_PM */
976 * Free a 0-order page
978 static void free_hot_cold_page(struct page
*page
, int cold
)
980 struct zone
*zone
= page_zone(page
);
981 struct per_cpu_pages
*pcp
;
985 page
->mapping
= NULL
;
986 if (free_pages_check(page
))
989 if (!PageHighMem(page
)) {
990 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
991 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
993 arch_free_page(page
, 0);
994 kernel_map_pages(page
, 1, 0);
996 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
997 local_irq_save(flags
);
998 __count_vm_event(PGFREE
);
1000 list_add_tail(&page
->lru
, &pcp
->list
);
1002 list_add(&page
->lru
, &pcp
->list
);
1003 set_page_private(page
, get_pageblock_migratetype(page
));
1005 if (pcp
->count
>= pcp
->high
) {
1006 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1007 pcp
->count
-= pcp
->batch
;
1009 local_irq_restore(flags
);
1013 void free_hot_page(struct page
*page
)
1015 free_hot_cold_page(page
, 0);
1018 void free_cold_page(struct page
*page
)
1020 free_hot_cold_page(page
, 1);
1024 * split_page takes a non-compound higher-order page, and splits it into
1025 * n (1<<order) sub-pages: page[0..n]
1026 * Each sub-page must be freed individually.
1028 * Note: this is probably too low level an operation for use in drivers.
1029 * Please consult with lkml before using this in your driver.
1031 void split_page(struct page
*page
, unsigned int order
)
1035 VM_BUG_ON(PageCompound(page
));
1036 VM_BUG_ON(!page_count(page
));
1037 for (i
= 1; i
< (1 << order
); i
++)
1038 set_page_refcounted(page
+ i
);
1042 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1043 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1046 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1047 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1049 unsigned long flags
;
1051 int cold
= !!(gfp_flags
& __GFP_COLD
);
1053 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1057 if (likely(order
== 0)) {
1058 struct per_cpu_pages
*pcp
;
1060 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1061 local_irq_save(flags
);
1063 pcp
->count
= rmqueue_bulk(zone
, 0,
1064 pcp
->batch
, &pcp
->list
, migratetype
);
1065 if (unlikely(!pcp
->count
))
1069 /* Find a page of the appropriate migrate type */
1071 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1072 if (page_private(page
) == migratetype
)
1075 list_for_each_entry(page
, &pcp
->list
, lru
)
1076 if (page_private(page
) == migratetype
)
1080 /* Allocate more to the pcp list if necessary */
1081 if (unlikely(&page
->lru
== &pcp
->list
)) {
1082 pcp
->count
+= rmqueue_bulk(zone
, 0,
1083 pcp
->batch
, &pcp
->list
, migratetype
);
1084 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1087 list_del(&page
->lru
);
1090 spin_lock_irqsave(&zone
->lock
, flags
);
1091 page
= __rmqueue(zone
, order
, migratetype
);
1092 spin_unlock(&zone
->lock
);
1097 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1098 zone_statistics(preferred_zone
, zone
);
1099 local_irq_restore(flags
);
1102 VM_BUG_ON(bad_range(zone
, page
));
1103 if (prep_new_page(page
, order
, gfp_flags
))
1108 local_irq_restore(flags
);
1113 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1114 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1115 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1116 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1117 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1118 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1119 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1121 #ifdef CONFIG_FAIL_PAGE_ALLOC
1123 static struct fail_page_alloc_attr
{
1124 struct fault_attr attr
;
1126 u32 ignore_gfp_highmem
;
1127 u32 ignore_gfp_wait
;
1130 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1132 struct dentry
*ignore_gfp_highmem_file
;
1133 struct dentry
*ignore_gfp_wait_file
;
1134 struct dentry
*min_order_file
;
1136 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1138 } fail_page_alloc
= {
1139 .attr
= FAULT_ATTR_INITIALIZER
,
1140 .ignore_gfp_wait
= 1,
1141 .ignore_gfp_highmem
= 1,
1145 static int __init
setup_fail_page_alloc(char *str
)
1147 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1149 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1151 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1153 if (order
< fail_page_alloc
.min_order
)
1155 if (gfp_mask
& __GFP_NOFAIL
)
1157 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1159 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1162 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1165 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1167 static int __init
fail_page_alloc_debugfs(void)
1169 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1173 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1177 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1179 fail_page_alloc
.ignore_gfp_wait_file
=
1180 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1181 &fail_page_alloc
.ignore_gfp_wait
);
1183 fail_page_alloc
.ignore_gfp_highmem_file
=
1184 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1185 &fail_page_alloc
.ignore_gfp_highmem
);
1186 fail_page_alloc
.min_order_file
=
1187 debugfs_create_u32("min-order", mode
, dir
,
1188 &fail_page_alloc
.min_order
);
1190 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1191 !fail_page_alloc
.ignore_gfp_highmem_file
||
1192 !fail_page_alloc
.min_order_file
) {
1194 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1195 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1196 debugfs_remove(fail_page_alloc
.min_order_file
);
1197 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1203 late_initcall(fail_page_alloc_debugfs
);
1205 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1207 #else /* CONFIG_FAIL_PAGE_ALLOC */
1209 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1214 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1217 * Return 1 if free pages are above 'mark'. This takes into account the order
1218 * of the allocation.
1220 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1221 int classzone_idx
, int alloc_flags
)
1223 /* free_pages my go negative - that's OK */
1225 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1228 if (alloc_flags
& ALLOC_HIGH
)
1230 if (alloc_flags
& ALLOC_HARDER
)
1233 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1235 for (o
= 0; o
< order
; o
++) {
1236 /* At the next order, this order's pages become unavailable */
1237 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1239 /* Require fewer higher order pages to be free */
1242 if (free_pages
<= min
)
1250 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1251 * skip over zones that are not allowed by the cpuset, or that have
1252 * been recently (in last second) found to be nearly full. See further
1253 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1254 * that have to skip over a lot of full or unallowed zones.
1256 * If the zonelist cache is present in the passed in zonelist, then
1257 * returns a pointer to the allowed node mask (either the current
1258 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1260 * If the zonelist cache is not available for this zonelist, does
1261 * nothing and returns NULL.
1263 * If the fullzones BITMAP in the zonelist cache is stale (more than
1264 * a second since last zap'd) then we zap it out (clear its bits.)
1266 * We hold off even calling zlc_setup, until after we've checked the
1267 * first zone in the zonelist, on the theory that most allocations will
1268 * be satisfied from that first zone, so best to examine that zone as
1269 * quickly as we can.
1271 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1273 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1274 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1276 zlc
= zonelist
->zlcache_ptr
;
1280 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1281 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1282 zlc
->last_full_zap
= jiffies
;
1285 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1286 &cpuset_current_mems_allowed
:
1287 &node_states
[N_HIGH_MEMORY
];
1288 return allowednodes
;
1292 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1293 * if it is worth looking at further for free memory:
1294 * 1) Check that the zone isn't thought to be full (doesn't have its
1295 * bit set in the zonelist_cache fullzones BITMAP).
1296 * 2) Check that the zones node (obtained from the zonelist_cache
1297 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1298 * Return true (non-zero) if zone is worth looking at further, or
1299 * else return false (zero) if it is not.
1301 * This check -ignores- the distinction between various watermarks,
1302 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1303 * found to be full for any variation of these watermarks, it will
1304 * be considered full for up to one second by all requests, unless
1305 * we are so low on memory on all allowed nodes that we are forced
1306 * into the second scan of the zonelist.
1308 * In the second scan we ignore this zonelist cache and exactly
1309 * apply the watermarks to all zones, even it is slower to do so.
1310 * We are low on memory in the second scan, and should leave no stone
1311 * unturned looking for a free page.
1313 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1314 nodemask_t
*allowednodes
)
1316 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1317 int i
; /* index of *z in zonelist zones */
1318 int n
; /* node that zone *z is on */
1320 zlc
= zonelist
->zlcache_ptr
;
1324 i
= z
- zonelist
->_zonerefs
;
1327 /* This zone is worth trying if it is allowed but not full */
1328 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1332 * Given 'z' scanning a zonelist, set the corresponding bit in
1333 * zlc->fullzones, so that subsequent attempts to allocate a page
1334 * from that zone don't waste time re-examining it.
1336 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1338 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1339 int i
; /* index of *z in zonelist zones */
1341 zlc
= zonelist
->zlcache_ptr
;
1345 i
= z
- zonelist
->_zonerefs
;
1347 set_bit(i
, zlc
->fullzones
);
1350 #else /* CONFIG_NUMA */
1352 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1357 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1358 nodemask_t
*allowednodes
)
1363 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1366 #endif /* CONFIG_NUMA */
1369 * get_page_from_freelist goes through the zonelist trying to allocate
1372 static struct page
*
1373 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1374 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1377 struct page
*page
= NULL
;
1379 struct zone
*zone
, *preferred_zone
;
1380 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1381 int zlc_active
= 0; /* set if using zonelist_cache */
1382 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1384 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1386 if (!preferred_zone
)
1389 classzone_idx
= zone_idx(preferred_zone
);
1393 * Scan zonelist, looking for a zone with enough free.
1394 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1396 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1397 high_zoneidx
, nodemask
) {
1398 if (NUMA_BUILD
&& zlc_active
&&
1399 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1401 if ((alloc_flags
& ALLOC_CPUSET
) &&
1402 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1405 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1407 if (alloc_flags
& ALLOC_WMARK_MIN
)
1408 mark
= zone
->pages_min
;
1409 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1410 mark
= zone
->pages_low
;
1412 mark
= zone
->pages_high
;
1413 if (!zone_watermark_ok(zone
, order
, mark
,
1414 classzone_idx
, alloc_flags
)) {
1415 if (!zone_reclaim_mode
||
1416 !zone_reclaim(zone
, gfp_mask
, order
))
1417 goto this_zone_full
;
1421 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1426 zlc_mark_zone_full(zonelist
, z
);
1428 if (NUMA_BUILD
&& !did_zlc_setup
) {
1429 /* we do zlc_setup after the first zone is tried */
1430 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1436 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1437 /* Disable zlc cache for second zonelist scan */
1445 * This is the 'heart' of the zoned buddy allocator.
1448 __alloc_pages_internal(gfp_t gfp_mask
, unsigned int order
,
1449 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1451 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1452 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1456 struct reclaim_state reclaim_state
;
1457 struct task_struct
*p
= current
;
1460 unsigned long did_some_progress
;
1461 unsigned long pages_reclaimed
= 0;
1463 might_sleep_if(wait
);
1465 if (should_fail_alloc_page(gfp_mask
, order
))
1469 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1471 if (unlikely(!z
->zone
)) {
1473 * Happens if we have an empty zonelist as a result of
1474 * GFP_THISNODE being used on a memoryless node
1479 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1480 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1485 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1486 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1487 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1488 * using a larger set of nodes after it has established that the
1489 * allowed per node queues are empty and that nodes are
1492 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1495 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1496 wakeup_kswapd(zone
, order
);
1499 * OK, we're below the kswapd watermark and have kicked background
1500 * reclaim. Now things get more complex, so set up alloc_flags according
1501 * to how we want to proceed.
1503 * The caller may dip into page reserves a bit more if the caller
1504 * cannot run direct reclaim, or if the caller has realtime scheduling
1505 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1506 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1508 alloc_flags
= ALLOC_WMARK_MIN
;
1509 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1510 alloc_flags
|= ALLOC_HARDER
;
1511 if (gfp_mask
& __GFP_HIGH
)
1512 alloc_flags
|= ALLOC_HIGH
;
1514 alloc_flags
|= ALLOC_CPUSET
;
1517 * Go through the zonelist again. Let __GFP_HIGH and allocations
1518 * coming from realtime tasks go deeper into reserves.
1520 * This is the last chance, in general, before the goto nopage.
1521 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1522 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1524 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1525 high_zoneidx
, alloc_flags
);
1529 /* This allocation should allow future memory freeing. */
1532 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1533 && !in_interrupt()) {
1534 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1536 /* go through the zonelist yet again, ignoring mins */
1537 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1538 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1541 if (gfp_mask
& __GFP_NOFAIL
) {
1542 congestion_wait(WRITE
, HZ
/50);
1549 /* Atomic allocations - we can't balance anything */
1555 /* We now go into synchronous reclaim */
1556 cpuset_memory_pressure_bump();
1557 p
->flags
|= PF_MEMALLOC
;
1558 reclaim_state
.reclaimed_slab
= 0;
1559 p
->reclaim_state
= &reclaim_state
;
1561 did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
);
1563 p
->reclaim_state
= NULL
;
1564 p
->flags
&= ~PF_MEMALLOC
;
1571 if (likely(did_some_progress
)) {
1572 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1573 zonelist
, high_zoneidx
, alloc_flags
);
1576 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1577 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1578 schedule_timeout_uninterruptible(1);
1583 * Go through the zonelist yet one more time, keep
1584 * very high watermark here, this is only to catch
1585 * a parallel oom killing, we must fail if we're still
1586 * under heavy pressure.
1588 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1589 order
, zonelist
, high_zoneidx
,
1590 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1592 clear_zonelist_oom(zonelist
, gfp_mask
);
1596 /* The OOM killer will not help higher order allocs so fail */
1597 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1598 clear_zonelist_oom(zonelist
, gfp_mask
);
1602 out_of_memory(zonelist
, gfp_mask
, order
);
1603 clear_zonelist_oom(zonelist
, gfp_mask
);
1608 * Don't let big-order allocations loop unless the caller explicitly
1609 * requests that. Wait for some write requests to complete then retry.
1611 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1612 * means __GFP_NOFAIL, but that may not be true in other
1615 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1616 * specified, then we retry until we no longer reclaim any pages
1617 * (above), or we've reclaimed an order of pages at least as
1618 * large as the allocation's order. In both cases, if the
1619 * allocation still fails, we stop retrying.
1621 pages_reclaimed
+= did_some_progress
;
1623 if (!(gfp_mask
& __GFP_NORETRY
)) {
1624 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1627 if (gfp_mask
& __GFP_REPEAT
&&
1628 pages_reclaimed
< (1 << order
))
1631 if (gfp_mask
& __GFP_NOFAIL
)
1635 congestion_wait(WRITE
, HZ
/50);
1640 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1641 printk(KERN_WARNING
"%s: page allocation failure."
1642 " order:%d, mode:0x%x\n",
1643 p
->comm
, order
, gfp_mask
);
1650 EXPORT_SYMBOL(__alloc_pages_internal
);
1653 * Common helper functions.
1655 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1658 page
= alloc_pages(gfp_mask
, order
);
1661 return (unsigned long) page_address(page
);
1664 EXPORT_SYMBOL(__get_free_pages
);
1666 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1671 * get_zeroed_page() returns a 32-bit address, which cannot represent
1674 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1676 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1678 return (unsigned long) page_address(page
);
1682 EXPORT_SYMBOL(get_zeroed_page
);
1684 void __pagevec_free(struct pagevec
*pvec
)
1686 int i
= pagevec_count(pvec
);
1689 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1692 void __free_pages(struct page
*page
, unsigned int order
)
1694 if (put_page_testzero(page
)) {
1696 free_hot_page(page
);
1698 __free_pages_ok(page
, order
);
1702 EXPORT_SYMBOL(__free_pages
);
1704 void free_pages(unsigned long addr
, unsigned int order
)
1707 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1708 __free_pages(virt_to_page((void *)addr
), order
);
1712 EXPORT_SYMBOL(free_pages
);
1715 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1716 * @size: the number of bytes to allocate
1717 * @gfp_mask: GFP flags for the allocation
1719 * This function is similar to alloc_pages(), except that it allocates the
1720 * minimum number of pages to satisfy the request. alloc_pages() can only
1721 * allocate memory in power-of-two pages.
1723 * This function is also limited by MAX_ORDER.
1725 * Memory allocated by this function must be released by free_pages_exact().
1727 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1729 unsigned int order
= get_order(size
);
1732 addr
= __get_free_pages(gfp_mask
, order
);
1734 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1735 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1737 split_page(virt_to_page(addr
), order
);
1738 while (used
< alloc_end
) {
1744 return (void *)addr
;
1746 EXPORT_SYMBOL(alloc_pages_exact
);
1749 * free_pages_exact - release memory allocated via alloc_pages_exact()
1750 * @virt: the value returned by alloc_pages_exact.
1751 * @size: size of allocation, same value as passed to alloc_pages_exact().
1753 * Release the memory allocated by a previous call to alloc_pages_exact.
1755 void free_pages_exact(void *virt
, size_t size
)
1757 unsigned long addr
= (unsigned long)virt
;
1758 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1760 while (addr
< end
) {
1765 EXPORT_SYMBOL(free_pages_exact
);
1767 static unsigned int nr_free_zone_pages(int offset
)
1772 /* Just pick one node, since fallback list is circular */
1773 unsigned int sum
= 0;
1775 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1777 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1778 unsigned long size
= zone
->present_pages
;
1779 unsigned long high
= zone
->pages_high
;
1788 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1790 unsigned int nr_free_buffer_pages(void)
1792 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1794 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1797 * Amount of free RAM allocatable within all zones
1799 unsigned int nr_free_pagecache_pages(void)
1801 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1804 static inline void show_node(struct zone
*zone
)
1807 printk("Node %d ", zone_to_nid(zone
));
1810 void si_meminfo(struct sysinfo
*val
)
1812 val
->totalram
= totalram_pages
;
1814 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1815 val
->bufferram
= nr_blockdev_pages();
1816 val
->totalhigh
= totalhigh_pages
;
1817 val
->freehigh
= nr_free_highpages();
1818 val
->mem_unit
= PAGE_SIZE
;
1821 EXPORT_SYMBOL(si_meminfo
);
1824 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1826 pg_data_t
*pgdat
= NODE_DATA(nid
);
1828 val
->totalram
= pgdat
->node_present_pages
;
1829 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1830 #ifdef CONFIG_HIGHMEM
1831 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1832 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1838 val
->mem_unit
= PAGE_SIZE
;
1842 #define K(x) ((x) << (PAGE_SHIFT-10))
1845 * Show free area list (used inside shift_scroll-lock stuff)
1846 * We also calculate the percentage fragmentation. We do this by counting the
1847 * memory on each free list with the exception of the first item on the list.
1849 void show_free_areas(void)
1854 for_each_zone(zone
) {
1855 if (!populated_zone(zone
))
1859 printk("%s per-cpu:\n", zone
->name
);
1861 for_each_online_cpu(cpu
) {
1862 struct per_cpu_pageset
*pageset
;
1864 pageset
= zone_pcp(zone
, cpu
);
1866 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1867 cpu
, pageset
->pcp
.high
,
1868 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1872 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1873 " inactive_file:%lu"
1874 //TODO: check/adjust line lengths
1875 #ifdef CONFIG_UNEVICTABLE_LRU
1878 " dirty:%lu writeback:%lu unstable:%lu\n"
1879 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1880 global_page_state(NR_ACTIVE_ANON
),
1881 global_page_state(NR_ACTIVE_FILE
),
1882 global_page_state(NR_INACTIVE_ANON
),
1883 global_page_state(NR_INACTIVE_FILE
),
1884 #ifdef CONFIG_UNEVICTABLE_LRU
1885 global_page_state(NR_UNEVICTABLE
),
1887 global_page_state(NR_FILE_DIRTY
),
1888 global_page_state(NR_WRITEBACK
),
1889 global_page_state(NR_UNSTABLE_NFS
),
1890 global_page_state(NR_FREE_PAGES
),
1891 global_page_state(NR_SLAB_RECLAIMABLE
) +
1892 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1893 global_page_state(NR_FILE_MAPPED
),
1894 global_page_state(NR_PAGETABLE
),
1895 global_page_state(NR_BOUNCE
));
1897 for_each_zone(zone
) {
1900 if (!populated_zone(zone
))
1909 " active_anon:%lukB"
1910 " inactive_anon:%lukB"
1911 " active_file:%lukB"
1912 " inactive_file:%lukB"
1913 #ifdef CONFIG_UNEVICTABLE_LRU
1914 " unevictable:%lukB"
1917 " pages_scanned:%lu"
1918 " all_unreclaimable? %s"
1921 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1924 K(zone
->pages_high
),
1925 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
1926 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
1927 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
1928 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
1929 #ifdef CONFIG_UNEVICTABLE_LRU
1930 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
1932 K(zone
->present_pages
),
1933 zone
->pages_scanned
,
1934 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1936 printk("lowmem_reserve[]:");
1937 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1938 printk(" %lu", zone
->lowmem_reserve
[i
]);
1942 for_each_zone(zone
) {
1943 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1945 if (!populated_zone(zone
))
1949 printk("%s: ", zone
->name
);
1951 spin_lock_irqsave(&zone
->lock
, flags
);
1952 for (order
= 0; order
< MAX_ORDER
; order
++) {
1953 nr
[order
] = zone
->free_area
[order
].nr_free
;
1954 total
+= nr
[order
] << order
;
1956 spin_unlock_irqrestore(&zone
->lock
, flags
);
1957 for (order
= 0; order
< MAX_ORDER
; order
++)
1958 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1959 printk("= %lukB\n", K(total
));
1962 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
1964 show_swap_cache_info();
1967 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
1969 zoneref
->zone
= zone
;
1970 zoneref
->zone_idx
= zone_idx(zone
);
1974 * Builds allocation fallback zone lists.
1976 * Add all populated zones of a node to the zonelist.
1978 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1979 int nr_zones
, enum zone_type zone_type
)
1983 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1988 zone
= pgdat
->node_zones
+ zone_type
;
1989 if (populated_zone(zone
)) {
1990 zoneref_set_zone(zone
,
1991 &zonelist
->_zonerefs
[nr_zones
++]);
1992 check_highest_zone(zone_type
);
1995 } while (zone_type
);
2002 * 0 = automatic detection of better ordering.
2003 * 1 = order by ([node] distance, -zonetype)
2004 * 2 = order by (-zonetype, [node] distance)
2006 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2007 * the same zonelist. So only NUMA can configure this param.
2009 #define ZONELIST_ORDER_DEFAULT 0
2010 #define ZONELIST_ORDER_NODE 1
2011 #define ZONELIST_ORDER_ZONE 2
2013 /* zonelist order in the kernel.
2014 * set_zonelist_order() will set this to NODE or ZONE.
2016 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2017 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2021 /* The value user specified ....changed by config */
2022 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2023 /* string for sysctl */
2024 #define NUMA_ZONELIST_ORDER_LEN 16
2025 char numa_zonelist_order
[16] = "default";
2028 * interface for configure zonelist ordering.
2029 * command line option "numa_zonelist_order"
2030 * = "[dD]efault - default, automatic configuration.
2031 * = "[nN]ode - order by node locality, then by zone within node
2032 * = "[zZ]one - order by zone, then by locality within zone
2035 static int __parse_numa_zonelist_order(char *s
)
2037 if (*s
== 'd' || *s
== 'D') {
2038 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2039 } else if (*s
== 'n' || *s
== 'N') {
2040 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2041 } else if (*s
== 'z' || *s
== 'Z') {
2042 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2045 "Ignoring invalid numa_zonelist_order value: "
2052 static __init
int setup_numa_zonelist_order(char *s
)
2055 return __parse_numa_zonelist_order(s
);
2058 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2061 * sysctl handler for numa_zonelist_order
2063 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2064 struct file
*file
, void __user
*buffer
, size_t *length
,
2067 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2071 strncpy(saved_string
, (char*)table
->data
,
2072 NUMA_ZONELIST_ORDER_LEN
);
2073 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2077 int oldval
= user_zonelist_order
;
2078 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2080 * bogus value. restore saved string
2082 strncpy((char*)table
->data
, saved_string
,
2083 NUMA_ZONELIST_ORDER_LEN
);
2084 user_zonelist_order
= oldval
;
2085 } else if (oldval
!= user_zonelist_order
)
2086 build_all_zonelists();
2092 #define MAX_NODE_LOAD (num_online_nodes())
2093 static int node_load
[MAX_NUMNODES
];
2096 * find_next_best_node - find the next node that should appear in a given node's fallback list
2097 * @node: node whose fallback list we're appending
2098 * @used_node_mask: nodemask_t of already used nodes
2100 * We use a number of factors to determine which is the next node that should
2101 * appear on a given node's fallback list. The node should not have appeared
2102 * already in @node's fallback list, and it should be the next closest node
2103 * according to the distance array (which contains arbitrary distance values
2104 * from each node to each node in the system), and should also prefer nodes
2105 * with no CPUs, since presumably they'll have very little allocation pressure
2106 * on them otherwise.
2107 * It returns -1 if no node is found.
2109 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2112 int min_val
= INT_MAX
;
2114 node_to_cpumask_ptr(tmp
, 0);
2116 /* Use the local node if we haven't already */
2117 if (!node_isset(node
, *used_node_mask
)) {
2118 node_set(node
, *used_node_mask
);
2122 for_each_node_state(n
, N_HIGH_MEMORY
) {
2124 /* Don't want a node to appear more than once */
2125 if (node_isset(n
, *used_node_mask
))
2128 /* Use the distance array to find the distance */
2129 val
= node_distance(node
, n
);
2131 /* Penalize nodes under us ("prefer the next node") */
2134 /* Give preference to headless and unused nodes */
2135 node_to_cpumask_ptr_next(tmp
, n
);
2136 if (!cpus_empty(*tmp
))
2137 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2139 /* Slight preference for less loaded node */
2140 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2141 val
+= node_load
[n
];
2143 if (val
< min_val
) {
2150 node_set(best_node
, *used_node_mask
);
2157 * Build zonelists ordered by node and zones within node.
2158 * This results in maximum locality--normal zone overflows into local
2159 * DMA zone, if any--but risks exhausting DMA zone.
2161 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2164 struct zonelist
*zonelist
;
2166 zonelist
= &pgdat
->node_zonelists
[0];
2167 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2169 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2171 zonelist
->_zonerefs
[j
].zone
= NULL
;
2172 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2176 * Build gfp_thisnode zonelists
2178 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2181 struct zonelist
*zonelist
;
2183 zonelist
= &pgdat
->node_zonelists
[1];
2184 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2185 zonelist
->_zonerefs
[j
].zone
= NULL
;
2186 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2190 * Build zonelists ordered by zone and nodes within zones.
2191 * This results in conserving DMA zone[s] until all Normal memory is
2192 * exhausted, but results in overflowing to remote node while memory
2193 * may still exist in local DMA zone.
2195 static int node_order
[MAX_NUMNODES
];
2197 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2200 int zone_type
; /* needs to be signed */
2202 struct zonelist
*zonelist
;
2204 zonelist
= &pgdat
->node_zonelists
[0];
2206 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2207 for (j
= 0; j
< nr_nodes
; j
++) {
2208 node
= node_order
[j
];
2209 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2210 if (populated_zone(z
)) {
2212 &zonelist
->_zonerefs
[pos
++]);
2213 check_highest_zone(zone_type
);
2217 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2218 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2221 static int default_zonelist_order(void)
2224 unsigned long low_kmem_size
,total_size
;
2228 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2229 * If they are really small and used heavily, the system can fall
2230 * into OOM very easily.
2231 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2233 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2236 for_each_online_node(nid
) {
2237 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2238 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2239 if (populated_zone(z
)) {
2240 if (zone_type
< ZONE_NORMAL
)
2241 low_kmem_size
+= z
->present_pages
;
2242 total_size
+= z
->present_pages
;
2246 if (!low_kmem_size
|| /* there are no DMA area. */
2247 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2248 return ZONELIST_ORDER_NODE
;
2250 * look into each node's config.
2251 * If there is a node whose DMA/DMA32 memory is very big area on
2252 * local memory, NODE_ORDER may be suitable.
2254 average_size
= total_size
/
2255 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2256 for_each_online_node(nid
) {
2259 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2260 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2261 if (populated_zone(z
)) {
2262 if (zone_type
< ZONE_NORMAL
)
2263 low_kmem_size
+= z
->present_pages
;
2264 total_size
+= z
->present_pages
;
2267 if (low_kmem_size
&&
2268 total_size
> average_size
&& /* ignore small node */
2269 low_kmem_size
> total_size
* 70/100)
2270 return ZONELIST_ORDER_NODE
;
2272 return ZONELIST_ORDER_ZONE
;
2275 static void set_zonelist_order(void)
2277 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2278 current_zonelist_order
= default_zonelist_order();
2280 current_zonelist_order
= user_zonelist_order
;
2283 static void build_zonelists(pg_data_t
*pgdat
)
2287 nodemask_t used_mask
;
2288 int local_node
, prev_node
;
2289 struct zonelist
*zonelist
;
2290 int order
= current_zonelist_order
;
2292 /* initialize zonelists */
2293 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2294 zonelist
= pgdat
->node_zonelists
+ i
;
2295 zonelist
->_zonerefs
[0].zone
= NULL
;
2296 zonelist
->_zonerefs
[0].zone_idx
= 0;
2299 /* NUMA-aware ordering of nodes */
2300 local_node
= pgdat
->node_id
;
2301 load
= num_online_nodes();
2302 prev_node
= local_node
;
2303 nodes_clear(used_mask
);
2305 memset(node_load
, 0, sizeof(node_load
));
2306 memset(node_order
, 0, sizeof(node_order
));
2309 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2310 int distance
= node_distance(local_node
, node
);
2313 * If another node is sufficiently far away then it is better
2314 * to reclaim pages in a zone before going off node.
2316 if (distance
> RECLAIM_DISTANCE
)
2317 zone_reclaim_mode
= 1;
2320 * We don't want to pressure a particular node.
2321 * So adding penalty to the first node in same
2322 * distance group to make it round-robin.
2324 if (distance
!= node_distance(local_node
, prev_node
))
2325 node_load
[node
] = load
;
2329 if (order
== ZONELIST_ORDER_NODE
)
2330 build_zonelists_in_node_order(pgdat
, node
);
2332 node_order
[j
++] = node
; /* remember order */
2335 if (order
== ZONELIST_ORDER_ZONE
) {
2336 /* calculate node order -- i.e., DMA last! */
2337 build_zonelists_in_zone_order(pgdat
, j
);
2340 build_thisnode_zonelists(pgdat
);
2343 /* Construct the zonelist performance cache - see further mmzone.h */
2344 static void build_zonelist_cache(pg_data_t
*pgdat
)
2346 struct zonelist
*zonelist
;
2347 struct zonelist_cache
*zlc
;
2350 zonelist
= &pgdat
->node_zonelists
[0];
2351 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2352 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2353 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2354 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2358 #else /* CONFIG_NUMA */
2360 static void set_zonelist_order(void)
2362 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2365 static void build_zonelists(pg_data_t
*pgdat
)
2367 int node
, local_node
;
2369 struct zonelist
*zonelist
;
2371 local_node
= pgdat
->node_id
;
2373 zonelist
= &pgdat
->node_zonelists
[0];
2374 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2377 * Now we build the zonelist so that it contains the zones
2378 * of all the other nodes.
2379 * We don't want to pressure a particular node, so when
2380 * building the zones for node N, we make sure that the
2381 * zones coming right after the local ones are those from
2382 * node N+1 (modulo N)
2384 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2385 if (!node_online(node
))
2387 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2390 for (node
= 0; node
< local_node
; node
++) {
2391 if (!node_online(node
))
2393 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2397 zonelist
->_zonerefs
[j
].zone
= NULL
;
2398 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2401 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2402 static void build_zonelist_cache(pg_data_t
*pgdat
)
2404 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2407 #endif /* CONFIG_NUMA */
2409 /* return values int ....just for stop_machine() */
2410 static int __build_all_zonelists(void *dummy
)
2414 for_each_online_node(nid
) {
2415 pg_data_t
*pgdat
= NODE_DATA(nid
);
2417 build_zonelists(pgdat
);
2418 build_zonelist_cache(pgdat
);
2423 void build_all_zonelists(void)
2425 set_zonelist_order();
2427 if (system_state
== SYSTEM_BOOTING
) {
2428 __build_all_zonelists(NULL
);
2429 mminit_verify_zonelist();
2430 cpuset_init_current_mems_allowed();
2432 /* we have to stop all cpus to guarantee there is no user
2434 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2435 /* cpuset refresh routine should be here */
2437 vm_total_pages
= nr_free_pagecache_pages();
2439 * Disable grouping by mobility if the number of pages in the
2440 * system is too low to allow the mechanism to work. It would be
2441 * more accurate, but expensive to check per-zone. This check is
2442 * made on memory-hotadd so a system can start with mobility
2443 * disabled and enable it later
2445 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2446 page_group_by_mobility_disabled
= 1;
2448 page_group_by_mobility_disabled
= 0;
2450 printk("Built %i zonelists in %s order, mobility grouping %s. "
2451 "Total pages: %ld\n",
2453 zonelist_order_name
[current_zonelist_order
],
2454 page_group_by_mobility_disabled
? "off" : "on",
2457 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2462 * Helper functions to size the waitqueue hash table.
2463 * Essentially these want to choose hash table sizes sufficiently
2464 * large so that collisions trying to wait on pages are rare.
2465 * But in fact, the number of active page waitqueues on typical
2466 * systems is ridiculously low, less than 200. So this is even
2467 * conservative, even though it seems large.
2469 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2470 * waitqueues, i.e. the size of the waitq table given the number of pages.
2472 #define PAGES_PER_WAITQUEUE 256
2474 #ifndef CONFIG_MEMORY_HOTPLUG
2475 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2477 unsigned long size
= 1;
2479 pages
/= PAGES_PER_WAITQUEUE
;
2481 while (size
< pages
)
2485 * Once we have dozens or even hundreds of threads sleeping
2486 * on IO we've got bigger problems than wait queue collision.
2487 * Limit the size of the wait table to a reasonable size.
2489 size
= min(size
, 4096UL);
2491 return max(size
, 4UL);
2495 * A zone's size might be changed by hot-add, so it is not possible to determine
2496 * a suitable size for its wait_table. So we use the maximum size now.
2498 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2500 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2501 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2502 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2504 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2505 * or more by the traditional way. (See above). It equals:
2507 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2508 * ia64(16K page size) : = ( 8G + 4M)byte.
2509 * powerpc (64K page size) : = (32G +16M)byte.
2511 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2518 * This is an integer logarithm so that shifts can be used later
2519 * to extract the more random high bits from the multiplicative
2520 * hash function before the remainder is taken.
2522 static inline unsigned long wait_table_bits(unsigned long size
)
2527 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2530 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2531 * of blocks reserved is based on zone->pages_min. The memory within the
2532 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2533 * higher will lead to a bigger reserve which will get freed as contiguous
2534 * blocks as reclaim kicks in
2536 static void setup_zone_migrate_reserve(struct zone
*zone
)
2538 unsigned long start_pfn
, pfn
, end_pfn
;
2540 unsigned long reserve
, block_migratetype
;
2542 /* Get the start pfn, end pfn and the number of blocks to reserve */
2543 start_pfn
= zone
->zone_start_pfn
;
2544 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2545 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2548 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2549 if (!pfn_valid(pfn
))
2551 page
= pfn_to_page(pfn
);
2553 /* Watch out for overlapping nodes */
2554 if (page_to_nid(page
) != zone_to_nid(zone
))
2557 /* Blocks with reserved pages will never free, skip them. */
2558 if (PageReserved(page
))
2561 block_migratetype
= get_pageblock_migratetype(page
);
2563 /* If this block is reserved, account for it */
2564 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2569 /* Suitable for reserving if this block is movable */
2570 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2571 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2572 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2578 * If the reserve is met and this is a previous reserved block,
2581 if (block_migratetype
== MIGRATE_RESERVE
) {
2582 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2583 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2589 * Initially all pages are reserved - free ones are freed
2590 * up by free_all_bootmem() once the early boot process is
2591 * done. Non-atomic initialization, single-pass.
2593 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2594 unsigned long start_pfn
, enum memmap_context context
)
2597 unsigned long end_pfn
= start_pfn
+ size
;
2601 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2602 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2604 * There can be holes in boot-time mem_map[]s
2605 * handed to this function. They do not
2606 * exist on hotplugged memory.
2608 if (context
== MEMMAP_EARLY
) {
2609 if (!early_pfn_valid(pfn
))
2611 if (!early_pfn_in_nid(pfn
, nid
))
2614 page
= pfn_to_page(pfn
);
2615 set_page_links(page
, zone
, nid
, pfn
);
2616 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2617 init_page_count(page
);
2618 reset_page_mapcount(page
);
2619 SetPageReserved(page
);
2621 * Mark the block movable so that blocks are reserved for
2622 * movable at startup. This will force kernel allocations
2623 * to reserve their blocks rather than leaking throughout
2624 * the address space during boot when many long-lived
2625 * kernel allocations are made. Later some blocks near
2626 * the start are marked MIGRATE_RESERVE by
2627 * setup_zone_migrate_reserve()
2629 * bitmap is created for zone's valid pfn range. but memmap
2630 * can be created for invalid pages (for alignment)
2631 * check here not to call set_pageblock_migratetype() against
2634 if ((z
->zone_start_pfn
<= pfn
)
2635 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2636 && !(pfn
& (pageblock_nr_pages
- 1)))
2637 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2639 INIT_LIST_HEAD(&page
->lru
);
2640 #ifdef WANT_PAGE_VIRTUAL
2641 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2642 if (!is_highmem_idx(zone
))
2643 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2648 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2651 for_each_migratetype_order(order
, t
) {
2652 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2653 zone
->free_area
[order
].nr_free
= 0;
2657 #ifndef __HAVE_ARCH_MEMMAP_INIT
2658 #define memmap_init(size, nid, zone, start_pfn) \
2659 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2662 static int zone_batchsize(struct zone
*zone
)
2667 * The per-cpu-pages pools are set to around 1000th of the
2668 * size of the zone. But no more than 1/2 of a meg.
2670 * OK, so we don't know how big the cache is. So guess.
2672 batch
= zone
->present_pages
/ 1024;
2673 if (batch
* PAGE_SIZE
> 512 * 1024)
2674 batch
= (512 * 1024) / PAGE_SIZE
;
2675 batch
/= 4; /* We effectively *= 4 below */
2680 * Clamp the batch to a 2^n - 1 value. Having a power
2681 * of 2 value was found to be more likely to have
2682 * suboptimal cache aliasing properties in some cases.
2684 * For example if 2 tasks are alternately allocating
2685 * batches of pages, one task can end up with a lot
2686 * of pages of one half of the possible page colors
2687 * and the other with pages of the other colors.
2689 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2694 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2696 struct per_cpu_pages
*pcp
;
2698 memset(p
, 0, sizeof(*p
));
2702 pcp
->high
= 6 * batch
;
2703 pcp
->batch
= max(1UL, 1 * batch
);
2704 INIT_LIST_HEAD(&pcp
->list
);
2708 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2709 * to the value high for the pageset p.
2712 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2715 struct per_cpu_pages
*pcp
;
2719 pcp
->batch
= max(1UL, high
/4);
2720 if ((high
/4) > (PAGE_SHIFT
* 8))
2721 pcp
->batch
= PAGE_SHIFT
* 8;
2727 * Boot pageset table. One per cpu which is going to be used for all
2728 * zones and all nodes. The parameters will be set in such a way
2729 * that an item put on a list will immediately be handed over to
2730 * the buddy list. This is safe since pageset manipulation is done
2731 * with interrupts disabled.
2733 * Some NUMA counter updates may also be caught by the boot pagesets.
2735 * The boot_pagesets must be kept even after bootup is complete for
2736 * unused processors and/or zones. They do play a role for bootstrapping
2737 * hotplugged processors.
2739 * zoneinfo_show() and maybe other functions do
2740 * not check if the processor is online before following the pageset pointer.
2741 * Other parts of the kernel may not check if the zone is available.
2743 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2746 * Dynamically allocate memory for the
2747 * per cpu pageset array in struct zone.
2749 static int __cpuinit
process_zones(int cpu
)
2751 struct zone
*zone
, *dzone
;
2752 int node
= cpu_to_node(cpu
);
2754 node_set_state(node
, N_CPU
); /* this node has a cpu */
2756 for_each_zone(zone
) {
2758 if (!populated_zone(zone
))
2761 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2763 if (!zone_pcp(zone
, cpu
))
2766 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2768 if (percpu_pagelist_fraction
)
2769 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2770 (zone
->present_pages
/ percpu_pagelist_fraction
));
2775 for_each_zone(dzone
) {
2776 if (!populated_zone(dzone
))
2780 kfree(zone_pcp(dzone
, cpu
));
2781 zone_pcp(dzone
, cpu
) = NULL
;
2786 static inline void free_zone_pagesets(int cpu
)
2790 for_each_zone(zone
) {
2791 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2793 /* Free per_cpu_pageset if it is slab allocated */
2794 if (pset
!= &boot_pageset
[cpu
])
2796 zone_pcp(zone
, cpu
) = NULL
;
2800 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2801 unsigned long action
,
2804 int cpu
= (long)hcpu
;
2805 int ret
= NOTIFY_OK
;
2808 case CPU_UP_PREPARE
:
2809 case CPU_UP_PREPARE_FROZEN
:
2810 if (process_zones(cpu
))
2813 case CPU_UP_CANCELED
:
2814 case CPU_UP_CANCELED_FROZEN
:
2816 case CPU_DEAD_FROZEN
:
2817 free_zone_pagesets(cpu
);
2825 static struct notifier_block __cpuinitdata pageset_notifier
=
2826 { &pageset_cpuup_callback
, NULL
, 0 };
2828 void __init
setup_per_cpu_pageset(void)
2832 /* Initialize per_cpu_pageset for cpu 0.
2833 * A cpuup callback will do this for every cpu
2834 * as it comes online
2836 err
= process_zones(smp_processor_id());
2838 register_cpu_notifier(&pageset_notifier
);
2843 static noinline __init_refok
2844 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2847 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2851 * The per-page waitqueue mechanism uses hashed waitqueues
2854 zone
->wait_table_hash_nr_entries
=
2855 wait_table_hash_nr_entries(zone_size_pages
);
2856 zone
->wait_table_bits
=
2857 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2858 alloc_size
= zone
->wait_table_hash_nr_entries
2859 * sizeof(wait_queue_head_t
);
2861 if (!slab_is_available()) {
2862 zone
->wait_table
= (wait_queue_head_t
*)
2863 alloc_bootmem_node(pgdat
, alloc_size
);
2866 * This case means that a zone whose size was 0 gets new memory
2867 * via memory hot-add.
2868 * But it may be the case that a new node was hot-added. In
2869 * this case vmalloc() will not be able to use this new node's
2870 * memory - this wait_table must be initialized to use this new
2871 * node itself as well.
2872 * To use this new node's memory, further consideration will be
2875 zone
->wait_table
= vmalloc(alloc_size
);
2877 if (!zone
->wait_table
)
2880 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2881 init_waitqueue_head(zone
->wait_table
+ i
);
2886 static __meminit
void zone_pcp_init(struct zone
*zone
)
2889 unsigned long batch
= zone_batchsize(zone
);
2891 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2893 /* Early boot. Slab allocator not functional yet */
2894 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2895 setup_pageset(&boot_pageset
[cpu
],0);
2897 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2900 if (zone
->present_pages
)
2901 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2902 zone
->name
, zone
->present_pages
, batch
);
2905 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2906 unsigned long zone_start_pfn
,
2908 enum memmap_context context
)
2910 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2912 ret
= zone_wait_table_init(zone
, size
);
2915 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2917 zone
->zone_start_pfn
= zone_start_pfn
;
2919 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
2920 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2922 (unsigned long)zone_idx(zone
),
2923 zone_start_pfn
, (zone_start_pfn
+ size
));
2925 zone_init_free_lists(zone
);
2930 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2932 * Basic iterator support. Return the first range of PFNs for a node
2933 * Note: nid == MAX_NUMNODES returns first region regardless of node
2935 static int __meminit
first_active_region_index_in_nid(int nid
)
2939 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2940 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2947 * Basic iterator support. Return the next active range of PFNs for a node
2948 * Note: nid == MAX_NUMNODES returns next region regardless of node
2950 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2952 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2953 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2959 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2961 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2962 * Architectures may implement their own version but if add_active_range()
2963 * was used and there are no special requirements, this is a convenient
2966 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2970 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2971 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2972 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2974 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2975 return early_node_map
[i
].nid
;
2980 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2982 /* Basic iterator support to walk early_node_map[] */
2983 #define for_each_active_range_index_in_nid(i, nid) \
2984 for (i = first_active_region_index_in_nid(nid); i != -1; \
2985 i = next_active_region_index_in_nid(i, nid))
2988 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2989 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2990 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2992 * If an architecture guarantees that all ranges registered with
2993 * add_active_ranges() contain no holes and may be freed, this
2994 * this function may be used instead of calling free_bootmem() manually.
2996 void __init
free_bootmem_with_active_regions(int nid
,
2997 unsigned long max_low_pfn
)
3001 for_each_active_range_index_in_nid(i
, nid
) {
3002 unsigned long size_pages
= 0;
3003 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3005 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3008 if (end_pfn
> max_low_pfn
)
3009 end_pfn
= max_low_pfn
;
3011 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3012 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3013 PFN_PHYS(early_node_map
[i
].start_pfn
),
3014 size_pages
<< PAGE_SHIFT
);
3018 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3023 for_each_active_range_index_in_nid(i
, nid
) {
3024 ret
= work_fn(early_node_map
[i
].start_pfn
,
3025 early_node_map
[i
].end_pfn
, data
);
3031 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3032 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3034 * If an architecture guarantees that all ranges registered with
3035 * add_active_ranges() contain no holes and may be freed, this
3036 * function may be used instead of calling memory_present() manually.
3038 void __init
sparse_memory_present_with_active_regions(int nid
)
3042 for_each_active_range_index_in_nid(i
, nid
)
3043 memory_present(early_node_map
[i
].nid
,
3044 early_node_map
[i
].start_pfn
,
3045 early_node_map
[i
].end_pfn
);
3049 * push_node_boundaries - Push node boundaries to at least the requested boundary
3050 * @nid: The nid of the node to push the boundary for
3051 * @start_pfn: The start pfn of the node
3052 * @end_pfn: The end pfn of the node
3054 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3055 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3056 * be hotplugged even though no physical memory exists. This function allows
3057 * an arch to push out the node boundaries so mem_map is allocated that can
3060 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3061 void __init
push_node_boundaries(unsigned int nid
,
3062 unsigned long start_pfn
, unsigned long end_pfn
)
3064 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3065 "Entering push_node_boundaries(%u, %lu, %lu)\n",
3066 nid
, start_pfn
, end_pfn
);
3068 /* Initialise the boundary for this node if necessary */
3069 if (node_boundary_end_pfn
[nid
] == 0)
3070 node_boundary_start_pfn
[nid
] = -1UL;
3072 /* Update the boundaries */
3073 if (node_boundary_start_pfn
[nid
] > start_pfn
)
3074 node_boundary_start_pfn
[nid
] = start_pfn
;
3075 if (node_boundary_end_pfn
[nid
] < end_pfn
)
3076 node_boundary_end_pfn
[nid
] = end_pfn
;
3079 /* If necessary, push the node boundary out for reserve hotadd */
3080 static void __meminit
account_node_boundary(unsigned int nid
,
3081 unsigned long *start_pfn
, unsigned long *end_pfn
)
3083 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3084 "Entering account_node_boundary(%u, %lu, %lu)\n",
3085 nid
, *start_pfn
, *end_pfn
);
3087 /* Return if boundary information has not been provided */
3088 if (node_boundary_end_pfn
[nid
] == 0)
3091 /* Check the boundaries and update if necessary */
3092 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
3093 *start_pfn
= node_boundary_start_pfn
[nid
];
3094 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
3095 *end_pfn
= node_boundary_end_pfn
[nid
];
3098 void __init
push_node_boundaries(unsigned int nid
,
3099 unsigned long start_pfn
, unsigned long end_pfn
) {}
3101 static void __meminit
account_node_boundary(unsigned int nid
,
3102 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3107 * get_pfn_range_for_nid - Return the start and end page frames for a node
3108 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3109 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3110 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3112 * It returns the start and end page frame of a node based on information
3113 * provided by an arch calling add_active_range(). If called for a node
3114 * with no available memory, a warning is printed and the start and end
3117 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3118 unsigned long *start_pfn
, unsigned long *end_pfn
)
3124 for_each_active_range_index_in_nid(i
, nid
) {
3125 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3126 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3129 if (*start_pfn
== -1UL)
3132 /* Push the node boundaries out if requested */
3133 account_node_boundary(nid
, start_pfn
, end_pfn
);
3137 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3138 * assumption is made that zones within a node are ordered in monotonic
3139 * increasing memory addresses so that the "highest" populated zone is used
3141 static void __init
find_usable_zone_for_movable(void)
3144 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3145 if (zone_index
== ZONE_MOVABLE
)
3148 if (arch_zone_highest_possible_pfn
[zone_index
] >
3149 arch_zone_lowest_possible_pfn
[zone_index
])
3153 VM_BUG_ON(zone_index
== -1);
3154 movable_zone
= zone_index
;
3158 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3159 * because it is sized independant of architecture. Unlike the other zones,
3160 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3161 * in each node depending on the size of each node and how evenly kernelcore
3162 * is distributed. This helper function adjusts the zone ranges
3163 * provided by the architecture for a given node by using the end of the
3164 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3165 * zones within a node are in order of monotonic increases memory addresses
3167 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3168 unsigned long zone_type
,
3169 unsigned long node_start_pfn
,
3170 unsigned long node_end_pfn
,
3171 unsigned long *zone_start_pfn
,
3172 unsigned long *zone_end_pfn
)
3174 /* Only adjust if ZONE_MOVABLE is on this node */
3175 if (zone_movable_pfn
[nid
]) {
3176 /* Size ZONE_MOVABLE */
3177 if (zone_type
== ZONE_MOVABLE
) {
3178 *zone_start_pfn
= zone_movable_pfn
[nid
];
3179 *zone_end_pfn
= min(node_end_pfn
,
3180 arch_zone_highest_possible_pfn
[movable_zone
]);
3182 /* Adjust for ZONE_MOVABLE starting within this range */
3183 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3184 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3185 *zone_end_pfn
= zone_movable_pfn
[nid
];
3187 /* Check if this whole range is within ZONE_MOVABLE */
3188 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3189 *zone_start_pfn
= *zone_end_pfn
;
3194 * Return the number of pages a zone spans in a node, including holes
3195 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3197 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3198 unsigned long zone_type
,
3199 unsigned long *ignored
)
3201 unsigned long node_start_pfn
, node_end_pfn
;
3202 unsigned long zone_start_pfn
, zone_end_pfn
;
3204 /* Get the start and end of the node and zone */
3205 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3206 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3207 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3208 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3209 node_start_pfn
, node_end_pfn
,
3210 &zone_start_pfn
, &zone_end_pfn
);
3212 /* Check that this node has pages within the zone's required range */
3213 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3216 /* Move the zone boundaries inside the node if necessary */
3217 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3218 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3220 /* Return the spanned pages */
3221 return zone_end_pfn
- zone_start_pfn
;
3225 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3226 * then all holes in the requested range will be accounted for.
3228 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3229 unsigned long range_start_pfn
,
3230 unsigned long range_end_pfn
)
3233 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3234 unsigned long start_pfn
;
3236 /* Find the end_pfn of the first active range of pfns in the node */
3237 i
= first_active_region_index_in_nid(nid
);
3241 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3243 /* Account for ranges before physical memory on this node */
3244 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3245 hole_pages
= prev_end_pfn
- range_start_pfn
;
3247 /* Find all holes for the zone within the node */
3248 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3250 /* No need to continue if prev_end_pfn is outside the zone */
3251 if (prev_end_pfn
>= range_end_pfn
)
3254 /* Make sure the end of the zone is not within the hole */
3255 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3256 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3258 /* Update the hole size cound and move on */
3259 if (start_pfn
> range_start_pfn
) {
3260 BUG_ON(prev_end_pfn
> start_pfn
);
3261 hole_pages
+= start_pfn
- prev_end_pfn
;
3263 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3266 /* Account for ranges past physical memory on this node */
3267 if (range_end_pfn
> prev_end_pfn
)
3268 hole_pages
+= range_end_pfn
-
3269 max(range_start_pfn
, prev_end_pfn
);
3275 * absent_pages_in_range - Return number of page frames in holes within a range
3276 * @start_pfn: The start PFN to start searching for holes
3277 * @end_pfn: The end PFN to stop searching for holes
3279 * It returns the number of pages frames in memory holes within a range.
3281 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3282 unsigned long end_pfn
)
3284 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3287 /* Return the number of page frames in holes in a zone on a node */
3288 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3289 unsigned long zone_type
,
3290 unsigned long *ignored
)
3292 unsigned long node_start_pfn
, node_end_pfn
;
3293 unsigned long zone_start_pfn
, zone_end_pfn
;
3295 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3296 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3298 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3301 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3302 node_start_pfn
, node_end_pfn
,
3303 &zone_start_pfn
, &zone_end_pfn
);
3304 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3308 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3309 unsigned long zone_type
,
3310 unsigned long *zones_size
)
3312 return zones_size
[zone_type
];
3315 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3316 unsigned long zone_type
,
3317 unsigned long *zholes_size
)
3322 return zholes_size
[zone_type
];
3327 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3328 unsigned long *zones_size
, unsigned long *zholes_size
)
3330 unsigned long realtotalpages
, totalpages
= 0;
3333 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3334 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3336 pgdat
->node_spanned_pages
= totalpages
;
3338 realtotalpages
= totalpages
;
3339 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3341 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3343 pgdat
->node_present_pages
= realtotalpages
;
3344 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3348 #ifndef CONFIG_SPARSEMEM
3350 * Calculate the size of the zone->blockflags rounded to an unsigned long
3351 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3352 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3353 * round what is now in bits to nearest long in bits, then return it in
3356 static unsigned long __init
usemap_size(unsigned long zonesize
)
3358 unsigned long usemapsize
;
3360 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3361 usemapsize
= usemapsize
>> pageblock_order
;
3362 usemapsize
*= NR_PAGEBLOCK_BITS
;
3363 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3365 return usemapsize
/ 8;
3368 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3369 struct zone
*zone
, unsigned long zonesize
)
3371 unsigned long usemapsize
= usemap_size(zonesize
);
3372 zone
->pageblock_flags
= NULL
;
3374 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3375 memset(zone
->pageblock_flags
, 0, usemapsize
);
3379 static void inline setup_usemap(struct pglist_data
*pgdat
,
3380 struct zone
*zone
, unsigned long zonesize
) {}
3381 #endif /* CONFIG_SPARSEMEM */
3383 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3385 /* Return a sensible default order for the pageblock size. */
3386 static inline int pageblock_default_order(void)
3388 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3389 return HUGETLB_PAGE_ORDER
;
3394 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3395 static inline void __init
set_pageblock_order(unsigned int order
)
3397 /* Check that pageblock_nr_pages has not already been setup */
3398 if (pageblock_order
)
3402 * Assume the largest contiguous order of interest is a huge page.
3403 * This value may be variable depending on boot parameters on IA64
3405 pageblock_order
= order
;
3407 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3410 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3411 * and pageblock_default_order() are unused as pageblock_order is set
3412 * at compile-time. See include/linux/pageblock-flags.h for the values of
3413 * pageblock_order based on the kernel config
3415 static inline int pageblock_default_order(unsigned int order
)
3419 #define set_pageblock_order(x) do {} while (0)
3421 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3424 * Set up the zone data structures:
3425 * - mark all pages reserved
3426 * - mark all memory queues empty
3427 * - clear the memory bitmaps
3429 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3430 unsigned long *zones_size
, unsigned long *zholes_size
)
3433 int nid
= pgdat
->node_id
;
3434 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3437 pgdat_resize_init(pgdat
);
3438 pgdat
->nr_zones
= 0;
3439 init_waitqueue_head(&pgdat
->kswapd_wait
);
3440 pgdat
->kswapd_max_order
= 0;
3442 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3443 struct zone
*zone
= pgdat
->node_zones
+ j
;
3444 unsigned long size
, realsize
, memmap_pages
;
3447 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3448 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3452 * Adjust realsize so that it accounts for how much memory
3453 * is used by this zone for memmap. This affects the watermark
3454 * and per-cpu initialisations
3457 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3458 if (realsize
>= memmap_pages
) {
3459 realsize
-= memmap_pages
;
3460 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3461 "%s zone: %lu pages used for memmap\n",
3462 zone_names
[j
], memmap_pages
);
3465 " %s zone: %lu pages exceeds realsize %lu\n",
3466 zone_names
[j
], memmap_pages
, realsize
);
3468 /* Account for reserved pages */
3469 if (j
== 0 && realsize
> dma_reserve
) {
3470 realsize
-= dma_reserve
;
3471 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3472 "%s zone: %lu pages reserved\n",
3473 zone_names
[0], dma_reserve
);
3476 if (!is_highmem_idx(j
))
3477 nr_kernel_pages
+= realsize
;
3478 nr_all_pages
+= realsize
;
3480 zone
->spanned_pages
= size
;
3481 zone
->present_pages
= realsize
;
3484 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3486 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3488 zone
->name
= zone_names
[j
];
3489 spin_lock_init(&zone
->lock
);
3490 spin_lock_init(&zone
->lru_lock
);
3491 zone_seqlock_init(zone
);
3492 zone
->zone_pgdat
= pgdat
;
3494 zone
->prev_priority
= DEF_PRIORITY
;
3496 zone_pcp_init(zone
);
3498 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3499 zone
->lru
[l
].nr_scan
= 0;
3501 zone
->recent_rotated
[0] = 0;
3502 zone
->recent_rotated
[1] = 0;
3503 zone
->recent_scanned
[0] = 0;
3504 zone
->recent_scanned
[1] = 0;
3505 zap_zone_vm_stats(zone
);
3510 set_pageblock_order(pageblock_default_order());
3511 setup_usemap(pgdat
, zone
, size
);
3512 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3513 size
, MEMMAP_EARLY
);
3515 memmap_init(size
, nid
, j
, zone_start_pfn
);
3516 zone_start_pfn
+= size
;
3520 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3522 /* Skip empty nodes */
3523 if (!pgdat
->node_spanned_pages
)
3526 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3527 /* ia64 gets its own node_mem_map, before this, without bootmem */
3528 if (!pgdat
->node_mem_map
) {
3529 unsigned long size
, start
, end
;
3533 * The zone's endpoints aren't required to be MAX_ORDER
3534 * aligned but the node_mem_map endpoints must be in order
3535 * for the buddy allocator to function correctly.
3537 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3538 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3539 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3540 size
= (end
- start
) * sizeof(struct page
);
3541 map
= alloc_remap(pgdat
->node_id
, size
);
3543 map
= alloc_bootmem_node(pgdat
, size
);
3544 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3546 #ifndef CONFIG_NEED_MULTIPLE_NODES
3548 * With no DISCONTIG, the global mem_map is just set as node 0's
3550 if (pgdat
== NODE_DATA(0)) {
3551 mem_map
= NODE_DATA(0)->node_mem_map
;
3552 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3553 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3554 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3555 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3558 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3561 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3562 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3564 pg_data_t
*pgdat
= NODE_DATA(nid
);
3566 pgdat
->node_id
= nid
;
3567 pgdat
->node_start_pfn
= node_start_pfn
;
3568 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3570 alloc_node_mem_map(pgdat
);
3571 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3572 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3573 nid
, (unsigned long)pgdat
,
3574 (unsigned long)pgdat
->node_mem_map
);
3577 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3580 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3582 #if MAX_NUMNODES > 1
3584 * Figure out the number of possible node ids.
3586 static void __init
setup_nr_node_ids(void)
3589 unsigned int highest
= 0;
3591 for_each_node_mask(node
, node_possible_map
)
3593 nr_node_ids
= highest
+ 1;
3596 static inline void setup_nr_node_ids(void)
3602 * add_active_range - Register a range of PFNs backed by physical memory
3603 * @nid: The node ID the range resides on
3604 * @start_pfn: The start PFN of the available physical memory
3605 * @end_pfn: The end PFN of the available physical memory
3607 * These ranges are stored in an early_node_map[] and later used by
3608 * free_area_init_nodes() to calculate zone sizes and holes. If the
3609 * range spans a memory hole, it is up to the architecture to ensure
3610 * the memory is not freed by the bootmem allocator. If possible
3611 * the range being registered will be merged with existing ranges.
3613 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3614 unsigned long end_pfn
)
3618 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3619 "Entering add_active_range(%d, %#lx, %#lx) "
3620 "%d entries of %d used\n",
3621 nid
, start_pfn
, end_pfn
,
3622 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3624 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3626 /* Merge with existing active regions if possible */
3627 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3628 if (early_node_map
[i
].nid
!= nid
)
3631 /* Skip if an existing region covers this new one */
3632 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3633 end_pfn
<= early_node_map
[i
].end_pfn
)
3636 /* Merge forward if suitable */
3637 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3638 end_pfn
> early_node_map
[i
].end_pfn
) {
3639 early_node_map
[i
].end_pfn
= end_pfn
;
3643 /* Merge backward if suitable */
3644 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3645 end_pfn
>= early_node_map
[i
].start_pfn
) {
3646 early_node_map
[i
].start_pfn
= start_pfn
;
3651 /* Check that early_node_map is large enough */
3652 if (i
>= MAX_ACTIVE_REGIONS
) {
3653 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3654 MAX_ACTIVE_REGIONS
);
3658 early_node_map
[i
].nid
= nid
;
3659 early_node_map
[i
].start_pfn
= start_pfn
;
3660 early_node_map
[i
].end_pfn
= end_pfn
;
3661 nr_nodemap_entries
= i
+ 1;
3665 * remove_active_range - Shrink an existing registered range of PFNs
3666 * @nid: The node id the range is on that should be shrunk
3667 * @start_pfn: The new PFN of the range
3668 * @end_pfn: The new PFN of the range
3670 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3671 * The map is kept near the end physical page range that has already been
3672 * registered. This function allows an arch to shrink an existing registered
3675 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3676 unsigned long end_pfn
)
3681 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3682 nid
, start_pfn
, end_pfn
);
3684 /* Find the old active region end and shrink */
3685 for_each_active_range_index_in_nid(i
, nid
) {
3686 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3687 early_node_map
[i
].end_pfn
<= end_pfn
) {
3689 early_node_map
[i
].start_pfn
= 0;
3690 early_node_map
[i
].end_pfn
= 0;
3694 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3695 early_node_map
[i
].end_pfn
> start_pfn
) {
3696 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3697 early_node_map
[i
].end_pfn
= start_pfn
;
3698 if (temp_end_pfn
> end_pfn
)
3699 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3702 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3703 early_node_map
[i
].end_pfn
> end_pfn
&&
3704 early_node_map
[i
].start_pfn
< end_pfn
) {
3705 early_node_map
[i
].start_pfn
= end_pfn
;
3713 /* remove the blank ones */
3714 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3715 if (early_node_map
[i
].nid
!= nid
)
3717 if (early_node_map
[i
].end_pfn
)
3719 /* we found it, get rid of it */
3720 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3721 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3722 sizeof(early_node_map
[j
]));
3723 j
= nr_nodemap_entries
- 1;
3724 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3725 nr_nodemap_entries
--;
3730 * remove_all_active_ranges - Remove all currently registered regions
3732 * During discovery, it may be found that a table like SRAT is invalid
3733 * and an alternative discovery method must be used. This function removes
3734 * all currently registered regions.
3736 void __init
remove_all_active_ranges(void)
3738 memset(early_node_map
, 0, sizeof(early_node_map
));
3739 nr_nodemap_entries
= 0;
3740 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3741 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3742 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3743 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3746 /* Compare two active node_active_regions */
3747 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3749 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3750 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3752 /* Done this way to avoid overflows */
3753 if (arange
->start_pfn
> brange
->start_pfn
)
3755 if (arange
->start_pfn
< brange
->start_pfn
)
3761 /* sort the node_map by start_pfn */
3762 static void __init
sort_node_map(void)
3764 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3765 sizeof(struct node_active_region
),
3766 cmp_node_active_region
, NULL
);
3769 /* Find the lowest pfn for a node */
3770 static unsigned long __init
find_min_pfn_for_node(int nid
)
3773 unsigned long min_pfn
= ULONG_MAX
;
3775 /* Assuming a sorted map, the first range found has the starting pfn */
3776 for_each_active_range_index_in_nid(i
, nid
)
3777 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3779 if (min_pfn
== ULONG_MAX
) {
3781 "Could not find start_pfn for node %d\n", nid
);
3789 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3791 * It returns the minimum PFN based on information provided via
3792 * add_active_range().
3794 unsigned long __init
find_min_pfn_with_active_regions(void)
3796 return find_min_pfn_for_node(MAX_NUMNODES
);
3800 * early_calculate_totalpages()
3801 * Sum pages in active regions for movable zone.
3802 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3804 static unsigned long __init
early_calculate_totalpages(void)
3807 unsigned long totalpages
= 0;
3809 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3810 unsigned long pages
= early_node_map
[i
].end_pfn
-
3811 early_node_map
[i
].start_pfn
;
3812 totalpages
+= pages
;
3814 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3820 * Find the PFN the Movable zone begins in each node. Kernel memory
3821 * is spread evenly between nodes as long as the nodes have enough
3822 * memory. When they don't, some nodes will have more kernelcore than
3825 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3828 unsigned long usable_startpfn
;
3829 unsigned long kernelcore_node
, kernelcore_remaining
;
3830 unsigned long totalpages
= early_calculate_totalpages();
3831 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3834 * If movablecore was specified, calculate what size of
3835 * kernelcore that corresponds so that memory usable for
3836 * any allocation type is evenly spread. If both kernelcore
3837 * and movablecore are specified, then the value of kernelcore
3838 * will be used for required_kernelcore if it's greater than
3839 * what movablecore would have allowed.
3841 if (required_movablecore
) {
3842 unsigned long corepages
;
3845 * Round-up so that ZONE_MOVABLE is at least as large as what
3846 * was requested by the user
3848 required_movablecore
=
3849 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3850 corepages
= totalpages
- required_movablecore
;
3852 required_kernelcore
= max(required_kernelcore
, corepages
);
3855 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3856 if (!required_kernelcore
)
3859 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3860 find_usable_zone_for_movable();
3861 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3864 /* Spread kernelcore memory as evenly as possible throughout nodes */
3865 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3866 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3868 * Recalculate kernelcore_node if the division per node
3869 * now exceeds what is necessary to satisfy the requested
3870 * amount of memory for the kernel
3872 if (required_kernelcore
< kernelcore_node
)
3873 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3876 * As the map is walked, we track how much memory is usable
3877 * by the kernel using kernelcore_remaining. When it is
3878 * 0, the rest of the node is usable by ZONE_MOVABLE
3880 kernelcore_remaining
= kernelcore_node
;
3882 /* Go through each range of PFNs within this node */
3883 for_each_active_range_index_in_nid(i
, nid
) {
3884 unsigned long start_pfn
, end_pfn
;
3885 unsigned long size_pages
;
3887 start_pfn
= max(early_node_map
[i
].start_pfn
,
3888 zone_movable_pfn
[nid
]);
3889 end_pfn
= early_node_map
[i
].end_pfn
;
3890 if (start_pfn
>= end_pfn
)
3893 /* Account for what is only usable for kernelcore */
3894 if (start_pfn
< usable_startpfn
) {
3895 unsigned long kernel_pages
;
3896 kernel_pages
= min(end_pfn
, usable_startpfn
)
3899 kernelcore_remaining
-= min(kernel_pages
,
3900 kernelcore_remaining
);
3901 required_kernelcore
-= min(kernel_pages
,
3902 required_kernelcore
);
3904 /* Continue if range is now fully accounted */
3905 if (end_pfn
<= usable_startpfn
) {
3908 * Push zone_movable_pfn to the end so
3909 * that if we have to rebalance
3910 * kernelcore across nodes, we will
3911 * not double account here
3913 zone_movable_pfn
[nid
] = end_pfn
;
3916 start_pfn
= usable_startpfn
;
3920 * The usable PFN range for ZONE_MOVABLE is from
3921 * start_pfn->end_pfn. Calculate size_pages as the
3922 * number of pages used as kernelcore
3924 size_pages
= end_pfn
- start_pfn
;
3925 if (size_pages
> kernelcore_remaining
)
3926 size_pages
= kernelcore_remaining
;
3927 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3930 * Some kernelcore has been met, update counts and
3931 * break if the kernelcore for this node has been
3934 required_kernelcore
-= min(required_kernelcore
,
3936 kernelcore_remaining
-= size_pages
;
3937 if (!kernelcore_remaining
)
3943 * If there is still required_kernelcore, we do another pass with one
3944 * less node in the count. This will push zone_movable_pfn[nid] further
3945 * along on the nodes that still have memory until kernelcore is
3949 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3952 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3953 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3954 zone_movable_pfn
[nid
] =
3955 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3958 /* Any regular memory on that node ? */
3959 static void check_for_regular_memory(pg_data_t
*pgdat
)
3961 #ifdef CONFIG_HIGHMEM
3962 enum zone_type zone_type
;
3964 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3965 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3966 if (zone
->present_pages
)
3967 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3973 * free_area_init_nodes - Initialise all pg_data_t and zone data
3974 * @max_zone_pfn: an array of max PFNs for each zone
3976 * This will call free_area_init_node() for each active node in the system.
3977 * Using the page ranges provided by add_active_range(), the size of each
3978 * zone in each node and their holes is calculated. If the maximum PFN
3979 * between two adjacent zones match, it is assumed that the zone is empty.
3980 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3981 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3982 * starts where the previous one ended. For example, ZONE_DMA32 starts
3983 * at arch_max_dma_pfn.
3985 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3990 /* Sort early_node_map as initialisation assumes it is sorted */
3993 /* Record where the zone boundaries are */
3994 memset(arch_zone_lowest_possible_pfn
, 0,
3995 sizeof(arch_zone_lowest_possible_pfn
));
3996 memset(arch_zone_highest_possible_pfn
, 0,
3997 sizeof(arch_zone_highest_possible_pfn
));
3998 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3999 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4000 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4001 if (i
== ZONE_MOVABLE
)
4003 arch_zone_lowest_possible_pfn
[i
] =
4004 arch_zone_highest_possible_pfn
[i
-1];
4005 arch_zone_highest_possible_pfn
[i
] =
4006 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4008 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4009 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4011 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4012 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4013 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4015 /* Print out the zone ranges */
4016 printk("Zone PFN ranges:\n");
4017 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4018 if (i
== ZONE_MOVABLE
)
4020 printk(" %-8s %0#10lx -> %0#10lx\n",
4022 arch_zone_lowest_possible_pfn
[i
],
4023 arch_zone_highest_possible_pfn
[i
]);
4026 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4027 printk("Movable zone start PFN for each node\n");
4028 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4029 if (zone_movable_pfn
[i
])
4030 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4033 /* Print out the early_node_map[] */
4034 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4035 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4036 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4037 early_node_map
[i
].start_pfn
,
4038 early_node_map
[i
].end_pfn
);
4040 /* Initialise every node */
4041 mminit_verify_pageflags_layout();
4042 setup_nr_node_ids();
4043 for_each_online_node(nid
) {
4044 pg_data_t
*pgdat
= NODE_DATA(nid
);
4045 free_area_init_node(nid
, NULL
,
4046 find_min_pfn_for_node(nid
), NULL
);
4048 /* Any memory on that node */
4049 if (pgdat
->node_present_pages
)
4050 node_set_state(nid
, N_HIGH_MEMORY
);
4051 check_for_regular_memory(pgdat
);
4055 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4057 unsigned long long coremem
;
4061 coremem
= memparse(p
, &p
);
4062 *core
= coremem
>> PAGE_SHIFT
;
4064 /* Paranoid check that UL is enough for the coremem value */
4065 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4071 * kernelcore=size sets the amount of memory for use for allocations that
4072 * cannot be reclaimed or migrated.
4074 static int __init
cmdline_parse_kernelcore(char *p
)
4076 return cmdline_parse_core(p
, &required_kernelcore
);
4080 * movablecore=size sets the amount of memory for use for allocations that
4081 * can be reclaimed or migrated.
4083 static int __init
cmdline_parse_movablecore(char *p
)
4085 return cmdline_parse_core(p
, &required_movablecore
);
4088 early_param("kernelcore", cmdline_parse_kernelcore
);
4089 early_param("movablecore", cmdline_parse_movablecore
);
4091 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4094 * set_dma_reserve - set the specified number of pages reserved in the first zone
4095 * @new_dma_reserve: The number of pages to mark reserved
4097 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4098 * In the DMA zone, a significant percentage may be consumed by kernel image
4099 * and other unfreeable allocations which can skew the watermarks badly. This
4100 * function may optionally be used to account for unfreeable pages in the
4101 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4102 * smaller per-cpu batchsize.
4104 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4106 dma_reserve
= new_dma_reserve
;
4109 #ifndef CONFIG_NEED_MULTIPLE_NODES
4110 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4111 EXPORT_SYMBOL(contig_page_data
);
4114 void __init
free_area_init(unsigned long *zones_size
)
4116 free_area_init_node(0, zones_size
,
4117 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4120 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4121 unsigned long action
, void *hcpu
)
4123 int cpu
= (unsigned long)hcpu
;
4125 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4129 * Spill the event counters of the dead processor
4130 * into the current processors event counters.
4131 * This artificially elevates the count of the current
4134 vm_events_fold_cpu(cpu
);
4137 * Zero the differential counters of the dead processor
4138 * so that the vm statistics are consistent.
4140 * This is only okay since the processor is dead and cannot
4141 * race with what we are doing.
4143 refresh_cpu_vm_stats(cpu
);
4148 void __init
page_alloc_init(void)
4150 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4154 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4155 * or min_free_kbytes changes.
4157 static void calculate_totalreserve_pages(void)
4159 struct pglist_data
*pgdat
;
4160 unsigned long reserve_pages
= 0;
4161 enum zone_type i
, j
;
4163 for_each_online_pgdat(pgdat
) {
4164 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4165 struct zone
*zone
= pgdat
->node_zones
+ i
;
4166 unsigned long max
= 0;
4168 /* Find valid and maximum lowmem_reserve in the zone */
4169 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4170 if (zone
->lowmem_reserve
[j
] > max
)
4171 max
= zone
->lowmem_reserve
[j
];
4174 /* we treat pages_high as reserved pages. */
4175 max
+= zone
->pages_high
;
4177 if (max
> zone
->present_pages
)
4178 max
= zone
->present_pages
;
4179 reserve_pages
+= max
;
4182 totalreserve_pages
= reserve_pages
;
4186 * setup_per_zone_lowmem_reserve - called whenever
4187 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4188 * has a correct pages reserved value, so an adequate number of
4189 * pages are left in the zone after a successful __alloc_pages().
4191 static void setup_per_zone_lowmem_reserve(void)
4193 struct pglist_data
*pgdat
;
4194 enum zone_type j
, idx
;
4196 for_each_online_pgdat(pgdat
) {
4197 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4198 struct zone
*zone
= pgdat
->node_zones
+ j
;
4199 unsigned long present_pages
= zone
->present_pages
;
4201 zone
->lowmem_reserve
[j
] = 0;
4205 struct zone
*lower_zone
;
4209 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4210 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4212 lower_zone
= pgdat
->node_zones
+ idx
;
4213 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4214 sysctl_lowmem_reserve_ratio
[idx
];
4215 present_pages
+= lower_zone
->present_pages
;
4220 /* update totalreserve_pages */
4221 calculate_totalreserve_pages();
4225 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4227 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4228 * with respect to min_free_kbytes.
4230 void setup_per_zone_pages_min(void)
4232 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4233 unsigned long lowmem_pages
= 0;
4235 unsigned long flags
;
4237 /* Calculate total number of !ZONE_HIGHMEM pages */
4238 for_each_zone(zone
) {
4239 if (!is_highmem(zone
))
4240 lowmem_pages
+= zone
->present_pages
;
4243 for_each_zone(zone
) {
4246 spin_lock_irqsave(&zone
->lru_lock
, flags
);
4247 tmp
= (u64
)pages_min
* zone
->present_pages
;
4248 do_div(tmp
, lowmem_pages
);
4249 if (is_highmem(zone
)) {
4251 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4252 * need highmem pages, so cap pages_min to a small
4255 * The (pages_high-pages_low) and (pages_low-pages_min)
4256 * deltas controls asynch page reclaim, and so should
4257 * not be capped for highmem.
4261 min_pages
= zone
->present_pages
/ 1024;
4262 if (min_pages
< SWAP_CLUSTER_MAX
)
4263 min_pages
= SWAP_CLUSTER_MAX
;
4264 if (min_pages
> 128)
4266 zone
->pages_min
= min_pages
;
4269 * If it's a lowmem zone, reserve a number of pages
4270 * proportionate to the zone's size.
4272 zone
->pages_min
= tmp
;
4275 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4276 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4277 setup_zone_migrate_reserve(zone
);
4278 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4281 /* update totalreserve_pages */
4282 calculate_totalreserve_pages();
4286 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4288 * The inactive anon list should be small enough that the VM never has to
4289 * do too much work, but large enough that each inactive page has a chance
4290 * to be referenced again before it is swapped out.
4292 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4293 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4294 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4295 * the anonymous pages are kept on the inactive list.
4298 * memory ratio inactive anon
4299 * -------------------------------------
4308 void setup_per_zone_inactive_ratio(void)
4312 for_each_zone(zone
) {
4313 unsigned int gb
, ratio
;
4315 /* Zone size in gigabytes */
4316 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4317 ratio
= int_sqrt(10 * gb
);
4321 zone
->inactive_ratio
= ratio
;
4326 * Initialise min_free_kbytes.
4328 * For small machines we want it small (128k min). For large machines
4329 * we want it large (64MB max). But it is not linear, because network
4330 * bandwidth does not increase linearly with machine size. We use
4332 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4333 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4349 static int __init
init_per_zone_pages_min(void)
4351 unsigned long lowmem_kbytes
;
4353 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4355 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4356 if (min_free_kbytes
< 128)
4357 min_free_kbytes
= 128;
4358 if (min_free_kbytes
> 65536)
4359 min_free_kbytes
= 65536;
4360 setup_per_zone_pages_min();
4361 setup_per_zone_lowmem_reserve();
4362 setup_per_zone_inactive_ratio();
4365 module_init(init_per_zone_pages_min
)
4368 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4369 * that we can call two helper functions whenever min_free_kbytes
4372 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4373 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4375 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4377 setup_per_zone_pages_min();
4382 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4383 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4388 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4393 zone
->min_unmapped_pages
= (zone
->present_pages
*
4394 sysctl_min_unmapped_ratio
) / 100;
4398 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4399 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4404 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4409 zone
->min_slab_pages
= (zone
->present_pages
*
4410 sysctl_min_slab_ratio
) / 100;
4416 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4417 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4418 * whenever sysctl_lowmem_reserve_ratio changes.
4420 * The reserve ratio obviously has absolutely no relation with the
4421 * pages_min watermarks. The lowmem reserve ratio can only make sense
4422 * if in function of the boot time zone sizes.
4424 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4425 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4427 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4428 setup_per_zone_lowmem_reserve();
4433 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4434 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4435 * can have before it gets flushed back to buddy allocator.
4438 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4439 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4445 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4446 if (!write
|| (ret
== -EINVAL
))
4448 for_each_zone(zone
) {
4449 for_each_online_cpu(cpu
) {
4451 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4452 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4458 int hashdist
= HASHDIST_DEFAULT
;
4461 static int __init
set_hashdist(char *str
)
4465 hashdist
= simple_strtoul(str
, &str
, 0);
4468 __setup("hashdist=", set_hashdist
);
4472 * allocate a large system hash table from bootmem
4473 * - it is assumed that the hash table must contain an exact power-of-2
4474 * quantity of entries
4475 * - limit is the number of hash buckets, not the total allocation size
4477 void *__init
alloc_large_system_hash(const char *tablename
,
4478 unsigned long bucketsize
,
4479 unsigned long numentries
,
4482 unsigned int *_hash_shift
,
4483 unsigned int *_hash_mask
,
4484 unsigned long limit
)
4486 unsigned long long max
= limit
;
4487 unsigned long log2qty
, size
;
4490 /* allow the kernel cmdline to have a say */
4492 /* round applicable memory size up to nearest megabyte */
4493 numentries
= nr_kernel_pages
;
4494 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4495 numentries
>>= 20 - PAGE_SHIFT
;
4496 numentries
<<= 20 - PAGE_SHIFT
;
4498 /* limit to 1 bucket per 2^scale bytes of low memory */
4499 if (scale
> PAGE_SHIFT
)
4500 numentries
>>= (scale
- PAGE_SHIFT
);
4502 numentries
<<= (PAGE_SHIFT
- scale
);
4504 /* Make sure we've got at least a 0-order allocation.. */
4505 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4506 numentries
= PAGE_SIZE
/ bucketsize
;
4508 numentries
= roundup_pow_of_two(numentries
);
4510 /* limit allocation size to 1/16 total memory by default */
4512 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4513 do_div(max
, bucketsize
);
4516 if (numentries
> max
)
4519 log2qty
= ilog2(numentries
);
4522 size
= bucketsize
<< log2qty
;
4523 if (flags
& HASH_EARLY
)
4524 table
= alloc_bootmem_nopanic(size
);
4526 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4528 unsigned long order
= get_order(size
);
4529 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4531 * If bucketsize is not a power-of-two, we may free
4532 * some pages at the end of hash table.
4535 unsigned long alloc_end
= (unsigned long)table
+
4536 (PAGE_SIZE
<< order
);
4537 unsigned long used
= (unsigned long)table
+
4539 split_page(virt_to_page(table
), order
);
4540 while (used
< alloc_end
) {
4546 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4549 panic("Failed to allocate %s hash table\n", tablename
);
4551 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4554 ilog2(size
) - PAGE_SHIFT
,
4558 *_hash_shift
= log2qty
;
4560 *_hash_mask
= (1 << log2qty
) - 1;
4565 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4566 struct page
*pfn_to_page(unsigned long pfn
)
4568 return __pfn_to_page(pfn
);
4570 unsigned long page_to_pfn(struct page
*page
)
4572 return __page_to_pfn(page
);
4574 EXPORT_SYMBOL(pfn_to_page
);
4575 EXPORT_SYMBOL(page_to_pfn
);
4576 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4578 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4579 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4582 #ifdef CONFIG_SPARSEMEM
4583 return __pfn_to_section(pfn
)->pageblock_flags
;
4585 return zone
->pageblock_flags
;
4586 #endif /* CONFIG_SPARSEMEM */
4589 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4591 #ifdef CONFIG_SPARSEMEM
4592 pfn
&= (PAGES_PER_SECTION
-1);
4593 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4595 pfn
= pfn
- zone
->zone_start_pfn
;
4596 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4597 #endif /* CONFIG_SPARSEMEM */
4601 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4602 * @page: The page within the block of interest
4603 * @start_bitidx: The first bit of interest to retrieve
4604 * @end_bitidx: The last bit of interest
4605 * returns pageblock_bits flags
4607 unsigned long get_pageblock_flags_group(struct page
*page
,
4608 int start_bitidx
, int end_bitidx
)
4611 unsigned long *bitmap
;
4612 unsigned long pfn
, bitidx
;
4613 unsigned long flags
= 0;
4614 unsigned long value
= 1;
4616 zone
= page_zone(page
);
4617 pfn
= page_to_pfn(page
);
4618 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4619 bitidx
= pfn_to_bitidx(zone
, pfn
);
4621 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4622 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4629 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4630 * @page: The page within the block of interest
4631 * @start_bitidx: The first bit of interest
4632 * @end_bitidx: The last bit of interest
4633 * @flags: The flags to set
4635 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4636 int start_bitidx
, int end_bitidx
)
4639 unsigned long *bitmap
;
4640 unsigned long pfn
, bitidx
;
4641 unsigned long value
= 1;
4643 zone
= page_zone(page
);
4644 pfn
= page_to_pfn(page
);
4645 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4646 bitidx
= pfn_to_bitidx(zone
, pfn
);
4647 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4648 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4650 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4652 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4654 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4658 * This is designed as sub function...plz see page_isolation.c also.
4659 * set/clear page block's type to be ISOLATE.
4660 * page allocater never alloc memory from ISOLATE block.
4663 int set_migratetype_isolate(struct page
*page
)
4666 unsigned long flags
;
4669 zone
= page_zone(page
);
4670 spin_lock_irqsave(&zone
->lock
, flags
);
4672 * In future, more migrate types will be able to be isolation target.
4674 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4676 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4677 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4680 spin_unlock_irqrestore(&zone
->lock
, flags
);
4686 void unset_migratetype_isolate(struct page
*page
)
4689 unsigned long flags
;
4690 zone
= page_zone(page
);
4691 spin_lock_irqsave(&zone
->lock
, flags
);
4692 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4694 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4695 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4697 spin_unlock_irqrestore(&zone
->lock
, flags
);
4700 #ifdef CONFIG_MEMORY_HOTREMOVE
4702 * All pages in the range must be isolated before calling this.
4705 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4711 unsigned long flags
;
4712 /* find the first valid pfn */
4713 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4718 zone
= page_zone(pfn_to_page(pfn
));
4719 spin_lock_irqsave(&zone
->lock
, flags
);
4721 while (pfn
< end_pfn
) {
4722 if (!pfn_valid(pfn
)) {
4726 page
= pfn_to_page(pfn
);
4727 BUG_ON(page_count(page
));
4728 BUG_ON(!PageBuddy(page
));
4729 order
= page_order(page
);
4730 #ifdef CONFIG_DEBUG_VM
4731 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4732 pfn
, 1 << order
, end_pfn
);
4734 list_del(&page
->lru
);
4735 rmv_page_order(page
);
4736 zone
->free_area
[order
].nr_free
--;
4737 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4739 for (i
= 0; i
< (1 << order
); i
++)
4740 SetPageReserved((page
+i
));
4741 pfn
+= (1 << order
);
4743 spin_unlock_irqrestore(&zone
->lock
, flags
);