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 if (unlikely(page_mapcount(page
) |
458 (page
->mapping
!= NULL
) |
459 (page_get_page_cgroup(page
) != NULL
) |
460 (page_count(page
) != 0) |
461 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)))
464 __ClearPageDirty(page
);
465 if (PageSwapBacked(page
))
466 __ClearPageSwapBacked(page
);
468 * For now, we report if PG_reserved was found set, but do not
469 * clear it, and do not free the page. But we shall soon need
470 * to do more, for when the ZERO_PAGE count wraps negative.
472 return PageReserved(page
);
476 * Frees a list of pages.
477 * Assumes all pages on list are in same zone, and of same order.
478 * count is the number of pages to free.
480 * If the zone was previously in an "all pages pinned" state then look to
481 * see if this freeing clears that state.
483 * And clear the zone's pages_scanned counter, to hold off the "all pages are
484 * pinned" detection logic.
486 static void free_pages_bulk(struct zone
*zone
, int count
,
487 struct list_head
*list
, int order
)
489 spin_lock(&zone
->lock
);
490 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
491 zone
->pages_scanned
= 0;
495 VM_BUG_ON(list_empty(list
));
496 page
= list_entry(list
->prev
, struct page
, lru
);
497 /* have to delete it as __free_one_page list manipulates */
498 list_del(&page
->lru
);
499 __free_one_page(page
, zone
, order
);
501 spin_unlock(&zone
->lock
);
504 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
506 spin_lock(&zone
->lock
);
507 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
508 zone
->pages_scanned
= 0;
509 __free_one_page(page
, zone
, order
);
510 spin_unlock(&zone
->lock
);
513 static void __free_pages_ok(struct page
*page
, unsigned int order
)
519 for (i
= 0 ; i
< (1 << order
) ; ++i
)
520 reserved
+= free_pages_check(page
+ i
);
524 if (!PageHighMem(page
)) {
525 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
526 debug_check_no_obj_freed(page_address(page
),
529 arch_free_page(page
, order
);
530 kernel_map_pages(page
, 1 << order
, 0);
532 local_irq_save(flags
);
533 __count_vm_events(PGFREE
, 1 << order
);
534 free_one_page(page_zone(page
), page
, order
);
535 local_irq_restore(flags
);
539 * permit the bootmem allocator to evade page validation on high-order frees
541 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
544 __ClearPageReserved(page
);
545 set_page_count(page
, 0);
546 set_page_refcounted(page
);
552 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
553 struct page
*p
= &page
[loop
];
555 if (loop
+ 1 < BITS_PER_LONG
)
557 __ClearPageReserved(p
);
558 set_page_count(p
, 0);
561 set_page_refcounted(page
);
562 __free_pages(page
, order
);
568 * The order of subdivision here is critical for the IO subsystem.
569 * Please do not alter this order without good reasons and regression
570 * testing. Specifically, as large blocks of memory are subdivided,
571 * the order in which smaller blocks are delivered depends on the order
572 * they're subdivided in this function. This is the primary factor
573 * influencing the order in which pages are delivered to the IO
574 * subsystem according to empirical testing, and this is also justified
575 * by considering the behavior of a buddy system containing a single
576 * large block of memory acted on by a series of small allocations.
577 * This behavior is a critical factor in sglist merging's success.
581 static inline void expand(struct zone
*zone
, struct page
*page
,
582 int low
, int high
, struct free_area
*area
,
585 unsigned long size
= 1 << high
;
591 VM_BUG_ON(bad_range(zone
, &page
[size
]));
592 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
594 set_page_order(&page
[size
], high
);
599 * This page is about to be returned from the page allocator
601 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
603 if (unlikely(page_mapcount(page
) |
604 (page
->mapping
!= NULL
) |
605 (page_get_page_cgroup(page
) != NULL
) |
606 (page_count(page
) != 0) |
607 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)))
611 * For now, we report if PG_reserved was found set, but do not
612 * clear it, and do not allocate the page: as a safety net.
614 if (PageReserved(page
))
617 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_reclaim
|
618 1 << PG_referenced
| 1 << PG_arch_1
|
619 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
);
620 set_page_private(page
, 0);
621 set_page_refcounted(page
);
623 arch_alloc_page(page
, order
);
624 kernel_map_pages(page
, 1 << order
, 1);
626 if (gfp_flags
& __GFP_ZERO
)
627 prep_zero_page(page
, order
, gfp_flags
);
629 if (order
&& (gfp_flags
& __GFP_COMP
))
630 prep_compound_page(page
, order
);
636 * Go through the free lists for the given migratetype and remove
637 * the smallest available page from the freelists
639 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
642 unsigned int current_order
;
643 struct free_area
* area
;
646 /* Find a page of the appropriate size in the preferred list */
647 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
648 area
= &(zone
->free_area
[current_order
]);
649 if (list_empty(&area
->free_list
[migratetype
]))
652 page
= list_entry(area
->free_list
[migratetype
].next
,
654 list_del(&page
->lru
);
655 rmv_page_order(page
);
657 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
658 expand(zone
, page
, order
, current_order
, area
, migratetype
);
667 * This array describes the order lists are fallen back to when
668 * the free lists for the desirable migrate type are depleted
670 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
671 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
672 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
673 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
674 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
678 * Move the free pages in a range to the free lists of the requested type.
679 * Note that start_page and end_pages are not aligned on a pageblock
680 * boundary. If alignment is required, use move_freepages_block()
682 static int move_freepages(struct zone
*zone
,
683 struct page
*start_page
, struct page
*end_page
,
690 #ifndef CONFIG_HOLES_IN_ZONE
692 * page_zone is not safe to call in this context when
693 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
694 * anyway as we check zone boundaries in move_freepages_block().
695 * Remove at a later date when no bug reports exist related to
696 * grouping pages by mobility
698 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
701 for (page
= start_page
; page
<= end_page
;) {
702 /* Make sure we are not inadvertently changing nodes */
703 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
705 if (!pfn_valid_within(page_to_pfn(page
))) {
710 if (!PageBuddy(page
)) {
715 order
= page_order(page
);
716 list_del(&page
->lru
);
718 &zone
->free_area
[order
].free_list
[migratetype
]);
720 pages_moved
+= 1 << order
;
726 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
729 unsigned long start_pfn
, end_pfn
;
730 struct page
*start_page
, *end_page
;
732 start_pfn
= page_to_pfn(page
);
733 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
734 start_page
= pfn_to_page(start_pfn
);
735 end_page
= start_page
+ pageblock_nr_pages
- 1;
736 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
738 /* Do not cross zone boundaries */
739 if (start_pfn
< zone
->zone_start_pfn
)
741 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
744 return move_freepages(zone
, start_page
, end_page
, migratetype
);
747 /* Remove an element from the buddy allocator from the fallback list */
748 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
749 int start_migratetype
)
751 struct free_area
* area
;
756 /* Find the largest possible block of pages in the other list */
757 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
759 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
760 migratetype
= fallbacks
[start_migratetype
][i
];
762 /* MIGRATE_RESERVE handled later if necessary */
763 if (migratetype
== MIGRATE_RESERVE
)
766 area
= &(zone
->free_area
[current_order
]);
767 if (list_empty(&area
->free_list
[migratetype
]))
770 page
= list_entry(area
->free_list
[migratetype
].next
,
775 * If breaking a large block of pages, move all free
776 * pages to the preferred allocation list. If falling
777 * back for a reclaimable kernel allocation, be more
778 * agressive about taking ownership of free pages
780 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
781 start_migratetype
== MIGRATE_RECLAIMABLE
) {
783 pages
= move_freepages_block(zone
, page
,
786 /* Claim the whole block if over half of it is free */
787 if (pages
>= (1 << (pageblock_order
-1)))
788 set_pageblock_migratetype(page
,
791 migratetype
= start_migratetype
;
794 /* Remove the page from the freelists */
795 list_del(&page
->lru
);
796 rmv_page_order(page
);
797 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
800 if (current_order
== pageblock_order
)
801 set_pageblock_migratetype(page
,
804 expand(zone
, page
, order
, current_order
, area
, migratetype
);
809 /* Use MIGRATE_RESERVE rather than fail an allocation */
810 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
814 * Do the hard work of removing an element from the buddy allocator.
815 * Call me with the zone->lock already held.
817 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
822 page
= __rmqueue_smallest(zone
, order
, migratetype
);
825 page
= __rmqueue_fallback(zone
, order
, migratetype
);
831 * Obtain a specified number of elements from the buddy allocator, all under
832 * a single hold of the lock, for efficiency. Add them to the supplied list.
833 * Returns the number of new pages which were placed at *list.
835 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
836 unsigned long count
, struct list_head
*list
,
841 spin_lock(&zone
->lock
);
842 for (i
= 0; i
< count
; ++i
) {
843 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
844 if (unlikely(page
== NULL
))
848 * Split buddy pages returned by expand() are received here
849 * in physical page order. The page is added to the callers and
850 * list and the list head then moves forward. From the callers
851 * perspective, the linked list is ordered by page number in
852 * some conditions. This is useful for IO devices that can
853 * merge IO requests if the physical pages are ordered
856 list_add(&page
->lru
, list
);
857 set_page_private(page
, migratetype
);
860 spin_unlock(&zone
->lock
);
866 * Called from the vmstat counter updater to drain pagesets of this
867 * currently executing processor on remote nodes after they have
870 * Note that this function must be called with the thread pinned to
871 * a single processor.
873 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
878 local_irq_save(flags
);
879 if (pcp
->count
>= pcp
->batch
)
880 to_drain
= pcp
->batch
;
882 to_drain
= pcp
->count
;
883 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
884 pcp
->count
-= to_drain
;
885 local_irq_restore(flags
);
890 * Drain pages of the indicated processor.
892 * The processor must either be the current processor and the
893 * thread pinned to the current processor or a processor that
896 static void drain_pages(unsigned int cpu
)
901 for_each_zone(zone
) {
902 struct per_cpu_pageset
*pset
;
903 struct per_cpu_pages
*pcp
;
905 if (!populated_zone(zone
))
908 pset
= zone_pcp(zone
, cpu
);
911 local_irq_save(flags
);
912 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
914 local_irq_restore(flags
);
919 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
921 void drain_local_pages(void *arg
)
923 drain_pages(smp_processor_id());
927 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
929 void drain_all_pages(void)
931 on_each_cpu(drain_local_pages
, NULL
, 1);
934 #ifdef CONFIG_HIBERNATION
936 void mark_free_pages(struct zone
*zone
)
938 unsigned long pfn
, max_zone_pfn
;
941 struct list_head
*curr
;
943 if (!zone
->spanned_pages
)
946 spin_lock_irqsave(&zone
->lock
, flags
);
948 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
949 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
950 if (pfn_valid(pfn
)) {
951 struct page
*page
= pfn_to_page(pfn
);
953 if (!swsusp_page_is_forbidden(page
))
954 swsusp_unset_page_free(page
);
957 for_each_migratetype_order(order
, t
) {
958 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
961 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
962 for (i
= 0; i
< (1UL << order
); i
++)
963 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
966 spin_unlock_irqrestore(&zone
->lock
, flags
);
968 #endif /* CONFIG_PM */
971 * Free a 0-order page
973 static void free_hot_cold_page(struct page
*page
, int cold
)
975 struct zone
*zone
= page_zone(page
);
976 struct per_cpu_pages
*pcp
;
980 page
->mapping
= NULL
;
981 if (free_pages_check(page
))
984 if (!PageHighMem(page
)) {
985 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
986 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
988 arch_free_page(page
, 0);
989 kernel_map_pages(page
, 1, 0);
991 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
992 local_irq_save(flags
);
993 __count_vm_event(PGFREE
);
995 list_add_tail(&page
->lru
, &pcp
->list
);
997 list_add(&page
->lru
, &pcp
->list
);
998 set_page_private(page
, get_pageblock_migratetype(page
));
1000 if (pcp
->count
>= pcp
->high
) {
1001 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1002 pcp
->count
-= pcp
->batch
;
1004 local_irq_restore(flags
);
1008 void free_hot_page(struct page
*page
)
1010 free_hot_cold_page(page
, 0);
1013 void free_cold_page(struct page
*page
)
1015 free_hot_cold_page(page
, 1);
1019 * split_page takes a non-compound higher-order page, and splits it into
1020 * n (1<<order) sub-pages: page[0..n]
1021 * Each sub-page must be freed individually.
1023 * Note: this is probably too low level an operation for use in drivers.
1024 * Please consult with lkml before using this in your driver.
1026 void split_page(struct page
*page
, unsigned int order
)
1030 VM_BUG_ON(PageCompound(page
));
1031 VM_BUG_ON(!page_count(page
));
1032 for (i
= 1; i
< (1 << order
); i
++)
1033 set_page_refcounted(page
+ i
);
1037 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1038 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1041 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1042 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1044 unsigned long flags
;
1046 int cold
= !!(gfp_flags
& __GFP_COLD
);
1048 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1052 if (likely(order
== 0)) {
1053 struct per_cpu_pages
*pcp
;
1055 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1056 local_irq_save(flags
);
1058 pcp
->count
= rmqueue_bulk(zone
, 0,
1059 pcp
->batch
, &pcp
->list
, migratetype
);
1060 if (unlikely(!pcp
->count
))
1064 /* Find a page of the appropriate migrate type */
1066 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1067 if (page_private(page
) == migratetype
)
1070 list_for_each_entry(page
, &pcp
->list
, lru
)
1071 if (page_private(page
) == migratetype
)
1075 /* Allocate more to the pcp list if necessary */
1076 if (unlikely(&page
->lru
== &pcp
->list
)) {
1077 pcp
->count
+= rmqueue_bulk(zone
, 0,
1078 pcp
->batch
, &pcp
->list
, migratetype
);
1079 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1082 list_del(&page
->lru
);
1085 spin_lock_irqsave(&zone
->lock
, flags
);
1086 page
= __rmqueue(zone
, order
, migratetype
);
1087 spin_unlock(&zone
->lock
);
1092 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1093 zone_statistics(preferred_zone
, zone
);
1094 local_irq_restore(flags
);
1097 VM_BUG_ON(bad_range(zone
, page
));
1098 if (prep_new_page(page
, order
, gfp_flags
))
1103 local_irq_restore(flags
);
1108 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1109 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1110 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1111 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1112 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1113 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1114 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1116 #ifdef CONFIG_FAIL_PAGE_ALLOC
1118 static struct fail_page_alloc_attr
{
1119 struct fault_attr attr
;
1121 u32 ignore_gfp_highmem
;
1122 u32 ignore_gfp_wait
;
1125 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1127 struct dentry
*ignore_gfp_highmem_file
;
1128 struct dentry
*ignore_gfp_wait_file
;
1129 struct dentry
*min_order_file
;
1131 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1133 } fail_page_alloc
= {
1134 .attr
= FAULT_ATTR_INITIALIZER
,
1135 .ignore_gfp_wait
= 1,
1136 .ignore_gfp_highmem
= 1,
1140 static int __init
setup_fail_page_alloc(char *str
)
1142 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1144 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1146 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1148 if (order
< fail_page_alloc
.min_order
)
1150 if (gfp_mask
& __GFP_NOFAIL
)
1152 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1154 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1157 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1160 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1162 static int __init
fail_page_alloc_debugfs(void)
1164 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1168 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1172 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1174 fail_page_alloc
.ignore_gfp_wait_file
=
1175 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1176 &fail_page_alloc
.ignore_gfp_wait
);
1178 fail_page_alloc
.ignore_gfp_highmem_file
=
1179 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1180 &fail_page_alloc
.ignore_gfp_highmem
);
1181 fail_page_alloc
.min_order_file
=
1182 debugfs_create_u32("min-order", mode
, dir
,
1183 &fail_page_alloc
.min_order
);
1185 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1186 !fail_page_alloc
.ignore_gfp_highmem_file
||
1187 !fail_page_alloc
.min_order_file
) {
1189 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1190 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1191 debugfs_remove(fail_page_alloc
.min_order_file
);
1192 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1198 late_initcall(fail_page_alloc_debugfs
);
1200 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1202 #else /* CONFIG_FAIL_PAGE_ALLOC */
1204 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1209 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1212 * Return 1 if free pages are above 'mark'. This takes into account the order
1213 * of the allocation.
1215 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1216 int classzone_idx
, int alloc_flags
)
1218 /* free_pages my go negative - that's OK */
1220 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1223 if (alloc_flags
& ALLOC_HIGH
)
1225 if (alloc_flags
& ALLOC_HARDER
)
1228 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1230 for (o
= 0; o
< order
; o
++) {
1231 /* At the next order, this order's pages become unavailable */
1232 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1234 /* Require fewer higher order pages to be free */
1237 if (free_pages
<= min
)
1245 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1246 * skip over zones that are not allowed by the cpuset, or that have
1247 * been recently (in last second) found to be nearly full. See further
1248 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1249 * that have to skip over a lot of full or unallowed zones.
1251 * If the zonelist cache is present in the passed in zonelist, then
1252 * returns a pointer to the allowed node mask (either the current
1253 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1255 * If the zonelist cache is not available for this zonelist, does
1256 * nothing and returns NULL.
1258 * If the fullzones BITMAP in the zonelist cache is stale (more than
1259 * a second since last zap'd) then we zap it out (clear its bits.)
1261 * We hold off even calling zlc_setup, until after we've checked the
1262 * first zone in the zonelist, on the theory that most allocations will
1263 * be satisfied from that first zone, so best to examine that zone as
1264 * quickly as we can.
1266 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1268 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1269 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1271 zlc
= zonelist
->zlcache_ptr
;
1275 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1276 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1277 zlc
->last_full_zap
= jiffies
;
1280 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1281 &cpuset_current_mems_allowed
:
1282 &node_states
[N_HIGH_MEMORY
];
1283 return allowednodes
;
1287 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1288 * if it is worth looking at further for free memory:
1289 * 1) Check that the zone isn't thought to be full (doesn't have its
1290 * bit set in the zonelist_cache fullzones BITMAP).
1291 * 2) Check that the zones node (obtained from the zonelist_cache
1292 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1293 * Return true (non-zero) if zone is worth looking at further, or
1294 * else return false (zero) if it is not.
1296 * This check -ignores- the distinction between various watermarks,
1297 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1298 * found to be full for any variation of these watermarks, it will
1299 * be considered full for up to one second by all requests, unless
1300 * we are so low on memory on all allowed nodes that we are forced
1301 * into the second scan of the zonelist.
1303 * In the second scan we ignore this zonelist cache and exactly
1304 * apply the watermarks to all zones, even it is slower to do so.
1305 * We are low on memory in the second scan, and should leave no stone
1306 * unturned looking for a free page.
1308 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1309 nodemask_t
*allowednodes
)
1311 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1312 int i
; /* index of *z in zonelist zones */
1313 int n
; /* node that zone *z is on */
1315 zlc
= zonelist
->zlcache_ptr
;
1319 i
= z
- zonelist
->_zonerefs
;
1322 /* This zone is worth trying if it is allowed but not full */
1323 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1327 * Given 'z' scanning a zonelist, set the corresponding bit in
1328 * zlc->fullzones, so that subsequent attempts to allocate a page
1329 * from that zone don't waste time re-examining it.
1331 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1333 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1334 int i
; /* index of *z in zonelist zones */
1336 zlc
= zonelist
->zlcache_ptr
;
1340 i
= z
- zonelist
->_zonerefs
;
1342 set_bit(i
, zlc
->fullzones
);
1345 #else /* CONFIG_NUMA */
1347 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1352 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1353 nodemask_t
*allowednodes
)
1358 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1361 #endif /* CONFIG_NUMA */
1364 * get_page_from_freelist goes through the zonelist trying to allocate
1367 static struct page
*
1368 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1369 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1372 struct page
*page
= NULL
;
1374 struct zone
*zone
, *preferred_zone
;
1375 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1376 int zlc_active
= 0; /* set if using zonelist_cache */
1377 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1379 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1381 if (!preferred_zone
)
1384 classzone_idx
= zone_idx(preferred_zone
);
1388 * Scan zonelist, looking for a zone with enough free.
1389 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1391 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1392 high_zoneidx
, nodemask
) {
1393 if (NUMA_BUILD
&& zlc_active
&&
1394 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1396 if ((alloc_flags
& ALLOC_CPUSET
) &&
1397 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1400 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1402 if (alloc_flags
& ALLOC_WMARK_MIN
)
1403 mark
= zone
->pages_min
;
1404 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1405 mark
= zone
->pages_low
;
1407 mark
= zone
->pages_high
;
1408 if (!zone_watermark_ok(zone
, order
, mark
,
1409 classzone_idx
, alloc_flags
)) {
1410 if (!zone_reclaim_mode
||
1411 !zone_reclaim(zone
, gfp_mask
, order
))
1412 goto this_zone_full
;
1416 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1421 zlc_mark_zone_full(zonelist
, z
);
1423 if (NUMA_BUILD
&& !did_zlc_setup
) {
1424 /* we do zlc_setup after the first zone is tried */
1425 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1431 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1432 /* Disable zlc cache for second zonelist scan */
1440 * This is the 'heart' of the zoned buddy allocator.
1443 __alloc_pages_internal(gfp_t gfp_mask
, unsigned int order
,
1444 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1446 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1447 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1451 struct reclaim_state reclaim_state
;
1452 struct task_struct
*p
= current
;
1455 unsigned long did_some_progress
;
1456 unsigned long pages_reclaimed
= 0;
1458 might_sleep_if(wait
);
1460 if (should_fail_alloc_page(gfp_mask
, order
))
1464 z
= zonelist
->_zonerefs
; /* the list of zones suitable for gfp_mask */
1466 if (unlikely(!z
->zone
)) {
1468 * Happens if we have an empty zonelist as a result of
1469 * GFP_THISNODE being used on a memoryless node
1474 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1475 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1480 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1481 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1482 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1483 * using a larger set of nodes after it has established that the
1484 * allowed per node queues are empty and that nodes are
1487 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1490 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1491 wakeup_kswapd(zone
, order
);
1494 * OK, we're below the kswapd watermark and have kicked background
1495 * reclaim. Now things get more complex, so set up alloc_flags according
1496 * to how we want to proceed.
1498 * The caller may dip into page reserves a bit more if the caller
1499 * cannot run direct reclaim, or if the caller has realtime scheduling
1500 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1501 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1503 alloc_flags
= ALLOC_WMARK_MIN
;
1504 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1505 alloc_flags
|= ALLOC_HARDER
;
1506 if (gfp_mask
& __GFP_HIGH
)
1507 alloc_flags
|= ALLOC_HIGH
;
1509 alloc_flags
|= ALLOC_CPUSET
;
1512 * Go through the zonelist again. Let __GFP_HIGH and allocations
1513 * coming from realtime tasks go deeper into reserves.
1515 * This is the last chance, in general, before the goto nopage.
1516 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1517 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1519 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1520 high_zoneidx
, alloc_flags
);
1524 /* This allocation should allow future memory freeing. */
1527 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1528 && !in_interrupt()) {
1529 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1531 /* go through the zonelist yet again, ignoring mins */
1532 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1533 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1536 if (gfp_mask
& __GFP_NOFAIL
) {
1537 congestion_wait(WRITE
, HZ
/50);
1544 /* Atomic allocations - we can't balance anything */
1550 /* We now go into synchronous reclaim */
1551 cpuset_memory_pressure_bump();
1552 p
->flags
|= PF_MEMALLOC
;
1553 reclaim_state
.reclaimed_slab
= 0;
1554 p
->reclaim_state
= &reclaim_state
;
1556 did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
);
1558 p
->reclaim_state
= NULL
;
1559 p
->flags
&= ~PF_MEMALLOC
;
1566 if (likely(did_some_progress
)) {
1567 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1568 zonelist
, high_zoneidx
, alloc_flags
);
1571 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1572 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1573 schedule_timeout_uninterruptible(1);
1578 * Go through the zonelist yet one more time, keep
1579 * very high watermark here, this is only to catch
1580 * a parallel oom killing, we must fail if we're still
1581 * under heavy pressure.
1583 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1584 order
, zonelist
, high_zoneidx
,
1585 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1587 clear_zonelist_oom(zonelist
, gfp_mask
);
1591 /* The OOM killer will not help higher order allocs so fail */
1592 if (order
> PAGE_ALLOC_COSTLY_ORDER
) {
1593 clear_zonelist_oom(zonelist
, gfp_mask
);
1597 out_of_memory(zonelist
, gfp_mask
, order
);
1598 clear_zonelist_oom(zonelist
, gfp_mask
);
1603 * Don't let big-order allocations loop unless the caller explicitly
1604 * requests that. Wait for some write requests to complete then retry.
1606 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1607 * means __GFP_NOFAIL, but that may not be true in other
1610 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1611 * specified, then we retry until we no longer reclaim any pages
1612 * (above), or we've reclaimed an order of pages at least as
1613 * large as the allocation's order. In both cases, if the
1614 * allocation still fails, we stop retrying.
1616 pages_reclaimed
+= did_some_progress
;
1618 if (!(gfp_mask
& __GFP_NORETRY
)) {
1619 if (order
<= PAGE_ALLOC_COSTLY_ORDER
) {
1622 if (gfp_mask
& __GFP_REPEAT
&&
1623 pages_reclaimed
< (1 << order
))
1626 if (gfp_mask
& __GFP_NOFAIL
)
1630 congestion_wait(WRITE
, HZ
/50);
1635 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1636 printk(KERN_WARNING
"%s: page allocation failure."
1637 " order:%d, mode:0x%x\n",
1638 p
->comm
, order
, gfp_mask
);
1645 EXPORT_SYMBOL(__alloc_pages_internal
);
1648 * Common helper functions.
1650 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1653 page
= alloc_pages(gfp_mask
, order
);
1656 return (unsigned long) page_address(page
);
1659 EXPORT_SYMBOL(__get_free_pages
);
1661 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1666 * get_zeroed_page() returns a 32-bit address, which cannot represent
1669 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1671 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1673 return (unsigned long) page_address(page
);
1677 EXPORT_SYMBOL(get_zeroed_page
);
1679 void __pagevec_free(struct pagevec
*pvec
)
1681 int i
= pagevec_count(pvec
);
1684 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1687 void __free_pages(struct page
*page
, unsigned int order
)
1689 if (put_page_testzero(page
)) {
1691 free_hot_page(page
);
1693 __free_pages_ok(page
, order
);
1697 EXPORT_SYMBOL(__free_pages
);
1699 void free_pages(unsigned long addr
, unsigned int order
)
1702 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1703 __free_pages(virt_to_page((void *)addr
), order
);
1707 EXPORT_SYMBOL(free_pages
);
1710 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1711 * @size: the number of bytes to allocate
1712 * @gfp_mask: GFP flags for the allocation
1714 * This function is similar to alloc_pages(), except that it allocates the
1715 * minimum number of pages to satisfy the request. alloc_pages() can only
1716 * allocate memory in power-of-two pages.
1718 * This function is also limited by MAX_ORDER.
1720 * Memory allocated by this function must be released by free_pages_exact().
1722 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1724 unsigned int order
= get_order(size
);
1727 addr
= __get_free_pages(gfp_mask
, order
);
1729 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1730 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1732 split_page(virt_to_page(addr
), order
);
1733 while (used
< alloc_end
) {
1739 return (void *)addr
;
1741 EXPORT_SYMBOL(alloc_pages_exact
);
1744 * free_pages_exact - release memory allocated via alloc_pages_exact()
1745 * @virt: the value returned by alloc_pages_exact.
1746 * @size: size of allocation, same value as passed to alloc_pages_exact().
1748 * Release the memory allocated by a previous call to alloc_pages_exact.
1750 void free_pages_exact(void *virt
, size_t size
)
1752 unsigned long addr
= (unsigned long)virt
;
1753 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1755 while (addr
< end
) {
1760 EXPORT_SYMBOL(free_pages_exact
);
1762 static unsigned int nr_free_zone_pages(int offset
)
1767 /* Just pick one node, since fallback list is circular */
1768 unsigned int sum
= 0;
1770 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1772 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1773 unsigned long size
= zone
->present_pages
;
1774 unsigned long high
= zone
->pages_high
;
1783 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1785 unsigned int nr_free_buffer_pages(void)
1787 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1789 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1792 * Amount of free RAM allocatable within all zones
1794 unsigned int nr_free_pagecache_pages(void)
1796 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1799 static inline void show_node(struct zone
*zone
)
1802 printk("Node %d ", zone_to_nid(zone
));
1805 void si_meminfo(struct sysinfo
*val
)
1807 val
->totalram
= totalram_pages
;
1809 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1810 val
->bufferram
= nr_blockdev_pages();
1811 val
->totalhigh
= totalhigh_pages
;
1812 val
->freehigh
= nr_free_highpages();
1813 val
->mem_unit
= PAGE_SIZE
;
1816 EXPORT_SYMBOL(si_meminfo
);
1819 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1821 pg_data_t
*pgdat
= NODE_DATA(nid
);
1823 val
->totalram
= pgdat
->node_present_pages
;
1824 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1825 #ifdef CONFIG_HIGHMEM
1826 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1827 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1833 val
->mem_unit
= PAGE_SIZE
;
1837 #define K(x) ((x) << (PAGE_SHIFT-10))
1840 * Show free area list (used inside shift_scroll-lock stuff)
1841 * We also calculate the percentage fragmentation. We do this by counting the
1842 * memory on each free list with the exception of the first item on the list.
1844 void show_free_areas(void)
1849 for_each_zone(zone
) {
1850 if (!populated_zone(zone
))
1854 printk("%s per-cpu:\n", zone
->name
);
1856 for_each_online_cpu(cpu
) {
1857 struct per_cpu_pageset
*pageset
;
1859 pageset
= zone_pcp(zone
, cpu
);
1861 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1862 cpu
, pageset
->pcp
.high
,
1863 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1867 printk("Active_anon:%lu active_file:%lu inactive_anon%lu\n"
1868 " inactive_file:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1869 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1870 global_page_state(NR_ACTIVE_ANON
),
1871 global_page_state(NR_ACTIVE_FILE
),
1872 global_page_state(NR_INACTIVE_ANON
),
1873 global_page_state(NR_INACTIVE_FILE
),
1874 global_page_state(NR_FILE_DIRTY
),
1875 global_page_state(NR_WRITEBACK
),
1876 global_page_state(NR_UNSTABLE_NFS
),
1877 global_page_state(NR_FREE_PAGES
),
1878 global_page_state(NR_SLAB_RECLAIMABLE
) +
1879 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1880 global_page_state(NR_FILE_MAPPED
),
1881 global_page_state(NR_PAGETABLE
),
1882 global_page_state(NR_BOUNCE
));
1884 for_each_zone(zone
) {
1887 if (!populated_zone(zone
))
1896 " active_anon:%lukB"
1897 " inactive_anon:%lukB"
1898 " active_file:%lukB"
1899 " inactive_file:%lukB"
1901 " pages_scanned:%lu"
1902 " all_unreclaimable? %s"
1905 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1908 K(zone
->pages_high
),
1909 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
1910 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
1911 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
1912 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
1913 K(zone
->present_pages
),
1914 zone
->pages_scanned
,
1915 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
1917 printk("lowmem_reserve[]:");
1918 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1919 printk(" %lu", zone
->lowmem_reserve
[i
]);
1923 for_each_zone(zone
) {
1924 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1926 if (!populated_zone(zone
))
1930 printk("%s: ", zone
->name
);
1932 spin_lock_irqsave(&zone
->lock
, flags
);
1933 for (order
= 0; order
< MAX_ORDER
; order
++) {
1934 nr
[order
] = zone
->free_area
[order
].nr_free
;
1935 total
+= nr
[order
] << order
;
1937 spin_unlock_irqrestore(&zone
->lock
, flags
);
1938 for (order
= 0; order
< MAX_ORDER
; order
++)
1939 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1940 printk("= %lukB\n", K(total
));
1943 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
1945 show_swap_cache_info();
1948 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
1950 zoneref
->zone
= zone
;
1951 zoneref
->zone_idx
= zone_idx(zone
);
1955 * Builds allocation fallback zone lists.
1957 * Add all populated zones of a node to the zonelist.
1959 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1960 int nr_zones
, enum zone_type zone_type
)
1964 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1969 zone
= pgdat
->node_zones
+ zone_type
;
1970 if (populated_zone(zone
)) {
1971 zoneref_set_zone(zone
,
1972 &zonelist
->_zonerefs
[nr_zones
++]);
1973 check_highest_zone(zone_type
);
1976 } while (zone_type
);
1983 * 0 = automatic detection of better ordering.
1984 * 1 = order by ([node] distance, -zonetype)
1985 * 2 = order by (-zonetype, [node] distance)
1987 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1988 * the same zonelist. So only NUMA can configure this param.
1990 #define ZONELIST_ORDER_DEFAULT 0
1991 #define ZONELIST_ORDER_NODE 1
1992 #define ZONELIST_ORDER_ZONE 2
1994 /* zonelist order in the kernel.
1995 * set_zonelist_order() will set this to NODE or ZONE.
1997 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1998 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2002 /* The value user specified ....changed by config */
2003 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2004 /* string for sysctl */
2005 #define NUMA_ZONELIST_ORDER_LEN 16
2006 char numa_zonelist_order
[16] = "default";
2009 * interface for configure zonelist ordering.
2010 * command line option "numa_zonelist_order"
2011 * = "[dD]efault - default, automatic configuration.
2012 * = "[nN]ode - order by node locality, then by zone within node
2013 * = "[zZ]one - order by zone, then by locality within zone
2016 static int __parse_numa_zonelist_order(char *s
)
2018 if (*s
== 'd' || *s
== 'D') {
2019 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2020 } else if (*s
== 'n' || *s
== 'N') {
2021 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2022 } else if (*s
== 'z' || *s
== 'Z') {
2023 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2026 "Ignoring invalid numa_zonelist_order value: "
2033 static __init
int setup_numa_zonelist_order(char *s
)
2036 return __parse_numa_zonelist_order(s
);
2039 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2042 * sysctl handler for numa_zonelist_order
2044 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2045 struct file
*file
, void __user
*buffer
, size_t *length
,
2048 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2052 strncpy(saved_string
, (char*)table
->data
,
2053 NUMA_ZONELIST_ORDER_LEN
);
2054 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2058 int oldval
= user_zonelist_order
;
2059 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2061 * bogus value. restore saved string
2063 strncpy((char*)table
->data
, saved_string
,
2064 NUMA_ZONELIST_ORDER_LEN
);
2065 user_zonelist_order
= oldval
;
2066 } else if (oldval
!= user_zonelist_order
)
2067 build_all_zonelists();
2073 #define MAX_NODE_LOAD (num_online_nodes())
2074 static int node_load
[MAX_NUMNODES
];
2077 * find_next_best_node - find the next node that should appear in a given node's fallback list
2078 * @node: node whose fallback list we're appending
2079 * @used_node_mask: nodemask_t of already used nodes
2081 * We use a number of factors to determine which is the next node that should
2082 * appear on a given node's fallback list. The node should not have appeared
2083 * already in @node's fallback list, and it should be the next closest node
2084 * according to the distance array (which contains arbitrary distance values
2085 * from each node to each node in the system), and should also prefer nodes
2086 * with no CPUs, since presumably they'll have very little allocation pressure
2087 * on them otherwise.
2088 * It returns -1 if no node is found.
2090 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2093 int min_val
= INT_MAX
;
2095 node_to_cpumask_ptr(tmp
, 0);
2097 /* Use the local node if we haven't already */
2098 if (!node_isset(node
, *used_node_mask
)) {
2099 node_set(node
, *used_node_mask
);
2103 for_each_node_state(n
, N_HIGH_MEMORY
) {
2105 /* Don't want a node to appear more than once */
2106 if (node_isset(n
, *used_node_mask
))
2109 /* Use the distance array to find the distance */
2110 val
= node_distance(node
, n
);
2112 /* Penalize nodes under us ("prefer the next node") */
2115 /* Give preference to headless and unused nodes */
2116 node_to_cpumask_ptr_next(tmp
, n
);
2117 if (!cpus_empty(*tmp
))
2118 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2120 /* Slight preference for less loaded node */
2121 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2122 val
+= node_load
[n
];
2124 if (val
< min_val
) {
2131 node_set(best_node
, *used_node_mask
);
2138 * Build zonelists ordered by node and zones within node.
2139 * This results in maximum locality--normal zone overflows into local
2140 * DMA zone, if any--but risks exhausting DMA zone.
2142 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2145 struct zonelist
*zonelist
;
2147 zonelist
= &pgdat
->node_zonelists
[0];
2148 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2150 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2152 zonelist
->_zonerefs
[j
].zone
= NULL
;
2153 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2157 * Build gfp_thisnode zonelists
2159 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2162 struct zonelist
*zonelist
;
2164 zonelist
= &pgdat
->node_zonelists
[1];
2165 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2166 zonelist
->_zonerefs
[j
].zone
= NULL
;
2167 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2171 * Build zonelists ordered by zone and nodes within zones.
2172 * This results in conserving DMA zone[s] until all Normal memory is
2173 * exhausted, but results in overflowing to remote node while memory
2174 * may still exist in local DMA zone.
2176 static int node_order
[MAX_NUMNODES
];
2178 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2181 int zone_type
; /* needs to be signed */
2183 struct zonelist
*zonelist
;
2185 zonelist
= &pgdat
->node_zonelists
[0];
2187 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2188 for (j
= 0; j
< nr_nodes
; j
++) {
2189 node
= node_order
[j
];
2190 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2191 if (populated_zone(z
)) {
2193 &zonelist
->_zonerefs
[pos
++]);
2194 check_highest_zone(zone_type
);
2198 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2199 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2202 static int default_zonelist_order(void)
2205 unsigned long low_kmem_size
,total_size
;
2209 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2210 * If they are really small and used heavily, the system can fall
2211 * into OOM very easily.
2212 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2214 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2217 for_each_online_node(nid
) {
2218 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2219 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2220 if (populated_zone(z
)) {
2221 if (zone_type
< ZONE_NORMAL
)
2222 low_kmem_size
+= z
->present_pages
;
2223 total_size
+= z
->present_pages
;
2227 if (!low_kmem_size
|| /* there are no DMA area. */
2228 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2229 return ZONELIST_ORDER_NODE
;
2231 * look into each node's config.
2232 * If there is a node whose DMA/DMA32 memory is very big area on
2233 * local memory, NODE_ORDER may be suitable.
2235 average_size
= total_size
/
2236 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2237 for_each_online_node(nid
) {
2240 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2241 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2242 if (populated_zone(z
)) {
2243 if (zone_type
< ZONE_NORMAL
)
2244 low_kmem_size
+= z
->present_pages
;
2245 total_size
+= z
->present_pages
;
2248 if (low_kmem_size
&&
2249 total_size
> average_size
&& /* ignore small node */
2250 low_kmem_size
> total_size
* 70/100)
2251 return ZONELIST_ORDER_NODE
;
2253 return ZONELIST_ORDER_ZONE
;
2256 static void set_zonelist_order(void)
2258 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2259 current_zonelist_order
= default_zonelist_order();
2261 current_zonelist_order
= user_zonelist_order
;
2264 static void build_zonelists(pg_data_t
*pgdat
)
2268 nodemask_t used_mask
;
2269 int local_node
, prev_node
;
2270 struct zonelist
*zonelist
;
2271 int order
= current_zonelist_order
;
2273 /* initialize zonelists */
2274 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2275 zonelist
= pgdat
->node_zonelists
+ i
;
2276 zonelist
->_zonerefs
[0].zone
= NULL
;
2277 zonelist
->_zonerefs
[0].zone_idx
= 0;
2280 /* NUMA-aware ordering of nodes */
2281 local_node
= pgdat
->node_id
;
2282 load
= num_online_nodes();
2283 prev_node
= local_node
;
2284 nodes_clear(used_mask
);
2286 memset(node_load
, 0, sizeof(node_load
));
2287 memset(node_order
, 0, sizeof(node_order
));
2290 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2291 int distance
= node_distance(local_node
, node
);
2294 * If another node is sufficiently far away then it is better
2295 * to reclaim pages in a zone before going off node.
2297 if (distance
> RECLAIM_DISTANCE
)
2298 zone_reclaim_mode
= 1;
2301 * We don't want to pressure a particular node.
2302 * So adding penalty to the first node in same
2303 * distance group to make it round-robin.
2305 if (distance
!= node_distance(local_node
, prev_node
))
2306 node_load
[node
] = load
;
2310 if (order
== ZONELIST_ORDER_NODE
)
2311 build_zonelists_in_node_order(pgdat
, node
);
2313 node_order
[j
++] = node
; /* remember order */
2316 if (order
== ZONELIST_ORDER_ZONE
) {
2317 /* calculate node order -- i.e., DMA last! */
2318 build_zonelists_in_zone_order(pgdat
, j
);
2321 build_thisnode_zonelists(pgdat
);
2324 /* Construct the zonelist performance cache - see further mmzone.h */
2325 static void build_zonelist_cache(pg_data_t
*pgdat
)
2327 struct zonelist
*zonelist
;
2328 struct zonelist_cache
*zlc
;
2331 zonelist
= &pgdat
->node_zonelists
[0];
2332 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2333 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2334 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2335 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2339 #else /* CONFIG_NUMA */
2341 static void set_zonelist_order(void)
2343 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2346 static void build_zonelists(pg_data_t
*pgdat
)
2348 int node
, local_node
;
2350 struct zonelist
*zonelist
;
2352 local_node
= pgdat
->node_id
;
2354 zonelist
= &pgdat
->node_zonelists
[0];
2355 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2358 * Now we build the zonelist so that it contains the zones
2359 * of all the other nodes.
2360 * We don't want to pressure a particular node, so when
2361 * building the zones for node N, we make sure that the
2362 * zones coming right after the local ones are those from
2363 * node N+1 (modulo N)
2365 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2366 if (!node_online(node
))
2368 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2371 for (node
= 0; node
< local_node
; node
++) {
2372 if (!node_online(node
))
2374 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2378 zonelist
->_zonerefs
[j
].zone
= NULL
;
2379 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2382 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2383 static void build_zonelist_cache(pg_data_t
*pgdat
)
2385 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2388 #endif /* CONFIG_NUMA */
2390 /* return values int ....just for stop_machine() */
2391 static int __build_all_zonelists(void *dummy
)
2395 for_each_online_node(nid
) {
2396 pg_data_t
*pgdat
= NODE_DATA(nid
);
2398 build_zonelists(pgdat
);
2399 build_zonelist_cache(pgdat
);
2404 void build_all_zonelists(void)
2406 set_zonelist_order();
2408 if (system_state
== SYSTEM_BOOTING
) {
2409 __build_all_zonelists(NULL
);
2410 mminit_verify_zonelist();
2411 cpuset_init_current_mems_allowed();
2413 /* we have to stop all cpus to guarantee there is no user
2415 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2416 /* cpuset refresh routine should be here */
2418 vm_total_pages
= nr_free_pagecache_pages();
2420 * Disable grouping by mobility if the number of pages in the
2421 * system is too low to allow the mechanism to work. It would be
2422 * more accurate, but expensive to check per-zone. This check is
2423 * made on memory-hotadd so a system can start with mobility
2424 * disabled and enable it later
2426 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2427 page_group_by_mobility_disabled
= 1;
2429 page_group_by_mobility_disabled
= 0;
2431 printk("Built %i zonelists in %s order, mobility grouping %s. "
2432 "Total pages: %ld\n",
2434 zonelist_order_name
[current_zonelist_order
],
2435 page_group_by_mobility_disabled
? "off" : "on",
2438 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2443 * Helper functions to size the waitqueue hash table.
2444 * Essentially these want to choose hash table sizes sufficiently
2445 * large so that collisions trying to wait on pages are rare.
2446 * But in fact, the number of active page waitqueues on typical
2447 * systems is ridiculously low, less than 200. So this is even
2448 * conservative, even though it seems large.
2450 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2451 * waitqueues, i.e. the size of the waitq table given the number of pages.
2453 #define PAGES_PER_WAITQUEUE 256
2455 #ifndef CONFIG_MEMORY_HOTPLUG
2456 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2458 unsigned long size
= 1;
2460 pages
/= PAGES_PER_WAITQUEUE
;
2462 while (size
< pages
)
2466 * Once we have dozens or even hundreds of threads sleeping
2467 * on IO we've got bigger problems than wait queue collision.
2468 * Limit the size of the wait table to a reasonable size.
2470 size
= min(size
, 4096UL);
2472 return max(size
, 4UL);
2476 * A zone's size might be changed by hot-add, so it is not possible to determine
2477 * a suitable size for its wait_table. So we use the maximum size now.
2479 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2481 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2482 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2483 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2485 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2486 * or more by the traditional way. (See above). It equals:
2488 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2489 * ia64(16K page size) : = ( 8G + 4M)byte.
2490 * powerpc (64K page size) : = (32G +16M)byte.
2492 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2499 * This is an integer logarithm so that shifts can be used later
2500 * to extract the more random high bits from the multiplicative
2501 * hash function before the remainder is taken.
2503 static inline unsigned long wait_table_bits(unsigned long size
)
2508 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2511 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2512 * of blocks reserved is based on zone->pages_min. The memory within the
2513 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2514 * higher will lead to a bigger reserve which will get freed as contiguous
2515 * blocks as reclaim kicks in
2517 static void setup_zone_migrate_reserve(struct zone
*zone
)
2519 unsigned long start_pfn
, pfn
, end_pfn
;
2521 unsigned long reserve
, block_migratetype
;
2523 /* Get the start pfn, end pfn and the number of blocks to reserve */
2524 start_pfn
= zone
->zone_start_pfn
;
2525 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2526 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2529 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2530 if (!pfn_valid(pfn
))
2532 page
= pfn_to_page(pfn
);
2534 /* Watch out for overlapping nodes */
2535 if (page_to_nid(page
) != zone_to_nid(zone
))
2538 /* Blocks with reserved pages will never free, skip them. */
2539 if (PageReserved(page
))
2542 block_migratetype
= get_pageblock_migratetype(page
);
2544 /* If this block is reserved, account for it */
2545 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2550 /* Suitable for reserving if this block is movable */
2551 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2552 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2553 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2559 * If the reserve is met and this is a previous reserved block,
2562 if (block_migratetype
== MIGRATE_RESERVE
) {
2563 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2564 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2570 * Initially all pages are reserved - free ones are freed
2571 * up by free_all_bootmem() once the early boot process is
2572 * done. Non-atomic initialization, single-pass.
2574 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2575 unsigned long start_pfn
, enum memmap_context context
)
2578 unsigned long end_pfn
= start_pfn
+ size
;
2582 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2583 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2585 * There can be holes in boot-time mem_map[]s
2586 * handed to this function. They do not
2587 * exist on hotplugged memory.
2589 if (context
== MEMMAP_EARLY
) {
2590 if (!early_pfn_valid(pfn
))
2592 if (!early_pfn_in_nid(pfn
, nid
))
2595 page
= pfn_to_page(pfn
);
2596 set_page_links(page
, zone
, nid
, pfn
);
2597 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2598 init_page_count(page
);
2599 reset_page_mapcount(page
);
2600 SetPageReserved(page
);
2602 * Mark the block movable so that blocks are reserved for
2603 * movable at startup. This will force kernel allocations
2604 * to reserve their blocks rather than leaking throughout
2605 * the address space during boot when many long-lived
2606 * kernel allocations are made. Later some blocks near
2607 * the start are marked MIGRATE_RESERVE by
2608 * setup_zone_migrate_reserve()
2610 * bitmap is created for zone's valid pfn range. but memmap
2611 * can be created for invalid pages (for alignment)
2612 * check here not to call set_pageblock_migratetype() against
2615 if ((z
->zone_start_pfn
<= pfn
)
2616 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2617 && !(pfn
& (pageblock_nr_pages
- 1)))
2618 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2620 INIT_LIST_HEAD(&page
->lru
);
2621 #ifdef WANT_PAGE_VIRTUAL
2622 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2623 if (!is_highmem_idx(zone
))
2624 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2629 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2632 for_each_migratetype_order(order
, t
) {
2633 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2634 zone
->free_area
[order
].nr_free
= 0;
2638 #ifndef __HAVE_ARCH_MEMMAP_INIT
2639 #define memmap_init(size, nid, zone, start_pfn) \
2640 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2643 static int zone_batchsize(struct zone
*zone
)
2648 * The per-cpu-pages pools are set to around 1000th of the
2649 * size of the zone. But no more than 1/2 of a meg.
2651 * OK, so we don't know how big the cache is. So guess.
2653 batch
= zone
->present_pages
/ 1024;
2654 if (batch
* PAGE_SIZE
> 512 * 1024)
2655 batch
= (512 * 1024) / PAGE_SIZE
;
2656 batch
/= 4; /* We effectively *= 4 below */
2661 * Clamp the batch to a 2^n - 1 value. Having a power
2662 * of 2 value was found to be more likely to have
2663 * suboptimal cache aliasing properties in some cases.
2665 * For example if 2 tasks are alternately allocating
2666 * batches of pages, one task can end up with a lot
2667 * of pages of one half of the possible page colors
2668 * and the other with pages of the other colors.
2670 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2675 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2677 struct per_cpu_pages
*pcp
;
2679 memset(p
, 0, sizeof(*p
));
2683 pcp
->high
= 6 * batch
;
2684 pcp
->batch
= max(1UL, 1 * batch
);
2685 INIT_LIST_HEAD(&pcp
->list
);
2689 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2690 * to the value high for the pageset p.
2693 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2696 struct per_cpu_pages
*pcp
;
2700 pcp
->batch
= max(1UL, high
/4);
2701 if ((high
/4) > (PAGE_SHIFT
* 8))
2702 pcp
->batch
= PAGE_SHIFT
* 8;
2708 * Boot pageset table. One per cpu which is going to be used for all
2709 * zones and all nodes. The parameters will be set in such a way
2710 * that an item put on a list will immediately be handed over to
2711 * the buddy list. This is safe since pageset manipulation is done
2712 * with interrupts disabled.
2714 * Some NUMA counter updates may also be caught by the boot pagesets.
2716 * The boot_pagesets must be kept even after bootup is complete for
2717 * unused processors and/or zones. They do play a role for bootstrapping
2718 * hotplugged processors.
2720 * zoneinfo_show() and maybe other functions do
2721 * not check if the processor is online before following the pageset pointer.
2722 * Other parts of the kernel may not check if the zone is available.
2724 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2727 * Dynamically allocate memory for the
2728 * per cpu pageset array in struct zone.
2730 static int __cpuinit
process_zones(int cpu
)
2732 struct zone
*zone
, *dzone
;
2733 int node
= cpu_to_node(cpu
);
2735 node_set_state(node
, N_CPU
); /* this node has a cpu */
2737 for_each_zone(zone
) {
2739 if (!populated_zone(zone
))
2742 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2744 if (!zone_pcp(zone
, cpu
))
2747 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2749 if (percpu_pagelist_fraction
)
2750 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2751 (zone
->present_pages
/ percpu_pagelist_fraction
));
2756 for_each_zone(dzone
) {
2757 if (!populated_zone(dzone
))
2761 kfree(zone_pcp(dzone
, cpu
));
2762 zone_pcp(dzone
, cpu
) = NULL
;
2767 static inline void free_zone_pagesets(int cpu
)
2771 for_each_zone(zone
) {
2772 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2774 /* Free per_cpu_pageset if it is slab allocated */
2775 if (pset
!= &boot_pageset
[cpu
])
2777 zone_pcp(zone
, cpu
) = NULL
;
2781 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2782 unsigned long action
,
2785 int cpu
= (long)hcpu
;
2786 int ret
= NOTIFY_OK
;
2789 case CPU_UP_PREPARE
:
2790 case CPU_UP_PREPARE_FROZEN
:
2791 if (process_zones(cpu
))
2794 case CPU_UP_CANCELED
:
2795 case CPU_UP_CANCELED_FROZEN
:
2797 case CPU_DEAD_FROZEN
:
2798 free_zone_pagesets(cpu
);
2806 static struct notifier_block __cpuinitdata pageset_notifier
=
2807 { &pageset_cpuup_callback
, NULL
, 0 };
2809 void __init
setup_per_cpu_pageset(void)
2813 /* Initialize per_cpu_pageset for cpu 0.
2814 * A cpuup callback will do this for every cpu
2815 * as it comes online
2817 err
= process_zones(smp_processor_id());
2819 register_cpu_notifier(&pageset_notifier
);
2824 static noinline __init_refok
2825 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2828 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2832 * The per-page waitqueue mechanism uses hashed waitqueues
2835 zone
->wait_table_hash_nr_entries
=
2836 wait_table_hash_nr_entries(zone_size_pages
);
2837 zone
->wait_table_bits
=
2838 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2839 alloc_size
= zone
->wait_table_hash_nr_entries
2840 * sizeof(wait_queue_head_t
);
2842 if (!slab_is_available()) {
2843 zone
->wait_table
= (wait_queue_head_t
*)
2844 alloc_bootmem_node(pgdat
, alloc_size
);
2847 * This case means that a zone whose size was 0 gets new memory
2848 * via memory hot-add.
2849 * But it may be the case that a new node was hot-added. In
2850 * this case vmalloc() will not be able to use this new node's
2851 * memory - this wait_table must be initialized to use this new
2852 * node itself as well.
2853 * To use this new node's memory, further consideration will be
2856 zone
->wait_table
= vmalloc(alloc_size
);
2858 if (!zone
->wait_table
)
2861 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2862 init_waitqueue_head(zone
->wait_table
+ i
);
2867 static __meminit
void zone_pcp_init(struct zone
*zone
)
2870 unsigned long batch
= zone_batchsize(zone
);
2872 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2874 /* Early boot. Slab allocator not functional yet */
2875 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2876 setup_pageset(&boot_pageset
[cpu
],0);
2878 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2881 if (zone
->present_pages
)
2882 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2883 zone
->name
, zone
->present_pages
, batch
);
2886 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2887 unsigned long zone_start_pfn
,
2889 enum memmap_context context
)
2891 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2893 ret
= zone_wait_table_init(zone
, size
);
2896 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2898 zone
->zone_start_pfn
= zone_start_pfn
;
2900 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
2901 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2903 (unsigned long)zone_idx(zone
),
2904 zone_start_pfn
, (zone_start_pfn
+ size
));
2906 zone_init_free_lists(zone
);
2911 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2913 * Basic iterator support. Return the first range of PFNs for a node
2914 * Note: nid == MAX_NUMNODES returns first region regardless of node
2916 static int __meminit
first_active_region_index_in_nid(int nid
)
2920 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2921 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2928 * Basic iterator support. Return the next active range of PFNs for a node
2929 * Note: nid == MAX_NUMNODES returns next region regardless of node
2931 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2933 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2934 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2940 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2942 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2943 * Architectures may implement their own version but if add_active_range()
2944 * was used and there are no special requirements, this is a convenient
2947 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2951 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2952 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2953 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2955 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2956 return early_node_map
[i
].nid
;
2961 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2963 /* Basic iterator support to walk early_node_map[] */
2964 #define for_each_active_range_index_in_nid(i, nid) \
2965 for (i = first_active_region_index_in_nid(nid); i != -1; \
2966 i = next_active_region_index_in_nid(i, nid))
2969 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2970 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2971 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2973 * If an architecture guarantees that all ranges registered with
2974 * add_active_ranges() contain no holes and may be freed, this
2975 * this function may be used instead of calling free_bootmem() manually.
2977 void __init
free_bootmem_with_active_regions(int nid
,
2978 unsigned long max_low_pfn
)
2982 for_each_active_range_index_in_nid(i
, nid
) {
2983 unsigned long size_pages
= 0;
2984 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2986 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2989 if (end_pfn
> max_low_pfn
)
2990 end_pfn
= max_low_pfn
;
2992 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2993 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2994 PFN_PHYS(early_node_map
[i
].start_pfn
),
2995 size_pages
<< PAGE_SHIFT
);
2999 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3004 for_each_active_range_index_in_nid(i
, nid
) {
3005 ret
= work_fn(early_node_map
[i
].start_pfn
,
3006 early_node_map
[i
].end_pfn
, data
);
3012 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3013 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3015 * If an architecture guarantees that all ranges registered with
3016 * add_active_ranges() contain no holes and may be freed, this
3017 * function may be used instead of calling memory_present() manually.
3019 void __init
sparse_memory_present_with_active_regions(int nid
)
3023 for_each_active_range_index_in_nid(i
, nid
)
3024 memory_present(early_node_map
[i
].nid
,
3025 early_node_map
[i
].start_pfn
,
3026 early_node_map
[i
].end_pfn
);
3030 * push_node_boundaries - Push node boundaries to at least the requested boundary
3031 * @nid: The nid of the node to push the boundary for
3032 * @start_pfn: The start pfn of the node
3033 * @end_pfn: The end pfn of the node
3035 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3036 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3037 * be hotplugged even though no physical memory exists. This function allows
3038 * an arch to push out the node boundaries so mem_map is allocated that can
3041 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3042 void __init
push_node_boundaries(unsigned int nid
,
3043 unsigned long start_pfn
, unsigned long end_pfn
)
3045 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3046 "Entering push_node_boundaries(%u, %lu, %lu)\n",
3047 nid
, start_pfn
, end_pfn
);
3049 /* Initialise the boundary for this node if necessary */
3050 if (node_boundary_end_pfn
[nid
] == 0)
3051 node_boundary_start_pfn
[nid
] = -1UL;
3053 /* Update the boundaries */
3054 if (node_boundary_start_pfn
[nid
] > start_pfn
)
3055 node_boundary_start_pfn
[nid
] = start_pfn
;
3056 if (node_boundary_end_pfn
[nid
] < end_pfn
)
3057 node_boundary_end_pfn
[nid
] = end_pfn
;
3060 /* If necessary, push the node boundary out for reserve hotadd */
3061 static void __meminit
account_node_boundary(unsigned int nid
,
3062 unsigned long *start_pfn
, unsigned long *end_pfn
)
3064 mminit_dprintk(MMINIT_TRACE
, "zoneboundary",
3065 "Entering account_node_boundary(%u, %lu, %lu)\n",
3066 nid
, *start_pfn
, *end_pfn
);
3068 /* Return if boundary information has not been provided */
3069 if (node_boundary_end_pfn
[nid
] == 0)
3072 /* Check the boundaries and update if necessary */
3073 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
3074 *start_pfn
= node_boundary_start_pfn
[nid
];
3075 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
3076 *end_pfn
= node_boundary_end_pfn
[nid
];
3079 void __init
push_node_boundaries(unsigned int nid
,
3080 unsigned long start_pfn
, unsigned long end_pfn
) {}
3082 static void __meminit
account_node_boundary(unsigned int nid
,
3083 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3088 * get_pfn_range_for_nid - Return the start and end page frames for a node
3089 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3090 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3091 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3093 * It returns the start and end page frame of a node based on information
3094 * provided by an arch calling add_active_range(). If called for a node
3095 * with no available memory, a warning is printed and the start and end
3098 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3099 unsigned long *start_pfn
, unsigned long *end_pfn
)
3105 for_each_active_range_index_in_nid(i
, nid
) {
3106 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3107 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3110 if (*start_pfn
== -1UL)
3113 /* Push the node boundaries out if requested */
3114 account_node_boundary(nid
, start_pfn
, end_pfn
);
3118 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3119 * assumption is made that zones within a node are ordered in monotonic
3120 * increasing memory addresses so that the "highest" populated zone is used
3122 static void __init
find_usable_zone_for_movable(void)
3125 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3126 if (zone_index
== ZONE_MOVABLE
)
3129 if (arch_zone_highest_possible_pfn
[zone_index
] >
3130 arch_zone_lowest_possible_pfn
[zone_index
])
3134 VM_BUG_ON(zone_index
== -1);
3135 movable_zone
= zone_index
;
3139 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3140 * because it is sized independant of architecture. Unlike the other zones,
3141 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3142 * in each node depending on the size of each node and how evenly kernelcore
3143 * is distributed. This helper function adjusts the zone ranges
3144 * provided by the architecture for a given node by using the end of the
3145 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3146 * zones within a node are in order of monotonic increases memory addresses
3148 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3149 unsigned long zone_type
,
3150 unsigned long node_start_pfn
,
3151 unsigned long node_end_pfn
,
3152 unsigned long *zone_start_pfn
,
3153 unsigned long *zone_end_pfn
)
3155 /* Only adjust if ZONE_MOVABLE is on this node */
3156 if (zone_movable_pfn
[nid
]) {
3157 /* Size ZONE_MOVABLE */
3158 if (zone_type
== ZONE_MOVABLE
) {
3159 *zone_start_pfn
= zone_movable_pfn
[nid
];
3160 *zone_end_pfn
= min(node_end_pfn
,
3161 arch_zone_highest_possible_pfn
[movable_zone
]);
3163 /* Adjust for ZONE_MOVABLE starting within this range */
3164 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3165 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3166 *zone_end_pfn
= zone_movable_pfn
[nid
];
3168 /* Check if this whole range is within ZONE_MOVABLE */
3169 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3170 *zone_start_pfn
= *zone_end_pfn
;
3175 * Return the number of pages a zone spans in a node, including holes
3176 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3178 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3179 unsigned long zone_type
,
3180 unsigned long *ignored
)
3182 unsigned long node_start_pfn
, node_end_pfn
;
3183 unsigned long zone_start_pfn
, zone_end_pfn
;
3185 /* Get the start and end of the node and zone */
3186 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3187 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3188 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3189 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3190 node_start_pfn
, node_end_pfn
,
3191 &zone_start_pfn
, &zone_end_pfn
);
3193 /* Check that this node has pages within the zone's required range */
3194 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3197 /* Move the zone boundaries inside the node if necessary */
3198 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3199 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3201 /* Return the spanned pages */
3202 return zone_end_pfn
- zone_start_pfn
;
3206 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3207 * then all holes in the requested range will be accounted for.
3209 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3210 unsigned long range_start_pfn
,
3211 unsigned long range_end_pfn
)
3214 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3215 unsigned long start_pfn
;
3217 /* Find the end_pfn of the first active range of pfns in the node */
3218 i
= first_active_region_index_in_nid(nid
);
3222 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3224 /* Account for ranges before physical memory on this node */
3225 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3226 hole_pages
= prev_end_pfn
- range_start_pfn
;
3228 /* Find all holes for the zone within the node */
3229 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3231 /* No need to continue if prev_end_pfn is outside the zone */
3232 if (prev_end_pfn
>= range_end_pfn
)
3235 /* Make sure the end of the zone is not within the hole */
3236 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3237 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3239 /* Update the hole size cound and move on */
3240 if (start_pfn
> range_start_pfn
) {
3241 BUG_ON(prev_end_pfn
> start_pfn
);
3242 hole_pages
+= start_pfn
- prev_end_pfn
;
3244 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3247 /* Account for ranges past physical memory on this node */
3248 if (range_end_pfn
> prev_end_pfn
)
3249 hole_pages
+= range_end_pfn
-
3250 max(range_start_pfn
, prev_end_pfn
);
3256 * absent_pages_in_range - Return number of page frames in holes within a range
3257 * @start_pfn: The start PFN to start searching for holes
3258 * @end_pfn: The end PFN to stop searching for holes
3260 * It returns the number of pages frames in memory holes within a range.
3262 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3263 unsigned long end_pfn
)
3265 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3268 /* Return the number of page frames in holes in a zone on a node */
3269 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3270 unsigned long zone_type
,
3271 unsigned long *ignored
)
3273 unsigned long node_start_pfn
, node_end_pfn
;
3274 unsigned long zone_start_pfn
, zone_end_pfn
;
3276 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3277 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3279 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3282 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3283 node_start_pfn
, node_end_pfn
,
3284 &zone_start_pfn
, &zone_end_pfn
);
3285 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3289 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3290 unsigned long zone_type
,
3291 unsigned long *zones_size
)
3293 return zones_size
[zone_type
];
3296 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3297 unsigned long zone_type
,
3298 unsigned long *zholes_size
)
3303 return zholes_size
[zone_type
];
3308 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3309 unsigned long *zones_size
, unsigned long *zholes_size
)
3311 unsigned long realtotalpages
, totalpages
= 0;
3314 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3315 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3317 pgdat
->node_spanned_pages
= totalpages
;
3319 realtotalpages
= totalpages
;
3320 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3322 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3324 pgdat
->node_present_pages
= realtotalpages
;
3325 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3329 #ifndef CONFIG_SPARSEMEM
3331 * Calculate the size of the zone->blockflags rounded to an unsigned long
3332 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3333 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3334 * round what is now in bits to nearest long in bits, then return it in
3337 static unsigned long __init
usemap_size(unsigned long zonesize
)
3339 unsigned long usemapsize
;
3341 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3342 usemapsize
= usemapsize
>> pageblock_order
;
3343 usemapsize
*= NR_PAGEBLOCK_BITS
;
3344 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3346 return usemapsize
/ 8;
3349 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3350 struct zone
*zone
, unsigned long zonesize
)
3352 unsigned long usemapsize
= usemap_size(zonesize
);
3353 zone
->pageblock_flags
= NULL
;
3355 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3356 memset(zone
->pageblock_flags
, 0, usemapsize
);
3360 static void inline setup_usemap(struct pglist_data
*pgdat
,
3361 struct zone
*zone
, unsigned long zonesize
) {}
3362 #endif /* CONFIG_SPARSEMEM */
3364 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3366 /* Return a sensible default order for the pageblock size. */
3367 static inline int pageblock_default_order(void)
3369 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3370 return HUGETLB_PAGE_ORDER
;
3375 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3376 static inline void __init
set_pageblock_order(unsigned int order
)
3378 /* Check that pageblock_nr_pages has not already been setup */
3379 if (pageblock_order
)
3383 * Assume the largest contiguous order of interest is a huge page.
3384 * This value may be variable depending on boot parameters on IA64
3386 pageblock_order
= order
;
3388 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3391 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3392 * and pageblock_default_order() are unused as pageblock_order is set
3393 * at compile-time. See include/linux/pageblock-flags.h for the values of
3394 * pageblock_order based on the kernel config
3396 static inline int pageblock_default_order(unsigned int order
)
3400 #define set_pageblock_order(x) do {} while (0)
3402 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3405 * Set up the zone data structures:
3406 * - mark all pages reserved
3407 * - mark all memory queues empty
3408 * - clear the memory bitmaps
3410 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3411 unsigned long *zones_size
, unsigned long *zholes_size
)
3414 int nid
= pgdat
->node_id
;
3415 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3418 pgdat_resize_init(pgdat
);
3419 pgdat
->nr_zones
= 0;
3420 init_waitqueue_head(&pgdat
->kswapd_wait
);
3421 pgdat
->kswapd_max_order
= 0;
3423 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3424 struct zone
*zone
= pgdat
->node_zones
+ j
;
3425 unsigned long size
, realsize
, memmap_pages
;
3428 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3429 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3433 * Adjust realsize so that it accounts for how much memory
3434 * is used by this zone for memmap. This affects the watermark
3435 * and per-cpu initialisations
3438 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3439 if (realsize
>= memmap_pages
) {
3440 realsize
-= memmap_pages
;
3441 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3442 "%s zone: %lu pages used for memmap\n",
3443 zone_names
[j
], memmap_pages
);
3446 " %s zone: %lu pages exceeds realsize %lu\n",
3447 zone_names
[j
], memmap_pages
, realsize
);
3449 /* Account for reserved pages */
3450 if (j
== 0 && realsize
> dma_reserve
) {
3451 realsize
-= dma_reserve
;
3452 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3453 "%s zone: %lu pages reserved\n",
3454 zone_names
[0], dma_reserve
);
3457 if (!is_highmem_idx(j
))
3458 nr_kernel_pages
+= realsize
;
3459 nr_all_pages
+= realsize
;
3461 zone
->spanned_pages
= size
;
3462 zone
->present_pages
= realsize
;
3465 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3467 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3469 zone
->name
= zone_names
[j
];
3470 spin_lock_init(&zone
->lock
);
3471 spin_lock_init(&zone
->lru_lock
);
3472 zone_seqlock_init(zone
);
3473 zone
->zone_pgdat
= pgdat
;
3475 zone
->prev_priority
= DEF_PRIORITY
;
3477 zone_pcp_init(zone
);
3479 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3480 zone
->lru
[l
].nr_scan
= 0;
3482 zone
->recent_rotated
[0] = 0;
3483 zone
->recent_rotated
[1] = 0;
3484 zone
->recent_scanned
[0] = 0;
3485 zone
->recent_scanned
[1] = 0;
3486 zap_zone_vm_stats(zone
);
3491 set_pageblock_order(pageblock_default_order());
3492 setup_usemap(pgdat
, zone
, size
);
3493 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3494 size
, MEMMAP_EARLY
);
3496 memmap_init(size
, nid
, j
, zone_start_pfn
);
3497 zone_start_pfn
+= size
;
3501 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3503 /* Skip empty nodes */
3504 if (!pgdat
->node_spanned_pages
)
3507 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3508 /* ia64 gets its own node_mem_map, before this, without bootmem */
3509 if (!pgdat
->node_mem_map
) {
3510 unsigned long size
, start
, end
;
3514 * The zone's endpoints aren't required to be MAX_ORDER
3515 * aligned but the node_mem_map endpoints must be in order
3516 * for the buddy allocator to function correctly.
3518 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3519 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3520 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3521 size
= (end
- start
) * sizeof(struct page
);
3522 map
= alloc_remap(pgdat
->node_id
, size
);
3524 map
= alloc_bootmem_node(pgdat
, size
);
3525 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3527 #ifndef CONFIG_NEED_MULTIPLE_NODES
3529 * With no DISCONTIG, the global mem_map is just set as node 0's
3531 if (pgdat
== NODE_DATA(0)) {
3532 mem_map
= NODE_DATA(0)->node_mem_map
;
3533 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3534 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3535 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3536 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3539 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3542 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3543 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3545 pg_data_t
*pgdat
= NODE_DATA(nid
);
3547 pgdat
->node_id
= nid
;
3548 pgdat
->node_start_pfn
= node_start_pfn
;
3549 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3551 alloc_node_mem_map(pgdat
);
3552 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3553 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3554 nid
, (unsigned long)pgdat
,
3555 (unsigned long)pgdat
->node_mem_map
);
3558 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3561 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3563 #if MAX_NUMNODES > 1
3565 * Figure out the number of possible node ids.
3567 static void __init
setup_nr_node_ids(void)
3570 unsigned int highest
= 0;
3572 for_each_node_mask(node
, node_possible_map
)
3574 nr_node_ids
= highest
+ 1;
3577 static inline void setup_nr_node_ids(void)
3583 * add_active_range - Register a range of PFNs backed by physical memory
3584 * @nid: The node ID the range resides on
3585 * @start_pfn: The start PFN of the available physical memory
3586 * @end_pfn: The end PFN of the available physical memory
3588 * These ranges are stored in an early_node_map[] and later used by
3589 * free_area_init_nodes() to calculate zone sizes and holes. If the
3590 * range spans a memory hole, it is up to the architecture to ensure
3591 * the memory is not freed by the bootmem allocator. If possible
3592 * the range being registered will be merged with existing ranges.
3594 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3595 unsigned long end_pfn
)
3599 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3600 "Entering add_active_range(%d, %#lx, %#lx) "
3601 "%d entries of %d used\n",
3602 nid
, start_pfn
, end_pfn
,
3603 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3605 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3607 /* Merge with existing active regions if possible */
3608 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3609 if (early_node_map
[i
].nid
!= nid
)
3612 /* Skip if an existing region covers this new one */
3613 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3614 end_pfn
<= early_node_map
[i
].end_pfn
)
3617 /* Merge forward if suitable */
3618 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3619 end_pfn
> early_node_map
[i
].end_pfn
) {
3620 early_node_map
[i
].end_pfn
= end_pfn
;
3624 /* Merge backward if suitable */
3625 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3626 end_pfn
>= early_node_map
[i
].start_pfn
) {
3627 early_node_map
[i
].start_pfn
= start_pfn
;
3632 /* Check that early_node_map is large enough */
3633 if (i
>= MAX_ACTIVE_REGIONS
) {
3634 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3635 MAX_ACTIVE_REGIONS
);
3639 early_node_map
[i
].nid
= nid
;
3640 early_node_map
[i
].start_pfn
= start_pfn
;
3641 early_node_map
[i
].end_pfn
= end_pfn
;
3642 nr_nodemap_entries
= i
+ 1;
3646 * remove_active_range - Shrink an existing registered range of PFNs
3647 * @nid: The node id the range is on that should be shrunk
3648 * @start_pfn: The new PFN of the range
3649 * @end_pfn: The new PFN of the range
3651 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3652 * The map is kept near the end physical page range that has already been
3653 * registered. This function allows an arch to shrink an existing registered
3656 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3657 unsigned long end_pfn
)
3662 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3663 nid
, start_pfn
, end_pfn
);
3665 /* Find the old active region end and shrink */
3666 for_each_active_range_index_in_nid(i
, nid
) {
3667 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3668 early_node_map
[i
].end_pfn
<= end_pfn
) {
3670 early_node_map
[i
].start_pfn
= 0;
3671 early_node_map
[i
].end_pfn
= 0;
3675 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3676 early_node_map
[i
].end_pfn
> start_pfn
) {
3677 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3678 early_node_map
[i
].end_pfn
= start_pfn
;
3679 if (temp_end_pfn
> end_pfn
)
3680 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3683 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3684 early_node_map
[i
].end_pfn
> end_pfn
&&
3685 early_node_map
[i
].start_pfn
< end_pfn
) {
3686 early_node_map
[i
].start_pfn
= end_pfn
;
3694 /* remove the blank ones */
3695 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3696 if (early_node_map
[i
].nid
!= nid
)
3698 if (early_node_map
[i
].end_pfn
)
3700 /* we found it, get rid of it */
3701 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3702 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3703 sizeof(early_node_map
[j
]));
3704 j
= nr_nodemap_entries
- 1;
3705 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3706 nr_nodemap_entries
--;
3711 * remove_all_active_ranges - Remove all currently registered regions
3713 * During discovery, it may be found that a table like SRAT is invalid
3714 * and an alternative discovery method must be used. This function removes
3715 * all currently registered regions.
3717 void __init
remove_all_active_ranges(void)
3719 memset(early_node_map
, 0, sizeof(early_node_map
));
3720 nr_nodemap_entries
= 0;
3721 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3722 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3723 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3724 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3727 /* Compare two active node_active_regions */
3728 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3730 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3731 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3733 /* Done this way to avoid overflows */
3734 if (arange
->start_pfn
> brange
->start_pfn
)
3736 if (arange
->start_pfn
< brange
->start_pfn
)
3742 /* sort the node_map by start_pfn */
3743 static void __init
sort_node_map(void)
3745 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3746 sizeof(struct node_active_region
),
3747 cmp_node_active_region
, NULL
);
3750 /* Find the lowest pfn for a node */
3751 static unsigned long __init
find_min_pfn_for_node(int nid
)
3754 unsigned long min_pfn
= ULONG_MAX
;
3756 /* Assuming a sorted map, the first range found has the starting pfn */
3757 for_each_active_range_index_in_nid(i
, nid
)
3758 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3760 if (min_pfn
== ULONG_MAX
) {
3762 "Could not find start_pfn for node %d\n", nid
);
3770 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3772 * It returns the minimum PFN based on information provided via
3773 * add_active_range().
3775 unsigned long __init
find_min_pfn_with_active_regions(void)
3777 return find_min_pfn_for_node(MAX_NUMNODES
);
3781 * early_calculate_totalpages()
3782 * Sum pages in active regions for movable zone.
3783 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3785 static unsigned long __init
early_calculate_totalpages(void)
3788 unsigned long totalpages
= 0;
3790 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3791 unsigned long pages
= early_node_map
[i
].end_pfn
-
3792 early_node_map
[i
].start_pfn
;
3793 totalpages
+= pages
;
3795 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3801 * Find the PFN the Movable zone begins in each node. Kernel memory
3802 * is spread evenly between nodes as long as the nodes have enough
3803 * memory. When they don't, some nodes will have more kernelcore than
3806 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3809 unsigned long usable_startpfn
;
3810 unsigned long kernelcore_node
, kernelcore_remaining
;
3811 unsigned long totalpages
= early_calculate_totalpages();
3812 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3815 * If movablecore was specified, calculate what size of
3816 * kernelcore that corresponds so that memory usable for
3817 * any allocation type is evenly spread. If both kernelcore
3818 * and movablecore are specified, then the value of kernelcore
3819 * will be used for required_kernelcore if it's greater than
3820 * what movablecore would have allowed.
3822 if (required_movablecore
) {
3823 unsigned long corepages
;
3826 * Round-up so that ZONE_MOVABLE is at least as large as what
3827 * was requested by the user
3829 required_movablecore
=
3830 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3831 corepages
= totalpages
- required_movablecore
;
3833 required_kernelcore
= max(required_kernelcore
, corepages
);
3836 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3837 if (!required_kernelcore
)
3840 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3841 find_usable_zone_for_movable();
3842 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3845 /* Spread kernelcore memory as evenly as possible throughout nodes */
3846 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3847 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3849 * Recalculate kernelcore_node if the division per node
3850 * now exceeds what is necessary to satisfy the requested
3851 * amount of memory for the kernel
3853 if (required_kernelcore
< kernelcore_node
)
3854 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3857 * As the map is walked, we track how much memory is usable
3858 * by the kernel using kernelcore_remaining. When it is
3859 * 0, the rest of the node is usable by ZONE_MOVABLE
3861 kernelcore_remaining
= kernelcore_node
;
3863 /* Go through each range of PFNs within this node */
3864 for_each_active_range_index_in_nid(i
, nid
) {
3865 unsigned long start_pfn
, end_pfn
;
3866 unsigned long size_pages
;
3868 start_pfn
= max(early_node_map
[i
].start_pfn
,
3869 zone_movable_pfn
[nid
]);
3870 end_pfn
= early_node_map
[i
].end_pfn
;
3871 if (start_pfn
>= end_pfn
)
3874 /* Account for what is only usable for kernelcore */
3875 if (start_pfn
< usable_startpfn
) {
3876 unsigned long kernel_pages
;
3877 kernel_pages
= min(end_pfn
, usable_startpfn
)
3880 kernelcore_remaining
-= min(kernel_pages
,
3881 kernelcore_remaining
);
3882 required_kernelcore
-= min(kernel_pages
,
3883 required_kernelcore
);
3885 /* Continue if range is now fully accounted */
3886 if (end_pfn
<= usable_startpfn
) {
3889 * Push zone_movable_pfn to the end so
3890 * that if we have to rebalance
3891 * kernelcore across nodes, we will
3892 * not double account here
3894 zone_movable_pfn
[nid
] = end_pfn
;
3897 start_pfn
= usable_startpfn
;
3901 * The usable PFN range for ZONE_MOVABLE is from
3902 * start_pfn->end_pfn. Calculate size_pages as the
3903 * number of pages used as kernelcore
3905 size_pages
= end_pfn
- start_pfn
;
3906 if (size_pages
> kernelcore_remaining
)
3907 size_pages
= kernelcore_remaining
;
3908 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3911 * Some kernelcore has been met, update counts and
3912 * break if the kernelcore for this node has been
3915 required_kernelcore
-= min(required_kernelcore
,
3917 kernelcore_remaining
-= size_pages
;
3918 if (!kernelcore_remaining
)
3924 * If there is still required_kernelcore, we do another pass with one
3925 * less node in the count. This will push zone_movable_pfn[nid] further
3926 * along on the nodes that still have memory until kernelcore is
3930 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3933 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3934 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3935 zone_movable_pfn
[nid
] =
3936 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3939 /* Any regular memory on that node ? */
3940 static void check_for_regular_memory(pg_data_t
*pgdat
)
3942 #ifdef CONFIG_HIGHMEM
3943 enum zone_type zone_type
;
3945 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3946 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3947 if (zone
->present_pages
)
3948 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3954 * free_area_init_nodes - Initialise all pg_data_t and zone data
3955 * @max_zone_pfn: an array of max PFNs for each zone
3957 * This will call free_area_init_node() for each active node in the system.
3958 * Using the page ranges provided by add_active_range(), the size of each
3959 * zone in each node and their holes is calculated. If the maximum PFN
3960 * between two adjacent zones match, it is assumed that the zone is empty.
3961 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3962 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3963 * starts where the previous one ended. For example, ZONE_DMA32 starts
3964 * at arch_max_dma_pfn.
3966 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3971 /* Sort early_node_map as initialisation assumes it is sorted */
3974 /* Record where the zone boundaries are */
3975 memset(arch_zone_lowest_possible_pfn
, 0,
3976 sizeof(arch_zone_lowest_possible_pfn
));
3977 memset(arch_zone_highest_possible_pfn
, 0,
3978 sizeof(arch_zone_highest_possible_pfn
));
3979 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3980 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3981 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3982 if (i
== ZONE_MOVABLE
)
3984 arch_zone_lowest_possible_pfn
[i
] =
3985 arch_zone_highest_possible_pfn
[i
-1];
3986 arch_zone_highest_possible_pfn
[i
] =
3987 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3989 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3990 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3992 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3993 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3994 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3996 /* Print out the zone ranges */
3997 printk("Zone PFN ranges:\n");
3998 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3999 if (i
== ZONE_MOVABLE
)
4001 printk(" %-8s %0#10lx -> %0#10lx\n",
4003 arch_zone_lowest_possible_pfn
[i
],
4004 arch_zone_highest_possible_pfn
[i
]);
4007 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4008 printk("Movable zone start PFN for each node\n");
4009 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4010 if (zone_movable_pfn
[i
])
4011 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4014 /* Print out the early_node_map[] */
4015 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4016 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4017 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4018 early_node_map
[i
].start_pfn
,
4019 early_node_map
[i
].end_pfn
);
4021 /* Initialise every node */
4022 mminit_verify_pageflags_layout();
4023 setup_nr_node_ids();
4024 for_each_online_node(nid
) {
4025 pg_data_t
*pgdat
= NODE_DATA(nid
);
4026 free_area_init_node(nid
, NULL
,
4027 find_min_pfn_for_node(nid
), NULL
);
4029 /* Any memory on that node */
4030 if (pgdat
->node_present_pages
)
4031 node_set_state(nid
, N_HIGH_MEMORY
);
4032 check_for_regular_memory(pgdat
);
4036 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4038 unsigned long long coremem
;
4042 coremem
= memparse(p
, &p
);
4043 *core
= coremem
>> PAGE_SHIFT
;
4045 /* Paranoid check that UL is enough for the coremem value */
4046 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4052 * kernelcore=size sets the amount of memory for use for allocations that
4053 * cannot be reclaimed or migrated.
4055 static int __init
cmdline_parse_kernelcore(char *p
)
4057 return cmdline_parse_core(p
, &required_kernelcore
);
4061 * movablecore=size sets the amount of memory for use for allocations that
4062 * can be reclaimed or migrated.
4064 static int __init
cmdline_parse_movablecore(char *p
)
4066 return cmdline_parse_core(p
, &required_movablecore
);
4069 early_param("kernelcore", cmdline_parse_kernelcore
);
4070 early_param("movablecore", cmdline_parse_movablecore
);
4072 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4075 * set_dma_reserve - set the specified number of pages reserved in the first zone
4076 * @new_dma_reserve: The number of pages to mark reserved
4078 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4079 * In the DMA zone, a significant percentage may be consumed by kernel image
4080 * and other unfreeable allocations which can skew the watermarks badly. This
4081 * function may optionally be used to account for unfreeable pages in the
4082 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4083 * smaller per-cpu batchsize.
4085 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4087 dma_reserve
= new_dma_reserve
;
4090 #ifndef CONFIG_NEED_MULTIPLE_NODES
4091 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4092 EXPORT_SYMBOL(contig_page_data
);
4095 void __init
free_area_init(unsigned long *zones_size
)
4097 free_area_init_node(0, zones_size
,
4098 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4101 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4102 unsigned long action
, void *hcpu
)
4104 int cpu
= (unsigned long)hcpu
;
4106 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4110 * Spill the event counters of the dead processor
4111 * into the current processors event counters.
4112 * This artificially elevates the count of the current
4115 vm_events_fold_cpu(cpu
);
4118 * Zero the differential counters of the dead processor
4119 * so that the vm statistics are consistent.
4121 * This is only okay since the processor is dead and cannot
4122 * race with what we are doing.
4124 refresh_cpu_vm_stats(cpu
);
4129 void __init
page_alloc_init(void)
4131 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4135 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4136 * or min_free_kbytes changes.
4138 static void calculate_totalreserve_pages(void)
4140 struct pglist_data
*pgdat
;
4141 unsigned long reserve_pages
= 0;
4142 enum zone_type i
, j
;
4144 for_each_online_pgdat(pgdat
) {
4145 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4146 struct zone
*zone
= pgdat
->node_zones
+ i
;
4147 unsigned long max
= 0;
4149 /* Find valid and maximum lowmem_reserve in the zone */
4150 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4151 if (zone
->lowmem_reserve
[j
] > max
)
4152 max
= zone
->lowmem_reserve
[j
];
4155 /* we treat pages_high as reserved pages. */
4156 max
+= zone
->pages_high
;
4158 if (max
> zone
->present_pages
)
4159 max
= zone
->present_pages
;
4160 reserve_pages
+= max
;
4163 totalreserve_pages
= reserve_pages
;
4167 * setup_per_zone_lowmem_reserve - called whenever
4168 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4169 * has a correct pages reserved value, so an adequate number of
4170 * pages are left in the zone after a successful __alloc_pages().
4172 static void setup_per_zone_lowmem_reserve(void)
4174 struct pglist_data
*pgdat
;
4175 enum zone_type j
, idx
;
4177 for_each_online_pgdat(pgdat
) {
4178 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4179 struct zone
*zone
= pgdat
->node_zones
+ j
;
4180 unsigned long present_pages
= zone
->present_pages
;
4182 zone
->lowmem_reserve
[j
] = 0;
4186 struct zone
*lower_zone
;
4190 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4191 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4193 lower_zone
= pgdat
->node_zones
+ idx
;
4194 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4195 sysctl_lowmem_reserve_ratio
[idx
];
4196 present_pages
+= lower_zone
->present_pages
;
4201 /* update totalreserve_pages */
4202 calculate_totalreserve_pages();
4206 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4208 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4209 * with respect to min_free_kbytes.
4211 void setup_per_zone_pages_min(void)
4213 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4214 unsigned long lowmem_pages
= 0;
4216 unsigned long flags
;
4218 /* Calculate total number of !ZONE_HIGHMEM pages */
4219 for_each_zone(zone
) {
4220 if (!is_highmem(zone
))
4221 lowmem_pages
+= zone
->present_pages
;
4224 for_each_zone(zone
) {
4227 spin_lock_irqsave(&zone
->lru_lock
, flags
);
4228 tmp
= (u64
)pages_min
* zone
->present_pages
;
4229 do_div(tmp
, lowmem_pages
);
4230 if (is_highmem(zone
)) {
4232 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4233 * need highmem pages, so cap pages_min to a small
4236 * The (pages_high-pages_low) and (pages_low-pages_min)
4237 * deltas controls asynch page reclaim, and so should
4238 * not be capped for highmem.
4242 min_pages
= zone
->present_pages
/ 1024;
4243 if (min_pages
< SWAP_CLUSTER_MAX
)
4244 min_pages
= SWAP_CLUSTER_MAX
;
4245 if (min_pages
> 128)
4247 zone
->pages_min
= min_pages
;
4250 * If it's a lowmem zone, reserve a number of pages
4251 * proportionate to the zone's size.
4253 zone
->pages_min
= tmp
;
4256 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4257 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4258 setup_zone_migrate_reserve(zone
);
4259 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4262 /* update totalreserve_pages */
4263 calculate_totalreserve_pages();
4267 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4269 * The inactive anon list should be small enough that the VM never has to
4270 * do too much work, but large enough that each inactive page has a chance
4271 * to be referenced again before it is swapped out.
4273 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4274 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4275 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4276 * the anonymous pages are kept on the inactive list.
4279 * memory ratio inactive anon
4280 * -------------------------------------
4289 void setup_per_zone_inactive_ratio(void)
4293 for_each_zone(zone
) {
4294 unsigned int gb
, ratio
;
4296 /* Zone size in gigabytes */
4297 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4298 ratio
= int_sqrt(10 * gb
);
4302 zone
->inactive_ratio
= ratio
;
4307 * Initialise min_free_kbytes.
4309 * For small machines we want it small (128k min). For large machines
4310 * we want it large (64MB max). But it is not linear, because network
4311 * bandwidth does not increase linearly with machine size. We use
4313 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4314 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4330 static int __init
init_per_zone_pages_min(void)
4332 unsigned long lowmem_kbytes
;
4334 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4336 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4337 if (min_free_kbytes
< 128)
4338 min_free_kbytes
= 128;
4339 if (min_free_kbytes
> 65536)
4340 min_free_kbytes
= 65536;
4341 setup_per_zone_pages_min();
4342 setup_per_zone_lowmem_reserve();
4343 setup_per_zone_inactive_ratio();
4346 module_init(init_per_zone_pages_min
)
4349 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4350 * that we can call two helper functions whenever min_free_kbytes
4353 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4354 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4356 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4358 setup_per_zone_pages_min();
4363 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4364 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4369 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4374 zone
->min_unmapped_pages
= (zone
->present_pages
*
4375 sysctl_min_unmapped_ratio
) / 100;
4379 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4380 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4385 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4390 zone
->min_slab_pages
= (zone
->present_pages
*
4391 sysctl_min_slab_ratio
) / 100;
4397 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4398 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4399 * whenever sysctl_lowmem_reserve_ratio changes.
4401 * The reserve ratio obviously has absolutely no relation with the
4402 * pages_min watermarks. The lowmem reserve ratio can only make sense
4403 * if in function of the boot time zone sizes.
4405 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4406 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4408 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4409 setup_per_zone_lowmem_reserve();
4414 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4415 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4416 * can have before it gets flushed back to buddy allocator.
4419 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4420 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4426 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4427 if (!write
|| (ret
== -EINVAL
))
4429 for_each_zone(zone
) {
4430 for_each_online_cpu(cpu
) {
4432 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4433 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4439 int hashdist
= HASHDIST_DEFAULT
;
4442 static int __init
set_hashdist(char *str
)
4446 hashdist
= simple_strtoul(str
, &str
, 0);
4449 __setup("hashdist=", set_hashdist
);
4453 * allocate a large system hash table from bootmem
4454 * - it is assumed that the hash table must contain an exact power-of-2
4455 * quantity of entries
4456 * - limit is the number of hash buckets, not the total allocation size
4458 void *__init
alloc_large_system_hash(const char *tablename
,
4459 unsigned long bucketsize
,
4460 unsigned long numentries
,
4463 unsigned int *_hash_shift
,
4464 unsigned int *_hash_mask
,
4465 unsigned long limit
)
4467 unsigned long long max
= limit
;
4468 unsigned long log2qty
, size
;
4471 /* allow the kernel cmdline to have a say */
4473 /* round applicable memory size up to nearest megabyte */
4474 numentries
= nr_kernel_pages
;
4475 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4476 numentries
>>= 20 - PAGE_SHIFT
;
4477 numentries
<<= 20 - PAGE_SHIFT
;
4479 /* limit to 1 bucket per 2^scale bytes of low memory */
4480 if (scale
> PAGE_SHIFT
)
4481 numentries
>>= (scale
- PAGE_SHIFT
);
4483 numentries
<<= (PAGE_SHIFT
- scale
);
4485 /* Make sure we've got at least a 0-order allocation.. */
4486 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4487 numentries
= PAGE_SIZE
/ bucketsize
;
4489 numentries
= roundup_pow_of_two(numentries
);
4491 /* limit allocation size to 1/16 total memory by default */
4493 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4494 do_div(max
, bucketsize
);
4497 if (numentries
> max
)
4500 log2qty
= ilog2(numentries
);
4503 size
= bucketsize
<< log2qty
;
4504 if (flags
& HASH_EARLY
)
4505 table
= alloc_bootmem_nopanic(size
);
4507 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4509 unsigned long order
= get_order(size
);
4510 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4512 * If bucketsize is not a power-of-two, we may free
4513 * some pages at the end of hash table.
4516 unsigned long alloc_end
= (unsigned long)table
+
4517 (PAGE_SIZE
<< order
);
4518 unsigned long used
= (unsigned long)table
+
4520 split_page(virt_to_page(table
), order
);
4521 while (used
< alloc_end
) {
4527 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4530 panic("Failed to allocate %s hash table\n", tablename
);
4532 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4535 ilog2(size
) - PAGE_SHIFT
,
4539 *_hash_shift
= log2qty
;
4541 *_hash_mask
= (1 << log2qty
) - 1;
4546 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4547 struct page
*pfn_to_page(unsigned long pfn
)
4549 return __pfn_to_page(pfn
);
4551 unsigned long page_to_pfn(struct page
*page
)
4553 return __page_to_pfn(page
);
4555 EXPORT_SYMBOL(pfn_to_page
);
4556 EXPORT_SYMBOL(page_to_pfn
);
4557 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4559 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4560 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4563 #ifdef CONFIG_SPARSEMEM
4564 return __pfn_to_section(pfn
)->pageblock_flags
;
4566 return zone
->pageblock_flags
;
4567 #endif /* CONFIG_SPARSEMEM */
4570 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4572 #ifdef CONFIG_SPARSEMEM
4573 pfn
&= (PAGES_PER_SECTION
-1);
4574 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4576 pfn
= pfn
- zone
->zone_start_pfn
;
4577 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4578 #endif /* CONFIG_SPARSEMEM */
4582 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4583 * @page: The page within the block of interest
4584 * @start_bitidx: The first bit of interest to retrieve
4585 * @end_bitidx: The last bit of interest
4586 * returns pageblock_bits flags
4588 unsigned long get_pageblock_flags_group(struct page
*page
,
4589 int start_bitidx
, int end_bitidx
)
4592 unsigned long *bitmap
;
4593 unsigned long pfn
, bitidx
;
4594 unsigned long flags
= 0;
4595 unsigned long value
= 1;
4597 zone
= page_zone(page
);
4598 pfn
= page_to_pfn(page
);
4599 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4600 bitidx
= pfn_to_bitidx(zone
, pfn
);
4602 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4603 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4610 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4611 * @page: The page within the block of interest
4612 * @start_bitidx: The first bit of interest
4613 * @end_bitidx: The last bit of interest
4614 * @flags: The flags to set
4616 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4617 int start_bitidx
, int end_bitidx
)
4620 unsigned long *bitmap
;
4621 unsigned long pfn
, bitidx
;
4622 unsigned long value
= 1;
4624 zone
= page_zone(page
);
4625 pfn
= page_to_pfn(page
);
4626 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4627 bitidx
= pfn_to_bitidx(zone
, pfn
);
4628 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4629 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4631 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4633 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4635 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4639 * This is designed as sub function...plz see page_isolation.c also.
4640 * set/clear page block's type to be ISOLATE.
4641 * page allocater never alloc memory from ISOLATE block.
4644 int set_migratetype_isolate(struct page
*page
)
4647 unsigned long flags
;
4650 zone
= page_zone(page
);
4651 spin_lock_irqsave(&zone
->lock
, flags
);
4653 * In future, more migrate types will be able to be isolation target.
4655 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4657 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4658 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4661 spin_unlock_irqrestore(&zone
->lock
, flags
);
4667 void unset_migratetype_isolate(struct page
*page
)
4670 unsigned long flags
;
4671 zone
= page_zone(page
);
4672 spin_lock_irqsave(&zone
->lock
, flags
);
4673 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4675 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4676 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4678 spin_unlock_irqrestore(&zone
->lock
, flags
);
4681 #ifdef CONFIG_MEMORY_HOTREMOVE
4683 * All pages in the range must be isolated before calling this.
4686 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4692 unsigned long flags
;
4693 /* find the first valid pfn */
4694 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4699 zone
= page_zone(pfn_to_page(pfn
));
4700 spin_lock_irqsave(&zone
->lock
, flags
);
4702 while (pfn
< end_pfn
) {
4703 if (!pfn_valid(pfn
)) {
4707 page
= pfn_to_page(pfn
);
4708 BUG_ON(page_count(page
));
4709 BUG_ON(!PageBuddy(page
));
4710 order
= page_order(page
);
4711 #ifdef CONFIG_DEBUG_VM
4712 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4713 pfn
, 1 << order
, end_pfn
);
4715 list_del(&page
->lru
);
4716 rmv_page_order(page
);
4717 zone
->free_area
[order
].nr_free
--;
4718 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4720 for (i
= 0; i
< (1 << order
); i
++)
4721 SetPageReserved((page
+i
));
4722 pfn
+= (1 << order
);
4724 spin_unlock_irqrestore(&zone
->lock
, flags
);