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/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
40 #include <linux/sort.h>
41 #include <linux/pfn.h>
42 #include <linux/backing-dev.h>
43 #include <linux/fault-inject.h>
45 #include <asm/tlbflush.h>
46 #include <asm/div64.h>
50 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
53 nodemask_t node_online_map __read_mostly
= { { [0] = 1UL } };
54 EXPORT_SYMBOL(node_online_map
);
55 nodemask_t node_possible_map __read_mostly
= NODE_MASK_ALL
;
56 EXPORT_SYMBOL(node_possible_map
);
57 unsigned long totalram_pages __read_mostly
;
58 unsigned long totalreserve_pages __read_mostly
;
60 int percpu_pagelist_fraction
;
62 static void __free_pages_ok(struct page
*page
, unsigned int order
);
65 * results with 256, 32 in the lowmem_reserve sysctl:
66 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
67 * 1G machine -> (16M dma, 784M normal, 224M high)
68 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
69 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
70 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
72 * TBD: should special case ZONE_DMA32 machines here - in those we normally
73 * don't need any ZONE_NORMAL reservation
75 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
76 #ifdef CONFIG_ZONE_DMA
79 #ifdef CONFIG_ZONE_DMA32
88 EXPORT_SYMBOL(totalram_pages
);
90 static char * const zone_names
[MAX_NR_ZONES
] = {
91 #ifdef CONFIG_ZONE_DMA
94 #ifdef CONFIG_ZONE_DMA32
104 int min_free_kbytes
= 1024;
106 unsigned long __meminitdata nr_kernel_pages
;
107 unsigned long __meminitdata nr_all_pages
;
108 static unsigned long __meminitdata dma_reserve
;
110 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
112 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
113 * ranges of memory (RAM) that may be registered with add_active_range().
114 * Ranges passed to add_active_range() will be merged if possible
115 * so the number of times add_active_range() can be called is
116 * related to the number of nodes and the number of holes
118 #ifdef CONFIG_MAX_ACTIVE_REGIONS
119 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
120 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
122 #if MAX_NUMNODES >= 32
123 /* If there can be many nodes, allow up to 50 holes per node */
124 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
126 /* By default, allow up to 256 distinct regions */
127 #define MAX_ACTIVE_REGIONS 256
131 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
132 static int __meminitdata nr_nodemap_entries
;
133 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
134 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
135 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
136 static unsigned long __meminitdata node_boundary_start_pfn
[MAX_NUMNODES
];
137 static unsigned long __meminitdata node_boundary_end_pfn
[MAX_NUMNODES
];
138 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
139 unsigned long __initdata required_kernelcore
;
140 unsigned long __initdata required_movablecore
;
141 unsigned long __initdata zone_movable_pfn
[MAX_NUMNODES
];
143 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
145 EXPORT_SYMBOL(movable_zone
);
146 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
149 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
150 EXPORT_SYMBOL(nr_node_ids
);
153 #ifdef CONFIG_DEBUG_VM
154 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
158 unsigned long pfn
= page_to_pfn(page
);
161 seq
= zone_span_seqbegin(zone
);
162 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
164 else if (pfn
< zone
->zone_start_pfn
)
166 } while (zone_span_seqretry(zone
, seq
));
171 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
173 if (!pfn_valid_within(page_to_pfn(page
)))
175 if (zone
!= page_zone(page
))
181 * Temporary debugging check for pages not lying within a given zone.
183 static int bad_range(struct zone
*zone
, struct page
*page
)
185 if (page_outside_zone_boundaries(zone
, page
))
187 if (!page_is_consistent(zone
, page
))
193 static inline int bad_range(struct zone
*zone
, struct page
*page
)
199 static void bad_page(struct page
*page
)
201 printk(KERN_EMERG
"Bad page state in process '%s'\n"
202 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
203 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
204 KERN_EMERG
"Backtrace:\n",
205 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
206 (unsigned long)page
->flags
, page
->mapping
,
207 page_mapcount(page
), page_count(page
));
209 page
->flags
&= ~(1 << PG_lru
|
219 set_page_count(page
, 0);
220 reset_page_mapcount(page
);
221 page
->mapping
= NULL
;
222 add_taint(TAINT_BAD_PAGE
);
226 * Higher-order pages are called "compound pages". They are structured thusly:
228 * The first PAGE_SIZE page is called the "head page".
230 * The remaining PAGE_SIZE pages are called "tail pages".
232 * All pages have PG_compound set. All pages have their ->private pointing at
233 * the head page (even the head page has this).
235 * The first tail page's ->lru.next holds the address of the compound page's
236 * put_page() function. Its ->lru.prev holds the order of allocation.
237 * This usage means that zero-order pages may not be compound.
240 static void free_compound_page(struct page
*page
)
242 __free_pages_ok(page
, compound_order(page
));
245 static void prep_compound_page(struct page
*page
, unsigned long order
)
248 int nr_pages
= 1 << order
;
250 set_compound_page_dtor(page
, free_compound_page
);
251 set_compound_order(page
, order
);
253 for (i
= 1; i
< nr_pages
; i
++) {
254 struct page
*p
= page
+ i
;
257 p
->first_page
= page
;
261 static void destroy_compound_page(struct page
*page
, unsigned long order
)
264 int nr_pages
= 1 << order
;
266 if (unlikely(compound_order(page
) != order
))
269 if (unlikely(!PageHead(page
)))
271 __ClearPageHead(page
);
272 for (i
= 1; i
< nr_pages
; i
++) {
273 struct page
*p
= page
+ i
;
275 if (unlikely(!PageTail(p
) |
276 (p
->first_page
!= page
)))
282 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
286 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
288 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
289 * and __GFP_HIGHMEM from hard or soft interrupt context.
291 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
292 for (i
= 0; i
< (1 << order
); i
++)
293 clear_highpage(page
+ i
);
297 * function for dealing with page's order in buddy system.
298 * zone->lock is already acquired when we use these.
299 * So, we don't need atomic page->flags operations here.
301 static inline unsigned long page_order(struct page
*page
)
303 return page_private(page
);
306 static inline void set_page_order(struct page
*page
, int order
)
308 set_page_private(page
, order
);
309 __SetPageBuddy(page
);
312 static inline void rmv_page_order(struct page
*page
)
314 __ClearPageBuddy(page
);
315 set_page_private(page
, 0);
319 * Locate the struct page for both the matching buddy in our
320 * pair (buddy1) and the combined O(n+1) page they form (page).
322 * 1) Any buddy B1 will have an order O twin B2 which satisfies
323 * the following equation:
325 * For example, if the starting buddy (buddy2) is #8 its order
327 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
329 * 2) Any buddy B will have an order O+1 parent P which
330 * satisfies the following equation:
333 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
335 static inline struct page
*
336 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
338 unsigned long buddy_idx
= page_idx
^ (1 << order
);
340 return page
+ (buddy_idx
- page_idx
);
343 static inline unsigned long
344 __find_combined_index(unsigned long page_idx
, unsigned int order
)
346 return (page_idx
& ~(1 << order
));
350 * This function checks whether a page is free && is the buddy
351 * we can do coalesce a page and its buddy if
352 * (a) the buddy is not in a hole &&
353 * (b) the buddy is in the buddy system &&
354 * (c) a page and its buddy have the same order &&
355 * (d) a page and its buddy are in the same zone.
357 * For recording whether a page is in the buddy system, we use PG_buddy.
358 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
360 * For recording page's order, we use page_private(page).
362 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
365 if (!pfn_valid_within(page_to_pfn(buddy
)))
368 if (page_zone_id(page
) != page_zone_id(buddy
))
371 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
372 BUG_ON(page_count(buddy
) != 0);
379 * Freeing function for a buddy system allocator.
381 * The concept of a buddy system is to maintain direct-mapped table
382 * (containing bit values) for memory blocks of various "orders".
383 * The bottom level table contains the map for the smallest allocatable
384 * units of memory (here, pages), and each level above it describes
385 * pairs of units from the levels below, hence, "buddies".
386 * At a high level, all that happens here is marking the table entry
387 * at the bottom level available, and propagating the changes upward
388 * as necessary, plus some accounting needed to play nicely with other
389 * parts of the VM system.
390 * At each level, we keep a list of pages, which are heads of continuous
391 * free pages of length of (1 << order) and marked with PG_buddy. Page's
392 * order is recorded in page_private(page) field.
393 * So when we are allocating or freeing one, we can derive the state of the
394 * other. That is, if we allocate a small block, and both were
395 * free, the remainder of the region must be split into blocks.
396 * If a block is freed, and its buddy is also free, then this
397 * triggers coalescing into a block of larger size.
402 static inline void __free_one_page(struct page
*page
,
403 struct zone
*zone
, unsigned int order
)
405 unsigned long page_idx
;
406 int order_size
= 1 << order
;
408 if (unlikely(PageCompound(page
)))
409 destroy_compound_page(page
, order
);
411 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
413 VM_BUG_ON(page_idx
& (order_size
- 1));
414 VM_BUG_ON(bad_range(zone
, page
));
416 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
417 while (order
< MAX_ORDER
-1) {
418 unsigned long combined_idx
;
419 struct free_area
*area
;
422 buddy
= __page_find_buddy(page
, page_idx
, order
);
423 if (!page_is_buddy(page
, buddy
, order
))
424 break; /* Move the buddy up one level. */
426 list_del(&buddy
->lru
);
427 area
= zone
->free_area
+ order
;
429 rmv_page_order(buddy
);
430 combined_idx
= __find_combined_index(page_idx
, order
);
431 page
= page
+ (combined_idx
- page_idx
);
432 page_idx
= combined_idx
;
435 set_page_order(page
, order
);
436 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
437 zone
->free_area
[order
].nr_free
++;
440 static inline int free_pages_check(struct page
*page
)
442 if (unlikely(page_mapcount(page
) |
443 (page
->mapping
!= NULL
) |
444 (page_count(page
) != 0) |
457 __ClearPageDirty(page
);
459 * For now, we report if PG_reserved was found set, but do not
460 * clear it, and do not free the page. But we shall soon need
461 * to do more, for when the ZERO_PAGE count wraps negative.
463 return PageReserved(page
);
467 * Frees a list of pages.
468 * Assumes all pages on list are in same zone, and of same order.
469 * count is the number of pages to free.
471 * If the zone was previously in an "all pages pinned" state then look to
472 * see if this freeing clears that state.
474 * And clear the zone's pages_scanned counter, to hold off the "all pages are
475 * pinned" detection logic.
477 static void free_pages_bulk(struct zone
*zone
, int count
,
478 struct list_head
*list
, int order
)
480 spin_lock(&zone
->lock
);
481 zone
->all_unreclaimable
= 0;
482 zone
->pages_scanned
= 0;
486 VM_BUG_ON(list_empty(list
));
487 page
= list_entry(list
->prev
, struct page
, lru
);
488 /* have to delete it as __free_one_page list manipulates */
489 list_del(&page
->lru
);
490 __free_one_page(page
, zone
, order
);
492 spin_unlock(&zone
->lock
);
495 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
497 spin_lock(&zone
->lock
);
498 zone
->all_unreclaimable
= 0;
499 zone
->pages_scanned
= 0;
500 __free_one_page(page
, zone
, order
);
501 spin_unlock(&zone
->lock
);
504 static void __free_pages_ok(struct page
*page
, unsigned int order
)
510 for (i
= 0 ; i
< (1 << order
) ; ++i
)
511 reserved
+= free_pages_check(page
+ i
);
515 if (!PageHighMem(page
))
516 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
517 arch_free_page(page
, order
);
518 kernel_map_pages(page
, 1 << order
, 0);
520 local_irq_save(flags
);
521 __count_vm_events(PGFREE
, 1 << order
);
522 free_one_page(page_zone(page
), page
, order
);
523 local_irq_restore(flags
);
527 * permit the bootmem allocator to evade page validation on high-order frees
529 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
532 __ClearPageReserved(page
);
533 set_page_count(page
, 0);
534 set_page_refcounted(page
);
540 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
541 struct page
*p
= &page
[loop
];
543 if (loop
+ 1 < BITS_PER_LONG
)
545 __ClearPageReserved(p
);
546 set_page_count(p
, 0);
549 set_page_refcounted(page
);
550 __free_pages(page
, order
);
556 * The order of subdivision here is critical for the IO subsystem.
557 * Please do not alter this order without good reasons and regression
558 * testing. Specifically, as large blocks of memory are subdivided,
559 * the order in which smaller blocks are delivered depends on the order
560 * they're subdivided in this function. This is the primary factor
561 * influencing the order in which pages are delivered to the IO
562 * subsystem according to empirical testing, and this is also justified
563 * by considering the behavior of a buddy system containing a single
564 * large block of memory acted on by a series of small allocations.
565 * This behavior is a critical factor in sglist merging's success.
569 static inline void expand(struct zone
*zone
, struct page
*page
,
570 int low
, int high
, struct free_area
*area
)
572 unsigned long size
= 1 << high
;
578 VM_BUG_ON(bad_range(zone
, &page
[size
]));
579 list_add(&page
[size
].lru
, &area
->free_list
);
581 set_page_order(&page
[size
], high
);
586 * This page is about to be returned from the page allocator
588 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
590 if (unlikely(page_mapcount(page
) |
591 (page
->mapping
!= NULL
) |
592 (page_count(page
) != 0) |
607 * For now, we report if PG_reserved was found set, but do not
608 * clear it, and do not allocate the page: as a safety net.
610 if (PageReserved(page
))
613 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_readahead
|
614 1 << PG_referenced
| 1 << PG_arch_1
|
615 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
);
616 set_page_private(page
, 0);
617 set_page_refcounted(page
);
619 arch_alloc_page(page
, order
);
620 kernel_map_pages(page
, 1 << order
, 1);
622 if (gfp_flags
& __GFP_ZERO
)
623 prep_zero_page(page
, order
, gfp_flags
);
625 if (order
&& (gfp_flags
& __GFP_COMP
))
626 prep_compound_page(page
, order
);
632 * Do the hard work of removing an element from the buddy allocator.
633 * Call me with the zone->lock already held.
635 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
637 struct free_area
* area
;
638 unsigned int current_order
;
641 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
642 area
= zone
->free_area
+ current_order
;
643 if (list_empty(&area
->free_list
))
646 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
647 list_del(&page
->lru
);
648 rmv_page_order(page
);
650 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
651 expand(zone
, page
, order
, current_order
, area
);
659 * Obtain a specified number of elements from the buddy allocator, all under
660 * a single hold of the lock, for efficiency. Add them to the supplied list.
661 * Returns the number of new pages which were placed at *list.
663 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
664 unsigned long count
, struct list_head
*list
)
668 spin_lock(&zone
->lock
);
669 for (i
= 0; i
< count
; ++i
) {
670 struct page
*page
= __rmqueue(zone
, order
);
671 if (unlikely(page
== NULL
))
673 list_add_tail(&page
->lru
, list
);
675 spin_unlock(&zone
->lock
);
681 * Called from the vmstat counter updater to drain pagesets of this
682 * currently executing processor on remote nodes after they have
685 * Note that this function must be called with the thread pinned to
686 * a single processor.
688 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
693 local_irq_save(flags
);
694 if (pcp
->count
>= pcp
->batch
)
695 to_drain
= pcp
->batch
;
697 to_drain
= pcp
->count
;
698 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
699 pcp
->count
-= to_drain
;
700 local_irq_restore(flags
);
704 static void __drain_pages(unsigned int cpu
)
710 for_each_zone(zone
) {
711 struct per_cpu_pageset
*pset
;
713 if (!populated_zone(zone
))
716 pset
= zone_pcp(zone
, cpu
);
717 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
718 struct per_cpu_pages
*pcp
;
721 local_irq_save(flags
);
722 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
724 local_irq_restore(flags
);
731 void mark_free_pages(struct zone
*zone
)
733 unsigned long pfn
, max_zone_pfn
;
736 struct list_head
*curr
;
738 if (!zone
->spanned_pages
)
741 spin_lock_irqsave(&zone
->lock
, flags
);
743 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
744 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
745 if (pfn_valid(pfn
)) {
746 struct page
*page
= pfn_to_page(pfn
);
748 if (!swsusp_page_is_forbidden(page
))
749 swsusp_unset_page_free(page
);
752 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
753 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
756 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
757 for (i
= 0; i
< (1UL << order
); i
++)
758 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
761 spin_unlock_irqrestore(&zone
->lock
, flags
);
765 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
767 void drain_local_pages(void)
771 local_irq_save(flags
);
772 __drain_pages(smp_processor_id());
773 local_irq_restore(flags
);
775 #endif /* CONFIG_PM */
778 * Free a 0-order page
780 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
782 struct zone
*zone
= page_zone(page
);
783 struct per_cpu_pages
*pcp
;
787 page
->mapping
= NULL
;
788 if (free_pages_check(page
))
791 if (!PageHighMem(page
))
792 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
793 arch_free_page(page
, 0);
794 kernel_map_pages(page
, 1, 0);
796 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
797 local_irq_save(flags
);
798 __count_vm_event(PGFREE
);
799 list_add(&page
->lru
, &pcp
->list
);
801 if (pcp
->count
>= pcp
->high
) {
802 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
803 pcp
->count
-= pcp
->batch
;
805 local_irq_restore(flags
);
809 void fastcall
free_hot_page(struct page
*page
)
811 free_hot_cold_page(page
, 0);
814 void fastcall
free_cold_page(struct page
*page
)
816 free_hot_cold_page(page
, 1);
820 * split_page takes a non-compound higher-order page, and splits it into
821 * n (1<<order) sub-pages: page[0..n]
822 * Each sub-page must be freed individually.
824 * Note: this is probably too low level an operation for use in drivers.
825 * Please consult with lkml before using this in your driver.
827 void split_page(struct page
*page
, unsigned int order
)
831 VM_BUG_ON(PageCompound(page
));
832 VM_BUG_ON(!page_count(page
));
833 for (i
= 1; i
< (1 << order
); i
++)
834 set_page_refcounted(page
+ i
);
838 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
839 * we cheat by calling it from here, in the order > 0 path. Saves a branch
842 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
843 struct zone
*zone
, int order
, gfp_t gfp_flags
)
847 int cold
= !!(gfp_flags
& __GFP_COLD
);
852 if (likely(order
== 0)) {
853 struct per_cpu_pages
*pcp
;
855 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
856 local_irq_save(flags
);
858 pcp
->count
= rmqueue_bulk(zone
, 0,
859 pcp
->batch
, &pcp
->list
);
860 if (unlikely(!pcp
->count
))
863 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
864 list_del(&page
->lru
);
867 spin_lock_irqsave(&zone
->lock
, flags
);
868 page
= __rmqueue(zone
, order
);
869 spin_unlock(&zone
->lock
);
874 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
875 zone_statistics(zonelist
, zone
);
876 local_irq_restore(flags
);
879 VM_BUG_ON(bad_range(zone
, page
));
880 if (prep_new_page(page
, order
, gfp_flags
))
885 local_irq_restore(flags
);
890 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
891 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
892 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
893 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
894 #define ALLOC_HARDER 0x10 /* try to alloc harder */
895 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
896 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
898 #ifdef CONFIG_FAIL_PAGE_ALLOC
900 static struct fail_page_alloc_attr
{
901 struct fault_attr attr
;
903 u32 ignore_gfp_highmem
;
907 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
909 struct dentry
*ignore_gfp_highmem_file
;
910 struct dentry
*ignore_gfp_wait_file
;
911 struct dentry
*min_order_file
;
913 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
915 } fail_page_alloc
= {
916 .attr
= FAULT_ATTR_INITIALIZER
,
917 .ignore_gfp_wait
= 1,
918 .ignore_gfp_highmem
= 1,
922 static int __init
setup_fail_page_alloc(char *str
)
924 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
926 __setup("fail_page_alloc=", setup_fail_page_alloc
);
928 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
930 if (order
< fail_page_alloc
.min_order
)
932 if (gfp_mask
& __GFP_NOFAIL
)
934 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
936 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
939 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
942 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
944 static int __init
fail_page_alloc_debugfs(void)
946 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
950 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
954 dir
= fail_page_alloc
.attr
.dentries
.dir
;
956 fail_page_alloc
.ignore_gfp_wait_file
=
957 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
958 &fail_page_alloc
.ignore_gfp_wait
);
960 fail_page_alloc
.ignore_gfp_highmem_file
=
961 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
962 &fail_page_alloc
.ignore_gfp_highmem
);
963 fail_page_alloc
.min_order_file
=
964 debugfs_create_u32("min-order", mode
, dir
,
965 &fail_page_alloc
.min_order
);
967 if (!fail_page_alloc
.ignore_gfp_wait_file
||
968 !fail_page_alloc
.ignore_gfp_highmem_file
||
969 !fail_page_alloc
.min_order_file
) {
971 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
972 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
973 debugfs_remove(fail_page_alloc
.min_order_file
);
974 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
980 late_initcall(fail_page_alloc_debugfs
);
982 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
984 #else /* CONFIG_FAIL_PAGE_ALLOC */
986 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
991 #endif /* CONFIG_FAIL_PAGE_ALLOC */
994 * Return 1 if free pages are above 'mark'. This takes into account the order
997 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
998 int classzone_idx
, int alloc_flags
)
1000 /* free_pages my go negative - that's OK */
1002 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1005 if (alloc_flags
& ALLOC_HIGH
)
1007 if (alloc_flags
& ALLOC_HARDER
)
1010 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1012 for (o
= 0; o
< order
; o
++) {
1013 /* At the next order, this order's pages become unavailable */
1014 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1016 /* Require fewer higher order pages to be free */
1019 if (free_pages
<= min
)
1027 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1028 * skip over zones that are not allowed by the cpuset, or that have
1029 * been recently (in last second) found to be nearly full. See further
1030 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1031 * that have to skip over alot of full or unallowed zones.
1033 * If the zonelist cache is present in the passed in zonelist, then
1034 * returns a pointer to the allowed node mask (either the current
1035 * tasks mems_allowed, or node_online_map.)
1037 * If the zonelist cache is not available for this zonelist, does
1038 * nothing and returns NULL.
1040 * If the fullzones BITMAP in the zonelist cache is stale (more than
1041 * a second since last zap'd) then we zap it out (clear its bits.)
1043 * We hold off even calling zlc_setup, until after we've checked the
1044 * first zone in the zonelist, on the theory that most allocations will
1045 * be satisfied from that first zone, so best to examine that zone as
1046 * quickly as we can.
1048 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1050 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1051 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1053 zlc
= zonelist
->zlcache_ptr
;
1057 if (jiffies
- zlc
->last_full_zap
> 1 * HZ
) {
1058 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1059 zlc
->last_full_zap
= jiffies
;
1062 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1063 &cpuset_current_mems_allowed
:
1065 return allowednodes
;
1069 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1070 * if it is worth looking at further for free memory:
1071 * 1) Check that the zone isn't thought to be full (doesn't have its
1072 * bit set in the zonelist_cache fullzones BITMAP).
1073 * 2) Check that the zones node (obtained from the zonelist_cache
1074 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1075 * Return true (non-zero) if zone is worth looking at further, or
1076 * else return false (zero) if it is not.
1078 * This check -ignores- the distinction between various watermarks,
1079 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1080 * found to be full for any variation of these watermarks, it will
1081 * be considered full for up to one second by all requests, unless
1082 * we are so low on memory on all allowed nodes that we are forced
1083 * into the second scan of the zonelist.
1085 * In the second scan we ignore this zonelist cache and exactly
1086 * apply the watermarks to all zones, even it is slower to do so.
1087 * We are low on memory in the second scan, and should leave no stone
1088 * unturned looking for a free page.
1090 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1091 nodemask_t
*allowednodes
)
1093 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1094 int i
; /* index of *z in zonelist zones */
1095 int n
; /* node that zone *z is on */
1097 zlc
= zonelist
->zlcache_ptr
;
1101 i
= z
- zonelist
->zones
;
1104 /* This zone is worth trying if it is allowed but not full */
1105 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1109 * Given 'z' scanning a zonelist, set the corresponding bit in
1110 * zlc->fullzones, so that subsequent attempts to allocate a page
1111 * from that zone don't waste time re-examining it.
1113 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1115 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1116 int i
; /* index of *z in zonelist zones */
1118 zlc
= zonelist
->zlcache_ptr
;
1122 i
= z
- zonelist
->zones
;
1124 set_bit(i
, zlc
->fullzones
);
1127 #else /* CONFIG_NUMA */
1129 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1134 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1135 nodemask_t
*allowednodes
)
1140 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1143 #endif /* CONFIG_NUMA */
1146 * get_page_from_freelist goes through the zonelist trying to allocate
1149 static struct page
*
1150 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
1151 struct zonelist
*zonelist
, int alloc_flags
)
1154 struct page
*page
= NULL
;
1155 int classzone_idx
= zone_idx(zonelist
->zones
[0]);
1157 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1158 int zlc_active
= 0; /* set if using zonelist_cache */
1159 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1163 * Scan zonelist, looking for a zone with enough free.
1164 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1166 z
= zonelist
->zones
;
1169 if (NUMA_BUILD
&& zlc_active
&&
1170 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1173 if (unlikely(NUMA_BUILD
&& (gfp_mask
& __GFP_THISNODE
) &&
1174 zone
->zone_pgdat
!= zonelist
->zones
[0]->zone_pgdat
))
1176 if ((alloc_flags
& ALLOC_CPUSET
) &&
1177 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1180 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1182 if (alloc_flags
& ALLOC_WMARK_MIN
)
1183 mark
= zone
->pages_min
;
1184 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1185 mark
= zone
->pages_low
;
1187 mark
= zone
->pages_high
;
1188 if (!zone_watermark_ok(zone
, order
, mark
,
1189 classzone_idx
, alloc_flags
)) {
1190 if (!zone_reclaim_mode
||
1191 !zone_reclaim(zone
, gfp_mask
, order
))
1192 goto this_zone_full
;
1196 page
= buffered_rmqueue(zonelist
, zone
, order
, gfp_mask
);
1201 zlc_mark_zone_full(zonelist
, z
);
1203 if (NUMA_BUILD
&& !did_zlc_setup
) {
1204 /* we do zlc_setup after the first zone is tried */
1205 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1209 } while (*(++z
) != NULL
);
1211 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1212 /* Disable zlc cache for second zonelist scan */
1220 * This is the 'heart' of the zoned buddy allocator.
1222 struct page
* fastcall
1223 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
1224 struct zonelist
*zonelist
)
1226 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1229 struct reclaim_state reclaim_state
;
1230 struct task_struct
*p
= current
;
1233 int did_some_progress
;
1235 might_sleep_if(wait
);
1237 if (should_fail_alloc_page(gfp_mask
, order
))
1241 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
1243 if (unlikely(*z
== NULL
)) {
1244 /* Should this ever happen?? */
1248 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1249 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1254 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1255 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1256 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1257 * using a larger set of nodes after it has established that the
1258 * allowed per node queues are empty and that nodes are
1261 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1264 for (z
= zonelist
->zones
; *z
; z
++)
1265 wakeup_kswapd(*z
, order
);
1268 * OK, we're below the kswapd watermark and have kicked background
1269 * reclaim. Now things get more complex, so set up alloc_flags according
1270 * to how we want to proceed.
1272 * The caller may dip into page reserves a bit more if the caller
1273 * cannot run direct reclaim, or if the caller has realtime scheduling
1274 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1275 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1277 alloc_flags
= ALLOC_WMARK_MIN
;
1278 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1279 alloc_flags
|= ALLOC_HARDER
;
1280 if (gfp_mask
& __GFP_HIGH
)
1281 alloc_flags
|= ALLOC_HIGH
;
1283 alloc_flags
|= ALLOC_CPUSET
;
1286 * Go through the zonelist again. Let __GFP_HIGH and allocations
1287 * coming from realtime tasks go deeper into reserves.
1289 * This is the last chance, in general, before the goto nopage.
1290 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1291 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1293 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1297 /* This allocation should allow future memory freeing. */
1300 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1301 && !in_interrupt()) {
1302 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1304 /* go through the zonelist yet again, ignoring mins */
1305 page
= get_page_from_freelist(gfp_mask
, order
,
1306 zonelist
, ALLOC_NO_WATERMARKS
);
1309 if (gfp_mask
& __GFP_NOFAIL
) {
1310 congestion_wait(WRITE
, HZ
/50);
1317 /* Atomic allocations - we can't balance anything */
1323 /* We now go into synchronous reclaim */
1324 cpuset_memory_pressure_bump();
1325 p
->flags
|= PF_MEMALLOC
;
1326 reclaim_state
.reclaimed_slab
= 0;
1327 p
->reclaim_state
= &reclaim_state
;
1329 did_some_progress
= try_to_free_pages(zonelist
->zones
, order
, gfp_mask
);
1331 p
->reclaim_state
= NULL
;
1332 p
->flags
&= ~PF_MEMALLOC
;
1336 if (likely(did_some_progress
)) {
1337 page
= get_page_from_freelist(gfp_mask
, order
,
1338 zonelist
, alloc_flags
);
1341 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1343 * Go through the zonelist yet one more time, keep
1344 * very high watermark here, this is only to catch
1345 * a parallel oom killing, we must fail if we're still
1346 * under heavy pressure.
1348 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1349 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1353 out_of_memory(zonelist
, gfp_mask
, order
);
1358 * Don't let big-order allocations loop unless the caller explicitly
1359 * requests that. Wait for some write requests to complete then retry.
1361 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1362 * <= 3, but that may not be true in other implementations.
1365 if (!(gfp_mask
& __GFP_NORETRY
)) {
1366 if ((order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
1367 (gfp_mask
& __GFP_REPEAT
))
1369 if (gfp_mask
& __GFP_NOFAIL
)
1373 congestion_wait(WRITE
, HZ
/50);
1378 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1379 printk(KERN_WARNING
"%s: page allocation failure."
1380 " order:%d, mode:0x%x\n",
1381 p
->comm
, order
, gfp_mask
);
1389 EXPORT_SYMBOL(__alloc_pages
);
1392 * Common helper functions.
1394 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1397 page
= alloc_pages(gfp_mask
, order
);
1400 return (unsigned long) page_address(page
);
1403 EXPORT_SYMBOL(__get_free_pages
);
1405 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1410 * get_zeroed_page() returns a 32-bit address, which cannot represent
1413 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1415 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1417 return (unsigned long) page_address(page
);
1421 EXPORT_SYMBOL(get_zeroed_page
);
1423 void __pagevec_free(struct pagevec
*pvec
)
1425 int i
= pagevec_count(pvec
);
1428 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1431 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1433 if (put_page_testzero(page
)) {
1435 free_hot_page(page
);
1437 __free_pages_ok(page
, order
);
1441 EXPORT_SYMBOL(__free_pages
);
1443 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1446 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1447 __free_pages(virt_to_page((void *)addr
), order
);
1451 EXPORT_SYMBOL(free_pages
);
1453 static unsigned int nr_free_zone_pages(int offset
)
1455 /* Just pick one node, since fallback list is circular */
1456 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1457 unsigned int sum
= 0;
1459 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1460 struct zone
**zonep
= zonelist
->zones
;
1463 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1464 unsigned long size
= zone
->present_pages
;
1465 unsigned long high
= zone
->pages_high
;
1474 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1476 unsigned int nr_free_buffer_pages(void)
1478 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1480 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1483 * Amount of free RAM allocatable within all zones
1485 unsigned int nr_free_pagecache_pages(void)
1487 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1490 static inline void show_node(struct zone
*zone
)
1493 printk("Node %d ", zone_to_nid(zone
));
1496 void si_meminfo(struct sysinfo
*val
)
1498 val
->totalram
= totalram_pages
;
1500 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1501 val
->bufferram
= nr_blockdev_pages();
1502 val
->totalhigh
= totalhigh_pages
;
1503 val
->freehigh
= nr_free_highpages();
1504 val
->mem_unit
= PAGE_SIZE
;
1507 EXPORT_SYMBOL(si_meminfo
);
1510 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1512 pg_data_t
*pgdat
= NODE_DATA(nid
);
1514 val
->totalram
= pgdat
->node_present_pages
;
1515 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1516 #ifdef CONFIG_HIGHMEM
1517 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1518 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1524 val
->mem_unit
= PAGE_SIZE
;
1528 #define K(x) ((x) << (PAGE_SHIFT-10))
1531 * Show free area list (used inside shift_scroll-lock stuff)
1532 * We also calculate the percentage fragmentation. We do this by counting the
1533 * memory on each free list with the exception of the first item on the list.
1535 void show_free_areas(void)
1540 for_each_zone(zone
) {
1541 if (!populated_zone(zone
))
1545 printk("%s per-cpu:\n", zone
->name
);
1547 for_each_online_cpu(cpu
) {
1548 struct per_cpu_pageset
*pageset
;
1550 pageset
= zone_pcp(zone
, cpu
);
1552 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1553 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1554 cpu
, pageset
->pcp
[0].high
,
1555 pageset
->pcp
[0].batch
, pageset
->pcp
[0].count
,
1556 pageset
->pcp
[1].high
, pageset
->pcp
[1].batch
,
1557 pageset
->pcp
[1].count
);
1561 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1562 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1563 global_page_state(NR_ACTIVE
),
1564 global_page_state(NR_INACTIVE
),
1565 global_page_state(NR_FILE_DIRTY
),
1566 global_page_state(NR_WRITEBACK
),
1567 global_page_state(NR_UNSTABLE_NFS
),
1568 global_page_state(NR_FREE_PAGES
),
1569 global_page_state(NR_SLAB_RECLAIMABLE
) +
1570 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1571 global_page_state(NR_FILE_MAPPED
),
1572 global_page_state(NR_PAGETABLE
),
1573 global_page_state(NR_BOUNCE
));
1575 for_each_zone(zone
) {
1578 if (!populated_zone(zone
))
1590 " pages_scanned:%lu"
1591 " all_unreclaimable? %s"
1594 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1597 K(zone
->pages_high
),
1598 K(zone_page_state(zone
, NR_ACTIVE
)),
1599 K(zone_page_state(zone
, NR_INACTIVE
)),
1600 K(zone
->present_pages
),
1601 zone
->pages_scanned
,
1602 (zone
->all_unreclaimable
? "yes" : "no")
1604 printk("lowmem_reserve[]:");
1605 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1606 printk(" %lu", zone
->lowmem_reserve
[i
]);
1610 for_each_zone(zone
) {
1611 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1613 if (!populated_zone(zone
))
1617 printk("%s: ", zone
->name
);
1619 spin_lock_irqsave(&zone
->lock
, flags
);
1620 for (order
= 0; order
< MAX_ORDER
; order
++) {
1621 nr
[order
] = zone
->free_area
[order
].nr_free
;
1622 total
+= nr
[order
] << order
;
1624 spin_unlock_irqrestore(&zone
->lock
, flags
);
1625 for (order
= 0; order
< MAX_ORDER
; order
++)
1626 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1627 printk("= %lukB\n", K(total
));
1630 show_swap_cache_info();
1634 * Builds allocation fallback zone lists.
1636 * Add all populated zones of a node to the zonelist.
1638 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1639 int nr_zones
, enum zone_type zone_type
)
1643 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1648 zone
= pgdat
->node_zones
+ zone_type
;
1649 if (populated_zone(zone
)) {
1650 zonelist
->zones
[nr_zones
++] = zone
;
1651 check_highest_zone(zone_type
);
1654 } while (zone_type
);
1661 * 0 = automatic detection of better ordering.
1662 * 1 = order by ([node] distance, -zonetype)
1663 * 2 = order by (-zonetype, [node] distance)
1665 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1666 * the same zonelist. So only NUMA can configure this param.
1668 #define ZONELIST_ORDER_DEFAULT 0
1669 #define ZONELIST_ORDER_NODE 1
1670 #define ZONELIST_ORDER_ZONE 2
1672 /* zonelist order in the kernel.
1673 * set_zonelist_order() will set this to NODE or ZONE.
1675 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1676 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
1680 /* The value user specified ....changed by config */
1681 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1682 /* string for sysctl */
1683 #define NUMA_ZONELIST_ORDER_LEN 16
1684 char numa_zonelist_order
[16] = "default";
1687 * interface for configure zonelist ordering.
1688 * command line option "numa_zonelist_order"
1689 * = "[dD]efault - default, automatic configuration.
1690 * = "[nN]ode - order by node locality, then by zone within node
1691 * = "[zZ]one - order by zone, then by locality within zone
1694 static int __parse_numa_zonelist_order(char *s
)
1696 if (*s
== 'd' || *s
== 'D') {
1697 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1698 } else if (*s
== 'n' || *s
== 'N') {
1699 user_zonelist_order
= ZONELIST_ORDER_NODE
;
1700 } else if (*s
== 'z' || *s
== 'Z') {
1701 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
1704 "Ignoring invalid numa_zonelist_order value: "
1711 static __init
int setup_numa_zonelist_order(char *s
)
1714 return __parse_numa_zonelist_order(s
);
1717 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
1720 * sysctl handler for numa_zonelist_order
1722 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
1723 struct file
*file
, void __user
*buffer
, size_t *length
,
1726 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
1730 strncpy(saved_string
, (char*)table
->data
,
1731 NUMA_ZONELIST_ORDER_LEN
);
1732 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
1736 int oldval
= user_zonelist_order
;
1737 if (__parse_numa_zonelist_order((char*)table
->data
)) {
1739 * bogus value. restore saved string
1741 strncpy((char*)table
->data
, saved_string
,
1742 NUMA_ZONELIST_ORDER_LEN
);
1743 user_zonelist_order
= oldval
;
1744 } else if (oldval
!= user_zonelist_order
)
1745 build_all_zonelists();
1751 #define MAX_NODE_LOAD (num_online_nodes())
1752 static int node_load
[MAX_NUMNODES
];
1755 * find_next_best_node - find the next node that should appear in a given node's fallback list
1756 * @node: node whose fallback list we're appending
1757 * @used_node_mask: nodemask_t of already used nodes
1759 * We use a number of factors to determine which is the next node that should
1760 * appear on a given node's fallback list. The node should not have appeared
1761 * already in @node's fallback list, and it should be the next closest node
1762 * according to the distance array (which contains arbitrary distance values
1763 * from each node to each node in the system), and should also prefer nodes
1764 * with no CPUs, since presumably they'll have very little allocation pressure
1765 * on them otherwise.
1766 * It returns -1 if no node is found.
1768 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1771 int min_val
= INT_MAX
;
1774 /* Use the local node if we haven't already */
1775 if (!node_isset(node
, *used_node_mask
)) {
1776 node_set(node
, *used_node_mask
);
1780 for_each_online_node(n
) {
1783 /* Don't want a node to appear more than once */
1784 if (node_isset(n
, *used_node_mask
))
1787 /* Use the distance array to find the distance */
1788 val
= node_distance(node
, n
);
1790 /* Penalize nodes under us ("prefer the next node") */
1793 /* Give preference to headless and unused nodes */
1794 tmp
= node_to_cpumask(n
);
1795 if (!cpus_empty(tmp
))
1796 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1798 /* Slight preference for less loaded node */
1799 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1800 val
+= node_load
[n
];
1802 if (val
< min_val
) {
1809 node_set(best_node
, *used_node_mask
);
1816 * Build zonelists ordered by node and zones within node.
1817 * This results in maximum locality--normal zone overflows into local
1818 * DMA zone, if any--but risks exhausting DMA zone.
1820 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
1824 struct zonelist
*zonelist
;
1826 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1827 zonelist
= pgdat
->node_zonelists
+ i
;
1828 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++)
1830 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1831 zonelist
->zones
[j
] = NULL
;
1836 * Build zonelists ordered by zone and nodes within zones.
1837 * This results in conserving DMA zone[s] until all Normal memory is
1838 * exhausted, but results in overflowing to remote node while memory
1839 * may still exist in local DMA zone.
1841 static int node_order
[MAX_NUMNODES
];
1843 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
1847 int zone_type
; /* needs to be signed */
1849 struct zonelist
*zonelist
;
1851 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1852 zonelist
= pgdat
->node_zonelists
+ i
;
1854 for (zone_type
= i
; zone_type
>= 0; zone_type
--) {
1855 for (j
= 0; j
< nr_nodes
; j
++) {
1856 node
= node_order
[j
];
1857 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
1858 if (populated_zone(z
)) {
1859 zonelist
->zones
[pos
++] = z
;
1860 check_highest_zone(zone_type
);
1864 zonelist
->zones
[pos
] = NULL
;
1868 static int default_zonelist_order(void)
1871 unsigned long low_kmem_size
,total_size
;
1875 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
1876 * If they are really small and used heavily, the system can fall
1877 * into OOM very easily.
1878 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
1880 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
1883 for_each_online_node(nid
) {
1884 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
1885 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
1886 if (populated_zone(z
)) {
1887 if (zone_type
< ZONE_NORMAL
)
1888 low_kmem_size
+= z
->present_pages
;
1889 total_size
+= z
->present_pages
;
1893 if (!low_kmem_size
|| /* there are no DMA area. */
1894 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
1895 return ZONELIST_ORDER_NODE
;
1897 * look into each node's config.
1898 * If there is a node whose DMA/DMA32 memory is very big area on
1899 * local memory, NODE_ORDER may be suitable.
1901 average_size
= total_size
/ (num_online_nodes() + 1);
1902 for_each_online_node(nid
) {
1905 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
1906 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
1907 if (populated_zone(z
)) {
1908 if (zone_type
< ZONE_NORMAL
)
1909 low_kmem_size
+= z
->present_pages
;
1910 total_size
+= z
->present_pages
;
1913 if (low_kmem_size
&&
1914 total_size
> average_size
&& /* ignore small node */
1915 low_kmem_size
> total_size
* 70/100)
1916 return ZONELIST_ORDER_NODE
;
1918 return ZONELIST_ORDER_ZONE
;
1921 static void set_zonelist_order(void)
1923 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
1924 current_zonelist_order
= default_zonelist_order();
1926 current_zonelist_order
= user_zonelist_order
;
1929 static void build_zonelists(pg_data_t
*pgdat
)
1933 nodemask_t used_mask
;
1934 int local_node
, prev_node
;
1935 struct zonelist
*zonelist
;
1936 int order
= current_zonelist_order
;
1938 /* initialize zonelists */
1939 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1940 zonelist
= pgdat
->node_zonelists
+ i
;
1941 zonelist
->zones
[0] = NULL
;
1944 /* NUMA-aware ordering of nodes */
1945 local_node
= pgdat
->node_id
;
1946 load
= num_online_nodes();
1947 prev_node
= local_node
;
1948 nodes_clear(used_mask
);
1950 memset(node_load
, 0, sizeof(node_load
));
1951 memset(node_order
, 0, sizeof(node_order
));
1954 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1955 int distance
= node_distance(local_node
, node
);
1958 * If another node is sufficiently far away then it is better
1959 * to reclaim pages in a zone before going off node.
1961 if (distance
> RECLAIM_DISTANCE
)
1962 zone_reclaim_mode
= 1;
1965 * We don't want to pressure a particular node.
1966 * So adding penalty to the first node in same
1967 * distance group to make it round-robin.
1969 if (distance
!= node_distance(local_node
, prev_node
))
1970 node_load
[node
] = load
;
1974 if (order
== ZONELIST_ORDER_NODE
)
1975 build_zonelists_in_node_order(pgdat
, node
);
1977 node_order
[j
++] = node
; /* remember order */
1980 if (order
== ZONELIST_ORDER_ZONE
) {
1981 /* calculate node order -- i.e., DMA last! */
1982 build_zonelists_in_zone_order(pgdat
, j
);
1986 /* Construct the zonelist performance cache - see further mmzone.h */
1987 static void build_zonelist_cache(pg_data_t
*pgdat
)
1991 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1992 struct zonelist
*zonelist
;
1993 struct zonelist_cache
*zlc
;
1996 zonelist
= pgdat
->node_zonelists
+ i
;
1997 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
1998 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1999 for (z
= zonelist
->zones
; *z
; z
++)
2000 zlc
->z_to_n
[z
- zonelist
->zones
] = zone_to_nid(*z
);
2005 #else /* CONFIG_NUMA */
2007 static void set_zonelist_order(void)
2009 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2012 static void build_zonelists(pg_data_t
*pgdat
)
2014 int node
, local_node
;
2017 local_node
= pgdat
->node_id
;
2018 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2019 struct zonelist
*zonelist
;
2021 zonelist
= pgdat
->node_zonelists
+ i
;
2023 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2025 * Now we build the zonelist so that it contains the zones
2026 * of all the other nodes.
2027 * We don't want to pressure a particular node, so when
2028 * building the zones for node N, we make sure that the
2029 * zones coming right after the local ones are those from
2030 * node N+1 (modulo N)
2032 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2033 if (!node_online(node
))
2035 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2037 for (node
= 0; node
< local_node
; node
++) {
2038 if (!node_online(node
))
2040 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2043 zonelist
->zones
[j
] = NULL
;
2047 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2048 static void build_zonelist_cache(pg_data_t
*pgdat
)
2052 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2053 pgdat
->node_zonelists
[i
].zlcache_ptr
= NULL
;
2056 #endif /* CONFIG_NUMA */
2058 /* return values int ....just for stop_machine_run() */
2059 static int __build_all_zonelists(void *dummy
)
2063 for_each_online_node(nid
) {
2064 build_zonelists(NODE_DATA(nid
));
2065 build_zonelist_cache(NODE_DATA(nid
));
2070 void build_all_zonelists(void)
2072 set_zonelist_order();
2074 if (system_state
== SYSTEM_BOOTING
) {
2075 __build_all_zonelists(NULL
);
2076 cpuset_init_current_mems_allowed();
2078 /* we have to stop all cpus to guaranntee there is no user
2080 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
2081 /* cpuset refresh routine should be here */
2083 vm_total_pages
= nr_free_pagecache_pages();
2084 printk("Built %i zonelists in %s order. Total pages: %ld\n",
2086 zonelist_order_name
[current_zonelist_order
],
2089 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2094 * Helper functions to size the waitqueue hash table.
2095 * Essentially these want to choose hash table sizes sufficiently
2096 * large so that collisions trying to wait on pages are rare.
2097 * But in fact, the number of active page waitqueues on typical
2098 * systems is ridiculously low, less than 200. So this is even
2099 * conservative, even though it seems large.
2101 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2102 * waitqueues, i.e. the size of the waitq table given the number of pages.
2104 #define PAGES_PER_WAITQUEUE 256
2106 #ifndef CONFIG_MEMORY_HOTPLUG
2107 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2109 unsigned long size
= 1;
2111 pages
/= PAGES_PER_WAITQUEUE
;
2113 while (size
< pages
)
2117 * Once we have dozens or even hundreds of threads sleeping
2118 * on IO we've got bigger problems than wait queue collision.
2119 * Limit the size of the wait table to a reasonable size.
2121 size
= min(size
, 4096UL);
2123 return max(size
, 4UL);
2127 * A zone's size might be changed by hot-add, so it is not possible to determine
2128 * a suitable size for its wait_table. So we use the maximum size now.
2130 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2132 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2133 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2134 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2136 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2137 * or more by the traditional way. (See above). It equals:
2139 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2140 * ia64(16K page size) : = ( 8G + 4M)byte.
2141 * powerpc (64K page size) : = (32G +16M)byte.
2143 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2150 * This is an integer logarithm so that shifts can be used later
2151 * to extract the more random high bits from the multiplicative
2152 * hash function before the remainder is taken.
2154 static inline unsigned long wait_table_bits(unsigned long size
)
2159 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2162 * Initially all pages are reserved - free ones are freed
2163 * up by free_all_bootmem() once the early boot process is
2164 * done. Non-atomic initialization, single-pass.
2166 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2167 unsigned long start_pfn
, enum memmap_context context
)
2170 unsigned long end_pfn
= start_pfn
+ size
;
2173 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2175 * There can be holes in boot-time mem_map[]s
2176 * handed to this function. They do not
2177 * exist on hotplugged memory.
2179 if (context
== MEMMAP_EARLY
) {
2180 if (!early_pfn_valid(pfn
))
2182 if (!early_pfn_in_nid(pfn
, nid
))
2185 page
= pfn_to_page(pfn
);
2186 set_page_links(page
, zone
, nid
, pfn
);
2187 init_page_count(page
);
2188 reset_page_mapcount(page
);
2189 SetPageReserved(page
);
2190 INIT_LIST_HEAD(&page
->lru
);
2191 #ifdef WANT_PAGE_VIRTUAL
2192 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2193 if (!is_highmem_idx(zone
))
2194 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2199 static void __meminit
zone_init_free_lists(struct pglist_data
*pgdat
,
2200 struct zone
*zone
, unsigned long size
)
2203 for (order
= 0; order
< MAX_ORDER
; order
++) {
2204 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
2205 zone
->free_area
[order
].nr_free
= 0;
2209 #ifndef __HAVE_ARCH_MEMMAP_INIT
2210 #define memmap_init(size, nid, zone, start_pfn) \
2211 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2214 static int __devinit
zone_batchsize(struct zone
*zone
)
2219 * The per-cpu-pages pools are set to around 1000th of the
2220 * size of the zone. But no more than 1/2 of a meg.
2222 * OK, so we don't know how big the cache is. So guess.
2224 batch
= zone
->present_pages
/ 1024;
2225 if (batch
* PAGE_SIZE
> 512 * 1024)
2226 batch
= (512 * 1024) / PAGE_SIZE
;
2227 batch
/= 4; /* We effectively *= 4 below */
2232 * Clamp the batch to a 2^n - 1 value. Having a power
2233 * of 2 value was found to be more likely to have
2234 * suboptimal cache aliasing properties in some cases.
2236 * For example if 2 tasks are alternately allocating
2237 * batches of pages, one task can end up with a lot
2238 * of pages of one half of the possible page colors
2239 * and the other with pages of the other colors.
2241 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2246 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2248 struct per_cpu_pages
*pcp
;
2250 memset(p
, 0, sizeof(*p
));
2252 pcp
= &p
->pcp
[0]; /* hot */
2254 pcp
->high
= 6 * batch
;
2255 pcp
->batch
= max(1UL, 1 * batch
);
2256 INIT_LIST_HEAD(&pcp
->list
);
2258 pcp
= &p
->pcp
[1]; /* cold*/
2260 pcp
->high
= 2 * batch
;
2261 pcp
->batch
= max(1UL, batch
/2);
2262 INIT_LIST_HEAD(&pcp
->list
);
2266 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2267 * to the value high for the pageset p.
2270 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2273 struct per_cpu_pages
*pcp
;
2275 pcp
= &p
->pcp
[0]; /* hot list */
2277 pcp
->batch
= max(1UL, high
/4);
2278 if ((high
/4) > (PAGE_SHIFT
* 8))
2279 pcp
->batch
= PAGE_SHIFT
* 8;
2285 * Boot pageset table. One per cpu which is going to be used for all
2286 * zones and all nodes. The parameters will be set in such a way
2287 * that an item put on a list will immediately be handed over to
2288 * the buddy list. This is safe since pageset manipulation is done
2289 * with interrupts disabled.
2291 * Some NUMA counter updates may also be caught by the boot pagesets.
2293 * The boot_pagesets must be kept even after bootup is complete for
2294 * unused processors and/or zones. They do play a role for bootstrapping
2295 * hotplugged processors.
2297 * zoneinfo_show() and maybe other functions do
2298 * not check if the processor is online before following the pageset pointer.
2299 * Other parts of the kernel may not check if the zone is available.
2301 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2304 * Dynamically allocate memory for the
2305 * per cpu pageset array in struct zone.
2307 static int __cpuinit
process_zones(int cpu
)
2309 struct zone
*zone
, *dzone
;
2311 for_each_zone(zone
) {
2313 if (!populated_zone(zone
))
2316 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2317 GFP_KERNEL
, cpu_to_node(cpu
));
2318 if (!zone_pcp(zone
, cpu
))
2321 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2323 if (percpu_pagelist_fraction
)
2324 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2325 (zone
->present_pages
/ percpu_pagelist_fraction
));
2330 for_each_zone(dzone
) {
2333 kfree(zone_pcp(dzone
, cpu
));
2334 zone_pcp(dzone
, cpu
) = NULL
;
2339 static inline void free_zone_pagesets(int cpu
)
2343 for_each_zone(zone
) {
2344 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2346 /* Free per_cpu_pageset if it is slab allocated */
2347 if (pset
!= &boot_pageset
[cpu
])
2349 zone_pcp(zone
, cpu
) = NULL
;
2353 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2354 unsigned long action
,
2357 int cpu
= (long)hcpu
;
2358 int ret
= NOTIFY_OK
;
2361 case CPU_UP_PREPARE
:
2362 case CPU_UP_PREPARE_FROZEN
:
2363 if (process_zones(cpu
))
2366 case CPU_UP_CANCELED
:
2367 case CPU_UP_CANCELED_FROZEN
:
2369 case CPU_DEAD_FROZEN
:
2370 free_zone_pagesets(cpu
);
2378 static struct notifier_block __cpuinitdata pageset_notifier
=
2379 { &pageset_cpuup_callback
, NULL
, 0 };
2381 void __init
setup_per_cpu_pageset(void)
2385 /* Initialize per_cpu_pageset for cpu 0.
2386 * A cpuup callback will do this for every cpu
2387 * as it comes online
2389 err
= process_zones(smp_processor_id());
2391 register_cpu_notifier(&pageset_notifier
);
2396 static noinline __init_refok
2397 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2400 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2404 * The per-page waitqueue mechanism uses hashed waitqueues
2407 zone
->wait_table_hash_nr_entries
=
2408 wait_table_hash_nr_entries(zone_size_pages
);
2409 zone
->wait_table_bits
=
2410 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2411 alloc_size
= zone
->wait_table_hash_nr_entries
2412 * sizeof(wait_queue_head_t
);
2414 if (system_state
== SYSTEM_BOOTING
) {
2415 zone
->wait_table
= (wait_queue_head_t
*)
2416 alloc_bootmem_node(pgdat
, alloc_size
);
2419 * This case means that a zone whose size was 0 gets new memory
2420 * via memory hot-add.
2421 * But it may be the case that a new node was hot-added. In
2422 * this case vmalloc() will not be able to use this new node's
2423 * memory - this wait_table must be initialized to use this new
2424 * node itself as well.
2425 * To use this new node's memory, further consideration will be
2428 zone
->wait_table
= (wait_queue_head_t
*)vmalloc(alloc_size
);
2430 if (!zone
->wait_table
)
2433 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2434 init_waitqueue_head(zone
->wait_table
+ i
);
2439 static __meminit
void zone_pcp_init(struct zone
*zone
)
2442 unsigned long batch
= zone_batchsize(zone
);
2444 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2446 /* Early boot. Slab allocator not functional yet */
2447 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2448 setup_pageset(&boot_pageset
[cpu
],0);
2450 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2453 if (zone
->present_pages
)
2454 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2455 zone
->name
, zone
->present_pages
, batch
);
2458 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2459 unsigned long zone_start_pfn
,
2461 enum memmap_context context
)
2463 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2465 ret
= zone_wait_table_init(zone
, size
);
2468 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2470 zone
->zone_start_pfn
= zone_start_pfn
;
2472 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
2474 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2479 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2481 * Basic iterator support. Return the first range of PFNs for a node
2482 * Note: nid == MAX_NUMNODES returns first region regardless of node
2484 static int __meminit
first_active_region_index_in_nid(int nid
)
2488 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2489 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2496 * Basic iterator support. Return the next active range of PFNs for a node
2497 * Note: nid == MAX_NUMNODES returns next region regardles of node
2499 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2501 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2502 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2508 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2510 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2511 * Architectures may implement their own version but if add_active_range()
2512 * was used and there are no special requirements, this is a convenient
2515 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2519 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2520 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2521 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2523 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2524 return early_node_map
[i
].nid
;
2529 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2531 /* Basic iterator support to walk early_node_map[] */
2532 #define for_each_active_range_index_in_nid(i, nid) \
2533 for (i = first_active_region_index_in_nid(nid); i != -1; \
2534 i = next_active_region_index_in_nid(i, nid))
2537 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2538 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2539 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2541 * If an architecture guarantees that all ranges registered with
2542 * add_active_ranges() contain no holes and may be freed, this
2543 * this function may be used instead of calling free_bootmem() manually.
2545 void __init
free_bootmem_with_active_regions(int nid
,
2546 unsigned long max_low_pfn
)
2550 for_each_active_range_index_in_nid(i
, nid
) {
2551 unsigned long size_pages
= 0;
2552 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2554 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2557 if (end_pfn
> max_low_pfn
)
2558 end_pfn
= max_low_pfn
;
2560 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2561 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2562 PFN_PHYS(early_node_map
[i
].start_pfn
),
2563 size_pages
<< PAGE_SHIFT
);
2568 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2569 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2571 * If an architecture guarantees that all ranges registered with
2572 * add_active_ranges() contain no holes and may be freed, this
2573 * function may be used instead of calling memory_present() manually.
2575 void __init
sparse_memory_present_with_active_regions(int nid
)
2579 for_each_active_range_index_in_nid(i
, nid
)
2580 memory_present(early_node_map
[i
].nid
,
2581 early_node_map
[i
].start_pfn
,
2582 early_node_map
[i
].end_pfn
);
2586 * push_node_boundaries - Push node boundaries to at least the requested boundary
2587 * @nid: The nid of the node to push the boundary for
2588 * @start_pfn: The start pfn of the node
2589 * @end_pfn: The end pfn of the node
2591 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2592 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2593 * be hotplugged even though no physical memory exists. This function allows
2594 * an arch to push out the node boundaries so mem_map is allocated that can
2597 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2598 void __init
push_node_boundaries(unsigned int nid
,
2599 unsigned long start_pfn
, unsigned long end_pfn
)
2601 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2602 nid
, start_pfn
, end_pfn
);
2604 /* Initialise the boundary for this node if necessary */
2605 if (node_boundary_end_pfn
[nid
] == 0)
2606 node_boundary_start_pfn
[nid
] = -1UL;
2608 /* Update the boundaries */
2609 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2610 node_boundary_start_pfn
[nid
] = start_pfn
;
2611 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2612 node_boundary_end_pfn
[nid
] = end_pfn
;
2615 /* If necessary, push the node boundary out for reserve hotadd */
2616 static void __meminit
account_node_boundary(unsigned int nid
,
2617 unsigned long *start_pfn
, unsigned long *end_pfn
)
2619 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2620 nid
, *start_pfn
, *end_pfn
);
2622 /* Return if boundary information has not been provided */
2623 if (node_boundary_end_pfn
[nid
] == 0)
2626 /* Check the boundaries and update if necessary */
2627 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
2628 *start_pfn
= node_boundary_start_pfn
[nid
];
2629 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
2630 *end_pfn
= node_boundary_end_pfn
[nid
];
2633 void __init
push_node_boundaries(unsigned int nid
,
2634 unsigned long start_pfn
, unsigned long end_pfn
) {}
2636 static void __meminit
account_node_boundary(unsigned int nid
,
2637 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
2642 * get_pfn_range_for_nid - Return the start and end page frames for a node
2643 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2644 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2645 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2647 * It returns the start and end page frame of a node based on information
2648 * provided by an arch calling add_active_range(). If called for a node
2649 * with no available memory, a warning is printed and the start and end
2652 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
2653 unsigned long *start_pfn
, unsigned long *end_pfn
)
2659 for_each_active_range_index_in_nid(i
, nid
) {
2660 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
2661 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
2664 if (*start_pfn
== -1UL) {
2665 printk(KERN_WARNING
"Node %u active with no memory\n", nid
);
2669 /* Push the node boundaries out if requested */
2670 account_node_boundary(nid
, start_pfn
, end_pfn
);
2674 * This finds a zone that can be used for ZONE_MOVABLE pages. The
2675 * assumption is made that zones within a node are ordered in monotonic
2676 * increasing memory addresses so that the "highest" populated zone is used
2678 void __init
find_usable_zone_for_movable(void)
2681 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
2682 if (zone_index
== ZONE_MOVABLE
)
2685 if (arch_zone_highest_possible_pfn
[zone_index
] >
2686 arch_zone_lowest_possible_pfn
[zone_index
])
2690 VM_BUG_ON(zone_index
== -1);
2691 movable_zone
= zone_index
;
2695 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
2696 * because it is sized independant of architecture. Unlike the other zones,
2697 * the starting point for ZONE_MOVABLE is not fixed. It may be different
2698 * in each node depending on the size of each node and how evenly kernelcore
2699 * is distributed. This helper function adjusts the zone ranges
2700 * provided by the architecture for a given node by using the end of the
2701 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
2702 * zones within a node are in order of monotonic increases memory addresses
2704 void __meminit
adjust_zone_range_for_zone_movable(int nid
,
2705 unsigned long zone_type
,
2706 unsigned long node_start_pfn
,
2707 unsigned long node_end_pfn
,
2708 unsigned long *zone_start_pfn
,
2709 unsigned long *zone_end_pfn
)
2711 /* Only adjust if ZONE_MOVABLE is on this node */
2712 if (zone_movable_pfn
[nid
]) {
2713 /* Size ZONE_MOVABLE */
2714 if (zone_type
== ZONE_MOVABLE
) {
2715 *zone_start_pfn
= zone_movable_pfn
[nid
];
2716 *zone_end_pfn
= min(node_end_pfn
,
2717 arch_zone_highest_possible_pfn
[movable_zone
]);
2719 /* Adjust for ZONE_MOVABLE starting within this range */
2720 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
2721 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
2722 *zone_end_pfn
= zone_movable_pfn
[nid
];
2724 /* Check if this whole range is within ZONE_MOVABLE */
2725 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
2726 *zone_start_pfn
= *zone_end_pfn
;
2731 * Return the number of pages a zone spans in a node, including holes
2732 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2734 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
2735 unsigned long zone_type
,
2736 unsigned long *ignored
)
2738 unsigned long node_start_pfn
, node_end_pfn
;
2739 unsigned long zone_start_pfn
, zone_end_pfn
;
2741 /* Get the start and end of the node and zone */
2742 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
2743 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
2744 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
2745 adjust_zone_range_for_zone_movable(nid
, zone_type
,
2746 node_start_pfn
, node_end_pfn
,
2747 &zone_start_pfn
, &zone_end_pfn
);
2749 /* Check that this node has pages within the zone's required range */
2750 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
2753 /* Move the zone boundaries inside the node if necessary */
2754 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
2755 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
2757 /* Return the spanned pages */
2758 return zone_end_pfn
- zone_start_pfn
;
2762 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
2763 * then all holes in the requested range will be accounted for.
2765 unsigned long __meminit
__absent_pages_in_range(int nid
,
2766 unsigned long range_start_pfn
,
2767 unsigned long range_end_pfn
)
2770 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
2771 unsigned long start_pfn
;
2773 /* Find the end_pfn of the first active range of pfns in the node */
2774 i
= first_active_region_index_in_nid(nid
);
2778 /* Account for ranges before physical memory on this node */
2779 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
2780 hole_pages
= early_node_map
[i
].start_pfn
- range_start_pfn
;
2782 prev_end_pfn
= early_node_map
[i
].start_pfn
;
2784 /* Find all holes for the zone within the node */
2785 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
2787 /* No need to continue if prev_end_pfn is outside the zone */
2788 if (prev_end_pfn
>= range_end_pfn
)
2791 /* Make sure the end of the zone is not within the hole */
2792 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
2793 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
2795 /* Update the hole size cound and move on */
2796 if (start_pfn
> range_start_pfn
) {
2797 BUG_ON(prev_end_pfn
> start_pfn
);
2798 hole_pages
+= start_pfn
- prev_end_pfn
;
2800 prev_end_pfn
= early_node_map
[i
].end_pfn
;
2803 /* Account for ranges past physical memory on this node */
2804 if (range_end_pfn
> prev_end_pfn
)
2805 hole_pages
+= range_end_pfn
-
2806 max(range_start_pfn
, prev_end_pfn
);
2812 * absent_pages_in_range - Return number of page frames in holes within a range
2813 * @start_pfn: The start PFN to start searching for holes
2814 * @end_pfn: The end PFN to stop searching for holes
2816 * It returns the number of pages frames in memory holes within a range.
2818 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
2819 unsigned long end_pfn
)
2821 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
2824 /* Return the number of page frames in holes in a zone on a node */
2825 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
2826 unsigned long zone_type
,
2827 unsigned long *ignored
)
2829 unsigned long node_start_pfn
, node_end_pfn
;
2830 unsigned long zone_start_pfn
, zone_end_pfn
;
2832 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
2833 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
2835 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
2838 adjust_zone_range_for_zone_movable(nid
, zone_type
,
2839 node_start_pfn
, node_end_pfn
,
2840 &zone_start_pfn
, &zone_end_pfn
);
2841 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
2845 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
2846 unsigned long zone_type
,
2847 unsigned long *zones_size
)
2849 return zones_size
[zone_type
];
2852 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
2853 unsigned long zone_type
,
2854 unsigned long *zholes_size
)
2859 return zholes_size
[zone_type
];
2864 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
2865 unsigned long *zones_size
, unsigned long *zholes_size
)
2867 unsigned long realtotalpages
, totalpages
= 0;
2870 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2871 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
2873 pgdat
->node_spanned_pages
= totalpages
;
2875 realtotalpages
= totalpages
;
2876 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2878 zone_absent_pages_in_node(pgdat
->node_id
, i
,
2880 pgdat
->node_present_pages
= realtotalpages
;
2881 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
2886 * Set up the zone data structures:
2887 * - mark all pages reserved
2888 * - mark all memory queues empty
2889 * - clear the memory bitmaps
2891 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
2892 unsigned long *zones_size
, unsigned long *zholes_size
)
2895 int nid
= pgdat
->node_id
;
2896 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
2899 pgdat_resize_init(pgdat
);
2900 pgdat
->nr_zones
= 0;
2901 init_waitqueue_head(&pgdat
->kswapd_wait
);
2902 pgdat
->kswapd_max_order
= 0;
2904 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2905 struct zone
*zone
= pgdat
->node_zones
+ j
;
2906 unsigned long size
, realsize
, memmap_pages
;
2908 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
2909 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
2913 * Adjust realsize so that it accounts for how much memory
2914 * is used by this zone for memmap. This affects the watermark
2915 * and per-cpu initialisations
2917 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
2918 if (realsize
>= memmap_pages
) {
2919 realsize
-= memmap_pages
;
2921 " %s zone: %lu pages used for memmap\n",
2922 zone_names
[j
], memmap_pages
);
2925 " %s zone: %lu pages exceeds realsize %lu\n",
2926 zone_names
[j
], memmap_pages
, realsize
);
2928 /* Account for reserved pages */
2929 if (j
== 0 && realsize
> dma_reserve
) {
2930 realsize
-= dma_reserve
;
2931 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
2932 zone_names
[0], dma_reserve
);
2935 if (!is_highmem_idx(j
))
2936 nr_kernel_pages
+= realsize
;
2937 nr_all_pages
+= realsize
;
2939 zone
->spanned_pages
= size
;
2940 zone
->present_pages
= realsize
;
2943 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
2945 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
2947 zone
->name
= zone_names
[j
];
2948 spin_lock_init(&zone
->lock
);
2949 spin_lock_init(&zone
->lru_lock
);
2950 zone_seqlock_init(zone
);
2951 zone
->zone_pgdat
= pgdat
;
2953 zone
->prev_priority
= DEF_PRIORITY
;
2955 zone_pcp_init(zone
);
2956 INIT_LIST_HEAD(&zone
->active_list
);
2957 INIT_LIST_HEAD(&zone
->inactive_list
);
2958 zone
->nr_scan_active
= 0;
2959 zone
->nr_scan_inactive
= 0;
2960 zap_zone_vm_stats(zone
);
2961 atomic_set(&zone
->reclaim_in_progress
, 0);
2965 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
2966 size
, MEMMAP_EARLY
);
2968 zone_start_pfn
+= size
;
2972 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
2974 /* Skip empty nodes */
2975 if (!pgdat
->node_spanned_pages
)
2978 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2979 /* ia64 gets its own node_mem_map, before this, without bootmem */
2980 if (!pgdat
->node_mem_map
) {
2981 unsigned long size
, start
, end
;
2985 * The zone's endpoints aren't required to be MAX_ORDER
2986 * aligned but the node_mem_map endpoints must be in order
2987 * for the buddy allocator to function correctly.
2989 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
2990 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
2991 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
2992 size
= (end
- start
) * sizeof(struct page
);
2993 map
= alloc_remap(pgdat
->node_id
, size
);
2995 map
= alloc_bootmem_node(pgdat
, size
);
2996 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
2998 #ifndef CONFIG_NEED_MULTIPLE_NODES
3000 * With no DISCONTIG, the global mem_map is just set as node 0's
3002 if (pgdat
== NODE_DATA(0)) {
3003 mem_map
= NODE_DATA(0)->node_mem_map
;
3004 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3005 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3006 mem_map
-= pgdat
->node_start_pfn
;
3007 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3010 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3013 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
3014 unsigned long *zones_size
, unsigned long node_start_pfn
,
3015 unsigned long *zholes_size
)
3017 pgdat
->node_id
= nid
;
3018 pgdat
->node_start_pfn
= node_start_pfn
;
3019 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3021 alloc_node_mem_map(pgdat
);
3023 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3026 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3028 #if MAX_NUMNODES > 1
3030 * Figure out the number of possible node ids.
3032 static void __init
setup_nr_node_ids(void)
3035 unsigned int highest
= 0;
3037 for_each_node_mask(node
, node_possible_map
)
3039 nr_node_ids
= highest
+ 1;
3042 static inline void setup_nr_node_ids(void)
3048 * add_active_range - Register a range of PFNs backed by physical memory
3049 * @nid: The node ID the range resides on
3050 * @start_pfn: The start PFN of the available physical memory
3051 * @end_pfn: The end PFN of the available physical memory
3053 * These ranges are stored in an early_node_map[] and later used by
3054 * free_area_init_nodes() to calculate zone sizes and holes. If the
3055 * range spans a memory hole, it is up to the architecture to ensure
3056 * the memory is not freed by the bootmem allocator. If possible
3057 * the range being registered will be merged with existing ranges.
3059 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3060 unsigned long end_pfn
)
3064 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
3065 "%d entries of %d used\n",
3066 nid
, start_pfn
, end_pfn
,
3067 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3069 /* Merge with existing active regions if possible */
3070 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3071 if (early_node_map
[i
].nid
!= nid
)
3074 /* Skip if an existing region covers this new one */
3075 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3076 end_pfn
<= early_node_map
[i
].end_pfn
)
3079 /* Merge forward if suitable */
3080 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3081 end_pfn
> early_node_map
[i
].end_pfn
) {
3082 early_node_map
[i
].end_pfn
= end_pfn
;
3086 /* Merge backward if suitable */
3087 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3088 end_pfn
>= early_node_map
[i
].start_pfn
) {
3089 early_node_map
[i
].start_pfn
= start_pfn
;
3094 /* Check that early_node_map is large enough */
3095 if (i
>= MAX_ACTIVE_REGIONS
) {
3096 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3097 MAX_ACTIVE_REGIONS
);
3101 early_node_map
[i
].nid
= nid
;
3102 early_node_map
[i
].start_pfn
= start_pfn
;
3103 early_node_map
[i
].end_pfn
= end_pfn
;
3104 nr_nodemap_entries
= i
+ 1;
3108 * shrink_active_range - Shrink an existing registered range of PFNs
3109 * @nid: The node id the range is on that should be shrunk
3110 * @old_end_pfn: The old end PFN of the range
3111 * @new_end_pfn: The new PFN of the range
3113 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3114 * The map is kept at the end physical page range that has already been
3115 * registered with add_active_range(). This function allows an arch to shrink
3116 * an existing registered range.
3118 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
3119 unsigned long new_end_pfn
)
3123 /* Find the old active region end and shrink */
3124 for_each_active_range_index_in_nid(i
, nid
)
3125 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
3126 early_node_map
[i
].end_pfn
= new_end_pfn
;
3132 * remove_all_active_ranges - Remove all currently registered regions
3134 * During discovery, it may be found that a table like SRAT is invalid
3135 * and an alternative discovery method must be used. This function removes
3136 * all currently registered regions.
3138 void __init
remove_all_active_ranges(void)
3140 memset(early_node_map
, 0, sizeof(early_node_map
));
3141 nr_nodemap_entries
= 0;
3142 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3143 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3144 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3145 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3148 /* Compare two active node_active_regions */
3149 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3151 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3152 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3154 /* Done this way to avoid overflows */
3155 if (arange
->start_pfn
> brange
->start_pfn
)
3157 if (arange
->start_pfn
< brange
->start_pfn
)
3163 /* sort the node_map by start_pfn */
3164 static void __init
sort_node_map(void)
3166 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3167 sizeof(struct node_active_region
),
3168 cmp_node_active_region
, NULL
);
3171 /* Find the lowest pfn for a node */
3172 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
3175 unsigned long min_pfn
= ULONG_MAX
;
3177 /* Assuming a sorted map, the first range found has the starting pfn */
3178 for_each_active_range_index_in_nid(i
, nid
)
3179 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3181 if (min_pfn
== ULONG_MAX
) {
3183 "Could not find start_pfn for node %lu\n", nid
);
3191 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3193 * It returns the minimum PFN based on information provided via
3194 * add_active_range().
3196 unsigned long __init
find_min_pfn_with_active_regions(void)
3198 return find_min_pfn_for_node(MAX_NUMNODES
);
3202 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3204 * It returns the maximum PFN based on information provided via
3205 * add_active_range().
3207 unsigned long __init
find_max_pfn_with_active_regions(void)
3210 unsigned long max_pfn
= 0;
3212 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3213 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
3218 unsigned long __init
early_calculate_totalpages(void)
3221 unsigned long totalpages
= 0;
3223 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3224 totalpages
+= early_node_map
[i
].end_pfn
-
3225 early_node_map
[i
].start_pfn
;
3231 * Find the PFN the Movable zone begins in each node. Kernel memory
3232 * is spread evenly between nodes as long as the nodes have enough
3233 * memory. When they don't, some nodes will have more kernelcore than
3236 void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3239 unsigned long usable_startpfn
;
3240 unsigned long kernelcore_node
, kernelcore_remaining
;
3241 int usable_nodes
= num_online_nodes();
3244 * If movablecore was specified, calculate what size of
3245 * kernelcore that corresponds so that memory usable for
3246 * any allocation type is evenly spread. If both kernelcore
3247 * and movablecore are specified, then the value of kernelcore
3248 * will be used for required_kernelcore if it's greater than
3249 * what movablecore would have allowed.
3251 if (required_movablecore
) {
3252 unsigned long totalpages
= early_calculate_totalpages();
3253 unsigned long corepages
;
3256 * Round-up so that ZONE_MOVABLE is at least as large as what
3257 * was requested by the user
3259 required_movablecore
=
3260 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3261 corepages
= totalpages
- required_movablecore
;
3263 required_kernelcore
= max(required_kernelcore
, corepages
);
3266 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3267 if (!required_kernelcore
)
3270 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3271 find_usable_zone_for_movable();
3272 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3275 /* Spread kernelcore memory as evenly as possible throughout nodes */
3276 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3277 for_each_online_node(nid
) {
3279 * Recalculate kernelcore_node if the division per node
3280 * now exceeds what is necessary to satisfy the requested
3281 * amount of memory for the kernel
3283 if (required_kernelcore
< kernelcore_node
)
3284 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3287 * As the map is walked, we track how much memory is usable
3288 * by the kernel using kernelcore_remaining. When it is
3289 * 0, the rest of the node is usable by ZONE_MOVABLE
3291 kernelcore_remaining
= kernelcore_node
;
3293 /* Go through each range of PFNs within this node */
3294 for_each_active_range_index_in_nid(i
, nid
) {
3295 unsigned long start_pfn
, end_pfn
;
3296 unsigned long size_pages
;
3298 start_pfn
= max(early_node_map
[i
].start_pfn
,
3299 zone_movable_pfn
[nid
]);
3300 end_pfn
= early_node_map
[i
].end_pfn
;
3301 if (start_pfn
>= end_pfn
)
3304 /* Account for what is only usable for kernelcore */
3305 if (start_pfn
< usable_startpfn
) {
3306 unsigned long kernel_pages
;
3307 kernel_pages
= min(end_pfn
, usable_startpfn
)
3310 kernelcore_remaining
-= min(kernel_pages
,
3311 kernelcore_remaining
);
3312 required_kernelcore
-= min(kernel_pages
,
3313 required_kernelcore
);
3315 /* Continue if range is now fully accounted */
3316 if (end_pfn
<= usable_startpfn
) {
3319 * Push zone_movable_pfn to the end so
3320 * that if we have to rebalance
3321 * kernelcore across nodes, we will
3322 * not double account here
3324 zone_movable_pfn
[nid
] = end_pfn
;
3327 start_pfn
= usable_startpfn
;
3331 * The usable PFN range for ZONE_MOVABLE is from
3332 * start_pfn->end_pfn. Calculate size_pages as the
3333 * number of pages used as kernelcore
3335 size_pages
= end_pfn
- start_pfn
;
3336 if (size_pages
> kernelcore_remaining
)
3337 size_pages
= kernelcore_remaining
;
3338 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3341 * Some kernelcore has been met, update counts and
3342 * break if the kernelcore for this node has been
3345 required_kernelcore
-= min(required_kernelcore
,
3347 kernelcore_remaining
-= size_pages
;
3348 if (!kernelcore_remaining
)
3354 * If there is still required_kernelcore, we do another pass with one
3355 * less node in the count. This will push zone_movable_pfn[nid] further
3356 * along on the nodes that still have memory until kernelcore is
3360 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3363 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3364 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3365 zone_movable_pfn
[nid
] =
3366 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3370 * free_area_init_nodes - Initialise all pg_data_t and zone data
3371 * @max_zone_pfn: an array of max PFNs for each zone
3373 * This will call free_area_init_node() for each active node in the system.
3374 * Using the page ranges provided by add_active_range(), the size of each
3375 * zone in each node and their holes is calculated. If the maximum PFN
3376 * between two adjacent zones match, it is assumed that the zone is empty.
3377 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3378 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3379 * starts where the previous one ended. For example, ZONE_DMA32 starts
3380 * at arch_max_dma_pfn.
3382 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3387 /* Sort early_node_map as initialisation assumes it is sorted */
3390 /* Record where the zone boundaries are */
3391 memset(arch_zone_lowest_possible_pfn
, 0,
3392 sizeof(arch_zone_lowest_possible_pfn
));
3393 memset(arch_zone_highest_possible_pfn
, 0,
3394 sizeof(arch_zone_highest_possible_pfn
));
3395 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3396 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3397 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3398 if (i
== ZONE_MOVABLE
)
3400 arch_zone_lowest_possible_pfn
[i
] =
3401 arch_zone_highest_possible_pfn
[i
-1];
3402 arch_zone_highest_possible_pfn
[i
] =
3403 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3405 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3406 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3408 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3409 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3410 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3412 /* Print out the zone ranges */
3413 printk("Zone PFN ranges:\n");
3414 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3415 if (i
== ZONE_MOVABLE
)
3417 printk(" %-8s %8lu -> %8lu\n",
3419 arch_zone_lowest_possible_pfn
[i
],
3420 arch_zone_highest_possible_pfn
[i
]);
3423 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3424 printk("Movable zone start PFN for each node\n");
3425 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
3426 if (zone_movable_pfn
[i
])
3427 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
3430 /* Print out the early_node_map[] */
3431 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
3432 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3433 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
3434 early_node_map
[i
].start_pfn
,
3435 early_node_map
[i
].end_pfn
);
3437 /* Initialise every node */
3438 setup_nr_node_ids();
3439 for_each_online_node(nid
) {
3440 pg_data_t
*pgdat
= NODE_DATA(nid
);
3441 free_area_init_node(nid
, pgdat
, NULL
,
3442 find_min_pfn_for_node(nid
), NULL
);
3446 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
3448 unsigned long long coremem
;
3452 coremem
= memparse(p
, &p
);
3453 *core
= coremem
>> PAGE_SHIFT
;
3455 /* Paranoid check that UL is enough for the coremem value */
3456 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
3462 * kernelcore=size sets the amount of memory for use for allocations that
3463 * cannot be reclaimed or migrated.
3465 static int __init
cmdline_parse_kernelcore(char *p
)
3467 return cmdline_parse_core(p
, &required_kernelcore
);
3471 * movablecore=size sets the amount of memory for use for allocations that
3472 * can be reclaimed or migrated.
3474 static int __init
cmdline_parse_movablecore(char *p
)
3476 return cmdline_parse_core(p
, &required_movablecore
);
3479 early_param("kernelcore", cmdline_parse_kernelcore
);
3480 early_param("movablecore", cmdline_parse_movablecore
);
3482 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3485 * set_dma_reserve - set the specified number of pages reserved in the first zone
3486 * @new_dma_reserve: The number of pages to mark reserved
3488 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3489 * In the DMA zone, a significant percentage may be consumed by kernel image
3490 * and other unfreeable allocations which can skew the watermarks badly. This
3491 * function may optionally be used to account for unfreeable pages in the
3492 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3493 * smaller per-cpu batchsize.
3495 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
3497 dma_reserve
= new_dma_reserve
;
3500 #ifndef CONFIG_NEED_MULTIPLE_NODES
3501 static bootmem_data_t contig_bootmem_data
;
3502 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
3504 EXPORT_SYMBOL(contig_page_data
);
3507 void __init
free_area_init(unsigned long *zones_size
)
3509 free_area_init_node(0, NODE_DATA(0), zones_size
,
3510 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
3513 static int page_alloc_cpu_notify(struct notifier_block
*self
,
3514 unsigned long action
, void *hcpu
)
3516 int cpu
= (unsigned long)hcpu
;
3518 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3519 local_irq_disable();
3521 vm_events_fold_cpu(cpu
);
3523 refresh_cpu_vm_stats(cpu
);
3528 void __init
page_alloc_init(void)
3530 hotcpu_notifier(page_alloc_cpu_notify
, 0);
3534 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3535 * or min_free_kbytes changes.
3537 static void calculate_totalreserve_pages(void)
3539 struct pglist_data
*pgdat
;
3540 unsigned long reserve_pages
= 0;
3541 enum zone_type i
, j
;
3543 for_each_online_pgdat(pgdat
) {
3544 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3545 struct zone
*zone
= pgdat
->node_zones
+ i
;
3546 unsigned long max
= 0;
3548 /* Find valid and maximum lowmem_reserve in the zone */
3549 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
3550 if (zone
->lowmem_reserve
[j
] > max
)
3551 max
= zone
->lowmem_reserve
[j
];
3554 /* we treat pages_high as reserved pages. */
3555 max
+= zone
->pages_high
;
3557 if (max
> zone
->present_pages
)
3558 max
= zone
->present_pages
;
3559 reserve_pages
+= max
;
3562 totalreserve_pages
= reserve_pages
;
3566 * setup_per_zone_lowmem_reserve - called whenever
3567 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3568 * has a correct pages reserved value, so an adequate number of
3569 * pages are left in the zone after a successful __alloc_pages().
3571 static void setup_per_zone_lowmem_reserve(void)
3573 struct pglist_data
*pgdat
;
3574 enum zone_type j
, idx
;
3576 for_each_online_pgdat(pgdat
) {
3577 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3578 struct zone
*zone
= pgdat
->node_zones
+ j
;
3579 unsigned long present_pages
= zone
->present_pages
;
3581 zone
->lowmem_reserve
[j
] = 0;
3585 struct zone
*lower_zone
;
3589 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
3590 sysctl_lowmem_reserve_ratio
[idx
] = 1;
3592 lower_zone
= pgdat
->node_zones
+ idx
;
3593 lower_zone
->lowmem_reserve
[j
] = present_pages
/
3594 sysctl_lowmem_reserve_ratio
[idx
];
3595 present_pages
+= lower_zone
->present_pages
;
3600 /* update totalreserve_pages */
3601 calculate_totalreserve_pages();
3605 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3607 * Ensures that the pages_{min,low,high} values for each zone are set correctly
3608 * with respect to min_free_kbytes.
3610 void setup_per_zone_pages_min(void)
3612 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
3613 unsigned long lowmem_pages
= 0;
3615 unsigned long flags
;
3617 /* Calculate total number of !ZONE_HIGHMEM pages */
3618 for_each_zone(zone
) {
3619 if (!is_highmem(zone
))
3620 lowmem_pages
+= zone
->present_pages
;
3623 for_each_zone(zone
) {
3626 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3627 tmp
= (u64
)pages_min
* zone
->present_pages
;
3628 do_div(tmp
, lowmem_pages
);
3629 if (is_highmem(zone
)) {
3631 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3632 * need highmem pages, so cap pages_min to a small
3635 * The (pages_high-pages_low) and (pages_low-pages_min)
3636 * deltas controls asynch page reclaim, and so should
3637 * not be capped for highmem.
3641 min_pages
= zone
->present_pages
/ 1024;
3642 if (min_pages
< SWAP_CLUSTER_MAX
)
3643 min_pages
= SWAP_CLUSTER_MAX
;
3644 if (min_pages
> 128)
3646 zone
->pages_min
= min_pages
;
3649 * If it's a lowmem zone, reserve a number of pages
3650 * proportionate to the zone's size.
3652 zone
->pages_min
= tmp
;
3655 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
3656 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
3657 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3660 /* update totalreserve_pages */
3661 calculate_totalreserve_pages();
3665 * Initialise min_free_kbytes.
3667 * For small machines we want it small (128k min). For large machines
3668 * we want it large (64MB max). But it is not linear, because network
3669 * bandwidth does not increase linearly with machine size. We use
3671 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
3672 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
3688 static int __init
init_per_zone_pages_min(void)
3690 unsigned long lowmem_kbytes
;
3692 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
3694 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
3695 if (min_free_kbytes
< 128)
3696 min_free_kbytes
= 128;
3697 if (min_free_kbytes
> 65536)
3698 min_free_kbytes
= 65536;
3699 setup_per_zone_pages_min();
3700 setup_per_zone_lowmem_reserve();
3703 module_init(init_per_zone_pages_min
)
3706 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
3707 * that we can call two helper functions whenever min_free_kbytes
3710 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
3711 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3713 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
3715 setup_per_zone_pages_min();
3720 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
3721 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3726 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3731 zone
->min_unmapped_pages
= (zone
->present_pages
*
3732 sysctl_min_unmapped_ratio
) / 100;
3736 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
3737 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3742 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3747 zone
->min_slab_pages
= (zone
->present_pages
*
3748 sysctl_min_slab_ratio
) / 100;
3754 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
3755 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
3756 * whenever sysctl_lowmem_reserve_ratio changes.
3758 * The reserve ratio obviously has absolutely no relation with the
3759 * pages_min watermarks. The lowmem reserve ratio can only make sense
3760 * if in function of the boot time zone sizes.
3762 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
3763 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3765 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3766 setup_per_zone_lowmem_reserve();
3771 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3772 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3773 * can have before it gets flushed back to buddy allocator.
3776 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
3777 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3783 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3784 if (!write
|| (ret
== -EINVAL
))
3786 for_each_zone(zone
) {
3787 for_each_online_cpu(cpu
) {
3789 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
3790 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
3796 int hashdist
= HASHDIST_DEFAULT
;
3799 static int __init
set_hashdist(char *str
)
3803 hashdist
= simple_strtoul(str
, &str
, 0);
3806 __setup("hashdist=", set_hashdist
);
3810 * allocate a large system hash table from bootmem
3811 * - it is assumed that the hash table must contain an exact power-of-2
3812 * quantity of entries
3813 * - limit is the number of hash buckets, not the total allocation size
3815 void *__init
alloc_large_system_hash(const char *tablename
,
3816 unsigned long bucketsize
,
3817 unsigned long numentries
,
3820 unsigned int *_hash_shift
,
3821 unsigned int *_hash_mask
,
3822 unsigned long limit
)
3824 unsigned long long max
= limit
;
3825 unsigned long log2qty
, size
;
3828 /* allow the kernel cmdline to have a say */
3830 /* round applicable memory size up to nearest megabyte */
3831 numentries
= nr_kernel_pages
;
3832 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
3833 numentries
>>= 20 - PAGE_SHIFT
;
3834 numentries
<<= 20 - PAGE_SHIFT
;
3836 /* limit to 1 bucket per 2^scale bytes of low memory */
3837 if (scale
> PAGE_SHIFT
)
3838 numentries
>>= (scale
- PAGE_SHIFT
);
3840 numentries
<<= (PAGE_SHIFT
- scale
);
3842 /* Make sure we've got at least a 0-order allocation.. */
3843 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
3844 numentries
= PAGE_SIZE
/ bucketsize
;
3846 numentries
= roundup_pow_of_two(numentries
);
3848 /* limit allocation size to 1/16 total memory by default */
3850 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
3851 do_div(max
, bucketsize
);
3854 if (numentries
> max
)
3857 log2qty
= ilog2(numentries
);
3860 size
= bucketsize
<< log2qty
;
3861 if (flags
& HASH_EARLY
)
3862 table
= alloc_bootmem(size
);
3864 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
3866 unsigned long order
;
3867 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
3869 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
3871 * If bucketsize is not a power-of-two, we may free
3872 * some pages at the end of hash table.
3875 unsigned long alloc_end
= (unsigned long)table
+
3876 (PAGE_SIZE
<< order
);
3877 unsigned long used
= (unsigned long)table
+
3879 split_page(virt_to_page(table
), order
);
3880 while (used
< alloc_end
) {
3886 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
3889 panic("Failed to allocate %s hash table\n", tablename
);
3891 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
3894 ilog2(size
) - PAGE_SHIFT
,
3898 *_hash_shift
= log2qty
;
3900 *_hash_mask
= (1 << log2qty
) - 1;
3905 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
3906 struct page
*pfn_to_page(unsigned long pfn
)
3908 return __pfn_to_page(pfn
);
3910 unsigned long page_to_pfn(struct page
*page
)
3912 return __page_to_pfn(page
);
3914 EXPORT_SYMBOL(pfn_to_page
);
3915 EXPORT_SYMBOL(page_to_pfn
);
3916 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */