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/config.h>
18 #include <linux/stddef.h>
20 #include <linux/swap.h>
21 #include <linux/interrupt.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/notifier.h>
32 #include <linux/topology.h>
33 #include <linux/sysctl.h>
34 #include <linux/cpu.h>
35 #include <linux/cpuset.h>
36 #include <linux/memory_hotplug.h>
37 #include <linux/nodemask.h>
38 #include <linux/vmalloc.h>
39 #include <linux/mempolicy.h>
41 #include <asm/tlbflush.h>
45 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
48 nodemask_t node_online_map __read_mostly
= { { [0] = 1UL } };
49 EXPORT_SYMBOL(node_online_map
);
50 nodemask_t node_possible_map __read_mostly
= NODE_MASK_ALL
;
51 EXPORT_SYMBOL(node_possible_map
);
52 struct pglist_data
*pgdat_list __read_mostly
;
53 unsigned long totalram_pages __read_mostly
;
54 unsigned long totalhigh_pages __read_mostly
;
56 int percpu_pagelist_fraction
;
58 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
);
59 static void __free_pages_ok(struct page
*page
, unsigned int order
);
62 * results with 256, 32 in the lowmem_reserve sysctl:
63 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
64 * 1G machine -> (16M dma, 784M normal, 224M high)
65 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
66 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
67 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
69 * TBD: should special case ZONE_DMA32 machines here - in those we normally
70 * don't need any ZONE_NORMAL reservation
72 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = { 256, 256, 32 };
74 EXPORT_SYMBOL(totalram_pages
);
77 * Used by page_zone() to look up the address of the struct zone whose
78 * id is encoded in the upper bits of page->flags
80 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
81 EXPORT_SYMBOL(zone_table
);
83 static char *zone_names
[MAX_NR_ZONES
] = { "DMA", "DMA32", "Normal", "HighMem" };
84 int min_free_kbytes
= 1024;
86 unsigned long __initdata nr_kernel_pages
;
87 unsigned long __initdata nr_all_pages
;
89 #ifdef CONFIG_DEBUG_VM
90 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
94 unsigned long pfn
= page_to_pfn(page
);
97 seq
= zone_span_seqbegin(zone
);
98 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
100 else if (pfn
< zone
->zone_start_pfn
)
102 } while (zone_span_seqretry(zone
, seq
));
107 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
109 #ifdef CONFIG_HOLES_IN_ZONE
110 if (!pfn_valid(page_to_pfn(page
)))
113 if (zone
!= page_zone(page
))
119 * Temporary debugging check for pages not lying within a given zone.
121 static int bad_range(struct zone
*zone
, struct page
*page
)
123 if (page_outside_zone_boundaries(zone
, page
))
125 if (!page_is_consistent(zone
, page
))
132 static inline int bad_range(struct zone
*zone
, struct page
*page
)
138 static void bad_page(struct page
*page
)
140 printk(KERN_EMERG
"Bad page state in process '%s'\n"
141 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
142 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
143 KERN_EMERG
"Backtrace:\n",
144 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
145 (unsigned long)page
->flags
, page
->mapping
,
146 page_mapcount(page
), page_count(page
));
148 page
->flags
&= ~(1 << PG_lru
|
157 set_page_count(page
, 0);
158 reset_page_mapcount(page
);
159 page
->mapping
= NULL
;
160 add_taint(TAINT_BAD_PAGE
);
164 * Higher-order pages are called "compound pages". They are structured thusly:
166 * The first PAGE_SIZE page is called the "head page".
168 * The remaining PAGE_SIZE pages are called "tail pages".
170 * All pages have PG_compound set. All pages have their ->private pointing at
171 * the head page (even the head page has this).
173 * The first tail page's ->lru.next holds the address of the compound page's
174 * put_page() function. Its ->lru.prev holds the order of allocation.
175 * This usage means that zero-order pages may not be compound.
178 static void free_compound_page(struct page
*page
)
180 __free_pages_ok(page
, (unsigned long)page
[1].lru
.prev
);
183 static void prep_compound_page(struct page
*page
, unsigned long order
)
186 int nr_pages
= 1 << order
;
188 page
[1].lru
.next
= (void *)free_compound_page
; /* set dtor */
189 page
[1].lru
.prev
= (void *)order
;
190 for (i
= 0; i
< nr_pages
; i
++) {
191 struct page
*p
= page
+ i
;
194 set_page_private(p
, (unsigned long)page
);
198 static void destroy_compound_page(struct page
*page
, unsigned long order
)
201 int nr_pages
= 1 << order
;
203 if (unlikely((unsigned long)page
[1].lru
.prev
!= order
))
206 for (i
= 0; i
< nr_pages
; i
++) {
207 struct page
*p
= page
+ i
;
209 if (unlikely(!PageCompound(p
) |
210 (page_private(p
) != (unsigned long)page
)))
212 ClearPageCompound(p
);
217 * function for dealing with page's order in buddy system.
218 * zone->lock is already acquired when we use these.
219 * So, we don't need atomic page->flags operations here.
221 static inline unsigned long page_order(struct page
*page
) {
222 return page_private(page
);
225 static inline void set_page_order(struct page
*page
, int order
) {
226 set_page_private(page
, order
);
227 __SetPagePrivate(page
);
230 static inline void rmv_page_order(struct page
*page
)
232 __ClearPagePrivate(page
);
233 set_page_private(page
, 0);
237 * Locate the struct page for both the matching buddy in our
238 * pair (buddy1) and the combined O(n+1) page they form (page).
240 * 1) Any buddy B1 will have an order O twin B2 which satisfies
241 * the following equation:
243 * For example, if the starting buddy (buddy2) is #8 its order
245 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
247 * 2) Any buddy B will have an order O+1 parent P which
248 * satisfies the following equation:
251 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
253 static inline struct page
*
254 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
256 unsigned long buddy_idx
= page_idx
^ (1 << order
);
258 return page
+ (buddy_idx
- page_idx
);
261 static inline unsigned long
262 __find_combined_index(unsigned long page_idx
, unsigned int order
)
264 return (page_idx
& ~(1 << order
));
268 * This function checks whether a page is free && is the buddy
269 * we can do coalesce a page and its buddy if
270 * (a) the buddy is not in a hole &&
271 * (b) the buddy is free &&
272 * (c) the buddy is on the buddy system &&
273 * (d) a page and its buddy have the same order.
274 * for recording page's order, we use page_private(page) and PG_private.
277 static inline int page_is_buddy(struct page
*page
, int order
)
279 #ifdef CONFIG_HOLES_IN_ZONE
280 if (!pfn_valid(page_to_pfn(page
)))
284 if (PagePrivate(page
) &&
285 (page_order(page
) == order
) &&
286 page_count(page
) == 0)
292 * Freeing function for a buddy system allocator.
294 * The concept of a buddy system is to maintain direct-mapped table
295 * (containing bit values) for memory blocks of various "orders".
296 * The bottom level table contains the map for the smallest allocatable
297 * units of memory (here, pages), and each level above it describes
298 * pairs of units from the levels below, hence, "buddies".
299 * At a high level, all that happens here is marking the table entry
300 * at the bottom level available, and propagating the changes upward
301 * as necessary, plus some accounting needed to play nicely with other
302 * parts of the VM system.
303 * At each level, we keep a list of pages, which are heads of continuous
304 * free pages of length of (1 << order) and marked with PG_Private.Page's
305 * order is recorded in page_private(page) field.
306 * So when we are allocating or freeing one, we can derive the state of the
307 * other. That is, if we allocate a small block, and both were
308 * free, the remainder of the region must be split into blocks.
309 * If a block is freed, and its buddy is also free, then this
310 * triggers coalescing into a block of larger size.
315 static inline void __free_one_page(struct page
*page
,
316 struct zone
*zone
, unsigned int order
)
318 unsigned long page_idx
;
319 int order_size
= 1 << order
;
321 if (unlikely(PageCompound(page
)))
322 destroy_compound_page(page
, order
);
324 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
326 BUG_ON(page_idx
& (order_size
- 1));
327 BUG_ON(bad_range(zone
, page
));
329 zone
->free_pages
+= order_size
;
330 while (order
< MAX_ORDER
-1) {
331 unsigned long combined_idx
;
332 struct free_area
*area
;
335 buddy
= __page_find_buddy(page
, page_idx
, order
);
336 if (!page_is_buddy(buddy
, order
))
337 break; /* Move the buddy up one level. */
339 list_del(&buddy
->lru
);
340 area
= zone
->free_area
+ order
;
342 rmv_page_order(buddy
);
343 combined_idx
= __find_combined_index(page_idx
, order
);
344 page
= page
+ (combined_idx
- page_idx
);
345 page_idx
= combined_idx
;
348 set_page_order(page
, order
);
349 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
350 zone
->free_area
[order
].nr_free
++;
353 static inline int free_pages_check(struct page
*page
)
355 if (unlikely(page_mapcount(page
) |
356 (page
->mapping
!= NULL
) |
357 (page_count(page
) != 0) |
367 1 << PG_reserved
))))
370 __ClearPageDirty(page
);
372 * For now, we report if PG_reserved was found set, but do not
373 * clear it, and do not free the page. But we shall soon need
374 * to do more, for when the ZERO_PAGE count wraps negative.
376 return PageReserved(page
);
380 * Frees a list of pages.
381 * Assumes all pages on list are in same zone, and of same order.
382 * count is the number of pages to free.
384 * If the zone was previously in an "all pages pinned" state then look to
385 * see if this freeing clears that state.
387 * And clear the zone's pages_scanned counter, to hold off the "all pages are
388 * pinned" detection logic.
390 static void free_pages_bulk(struct zone
*zone
, int count
,
391 struct list_head
*list
, int order
)
393 spin_lock(&zone
->lock
);
394 zone
->all_unreclaimable
= 0;
395 zone
->pages_scanned
= 0;
399 BUG_ON(list_empty(list
));
400 page
= list_entry(list
->prev
, struct page
, lru
);
401 /* have to delete it as __free_one_page list manipulates */
402 list_del(&page
->lru
);
403 __free_one_page(page
, zone
, order
);
405 spin_unlock(&zone
->lock
);
408 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
411 list_add(&page
->lru
, &list
);
412 free_pages_bulk(zone
, 1, &list
, order
);
415 static void __free_pages_ok(struct page
*page
, unsigned int order
)
421 arch_free_page(page
, order
);
422 if (!PageHighMem(page
))
423 mutex_debug_check_no_locks_freed(page_address(page
),
427 for (i
= 1 ; i
< (1 << order
) ; ++i
)
428 __put_page(page
+ i
);
431 for (i
= 0 ; i
< (1 << order
) ; ++i
)
432 reserved
+= free_pages_check(page
+ i
);
436 kernel_map_pages(page
, 1 << order
, 0);
437 local_irq_save(flags
);
438 __mod_page_state(pgfree
, 1 << order
);
439 free_one_page(page_zone(page
), page
, order
);
440 local_irq_restore(flags
);
444 * permit the bootmem allocator to evade page validation on high-order frees
446 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
449 __ClearPageReserved(page
);
450 set_page_count(page
, 0);
452 free_hot_cold_page(page
, 0);
457 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
458 struct page
*p
= &page
[loop
];
460 if (loop
+ 16 < BITS_PER_LONG
)
462 __ClearPageReserved(p
);
463 set_page_count(p
, 0);
466 arch_free_page(page
, order
);
468 mod_page_state(pgfree
, 1 << order
);
470 list_add(&page
->lru
, &list
);
471 kernel_map_pages(page
, 1 << order
, 0);
472 free_pages_bulk(page_zone(page
), 1, &list
, order
);
478 * The order of subdivision here is critical for the IO subsystem.
479 * Please do not alter this order without good reasons and regression
480 * testing. Specifically, as large blocks of memory are subdivided,
481 * the order in which smaller blocks are delivered depends on the order
482 * they're subdivided in this function. This is the primary factor
483 * influencing the order in which pages are delivered to the IO
484 * subsystem according to empirical testing, and this is also justified
485 * by considering the behavior of a buddy system containing a single
486 * large block of memory acted on by a series of small allocations.
487 * This behavior is a critical factor in sglist merging's success.
491 static inline void expand(struct zone
*zone
, struct page
*page
,
492 int low
, int high
, struct free_area
*area
)
494 unsigned long size
= 1 << high
;
500 BUG_ON(bad_range(zone
, &page
[size
]));
501 list_add(&page
[size
].lru
, &area
->free_list
);
503 set_page_order(&page
[size
], high
);
508 * This page is about to be returned from the page allocator
510 static int prep_new_page(struct page
*page
, int order
)
512 if (unlikely(page_mapcount(page
) |
513 (page
->mapping
!= NULL
) |
514 (page_count(page
) != 0) |
525 1 << PG_reserved
))))
529 * For now, we report if PG_reserved was found set, but do not
530 * clear it, and do not allocate the page: as a safety net.
532 if (PageReserved(page
))
535 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
536 1 << PG_referenced
| 1 << PG_arch_1
|
537 1 << PG_checked
| 1 << PG_mappedtodisk
);
538 set_page_private(page
, 0);
539 set_page_refs(page
, order
);
540 kernel_map_pages(page
, 1 << order
, 1);
545 * Do the hard work of removing an element from the buddy allocator.
546 * Call me with the zone->lock already held.
548 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
550 struct free_area
* area
;
551 unsigned int current_order
;
554 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
555 area
= zone
->free_area
+ current_order
;
556 if (list_empty(&area
->free_list
))
559 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
560 list_del(&page
->lru
);
561 rmv_page_order(page
);
563 zone
->free_pages
-= 1UL << order
;
564 expand(zone
, page
, order
, current_order
, area
);
572 * Obtain a specified number of elements from the buddy allocator, all under
573 * a single hold of the lock, for efficiency. Add them to the supplied list.
574 * Returns the number of new pages which were placed at *list.
576 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
577 unsigned long count
, struct list_head
*list
)
581 spin_lock(&zone
->lock
);
582 for (i
= 0; i
< count
; ++i
) {
583 struct page
*page
= __rmqueue(zone
, order
);
584 if (unlikely(page
== NULL
))
586 list_add_tail(&page
->lru
, list
);
588 spin_unlock(&zone
->lock
);
594 * Called from the slab reaper to drain pagesets on a particular node that
595 * belong to the currently executing processor.
597 void drain_node_pages(int nodeid
)
602 local_irq_save(flags
);
603 for (z
= 0; z
< MAX_NR_ZONES
; z
++) {
604 struct zone
*zone
= NODE_DATA(nodeid
)->node_zones
+ z
;
605 struct per_cpu_pageset
*pset
;
607 pset
= zone_pcp(zone
, smp_processor_id());
608 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
609 struct per_cpu_pages
*pcp
;
612 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
616 local_irq_restore(flags
);
620 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
621 static void __drain_pages(unsigned int cpu
)
627 for_each_zone(zone
) {
628 struct per_cpu_pageset
*pset
;
630 pset
= zone_pcp(zone
, cpu
);
631 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
632 struct per_cpu_pages
*pcp
;
635 local_irq_save(flags
);
636 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
638 local_irq_restore(flags
);
642 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
646 void mark_free_pages(struct zone
*zone
)
648 unsigned long zone_pfn
, flags
;
650 struct list_head
*curr
;
652 if (!zone
->spanned_pages
)
655 spin_lock_irqsave(&zone
->lock
, flags
);
656 for (zone_pfn
= 0; zone_pfn
< zone
->spanned_pages
; ++zone_pfn
)
657 ClearPageNosaveFree(pfn_to_page(zone_pfn
+ zone
->zone_start_pfn
));
659 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
660 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
661 unsigned long start_pfn
, i
;
663 start_pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
665 for (i
=0; i
< (1<<order
); i
++)
666 SetPageNosaveFree(pfn_to_page(start_pfn
+i
));
668 spin_unlock_irqrestore(&zone
->lock
, flags
);
672 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
674 void drain_local_pages(void)
678 local_irq_save(flags
);
679 __drain_pages(smp_processor_id());
680 local_irq_restore(flags
);
682 #endif /* CONFIG_PM */
684 static void zone_statistics(struct zonelist
*zonelist
, struct zone
*z
, int cpu
)
687 pg_data_t
*pg
= z
->zone_pgdat
;
688 pg_data_t
*orig
= zonelist
->zones
[0]->zone_pgdat
;
689 struct per_cpu_pageset
*p
;
691 p
= zone_pcp(z
, cpu
);
696 zone_pcp(zonelist
->zones
[0], cpu
)->numa_foreign
++;
698 if (pg
== NODE_DATA(numa_node_id()))
706 * Free a 0-order page
708 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
710 struct zone
*zone
= page_zone(page
);
711 struct per_cpu_pages
*pcp
;
714 arch_free_page(page
, 0);
717 page
->mapping
= NULL
;
718 if (free_pages_check(page
))
721 kernel_map_pages(page
, 1, 0);
723 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
724 local_irq_save(flags
);
725 __inc_page_state(pgfree
);
726 list_add(&page
->lru
, &pcp
->list
);
728 if (pcp
->count
>= pcp
->high
) {
729 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
730 pcp
->count
-= pcp
->batch
;
732 local_irq_restore(flags
);
736 void fastcall
free_hot_page(struct page
*page
)
738 free_hot_cold_page(page
, 0);
741 void fastcall
free_cold_page(struct page
*page
)
743 free_hot_cold_page(page
, 1);
746 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
750 BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
751 for(i
= 0; i
< (1 << order
); i
++)
752 clear_highpage(page
+ i
);
756 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
757 * we cheat by calling it from here, in the order > 0 path. Saves a branch
760 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
761 struct zone
*zone
, int order
, gfp_t gfp_flags
)
765 int cold
= !!(gfp_flags
& __GFP_COLD
);
770 if (likely(order
== 0)) {
771 struct per_cpu_pages
*pcp
;
773 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
774 local_irq_save(flags
);
776 pcp
->count
+= rmqueue_bulk(zone
, 0,
777 pcp
->batch
, &pcp
->list
);
778 if (unlikely(!pcp
->count
))
781 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
782 list_del(&page
->lru
);
785 spin_lock_irqsave(&zone
->lock
, flags
);
786 page
= __rmqueue(zone
, order
);
787 spin_unlock(&zone
->lock
);
792 __mod_page_state_zone(zone
, pgalloc
, 1 << order
);
793 zone_statistics(zonelist
, zone
, cpu
);
794 local_irq_restore(flags
);
797 BUG_ON(bad_range(zone
, page
));
798 if (prep_new_page(page
, order
))
801 if (gfp_flags
& __GFP_ZERO
)
802 prep_zero_page(page
, order
, gfp_flags
);
804 if (order
&& (gfp_flags
& __GFP_COMP
))
805 prep_compound_page(page
, order
);
809 local_irq_restore(flags
);
814 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
815 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
816 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
817 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
818 #define ALLOC_HARDER 0x10 /* try to alloc harder */
819 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
820 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
823 * Return 1 if free pages are above 'mark'. This takes into account the order
826 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
827 int classzone_idx
, int alloc_flags
)
829 /* free_pages my go negative - that's OK */
830 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
833 if (alloc_flags
& ALLOC_HIGH
)
835 if (alloc_flags
& ALLOC_HARDER
)
838 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
840 for (o
= 0; o
< order
; o
++) {
841 /* At the next order, this order's pages become unavailable */
842 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
844 /* Require fewer higher order pages to be free */
847 if (free_pages
<= min
)
854 * get_page_from_freeliest goes through the zonelist trying to allocate
858 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
859 struct zonelist
*zonelist
, int alloc_flags
)
861 struct zone
**z
= zonelist
->zones
;
862 struct page
*page
= NULL
;
863 int classzone_idx
= zone_idx(*z
);
866 * Go through the zonelist once, looking for a zone with enough free.
867 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
870 if ((alloc_flags
& ALLOC_CPUSET
) &&
871 !cpuset_zone_allowed(*z
, gfp_mask
))
874 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
876 if (alloc_flags
& ALLOC_WMARK_MIN
)
877 mark
= (*z
)->pages_min
;
878 else if (alloc_flags
& ALLOC_WMARK_LOW
)
879 mark
= (*z
)->pages_low
;
881 mark
= (*z
)->pages_high
;
882 if (!zone_watermark_ok(*z
, order
, mark
,
883 classzone_idx
, alloc_flags
))
884 if (!zone_reclaim_mode
||
885 !zone_reclaim(*z
, gfp_mask
, order
))
889 page
= buffered_rmqueue(zonelist
, *z
, order
, gfp_mask
);
893 } while (*(++z
) != NULL
);
898 * This is the 'heart' of the zoned buddy allocator.
900 struct page
* fastcall
901 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
902 struct zonelist
*zonelist
)
904 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
907 struct reclaim_state reclaim_state
;
908 struct task_struct
*p
= current
;
911 int did_some_progress
;
913 might_sleep_if(wait
);
916 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
918 if (unlikely(*z
== NULL
)) {
919 /* Should this ever happen?? */
923 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
924 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
929 wakeup_kswapd(*z
, order
);
933 * OK, we're below the kswapd watermark and have kicked background
934 * reclaim. Now things get more complex, so set up alloc_flags according
935 * to how we want to proceed.
937 * The caller may dip into page reserves a bit more if the caller
938 * cannot run direct reclaim, or if the caller has realtime scheduling
939 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
940 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
942 alloc_flags
= ALLOC_WMARK_MIN
;
943 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
944 alloc_flags
|= ALLOC_HARDER
;
945 if (gfp_mask
& __GFP_HIGH
)
946 alloc_flags
|= ALLOC_HIGH
;
947 alloc_flags
|= ALLOC_CPUSET
;
950 * Go through the zonelist again. Let __GFP_HIGH and allocations
951 * coming from realtime tasks go deeper into reserves.
953 * This is the last chance, in general, before the goto nopage.
954 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
955 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
957 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
961 /* This allocation should allow future memory freeing. */
963 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
964 && !in_interrupt()) {
965 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
967 /* go through the zonelist yet again, ignoring mins */
968 page
= get_page_from_freelist(gfp_mask
, order
,
969 zonelist
, ALLOC_NO_WATERMARKS
);
972 if (gfp_mask
& __GFP_NOFAIL
) {
973 blk_congestion_wait(WRITE
, HZ
/50);
980 /* Atomic allocations - we can't balance anything */
987 /* We now go into synchronous reclaim */
988 cpuset_memory_pressure_bump();
989 p
->flags
|= PF_MEMALLOC
;
990 reclaim_state
.reclaimed_slab
= 0;
991 p
->reclaim_state
= &reclaim_state
;
993 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
995 p
->reclaim_state
= NULL
;
996 p
->flags
&= ~PF_MEMALLOC
;
1000 if (likely(did_some_progress
)) {
1001 page
= get_page_from_freelist(gfp_mask
, order
,
1002 zonelist
, alloc_flags
);
1005 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1007 * Go through the zonelist yet one more time, keep
1008 * very high watermark here, this is only to catch
1009 * a parallel oom killing, we must fail if we're still
1010 * under heavy pressure.
1012 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1013 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1017 out_of_memory(zonelist
, gfp_mask
, order
);
1022 * Don't let big-order allocations loop unless the caller explicitly
1023 * requests that. Wait for some write requests to complete then retry.
1025 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1026 * <= 3, but that may not be true in other implementations.
1029 if (!(gfp_mask
& __GFP_NORETRY
)) {
1030 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
1032 if (gfp_mask
& __GFP_NOFAIL
)
1036 blk_congestion_wait(WRITE
, HZ
/50);
1041 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1042 printk(KERN_WARNING
"%s: page allocation failure."
1043 " order:%d, mode:0x%x\n",
1044 p
->comm
, order
, gfp_mask
);
1052 EXPORT_SYMBOL(__alloc_pages
);
1055 * Common helper functions.
1057 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1060 page
= alloc_pages(gfp_mask
, order
);
1063 return (unsigned long) page_address(page
);
1066 EXPORT_SYMBOL(__get_free_pages
);
1068 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1073 * get_zeroed_page() returns a 32-bit address, which cannot represent
1076 BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1078 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1080 return (unsigned long) page_address(page
);
1084 EXPORT_SYMBOL(get_zeroed_page
);
1086 void __pagevec_free(struct pagevec
*pvec
)
1088 int i
= pagevec_count(pvec
);
1091 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1094 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1096 if (put_page_testzero(page
)) {
1098 free_hot_page(page
);
1100 __free_pages_ok(page
, order
);
1104 EXPORT_SYMBOL(__free_pages
);
1106 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1109 BUG_ON(!virt_addr_valid((void *)addr
));
1110 __free_pages(virt_to_page((void *)addr
), order
);
1114 EXPORT_SYMBOL(free_pages
);
1117 * Total amount of free (allocatable) RAM:
1119 unsigned int nr_free_pages(void)
1121 unsigned int sum
= 0;
1125 sum
+= zone
->free_pages
;
1130 EXPORT_SYMBOL(nr_free_pages
);
1133 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1135 unsigned int i
, sum
= 0;
1137 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1138 sum
+= pgdat
->node_zones
[i
].free_pages
;
1144 static unsigned int nr_free_zone_pages(int offset
)
1146 /* Just pick one node, since fallback list is circular */
1147 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1148 unsigned int sum
= 0;
1150 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1151 struct zone
**zonep
= zonelist
->zones
;
1154 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1155 unsigned long size
= zone
->present_pages
;
1156 unsigned long high
= zone
->pages_high
;
1165 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1167 unsigned int nr_free_buffer_pages(void)
1169 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1173 * Amount of free RAM allocatable within all zones
1175 unsigned int nr_free_pagecache_pages(void)
1177 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1180 #ifdef CONFIG_HIGHMEM
1181 unsigned int nr_free_highpages (void)
1184 unsigned int pages
= 0;
1186 for_each_pgdat(pgdat
)
1187 pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1194 static void show_node(struct zone
*zone
)
1196 printk("Node %d ", zone
->zone_pgdat
->node_id
);
1199 #define show_node(zone) do { } while (0)
1203 * Accumulate the page_state information across all CPUs.
1204 * The result is unavoidably approximate - it can change
1205 * during and after execution of this function.
1207 static DEFINE_PER_CPU(struct page_state
, page_states
) = {0};
1209 atomic_t nr_pagecache
= ATOMIC_INIT(0);
1210 EXPORT_SYMBOL(nr_pagecache
);
1212 DEFINE_PER_CPU(long, nr_pagecache_local
) = 0;
1215 static void __get_page_state(struct page_state
*ret
, int nr
, cpumask_t
*cpumask
)
1219 memset(ret
, 0, nr
* sizeof(unsigned long));
1220 cpus_and(*cpumask
, *cpumask
, cpu_online_map
);
1222 cpu
= first_cpu(*cpumask
);
1223 while (cpu
< NR_CPUS
) {
1224 unsigned long *in
, *out
, off
;
1226 if (!cpu_isset(cpu
, *cpumask
))
1229 in
= (unsigned long *)&per_cpu(page_states
, cpu
);
1231 cpu
= next_cpu(cpu
, *cpumask
);
1233 if (likely(cpu
< NR_CPUS
))
1234 prefetch(&per_cpu(page_states
, cpu
));
1236 out
= (unsigned long *)ret
;
1237 for (off
= 0; off
< nr
; off
++)
1242 void get_page_state_node(struct page_state
*ret
, int node
)
1245 cpumask_t mask
= node_to_cpumask(node
);
1247 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1248 nr
/= sizeof(unsigned long);
1250 __get_page_state(ret
, nr
+1, &mask
);
1253 void get_page_state(struct page_state
*ret
)
1256 cpumask_t mask
= CPU_MASK_ALL
;
1258 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1259 nr
/= sizeof(unsigned long);
1261 __get_page_state(ret
, nr
+ 1, &mask
);
1264 void get_full_page_state(struct page_state
*ret
)
1266 cpumask_t mask
= CPU_MASK_ALL
;
1268 __get_page_state(ret
, sizeof(*ret
) / sizeof(unsigned long), &mask
);
1271 unsigned long read_page_state_offset(unsigned long offset
)
1273 unsigned long ret
= 0;
1276 for_each_online_cpu(cpu
) {
1279 in
= (unsigned long)&per_cpu(page_states
, cpu
) + offset
;
1280 ret
+= *((unsigned long *)in
);
1285 void __mod_page_state_offset(unsigned long offset
, unsigned long delta
)
1289 ptr
= &__get_cpu_var(page_states
);
1290 *(unsigned long *)(ptr
+ offset
) += delta
;
1292 EXPORT_SYMBOL(__mod_page_state_offset
);
1294 void mod_page_state_offset(unsigned long offset
, unsigned long delta
)
1296 unsigned long flags
;
1299 local_irq_save(flags
);
1300 ptr
= &__get_cpu_var(page_states
);
1301 *(unsigned long *)(ptr
+ offset
) += delta
;
1302 local_irq_restore(flags
);
1304 EXPORT_SYMBOL(mod_page_state_offset
);
1306 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
1307 unsigned long *free
, struct pglist_data
*pgdat
)
1309 struct zone
*zones
= pgdat
->node_zones
;
1315 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1316 *active
+= zones
[i
].nr_active
;
1317 *inactive
+= zones
[i
].nr_inactive
;
1318 *free
+= zones
[i
].free_pages
;
1322 void get_zone_counts(unsigned long *active
,
1323 unsigned long *inactive
, unsigned long *free
)
1325 struct pglist_data
*pgdat
;
1330 for_each_pgdat(pgdat
) {
1331 unsigned long l
, m
, n
;
1332 __get_zone_counts(&l
, &m
, &n
, pgdat
);
1339 void si_meminfo(struct sysinfo
*val
)
1341 val
->totalram
= totalram_pages
;
1343 val
->freeram
= nr_free_pages();
1344 val
->bufferram
= nr_blockdev_pages();
1345 #ifdef CONFIG_HIGHMEM
1346 val
->totalhigh
= totalhigh_pages
;
1347 val
->freehigh
= nr_free_highpages();
1352 val
->mem_unit
= PAGE_SIZE
;
1355 EXPORT_SYMBOL(si_meminfo
);
1358 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1360 pg_data_t
*pgdat
= NODE_DATA(nid
);
1362 val
->totalram
= pgdat
->node_present_pages
;
1363 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1364 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1365 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1366 val
->mem_unit
= PAGE_SIZE
;
1370 #define K(x) ((x) << (PAGE_SHIFT-10))
1373 * Show free area list (used inside shift_scroll-lock stuff)
1374 * We also calculate the percentage fragmentation. We do this by counting the
1375 * memory on each free list with the exception of the first item on the list.
1377 void show_free_areas(void)
1379 struct page_state ps
;
1380 int cpu
, temperature
;
1381 unsigned long active
;
1382 unsigned long inactive
;
1386 for_each_zone(zone
) {
1388 printk("%s per-cpu:", zone
->name
);
1390 if (!populated_zone(zone
)) {
1396 for_each_online_cpu(cpu
) {
1397 struct per_cpu_pageset
*pageset
;
1399 pageset
= zone_pcp(zone
, cpu
);
1401 for (temperature
= 0; temperature
< 2; temperature
++)
1402 printk("cpu %d %s: high %d, batch %d used:%d\n",
1404 temperature
? "cold" : "hot",
1405 pageset
->pcp
[temperature
].high
,
1406 pageset
->pcp
[temperature
].batch
,
1407 pageset
->pcp
[temperature
].count
);
1411 get_page_state(&ps
);
1412 get_zone_counts(&active
, &inactive
, &free
);
1414 printk("Free pages: %11ukB (%ukB HighMem)\n",
1416 K(nr_free_highpages()));
1418 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1419 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1428 ps
.nr_page_table_pages
);
1430 for_each_zone(zone
) {
1442 " pages_scanned:%lu"
1443 " all_unreclaimable? %s"
1446 K(zone
->free_pages
),
1449 K(zone
->pages_high
),
1451 K(zone
->nr_inactive
),
1452 K(zone
->present_pages
),
1453 zone
->pages_scanned
,
1454 (zone
->all_unreclaimable
? "yes" : "no")
1456 printk("lowmem_reserve[]:");
1457 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1458 printk(" %lu", zone
->lowmem_reserve
[i
]);
1462 for_each_zone(zone
) {
1463 unsigned long nr
, flags
, order
, total
= 0;
1466 printk("%s: ", zone
->name
);
1467 if (!populated_zone(zone
)) {
1472 spin_lock_irqsave(&zone
->lock
, flags
);
1473 for (order
= 0; order
< MAX_ORDER
; order
++) {
1474 nr
= zone
->free_area
[order
].nr_free
;
1475 total
+= nr
<< order
;
1476 printk("%lu*%lukB ", nr
, K(1UL) << order
);
1478 spin_unlock_irqrestore(&zone
->lock
, flags
);
1479 printk("= %lukB\n", K(total
));
1482 show_swap_cache_info();
1486 * Builds allocation fallback zone lists.
1488 * Add all populated zones of a node to the zonelist.
1490 static int __init
build_zonelists_node(pg_data_t
*pgdat
,
1491 struct zonelist
*zonelist
, int nr_zones
, int zone_type
)
1495 BUG_ON(zone_type
> ZONE_HIGHMEM
);
1498 zone
= pgdat
->node_zones
+ zone_type
;
1499 if (populated_zone(zone
)) {
1500 #ifndef CONFIG_HIGHMEM
1501 BUG_ON(zone_type
> ZONE_NORMAL
);
1503 zonelist
->zones
[nr_zones
++] = zone
;
1504 check_highest_zone(zone_type
);
1508 } while (zone_type
>= 0);
1512 static inline int highest_zone(int zone_bits
)
1514 int res
= ZONE_NORMAL
;
1515 if (zone_bits
& (__force
int)__GFP_HIGHMEM
)
1517 if (zone_bits
& (__force
int)__GFP_DMA32
)
1519 if (zone_bits
& (__force
int)__GFP_DMA
)
1525 #define MAX_NODE_LOAD (num_online_nodes())
1526 static int __initdata node_load
[MAX_NUMNODES
];
1528 * find_next_best_node - find the next node that should appear in a given node's fallback list
1529 * @node: node whose fallback list we're appending
1530 * @used_node_mask: nodemask_t of already used nodes
1532 * We use a number of factors to determine which is the next node that should
1533 * appear on a given node's fallback list. The node should not have appeared
1534 * already in @node's fallback list, and it should be the next closest node
1535 * according to the distance array (which contains arbitrary distance values
1536 * from each node to each node in the system), and should also prefer nodes
1537 * with no CPUs, since presumably they'll have very little allocation pressure
1538 * on them otherwise.
1539 * It returns -1 if no node is found.
1541 static int __init
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1544 int min_val
= INT_MAX
;
1547 /* Use the local node if we haven't already */
1548 if (!node_isset(node
, *used_node_mask
)) {
1549 node_set(node
, *used_node_mask
);
1553 for_each_online_node(n
) {
1556 /* Don't want a node to appear more than once */
1557 if (node_isset(n
, *used_node_mask
))
1560 /* Use the distance array to find the distance */
1561 val
= node_distance(node
, n
);
1563 /* Penalize nodes under us ("prefer the next node") */
1566 /* Give preference to headless and unused nodes */
1567 tmp
= node_to_cpumask(n
);
1568 if (!cpus_empty(tmp
))
1569 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1571 /* Slight preference for less loaded node */
1572 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1573 val
+= node_load
[n
];
1575 if (val
< min_val
) {
1582 node_set(best_node
, *used_node_mask
);
1587 static void __init
build_zonelists(pg_data_t
*pgdat
)
1589 int i
, j
, k
, node
, local_node
;
1590 int prev_node
, load
;
1591 struct zonelist
*zonelist
;
1592 nodemask_t used_mask
;
1594 /* initialize zonelists */
1595 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1596 zonelist
= pgdat
->node_zonelists
+ i
;
1597 zonelist
->zones
[0] = NULL
;
1600 /* NUMA-aware ordering of nodes */
1601 local_node
= pgdat
->node_id
;
1602 load
= num_online_nodes();
1603 prev_node
= local_node
;
1604 nodes_clear(used_mask
);
1605 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1606 int distance
= node_distance(local_node
, node
);
1609 * If another node is sufficiently far away then it is better
1610 * to reclaim pages in a zone before going off node.
1612 if (distance
> RECLAIM_DISTANCE
)
1613 zone_reclaim_mode
= 1;
1616 * We don't want to pressure a particular node.
1617 * So adding penalty to the first node in same
1618 * distance group to make it round-robin.
1621 if (distance
!= node_distance(local_node
, prev_node
))
1622 node_load
[node
] += load
;
1625 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1626 zonelist
= pgdat
->node_zonelists
+ i
;
1627 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1629 k
= highest_zone(i
);
1631 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1632 zonelist
->zones
[j
] = NULL
;
1637 #else /* CONFIG_NUMA */
1639 static void __init
build_zonelists(pg_data_t
*pgdat
)
1641 int i
, j
, k
, node
, local_node
;
1643 local_node
= pgdat
->node_id
;
1644 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1645 struct zonelist
*zonelist
;
1647 zonelist
= pgdat
->node_zonelists
+ i
;
1650 k
= highest_zone(i
);
1651 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1653 * Now we build the zonelist so that it contains the zones
1654 * of all the other nodes.
1655 * We don't want to pressure a particular node, so when
1656 * building the zones for node N, we make sure that the
1657 * zones coming right after the local ones are those from
1658 * node N+1 (modulo N)
1660 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1661 if (!node_online(node
))
1663 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1665 for (node
= 0; node
< local_node
; node
++) {
1666 if (!node_online(node
))
1668 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1671 zonelist
->zones
[j
] = NULL
;
1675 #endif /* CONFIG_NUMA */
1677 void __init
build_all_zonelists(void)
1681 for_each_online_node(i
)
1682 build_zonelists(NODE_DATA(i
));
1683 printk("Built %i zonelists\n", num_online_nodes());
1684 cpuset_init_current_mems_allowed();
1688 * Helper functions to size the waitqueue hash table.
1689 * Essentially these want to choose hash table sizes sufficiently
1690 * large so that collisions trying to wait on pages are rare.
1691 * But in fact, the number of active page waitqueues on typical
1692 * systems is ridiculously low, less than 200. So this is even
1693 * conservative, even though it seems large.
1695 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1696 * waitqueues, i.e. the size of the waitq table given the number of pages.
1698 #define PAGES_PER_WAITQUEUE 256
1700 static inline unsigned long wait_table_size(unsigned long pages
)
1702 unsigned long size
= 1;
1704 pages
/= PAGES_PER_WAITQUEUE
;
1706 while (size
< pages
)
1710 * Once we have dozens or even hundreds of threads sleeping
1711 * on IO we've got bigger problems than wait queue collision.
1712 * Limit the size of the wait table to a reasonable size.
1714 size
= min(size
, 4096UL);
1716 return max(size
, 4UL);
1720 * This is an integer logarithm so that shifts can be used later
1721 * to extract the more random high bits from the multiplicative
1722 * hash function before the remainder is taken.
1724 static inline unsigned long wait_table_bits(unsigned long size
)
1729 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1731 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1732 unsigned long *zones_size
, unsigned long *zholes_size
)
1734 unsigned long realtotalpages
, totalpages
= 0;
1737 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1738 totalpages
+= zones_size
[i
];
1739 pgdat
->node_spanned_pages
= totalpages
;
1741 realtotalpages
= totalpages
;
1743 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1744 realtotalpages
-= zholes_size
[i
];
1745 pgdat
->node_present_pages
= realtotalpages
;
1746 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1751 * Initially all pages are reserved - free ones are freed
1752 * up by free_all_bootmem() once the early boot process is
1753 * done. Non-atomic initialization, single-pass.
1755 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1756 unsigned long start_pfn
)
1759 unsigned long end_pfn
= start_pfn
+ size
;
1762 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1763 if (!early_pfn_valid(pfn
))
1765 page
= pfn_to_page(pfn
);
1766 set_page_links(page
, zone
, nid
, pfn
);
1767 set_page_count(page
, 1);
1768 reset_page_mapcount(page
);
1769 SetPageReserved(page
);
1770 INIT_LIST_HEAD(&page
->lru
);
1771 #ifdef WANT_PAGE_VIRTUAL
1772 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1773 if (!is_highmem_idx(zone
))
1774 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1779 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1783 for (order
= 0; order
< MAX_ORDER
; order
++) {
1784 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1785 zone
->free_area
[order
].nr_free
= 0;
1789 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1790 void zonetable_add(struct zone
*zone
, int nid
, int zid
, unsigned long pfn
,
1793 unsigned long snum
= pfn_to_section_nr(pfn
);
1794 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1797 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1799 for (; snum
<= end
; snum
++)
1800 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1803 #ifndef __HAVE_ARCH_MEMMAP_INIT
1804 #define memmap_init(size, nid, zone, start_pfn) \
1805 memmap_init_zone((size), (nid), (zone), (start_pfn))
1808 static int __cpuinit
zone_batchsize(struct zone
*zone
)
1813 * The per-cpu-pages pools are set to around 1000th of the
1814 * size of the zone. But no more than 1/2 of a meg.
1816 * OK, so we don't know how big the cache is. So guess.
1818 batch
= zone
->present_pages
/ 1024;
1819 if (batch
* PAGE_SIZE
> 512 * 1024)
1820 batch
= (512 * 1024) / PAGE_SIZE
;
1821 batch
/= 4; /* We effectively *= 4 below */
1826 * Clamp the batch to a 2^n - 1 value. Having a power
1827 * of 2 value was found to be more likely to have
1828 * suboptimal cache aliasing properties in some cases.
1830 * For example if 2 tasks are alternately allocating
1831 * batches of pages, one task can end up with a lot
1832 * of pages of one half of the possible page colors
1833 * and the other with pages of the other colors.
1835 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1840 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1842 struct per_cpu_pages
*pcp
;
1844 memset(p
, 0, sizeof(*p
));
1846 pcp
= &p
->pcp
[0]; /* hot */
1848 pcp
->high
= 6 * batch
;
1849 pcp
->batch
= max(1UL, 1 * batch
);
1850 INIT_LIST_HEAD(&pcp
->list
);
1852 pcp
= &p
->pcp
[1]; /* cold*/
1854 pcp
->high
= 2 * batch
;
1855 pcp
->batch
= max(1UL, batch
/2);
1856 INIT_LIST_HEAD(&pcp
->list
);
1860 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1861 * to the value high for the pageset p.
1864 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
1867 struct per_cpu_pages
*pcp
;
1869 pcp
= &p
->pcp
[0]; /* hot list */
1871 pcp
->batch
= max(1UL, high
/4);
1872 if ((high
/4) > (PAGE_SHIFT
* 8))
1873 pcp
->batch
= PAGE_SHIFT
* 8;
1879 * Boot pageset table. One per cpu which is going to be used for all
1880 * zones and all nodes. The parameters will be set in such a way
1881 * that an item put on a list will immediately be handed over to
1882 * the buddy list. This is safe since pageset manipulation is done
1883 * with interrupts disabled.
1885 * Some NUMA counter updates may also be caught by the boot pagesets.
1887 * The boot_pagesets must be kept even after bootup is complete for
1888 * unused processors and/or zones. They do play a role for bootstrapping
1889 * hotplugged processors.
1891 * zoneinfo_show() and maybe other functions do
1892 * not check if the processor is online before following the pageset pointer.
1893 * Other parts of the kernel may not check if the zone is available.
1895 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
1898 * Dynamically allocate memory for the
1899 * per cpu pageset array in struct zone.
1901 static int __cpuinit
process_zones(int cpu
)
1903 struct zone
*zone
, *dzone
;
1905 for_each_zone(zone
) {
1907 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
1908 GFP_KERNEL
, cpu_to_node(cpu
));
1909 if (!zone_pcp(zone
, cpu
))
1912 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
1914 if (percpu_pagelist_fraction
)
1915 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
1916 (zone
->present_pages
/ percpu_pagelist_fraction
));
1921 for_each_zone(dzone
) {
1924 kfree(zone_pcp(dzone
, cpu
));
1925 zone_pcp(dzone
, cpu
) = NULL
;
1930 static inline void free_zone_pagesets(int cpu
)
1934 for_each_zone(zone
) {
1935 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1937 zone_pcp(zone
, cpu
) = NULL
;
1942 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1943 unsigned long action
,
1946 int cpu
= (long)hcpu
;
1947 int ret
= NOTIFY_OK
;
1950 case CPU_UP_PREPARE
:
1951 if (process_zones(cpu
))
1954 case CPU_UP_CANCELED
:
1956 free_zone_pagesets(cpu
);
1964 static struct notifier_block pageset_notifier
=
1965 { &pageset_cpuup_callback
, NULL
, 0 };
1967 void __init
setup_per_cpu_pageset(void)
1971 /* Initialize per_cpu_pageset for cpu 0.
1972 * A cpuup callback will do this for every cpu
1973 * as it comes online
1975 err
= process_zones(smp_processor_id());
1977 register_cpu_notifier(&pageset_notifier
);
1983 void zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1986 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1989 * The per-page waitqueue mechanism uses hashed waitqueues
1992 zone
->wait_table_size
= wait_table_size(zone_size_pages
);
1993 zone
->wait_table_bits
= wait_table_bits(zone
->wait_table_size
);
1994 zone
->wait_table
= (wait_queue_head_t
*)
1995 alloc_bootmem_node(pgdat
, zone
->wait_table_size
1996 * sizeof(wait_queue_head_t
));
1998 for(i
= 0; i
< zone
->wait_table_size
; ++i
)
1999 init_waitqueue_head(zone
->wait_table
+ i
);
2002 static __meminit
void zone_pcp_init(struct zone
*zone
)
2005 unsigned long batch
= zone_batchsize(zone
);
2007 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2009 /* Early boot. Slab allocator not functional yet */
2010 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2011 setup_pageset(&boot_pageset
[cpu
],0);
2013 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2016 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2017 zone
->name
, zone
->present_pages
, batch
);
2020 static __meminit
void init_currently_empty_zone(struct zone
*zone
,
2021 unsigned long zone_start_pfn
, unsigned long size
)
2023 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2025 zone_wait_table_init(zone
, size
);
2026 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2028 zone
->zone_mem_map
= pfn_to_page(zone_start_pfn
);
2029 zone
->zone_start_pfn
= zone_start_pfn
;
2031 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
2033 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2037 * Set up the zone data structures:
2038 * - mark all pages reserved
2039 * - mark all memory queues empty
2040 * - clear the memory bitmaps
2042 static void __init
free_area_init_core(struct pglist_data
*pgdat
,
2043 unsigned long *zones_size
, unsigned long *zholes_size
)
2046 int nid
= pgdat
->node_id
;
2047 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
2049 pgdat_resize_init(pgdat
);
2050 pgdat
->nr_zones
= 0;
2051 init_waitqueue_head(&pgdat
->kswapd_wait
);
2052 pgdat
->kswapd_max_order
= 0;
2054 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2055 struct zone
*zone
= pgdat
->node_zones
+ j
;
2056 unsigned long size
, realsize
;
2058 realsize
= size
= zones_size
[j
];
2060 realsize
-= zholes_size
[j
];
2062 if (j
< ZONE_HIGHMEM
)
2063 nr_kernel_pages
+= realsize
;
2064 nr_all_pages
+= realsize
;
2066 zone
->spanned_pages
= size
;
2067 zone
->present_pages
= realsize
;
2068 zone
->name
= zone_names
[j
];
2069 spin_lock_init(&zone
->lock
);
2070 spin_lock_init(&zone
->lru_lock
);
2071 zone_seqlock_init(zone
);
2072 zone
->zone_pgdat
= pgdat
;
2073 zone
->free_pages
= 0;
2075 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
2077 zone_pcp_init(zone
);
2078 INIT_LIST_HEAD(&zone
->active_list
);
2079 INIT_LIST_HEAD(&zone
->inactive_list
);
2080 zone
->nr_scan_active
= 0;
2081 zone
->nr_scan_inactive
= 0;
2082 zone
->nr_active
= 0;
2083 zone
->nr_inactive
= 0;
2084 atomic_set(&zone
->reclaim_in_progress
, 0);
2088 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2089 init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2090 zone_start_pfn
+= size
;
2094 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2096 /* Skip empty nodes */
2097 if (!pgdat
->node_spanned_pages
)
2100 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2101 /* ia64 gets its own node_mem_map, before this, without bootmem */
2102 if (!pgdat
->node_mem_map
) {
2106 size
= (pgdat
->node_spanned_pages
+ 1) * sizeof(struct page
);
2107 map
= alloc_remap(pgdat
->node_id
, size
);
2109 map
= alloc_bootmem_node(pgdat
, size
);
2110 pgdat
->node_mem_map
= map
;
2112 #ifdef CONFIG_FLATMEM
2114 * With no DISCONTIG, the global mem_map is just set as node 0's
2116 if (pgdat
== NODE_DATA(0))
2117 mem_map
= NODE_DATA(0)->node_mem_map
;
2119 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2122 void __init
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2123 unsigned long *zones_size
, unsigned long node_start_pfn
,
2124 unsigned long *zholes_size
)
2126 pgdat
->node_id
= nid
;
2127 pgdat
->node_start_pfn
= node_start_pfn
;
2128 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
2130 alloc_node_mem_map(pgdat
);
2132 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2135 #ifndef CONFIG_NEED_MULTIPLE_NODES
2136 static bootmem_data_t contig_bootmem_data
;
2137 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2139 EXPORT_SYMBOL(contig_page_data
);
2142 void __init
free_area_init(unsigned long *zones_size
)
2144 free_area_init_node(0, NODE_DATA(0), zones_size
,
2145 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2148 #ifdef CONFIG_PROC_FS
2150 #include <linux/seq_file.h>
2152 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
2157 for (pgdat
= pgdat_list
; pgdat
&& node
; pgdat
= pgdat
->pgdat_next
)
2163 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2165 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2168 return pgdat
->pgdat_next
;
2171 static void frag_stop(struct seq_file
*m
, void *arg
)
2176 * This walks the free areas for each zone.
2178 static int frag_show(struct seq_file
*m
, void *arg
)
2180 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2182 struct zone
*node_zones
= pgdat
->node_zones
;
2183 unsigned long flags
;
2186 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
2187 if (!populated_zone(zone
))
2190 spin_lock_irqsave(&zone
->lock
, flags
);
2191 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
2192 for (order
= 0; order
< MAX_ORDER
; ++order
)
2193 seq_printf(m
, "%6lu ", zone
->free_area
[order
].nr_free
);
2194 spin_unlock_irqrestore(&zone
->lock
, flags
);
2200 struct seq_operations fragmentation_op
= {
2201 .start
= frag_start
,
2208 * Output information about zones in @pgdat.
2210 static int zoneinfo_show(struct seq_file
*m
, void *arg
)
2212 pg_data_t
*pgdat
= arg
;
2214 struct zone
*node_zones
= pgdat
->node_zones
;
2215 unsigned long flags
;
2217 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; zone
++) {
2220 if (!populated_zone(zone
))
2223 spin_lock_irqsave(&zone
->lock
, flags
);
2224 seq_printf(m
, "Node %d, zone %8s", pgdat
->node_id
, zone
->name
);
2232 "\n scanned %lu (a: %lu i: %lu)"
2241 zone
->pages_scanned
,
2242 zone
->nr_scan_active
, zone
->nr_scan_inactive
,
2243 zone
->spanned_pages
,
2244 zone
->present_pages
);
2246 "\n protection: (%lu",
2247 zone
->lowmem_reserve
[0]);
2248 for (i
= 1; i
< ARRAY_SIZE(zone
->lowmem_reserve
); i
++)
2249 seq_printf(m
, ", %lu", zone
->lowmem_reserve
[i
]);
2253 for_each_online_cpu(i
) {
2254 struct per_cpu_pageset
*pageset
;
2257 pageset
= zone_pcp(zone
, i
);
2258 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2259 if (pageset
->pcp
[j
].count
)
2262 if (j
== ARRAY_SIZE(pageset
->pcp
))
2264 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2266 "\n cpu: %i pcp: %i"
2271 pageset
->pcp
[j
].count
,
2272 pageset
->pcp
[j
].high
,
2273 pageset
->pcp
[j
].batch
);
2279 "\n numa_foreign: %lu"
2280 "\n interleave_hit: %lu"
2281 "\n local_node: %lu"
2282 "\n other_node: %lu",
2285 pageset
->numa_foreign
,
2286 pageset
->interleave_hit
,
2287 pageset
->local_node
,
2288 pageset
->other_node
);
2292 "\n all_unreclaimable: %u"
2293 "\n prev_priority: %i"
2294 "\n temp_priority: %i"
2295 "\n start_pfn: %lu",
2296 zone
->all_unreclaimable
,
2297 zone
->prev_priority
,
2298 zone
->temp_priority
,
2299 zone
->zone_start_pfn
);
2300 spin_unlock_irqrestore(&zone
->lock
, flags
);
2306 struct seq_operations zoneinfo_op
= {
2307 .start
= frag_start
, /* iterate over all zones. The same as in
2311 .show
= zoneinfo_show
,
2314 static char *vmstat_text
[] = {
2318 "nr_page_table_pages",
2349 "pgscan_kswapd_high",
2350 "pgscan_kswapd_normal",
2351 "pgscan_kswapd_dma32",
2352 "pgscan_kswapd_dma",
2354 "pgscan_direct_high",
2355 "pgscan_direct_normal",
2356 "pgscan_direct_dma32",
2357 "pgscan_direct_dma",
2362 "kswapd_inodesteal",
2370 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
2372 struct page_state
*ps
;
2374 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2377 ps
= kmalloc(sizeof(*ps
), GFP_KERNEL
);
2380 return ERR_PTR(-ENOMEM
);
2381 get_full_page_state(ps
);
2382 ps
->pgpgin
/= 2; /* sectors -> kbytes */
2384 return (unsigned long *)ps
+ *pos
;
2387 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2390 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2392 return (unsigned long *)m
->private + *pos
;
2395 static int vmstat_show(struct seq_file
*m
, void *arg
)
2397 unsigned long *l
= arg
;
2398 unsigned long off
= l
- (unsigned long *)m
->private;
2400 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
2404 static void vmstat_stop(struct seq_file
*m
, void *arg
)
2410 struct seq_operations vmstat_op
= {
2411 .start
= vmstat_start
,
2412 .next
= vmstat_next
,
2413 .stop
= vmstat_stop
,
2414 .show
= vmstat_show
,
2417 #endif /* CONFIG_PROC_FS */
2419 #ifdef CONFIG_HOTPLUG_CPU
2420 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2421 unsigned long action
, void *hcpu
)
2423 int cpu
= (unsigned long)hcpu
;
2425 unsigned long *src
, *dest
;
2427 if (action
== CPU_DEAD
) {
2430 /* Drain local pagecache count. */
2431 count
= &per_cpu(nr_pagecache_local
, cpu
);
2432 atomic_add(*count
, &nr_pagecache
);
2434 local_irq_disable();
2437 /* Add dead cpu's page_states to our own. */
2438 dest
= (unsigned long *)&__get_cpu_var(page_states
);
2439 src
= (unsigned long *)&per_cpu(page_states
, cpu
);
2441 for (i
= 0; i
< sizeof(struct page_state
)/sizeof(unsigned long);
2451 #endif /* CONFIG_HOTPLUG_CPU */
2453 void __init
page_alloc_init(void)
2455 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2459 * setup_per_zone_lowmem_reserve - called whenever
2460 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2461 * has a correct pages reserved value, so an adequate number of
2462 * pages are left in the zone after a successful __alloc_pages().
2464 static void setup_per_zone_lowmem_reserve(void)
2466 struct pglist_data
*pgdat
;
2469 for_each_pgdat(pgdat
) {
2470 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2471 struct zone
*zone
= pgdat
->node_zones
+ j
;
2472 unsigned long present_pages
= zone
->present_pages
;
2474 zone
->lowmem_reserve
[j
] = 0;
2476 for (idx
= j
-1; idx
>= 0; idx
--) {
2477 struct zone
*lower_zone
;
2479 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2480 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2482 lower_zone
= pgdat
->node_zones
+ idx
;
2483 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2484 sysctl_lowmem_reserve_ratio
[idx
];
2485 present_pages
+= lower_zone
->present_pages
;
2492 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2493 * that the pages_{min,low,high} values for each zone are set correctly
2494 * with respect to min_free_kbytes.
2496 void setup_per_zone_pages_min(void)
2498 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2499 unsigned long lowmem_pages
= 0;
2501 unsigned long flags
;
2503 /* Calculate total number of !ZONE_HIGHMEM pages */
2504 for_each_zone(zone
) {
2505 if (!is_highmem(zone
))
2506 lowmem_pages
+= zone
->present_pages
;
2509 for_each_zone(zone
) {
2511 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2512 tmp
= (pages_min
* zone
->present_pages
) / lowmem_pages
;
2513 if (is_highmem(zone
)) {
2515 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2516 * need highmem pages, so cap pages_min to a small
2519 * The (pages_high-pages_low) and (pages_low-pages_min)
2520 * deltas controls asynch page reclaim, and so should
2521 * not be capped for highmem.
2525 min_pages
= zone
->present_pages
/ 1024;
2526 if (min_pages
< SWAP_CLUSTER_MAX
)
2527 min_pages
= SWAP_CLUSTER_MAX
;
2528 if (min_pages
> 128)
2530 zone
->pages_min
= min_pages
;
2533 * If it's a lowmem zone, reserve a number of pages
2534 * proportionate to the zone's size.
2536 zone
->pages_min
= tmp
;
2539 zone
->pages_low
= zone
->pages_min
+ tmp
/ 4;
2540 zone
->pages_high
= zone
->pages_min
+ tmp
/ 2;
2541 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2546 * Initialise min_free_kbytes.
2548 * For small machines we want it small (128k min). For large machines
2549 * we want it large (64MB max). But it is not linear, because network
2550 * bandwidth does not increase linearly with machine size. We use
2552 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2553 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2569 static int __init
init_per_zone_pages_min(void)
2571 unsigned long lowmem_kbytes
;
2573 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2575 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2576 if (min_free_kbytes
< 128)
2577 min_free_kbytes
= 128;
2578 if (min_free_kbytes
> 65536)
2579 min_free_kbytes
= 65536;
2580 setup_per_zone_pages_min();
2581 setup_per_zone_lowmem_reserve();
2584 module_init(init_per_zone_pages_min
)
2587 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2588 * that we can call two helper functions whenever min_free_kbytes
2591 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2592 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2594 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2595 setup_per_zone_pages_min();
2600 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2601 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2602 * whenever sysctl_lowmem_reserve_ratio changes.
2604 * The reserve ratio obviously has absolutely no relation with the
2605 * pages_min watermarks. The lowmem reserve ratio can only make sense
2606 * if in function of the boot time zone sizes.
2608 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2609 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2611 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2612 setup_per_zone_lowmem_reserve();
2617 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2618 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2619 * can have before it gets flushed back to buddy allocator.
2622 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
2623 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2629 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2630 if (!write
|| (ret
== -EINVAL
))
2632 for_each_zone(zone
) {
2633 for_each_online_cpu(cpu
) {
2635 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
2636 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
2642 __initdata
int hashdist
= HASHDIST_DEFAULT
;
2645 static int __init
set_hashdist(char *str
)
2649 hashdist
= simple_strtoul(str
, &str
, 0);
2652 __setup("hashdist=", set_hashdist
);
2656 * allocate a large system hash table from bootmem
2657 * - it is assumed that the hash table must contain an exact power-of-2
2658 * quantity of entries
2659 * - limit is the number of hash buckets, not the total allocation size
2661 void *__init
alloc_large_system_hash(const char *tablename
,
2662 unsigned long bucketsize
,
2663 unsigned long numentries
,
2666 unsigned int *_hash_shift
,
2667 unsigned int *_hash_mask
,
2668 unsigned long limit
)
2670 unsigned long long max
= limit
;
2671 unsigned long log2qty
, size
;
2674 /* allow the kernel cmdline to have a say */
2676 /* round applicable memory size up to nearest megabyte */
2677 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2678 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2679 numentries
>>= 20 - PAGE_SHIFT
;
2680 numentries
<<= 20 - PAGE_SHIFT
;
2682 /* limit to 1 bucket per 2^scale bytes of low memory */
2683 if (scale
> PAGE_SHIFT
)
2684 numentries
>>= (scale
- PAGE_SHIFT
);
2686 numentries
<<= (PAGE_SHIFT
- scale
);
2688 /* rounded up to nearest power of 2 in size */
2689 numentries
= 1UL << (long_log2(numentries
) + 1);
2691 /* limit allocation size to 1/16 total memory by default */
2693 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2694 do_div(max
, bucketsize
);
2697 if (numentries
> max
)
2700 log2qty
= long_log2(numentries
);
2703 size
= bucketsize
<< log2qty
;
2704 if (flags
& HASH_EARLY
)
2705 table
= alloc_bootmem(size
);
2707 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
2709 unsigned long order
;
2710 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
2712 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
2714 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2717 panic("Failed to allocate %s hash table\n", tablename
);
2719 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2722 long_log2(size
) - PAGE_SHIFT
,
2726 *_hash_shift
= log2qty
;
2728 *_hash_mask
= (1 << log2qty
) - 1;