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
);
61 * results with 256, 32 in the lowmem_reserve sysctl:
62 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
63 * 1G machine -> (16M dma, 784M normal, 224M high)
64 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
65 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
66 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
68 * TBD: should special case ZONE_DMA32 machines here - in those we normally
69 * don't need any ZONE_NORMAL reservation
71 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = { 256, 256, 32 };
73 EXPORT_SYMBOL(totalram_pages
);
76 * Used by page_zone() to look up the address of the struct zone whose
77 * id is encoded in the upper bits of page->flags
79 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
80 EXPORT_SYMBOL(zone_table
);
82 static char *zone_names
[MAX_NR_ZONES
] = { "DMA", "DMA32", "Normal", "HighMem" };
83 int min_free_kbytes
= 1024;
85 unsigned long __initdata nr_kernel_pages
;
86 unsigned long __initdata nr_all_pages
;
88 #ifdef CONFIG_DEBUG_VM
89 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
93 unsigned long pfn
= page_to_pfn(page
);
96 seq
= zone_span_seqbegin(zone
);
97 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
99 else if (pfn
< zone
->zone_start_pfn
)
101 } while (zone_span_seqretry(zone
, seq
));
106 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
108 #ifdef CONFIG_HOLES_IN_ZONE
109 if (!pfn_valid(page_to_pfn(page
)))
112 if (zone
!= page_zone(page
))
118 * Temporary debugging check for pages not lying within a given zone.
120 static int bad_range(struct zone
*zone
, struct page
*page
)
122 if (page_outside_zone_boundaries(zone
, page
))
124 if (!page_is_consistent(zone
, page
))
131 static inline int bad_range(struct zone
*zone
, struct page
*page
)
137 static void bad_page(struct page
*page
)
139 printk(KERN_EMERG
"Bad page state in process '%s'\n"
140 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
141 "Trying to fix it up, but a reboot is needed\n"
143 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
144 (unsigned long)page
->flags
, page
->mapping
,
145 page_mapcount(page
), page_count(page
));
147 page
->flags
&= ~(1 << PG_lru
|
156 set_page_count(page
, 0);
157 reset_page_mapcount(page
);
158 page
->mapping
= NULL
;
159 add_taint(TAINT_BAD_PAGE
);
163 * Higher-order pages are called "compound pages". They are structured thusly:
165 * The first PAGE_SIZE page is called the "head page".
167 * The remaining PAGE_SIZE pages are called "tail pages".
169 * All pages have PG_compound set. All pages have their ->private pointing at
170 * the head page (even the head page has this).
172 * The first tail page's ->mapping, if non-zero, holds the address of the
173 * compound page's put_page() function.
175 * The order of the allocation is stored in the first tail page's ->index
176 * This is only for debug at present. This usage means that zero-order pages
177 * may not be compound.
179 static void prep_compound_page(struct page
*page
, unsigned long order
)
182 int nr_pages
= 1 << order
;
184 page
[1].mapping
= NULL
;
185 page
[1].index
= order
;
186 for (i
= 0; i
< nr_pages
; i
++) {
187 struct page
*p
= page
+ i
;
190 set_page_private(p
, (unsigned long)page
);
194 static void destroy_compound_page(struct page
*page
, unsigned long order
)
197 int nr_pages
= 1 << order
;
199 if (unlikely(page
[1].index
!= order
))
202 for (i
= 0; i
< nr_pages
; i
++) {
203 struct page
*p
= page
+ i
;
205 if (unlikely(!PageCompound(p
) |
206 (page_private(p
) != (unsigned long)page
)))
208 ClearPageCompound(p
);
213 * function for dealing with page's order in buddy system.
214 * zone->lock is already acquired when we use these.
215 * So, we don't need atomic page->flags operations here.
217 static inline unsigned long page_order(struct page
*page
) {
218 return page_private(page
);
221 static inline void set_page_order(struct page
*page
, int order
) {
222 set_page_private(page
, order
);
223 __SetPagePrivate(page
);
226 static inline void rmv_page_order(struct page
*page
)
228 __ClearPagePrivate(page
);
229 set_page_private(page
, 0);
233 * Locate the struct page for both the matching buddy in our
234 * pair (buddy1) and the combined O(n+1) page they form (page).
236 * 1) Any buddy B1 will have an order O twin B2 which satisfies
237 * the following equation:
239 * For example, if the starting buddy (buddy2) is #8 its order
241 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
243 * 2) Any buddy B will have an order O+1 parent P which
244 * satisfies the following equation:
247 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
249 static inline struct page
*
250 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
252 unsigned long buddy_idx
= page_idx
^ (1 << order
);
254 return page
+ (buddy_idx
- page_idx
);
257 static inline unsigned long
258 __find_combined_index(unsigned long page_idx
, unsigned int order
)
260 return (page_idx
& ~(1 << order
));
264 * This function checks whether a page is free && is the buddy
265 * we can do coalesce a page and its buddy if
266 * (a) the buddy is not in a hole &&
267 * (b) the buddy is free &&
268 * (c) the buddy is on the buddy system &&
269 * (d) a page and its buddy have the same order.
270 * for recording page's order, we use page_private(page) and PG_private.
273 static inline int page_is_buddy(struct page
*page
, int order
)
275 #ifdef CONFIG_HOLES_IN_ZONE
276 if (!pfn_valid(page_to_pfn(page
)))
280 if (PagePrivate(page
) &&
281 (page_order(page
) == order
) &&
282 page_count(page
) == 0)
288 * Freeing function for a buddy system allocator.
290 * The concept of a buddy system is to maintain direct-mapped table
291 * (containing bit values) for memory blocks of various "orders".
292 * The bottom level table contains the map for the smallest allocatable
293 * units of memory (here, pages), and each level above it describes
294 * pairs of units from the levels below, hence, "buddies".
295 * At a high level, all that happens here is marking the table entry
296 * at the bottom level available, and propagating the changes upward
297 * as necessary, plus some accounting needed to play nicely with other
298 * parts of the VM system.
299 * At each level, we keep a list of pages, which are heads of continuous
300 * free pages of length of (1 << order) and marked with PG_Private.Page's
301 * order is recorded in page_private(page) field.
302 * So when we are allocating or freeing one, we can derive the state of the
303 * other. That is, if we allocate a small block, and both were
304 * free, the remainder of the region must be split into blocks.
305 * If a block is freed, and its buddy is also free, then this
306 * triggers coalescing into a block of larger size.
311 static inline void __free_pages_bulk (struct page
*page
,
312 struct zone
*zone
, unsigned int order
)
314 unsigned long page_idx
;
315 int order_size
= 1 << order
;
317 if (unlikely(PageCompound(page
)))
318 destroy_compound_page(page
, order
);
320 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
322 BUG_ON(page_idx
& (order_size
- 1));
323 BUG_ON(bad_range(zone
, page
));
325 zone
->free_pages
+= order_size
;
326 while (order
< MAX_ORDER
-1) {
327 unsigned long combined_idx
;
328 struct free_area
*area
;
331 buddy
= __page_find_buddy(page
, page_idx
, order
);
332 if (!page_is_buddy(buddy
, order
))
333 break; /* Move the buddy up one level. */
335 list_del(&buddy
->lru
);
336 area
= zone
->free_area
+ order
;
338 rmv_page_order(buddy
);
339 combined_idx
= __find_combined_index(page_idx
, order
);
340 page
= page
+ (combined_idx
- page_idx
);
341 page_idx
= combined_idx
;
344 set_page_order(page
, order
);
345 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
346 zone
->free_area
[order
].nr_free
++;
349 static inline int free_pages_check(struct page
*page
)
351 if (unlikely(page_mapcount(page
) |
352 (page
->mapping
!= NULL
) |
353 (page_count(page
) != 0) |
363 1 << PG_reserved
))))
366 __ClearPageDirty(page
);
368 * For now, we report if PG_reserved was found set, but do not
369 * clear it, and do not free the page. But we shall soon need
370 * to do more, for when the ZERO_PAGE count wraps negative.
372 return PageReserved(page
);
376 * Frees a list of pages.
377 * Assumes all pages on list are in same zone, and of same order.
378 * count is the number of pages to free.
380 * If the zone was previously in an "all pages pinned" state then look to
381 * see if this freeing clears that state.
383 * And clear the zone's pages_scanned counter, to hold off the "all pages are
384 * pinned" detection logic.
387 free_pages_bulk(struct zone
*zone
, int count
,
388 struct list_head
*list
, unsigned int order
)
390 struct page
*page
= NULL
;
393 spin_lock(&zone
->lock
);
394 zone
->all_unreclaimable
= 0;
395 zone
->pages_scanned
= 0;
396 while (!list_empty(list
) && count
--) {
397 page
= list_entry(list
->prev
, struct page
, lru
);
398 /* have to delete it as __free_pages_bulk list manipulates */
399 list_del(&page
->lru
);
400 __free_pages_bulk(page
, zone
, order
);
403 spin_unlock(&zone
->lock
);
407 void __free_pages_ok(struct page
*page
, unsigned int order
)
414 arch_free_page(page
, order
);
418 for (i
= 1 ; i
< (1 << order
) ; ++i
)
419 __put_page(page
+ i
);
422 for (i
= 0 ; i
< (1 << order
) ; ++i
)
423 reserved
+= free_pages_check(page
+ i
);
427 list_add(&page
->lru
, &list
);
428 kernel_map_pages(page
, 1<<order
, 0);
429 local_irq_save(flags
);
430 __mod_page_state(pgfree
, 1 << order
);
431 free_pages_bulk(page_zone(page
), 1, &list
, order
);
432 local_irq_restore(flags
);
436 * permit the bootmem allocator to evade page validation on high-order frees
438 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
441 __ClearPageReserved(page
);
442 set_page_count(page
, 0);
444 free_hot_cold_page(page
, 0);
449 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
450 struct page
*p
= &page
[loop
];
452 if (loop
+ 16 < BITS_PER_LONG
)
454 __ClearPageReserved(p
);
455 set_page_count(p
, 0);
458 arch_free_page(page
, order
);
460 mod_page_state(pgfree
, 1 << order
);
462 list_add(&page
->lru
, &list
);
463 kernel_map_pages(page
, 1 << order
, 0);
464 free_pages_bulk(page_zone(page
), 1, &list
, order
);
470 * The order of subdivision here is critical for the IO subsystem.
471 * Please do not alter this order without good reasons and regression
472 * testing. Specifically, as large blocks of memory are subdivided,
473 * the order in which smaller blocks are delivered depends on the order
474 * they're subdivided in this function. This is the primary factor
475 * influencing the order in which pages are delivered to the IO
476 * subsystem according to empirical testing, and this is also justified
477 * by considering the behavior of a buddy system containing a single
478 * large block of memory acted on by a series of small allocations.
479 * This behavior is a critical factor in sglist merging's success.
483 static inline void expand(struct zone
*zone
, struct page
*page
,
484 int low
, int high
, struct free_area
*area
)
486 unsigned long size
= 1 << high
;
492 BUG_ON(bad_range(zone
, &page
[size
]));
493 list_add(&page
[size
].lru
, &area
->free_list
);
495 set_page_order(&page
[size
], high
);
500 * This page is about to be returned from the page allocator
502 static int prep_new_page(struct page
*page
, int order
)
504 if (unlikely(page_mapcount(page
) |
505 (page
->mapping
!= NULL
) |
506 (page_count(page
) != 0) |
517 1 << PG_reserved
))))
521 * For now, we report if PG_reserved was found set, but do not
522 * clear it, and do not allocate the page: as a safety net.
524 if (PageReserved(page
))
527 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
528 1 << PG_referenced
| 1 << PG_arch_1
|
529 1 << PG_checked
| 1 << PG_mappedtodisk
);
530 set_page_private(page
, 0);
531 set_page_refs(page
, order
);
532 kernel_map_pages(page
, 1 << order
, 1);
537 * Do the hard work of removing an element from the buddy allocator.
538 * Call me with the zone->lock already held.
540 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
542 struct free_area
* area
;
543 unsigned int current_order
;
546 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
547 area
= zone
->free_area
+ current_order
;
548 if (list_empty(&area
->free_list
))
551 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
552 list_del(&page
->lru
);
553 rmv_page_order(page
);
555 zone
->free_pages
-= 1UL << order
;
556 expand(zone
, page
, order
, current_order
, area
);
564 * Obtain a specified number of elements from the buddy allocator, all under
565 * a single hold of the lock, for efficiency. Add them to the supplied list.
566 * Returns the number of new pages which were placed at *list.
568 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
569 unsigned long count
, struct list_head
*list
)
573 spin_lock(&zone
->lock
);
574 for (i
= 0; i
< count
; ++i
) {
575 struct page
*page
= __rmqueue(zone
, order
);
576 if (unlikely(page
== NULL
))
578 list_add_tail(&page
->lru
, list
);
580 spin_unlock(&zone
->lock
);
585 /* Called from the slab reaper to drain remote pagesets */
586 void drain_remote_pages(void)
592 local_irq_save(flags
);
593 for_each_zone(zone
) {
594 struct per_cpu_pageset
*pset
;
596 /* Do not drain local pagesets */
597 if (zone
->zone_pgdat
->node_id
== numa_node_id())
600 pset
= zone
->pageset
[smp_processor_id()];
601 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
602 struct per_cpu_pages
*pcp
;
606 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
610 local_irq_restore(flags
);
614 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
615 static void __drain_pages(unsigned int cpu
)
621 for_each_zone(zone
) {
622 struct per_cpu_pageset
*pset
;
624 pset
= zone_pcp(zone
, cpu
);
625 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
626 struct per_cpu_pages
*pcp
;
629 local_irq_save(flags
);
630 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
632 local_irq_restore(flags
);
636 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
640 void mark_free_pages(struct zone
*zone
)
642 unsigned long zone_pfn
, flags
;
644 struct list_head
*curr
;
646 if (!zone
->spanned_pages
)
649 spin_lock_irqsave(&zone
->lock
, flags
);
650 for (zone_pfn
= 0; zone_pfn
< zone
->spanned_pages
; ++zone_pfn
)
651 ClearPageNosaveFree(pfn_to_page(zone_pfn
+ zone
->zone_start_pfn
));
653 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
654 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
655 unsigned long start_pfn
, i
;
657 start_pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
659 for (i
=0; i
< (1<<order
); i
++)
660 SetPageNosaveFree(pfn_to_page(start_pfn
+i
));
662 spin_unlock_irqrestore(&zone
->lock
, flags
);
666 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
668 void drain_local_pages(void)
672 local_irq_save(flags
);
673 __drain_pages(smp_processor_id());
674 local_irq_restore(flags
);
676 #endif /* CONFIG_PM */
678 static void zone_statistics(struct zonelist
*zonelist
, struct zone
*z
, int cpu
)
681 pg_data_t
*pg
= z
->zone_pgdat
;
682 pg_data_t
*orig
= zonelist
->zones
[0]->zone_pgdat
;
683 struct per_cpu_pageset
*p
;
685 p
= zone_pcp(z
, cpu
);
690 zone_pcp(zonelist
->zones
[0], cpu
)->numa_foreign
++;
692 if (pg
== NODE_DATA(numa_node_id()))
700 * Free a 0-order page
702 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
704 struct zone
*zone
= page_zone(page
);
705 struct per_cpu_pages
*pcp
;
708 arch_free_page(page
, 0);
711 page
->mapping
= NULL
;
712 if (free_pages_check(page
))
715 kernel_map_pages(page
, 1, 0);
717 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
718 local_irq_save(flags
);
719 __inc_page_state(pgfree
);
720 list_add(&page
->lru
, &pcp
->list
);
722 if (pcp
->count
>= pcp
->high
)
723 pcp
->count
-= free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
724 local_irq_restore(flags
);
728 void fastcall
free_hot_page(struct page
*page
)
730 free_hot_cold_page(page
, 0);
733 void fastcall
free_cold_page(struct page
*page
)
735 free_hot_cold_page(page
, 1);
738 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
742 BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
743 for(i
= 0; i
< (1 << order
); i
++)
744 clear_highpage(page
+ i
);
748 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
749 * we cheat by calling it from here, in the order > 0 path. Saves a branch
752 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
753 struct zone
*zone
, int order
, gfp_t gfp_flags
)
757 int cold
= !!(gfp_flags
& __GFP_COLD
);
763 struct per_cpu_pages
*pcp
;
765 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
766 local_irq_save(flags
);
768 pcp
->count
+= rmqueue_bulk(zone
, 0,
769 pcp
->batch
, &pcp
->list
);
770 if (unlikely(!pcp
->count
))
773 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
774 list_del(&page
->lru
);
777 spin_lock_irqsave(&zone
->lock
, flags
);
778 page
= __rmqueue(zone
, order
);
779 spin_unlock(&zone
->lock
);
784 __mod_page_state_zone(zone
, pgalloc
, 1 << order
);
785 zone_statistics(zonelist
, zone
, cpu
);
786 local_irq_restore(flags
);
789 BUG_ON(bad_range(zone
, page
));
790 if (prep_new_page(page
, order
))
793 if (gfp_flags
& __GFP_ZERO
)
794 prep_zero_page(page
, order
, gfp_flags
);
796 if (order
&& (gfp_flags
& __GFP_COMP
))
797 prep_compound_page(page
, order
);
801 local_irq_restore(flags
);
806 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
807 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
808 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
809 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
810 #define ALLOC_HARDER 0x10 /* try to alloc harder */
811 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
812 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
815 * Return 1 if free pages are above 'mark'. This takes into account the order
818 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
819 int classzone_idx
, int alloc_flags
)
821 /* free_pages my go negative - that's OK */
822 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
825 if (alloc_flags
& ALLOC_HIGH
)
827 if (alloc_flags
& ALLOC_HARDER
)
830 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
832 for (o
= 0; o
< order
; o
++) {
833 /* At the next order, this order's pages become unavailable */
834 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
836 /* Require fewer higher order pages to be free */
839 if (free_pages
<= min
)
846 * get_page_from_freeliest goes through the zonelist trying to allocate
850 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
851 struct zonelist
*zonelist
, int alloc_flags
)
853 struct zone
**z
= zonelist
->zones
;
854 struct page
*page
= NULL
;
855 int classzone_idx
= zone_idx(*z
);
858 * Go through the zonelist once, looking for a zone with enough free.
859 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
862 if ((alloc_flags
& ALLOC_CPUSET
) &&
863 !cpuset_zone_allowed(*z
, gfp_mask
))
866 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
868 if (alloc_flags
& ALLOC_WMARK_MIN
)
869 mark
= (*z
)->pages_min
;
870 else if (alloc_flags
& ALLOC_WMARK_LOW
)
871 mark
= (*z
)->pages_low
;
873 mark
= (*z
)->pages_high
;
874 if (!zone_watermark_ok(*z
, order
, mark
,
875 classzone_idx
, alloc_flags
))
879 page
= buffered_rmqueue(zonelist
, *z
, order
, gfp_mask
);
883 } while (*(++z
) != NULL
);
888 * This is the 'heart' of the zoned buddy allocator.
890 struct page
* fastcall
891 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
892 struct zonelist
*zonelist
)
894 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
897 struct reclaim_state reclaim_state
;
898 struct task_struct
*p
= current
;
901 int did_some_progress
;
903 might_sleep_if(wait
);
906 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
908 if (unlikely(*z
== NULL
)) {
909 /* Should this ever happen?? */
913 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
914 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
919 wakeup_kswapd(*z
, order
);
923 * OK, we're below the kswapd watermark and have kicked background
924 * reclaim. Now things get more complex, so set up alloc_flags according
925 * to how we want to proceed.
927 * The caller may dip into page reserves a bit more if the caller
928 * cannot run direct reclaim, or if the caller has realtime scheduling
931 alloc_flags
= ALLOC_WMARK_MIN
;
932 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
933 alloc_flags
|= ALLOC_HARDER
;
934 if (gfp_mask
& __GFP_HIGH
)
935 alloc_flags
|= ALLOC_HIGH
;
936 alloc_flags
|= ALLOC_CPUSET
;
939 * Go through the zonelist again. Let __GFP_HIGH and allocations
940 * coming from realtime tasks go deeper into reserves.
942 * This is the last chance, in general, before the goto nopage.
943 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
944 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
946 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
950 /* This allocation should allow future memory freeing. */
952 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
953 && !in_interrupt()) {
954 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
956 /* go through the zonelist yet again, ignoring mins */
957 page
= get_page_from_freelist(gfp_mask
, order
,
958 zonelist
, ALLOC_NO_WATERMARKS
);
961 if (gfp_mask
& __GFP_NOFAIL
) {
962 blk_congestion_wait(WRITE
, HZ
/50);
969 /* Atomic allocations - we can't balance anything */
976 /* We now go into synchronous reclaim */
977 p
->flags
|= PF_MEMALLOC
;
978 reclaim_state
.reclaimed_slab
= 0;
979 p
->reclaim_state
= &reclaim_state
;
981 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
983 p
->reclaim_state
= NULL
;
984 p
->flags
&= ~PF_MEMALLOC
;
988 if (likely(did_some_progress
)) {
989 page
= get_page_from_freelist(gfp_mask
, order
,
990 zonelist
, alloc_flags
);
993 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
995 * Go through the zonelist yet one more time, keep
996 * very high watermark here, this is only to catch
997 * a parallel oom killing, we must fail if we're still
998 * under heavy pressure.
1000 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1001 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1005 out_of_memory(gfp_mask
, order
);
1010 * Don't let big-order allocations loop unless the caller explicitly
1011 * requests that. Wait for some write requests to complete then retry.
1013 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1014 * <= 3, but that may not be true in other implementations.
1017 if (!(gfp_mask
& __GFP_NORETRY
)) {
1018 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
1020 if (gfp_mask
& __GFP_NOFAIL
)
1024 blk_congestion_wait(WRITE
, HZ
/50);
1029 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1030 printk(KERN_WARNING
"%s: page allocation failure."
1031 " order:%d, mode:0x%x\n",
1032 p
->comm
, order
, gfp_mask
);
1040 EXPORT_SYMBOL(__alloc_pages
);
1043 * Common helper functions.
1045 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1048 page
= alloc_pages(gfp_mask
, order
);
1051 return (unsigned long) page_address(page
);
1054 EXPORT_SYMBOL(__get_free_pages
);
1056 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1061 * get_zeroed_page() returns a 32-bit address, which cannot represent
1064 BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1066 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1068 return (unsigned long) page_address(page
);
1072 EXPORT_SYMBOL(get_zeroed_page
);
1074 void __pagevec_free(struct pagevec
*pvec
)
1076 int i
= pagevec_count(pvec
);
1079 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1082 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1084 if (put_page_testzero(page
)) {
1086 free_hot_page(page
);
1088 __free_pages_ok(page
, order
);
1092 EXPORT_SYMBOL(__free_pages
);
1094 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1097 BUG_ON(!virt_addr_valid((void *)addr
));
1098 __free_pages(virt_to_page((void *)addr
), order
);
1102 EXPORT_SYMBOL(free_pages
);
1105 * Total amount of free (allocatable) RAM:
1107 unsigned int nr_free_pages(void)
1109 unsigned int sum
= 0;
1113 sum
+= zone
->free_pages
;
1118 EXPORT_SYMBOL(nr_free_pages
);
1121 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1123 unsigned int i
, sum
= 0;
1125 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1126 sum
+= pgdat
->node_zones
[i
].free_pages
;
1132 static unsigned int nr_free_zone_pages(int offset
)
1134 /* Just pick one node, since fallback list is circular */
1135 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1136 unsigned int sum
= 0;
1138 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1139 struct zone
**zonep
= zonelist
->zones
;
1142 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1143 unsigned long size
= zone
->present_pages
;
1144 unsigned long high
= zone
->pages_high
;
1153 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1155 unsigned int nr_free_buffer_pages(void)
1157 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1161 * Amount of free RAM allocatable within all zones
1163 unsigned int nr_free_pagecache_pages(void)
1165 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1168 #ifdef CONFIG_HIGHMEM
1169 unsigned int nr_free_highpages (void)
1172 unsigned int pages
= 0;
1174 for_each_pgdat(pgdat
)
1175 pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1182 static void show_node(struct zone
*zone
)
1184 printk("Node %d ", zone
->zone_pgdat
->node_id
);
1187 #define show_node(zone) do { } while (0)
1191 * Accumulate the page_state information across all CPUs.
1192 * The result is unavoidably approximate - it can change
1193 * during and after execution of this function.
1195 static DEFINE_PER_CPU(struct page_state
, page_states
) = {0};
1197 atomic_t nr_pagecache
= ATOMIC_INIT(0);
1198 EXPORT_SYMBOL(nr_pagecache
);
1200 DEFINE_PER_CPU(long, nr_pagecache_local
) = 0;
1203 static void __get_page_state(struct page_state
*ret
, int nr
, cpumask_t
*cpumask
)
1207 memset(ret
, 0, sizeof(*ret
));
1208 cpus_and(*cpumask
, *cpumask
, cpu_online_map
);
1210 cpu
= first_cpu(*cpumask
);
1211 while (cpu
< NR_CPUS
) {
1212 unsigned long *in
, *out
, off
;
1214 in
= (unsigned long *)&per_cpu(page_states
, cpu
);
1216 cpu
= next_cpu(cpu
, *cpumask
);
1219 prefetch(&per_cpu(page_states
, cpu
));
1221 out
= (unsigned long *)ret
;
1222 for (off
= 0; off
< nr
; off
++)
1227 void get_page_state_node(struct page_state
*ret
, int node
)
1230 cpumask_t mask
= node_to_cpumask(node
);
1232 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1233 nr
/= sizeof(unsigned long);
1235 __get_page_state(ret
, nr
+1, &mask
);
1238 void get_page_state(struct page_state
*ret
)
1241 cpumask_t mask
= CPU_MASK_ALL
;
1243 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1244 nr
/= sizeof(unsigned long);
1246 __get_page_state(ret
, nr
+ 1, &mask
);
1249 void get_full_page_state(struct page_state
*ret
)
1251 cpumask_t mask
= CPU_MASK_ALL
;
1253 __get_page_state(ret
, sizeof(*ret
) / sizeof(unsigned long), &mask
);
1256 unsigned long read_page_state_offset(unsigned long offset
)
1258 unsigned long ret
= 0;
1261 for_each_online_cpu(cpu
) {
1264 in
= (unsigned long)&per_cpu(page_states
, cpu
) + offset
;
1265 ret
+= *((unsigned long *)in
);
1270 void __mod_page_state_offset(unsigned long offset
, unsigned long delta
)
1274 ptr
= &__get_cpu_var(page_states
);
1275 *(unsigned long *)(ptr
+ offset
) += delta
;
1277 EXPORT_SYMBOL(__mod_page_state_offset
);
1279 void mod_page_state_offset(unsigned long offset
, unsigned long delta
)
1281 unsigned long flags
;
1284 local_irq_save(flags
);
1285 ptr
= &__get_cpu_var(page_states
);
1286 *(unsigned long *)(ptr
+ offset
) += delta
;
1287 local_irq_restore(flags
);
1289 EXPORT_SYMBOL(mod_page_state_offset
);
1291 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
1292 unsigned long *free
, struct pglist_data
*pgdat
)
1294 struct zone
*zones
= pgdat
->node_zones
;
1300 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1301 *active
+= zones
[i
].nr_active
;
1302 *inactive
+= zones
[i
].nr_inactive
;
1303 *free
+= zones
[i
].free_pages
;
1307 void get_zone_counts(unsigned long *active
,
1308 unsigned long *inactive
, unsigned long *free
)
1310 struct pglist_data
*pgdat
;
1315 for_each_pgdat(pgdat
) {
1316 unsigned long l
, m
, n
;
1317 __get_zone_counts(&l
, &m
, &n
, pgdat
);
1324 void si_meminfo(struct sysinfo
*val
)
1326 val
->totalram
= totalram_pages
;
1328 val
->freeram
= nr_free_pages();
1329 val
->bufferram
= nr_blockdev_pages();
1330 #ifdef CONFIG_HIGHMEM
1331 val
->totalhigh
= totalhigh_pages
;
1332 val
->freehigh
= nr_free_highpages();
1337 val
->mem_unit
= PAGE_SIZE
;
1340 EXPORT_SYMBOL(si_meminfo
);
1343 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1345 pg_data_t
*pgdat
= NODE_DATA(nid
);
1347 val
->totalram
= pgdat
->node_present_pages
;
1348 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1349 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1350 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1351 val
->mem_unit
= PAGE_SIZE
;
1355 #define K(x) ((x) << (PAGE_SHIFT-10))
1358 * Show free area list (used inside shift_scroll-lock stuff)
1359 * We also calculate the percentage fragmentation. We do this by counting the
1360 * memory on each free list with the exception of the first item on the list.
1362 void show_free_areas(void)
1364 struct page_state ps
;
1365 int cpu
, temperature
;
1366 unsigned long active
;
1367 unsigned long inactive
;
1371 for_each_zone(zone
) {
1373 printk("%s per-cpu:", zone
->name
);
1375 if (!populated_zone(zone
)) {
1381 for_each_online_cpu(cpu
) {
1382 struct per_cpu_pageset
*pageset
;
1384 pageset
= zone_pcp(zone
, cpu
);
1386 for (temperature
= 0; temperature
< 2; temperature
++)
1387 printk("cpu %d %s: high %d, batch %d used:%d\n",
1389 temperature
? "cold" : "hot",
1390 pageset
->pcp
[temperature
].high
,
1391 pageset
->pcp
[temperature
].batch
,
1392 pageset
->pcp
[temperature
].count
);
1396 get_page_state(&ps
);
1397 get_zone_counts(&active
, &inactive
, &free
);
1399 printk("Free pages: %11ukB (%ukB HighMem)\n",
1401 K(nr_free_highpages()));
1403 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1404 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1413 ps
.nr_page_table_pages
);
1415 for_each_zone(zone
) {
1427 " pages_scanned:%lu"
1428 " all_unreclaimable? %s"
1431 K(zone
->free_pages
),
1434 K(zone
->pages_high
),
1436 K(zone
->nr_inactive
),
1437 K(zone
->present_pages
),
1438 zone
->pages_scanned
,
1439 (zone
->all_unreclaimable
? "yes" : "no")
1441 printk("lowmem_reserve[]:");
1442 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1443 printk(" %lu", zone
->lowmem_reserve
[i
]);
1447 for_each_zone(zone
) {
1448 unsigned long nr
, flags
, order
, total
= 0;
1451 printk("%s: ", zone
->name
);
1452 if (!populated_zone(zone
)) {
1457 spin_lock_irqsave(&zone
->lock
, flags
);
1458 for (order
= 0; order
< MAX_ORDER
; order
++) {
1459 nr
= zone
->free_area
[order
].nr_free
;
1460 total
+= nr
<< order
;
1461 printk("%lu*%lukB ", nr
, K(1UL) << order
);
1463 spin_unlock_irqrestore(&zone
->lock
, flags
);
1464 printk("= %lukB\n", K(total
));
1467 show_swap_cache_info();
1471 * Builds allocation fallback zone lists.
1473 * Add all populated zones of a node to the zonelist.
1475 static int __init
build_zonelists_node(pg_data_t
*pgdat
,
1476 struct zonelist
*zonelist
, int nr_zones
, int zone_type
)
1480 BUG_ON(zone_type
> ZONE_HIGHMEM
);
1483 zone
= pgdat
->node_zones
+ zone_type
;
1484 if (populated_zone(zone
)) {
1485 #ifndef CONFIG_HIGHMEM
1486 BUG_ON(zone_type
> ZONE_NORMAL
);
1488 zonelist
->zones
[nr_zones
++] = zone
;
1489 check_highest_zone(zone_type
);
1493 } while (zone_type
>= 0);
1497 static inline int highest_zone(int zone_bits
)
1499 int res
= ZONE_NORMAL
;
1500 if (zone_bits
& (__force
int)__GFP_HIGHMEM
)
1502 if (zone_bits
& (__force
int)__GFP_DMA32
)
1504 if (zone_bits
& (__force
int)__GFP_DMA
)
1510 #define MAX_NODE_LOAD (num_online_nodes())
1511 static int __initdata node_load
[MAX_NUMNODES
];
1513 * find_next_best_node - find the next node that should appear in a given node's fallback list
1514 * @node: node whose fallback list we're appending
1515 * @used_node_mask: nodemask_t of already used nodes
1517 * We use a number of factors to determine which is the next node that should
1518 * appear on a given node's fallback list. The node should not have appeared
1519 * already in @node's fallback list, and it should be the next closest node
1520 * according to the distance array (which contains arbitrary distance values
1521 * from each node to each node in the system), and should also prefer nodes
1522 * with no CPUs, since presumably they'll have very little allocation pressure
1523 * on them otherwise.
1524 * It returns -1 if no node is found.
1526 static int __init
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1529 int min_val
= INT_MAX
;
1532 for_each_online_node(i
) {
1535 /* Start from local node */
1536 n
= (node
+i
) % num_online_nodes();
1538 /* Don't want a node to appear more than once */
1539 if (node_isset(n
, *used_node_mask
))
1542 /* Use the local node if we haven't already */
1543 if (!node_isset(node
, *used_node_mask
)) {
1548 /* Use the distance array to find the distance */
1549 val
= node_distance(node
, n
);
1551 /* Give preference to headless and unused nodes */
1552 tmp
= node_to_cpumask(n
);
1553 if (!cpus_empty(tmp
))
1554 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1556 /* Slight preference for less loaded node */
1557 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1558 val
+= node_load
[n
];
1560 if (val
< min_val
) {
1567 node_set(best_node
, *used_node_mask
);
1572 static void __init
build_zonelists(pg_data_t
*pgdat
)
1574 int i
, j
, k
, node
, local_node
;
1575 int prev_node
, load
;
1576 struct zonelist
*zonelist
;
1577 nodemask_t used_mask
;
1579 /* initialize zonelists */
1580 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1581 zonelist
= pgdat
->node_zonelists
+ i
;
1582 zonelist
->zones
[0] = NULL
;
1585 /* NUMA-aware ordering of nodes */
1586 local_node
= pgdat
->node_id
;
1587 load
= num_online_nodes();
1588 prev_node
= local_node
;
1589 nodes_clear(used_mask
);
1590 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1592 * We don't want to pressure a particular node.
1593 * So adding penalty to the first node in same
1594 * distance group to make it round-robin.
1596 if (node_distance(local_node
, node
) !=
1597 node_distance(local_node
, prev_node
))
1598 node_load
[node
] += load
;
1601 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1602 zonelist
= pgdat
->node_zonelists
+ i
;
1603 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1605 k
= highest_zone(i
);
1607 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1608 zonelist
->zones
[j
] = NULL
;
1613 #else /* CONFIG_NUMA */
1615 static void __init
build_zonelists(pg_data_t
*pgdat
)
1617 int i
, j
, k
, node
, local_node
;
1619 local_node
= pgdat
->node_id
;
1620 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1621 struct zonelist
*zonelist
;
1623 zonelist
= pgdat
->node_zonelists
+ i
;
1626 k
= highest_zone(i
);
1627 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1629 * Now we build the zonelist so that it contains the zones
1630 * of all the other nodes.
1631 * We don't want to pressure a particular node, so when
1632 * building the zones for node N, we make sure that the
1633 * zones coming right after the local ones are those from
1634 * node N+1 (modulo N)
1636 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1637 if (!node_online(node
))
1639 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1641 for (node
= 0; node
< local_node
; node
++) {
1642 if (!node_online(node
))
1644 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1647 zonelist
->zones
[j
] = NULL
;
1651 #endif /* CONFIG_NUMA */
1653 void __init
build_all_zonelists(void)
1657 for_each_online_node(i
)
1658 build_zonelists(NODE_DATA(i
));
1659 printk("Built %i zonelists\n", num_online_nodes());
1660 cpuset_init_current_mems_allowed();
1664 * Helper functions to size the waitqueue hash table.
1665 * Essentially these want to choose hash table sizes sufficiently
1666 * large so that collisions trying to wait on pages are rare.
1667 * But in fact, the number of active page waitqueues on typical
1668 * systems is ridiculously low, less than 200. So this is even
1669 * conservative, even though it seems large.
1671 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1672 * waitqueues, i.e. the size of the waitq table given the number of pages.
1674 #define PAGES_PER_WAITQUEUE 256
1676 static inline unsigned long wait_table_size(unsigned long pages
)
1678 unsigned long size
= 1;
1680 pages
/= PAGES_PER_WAITQUEUE
;
1682 while (size
< pages
)
1686 * Once we have dozens or even hundreds of threads sleeping
1687 * on IO we've got bigger problems than wait queue collision.
1688 * Limit the size of the wait table to a reasonable size.
1690 size
= min(size
, 4096UL);
1692 return max(size
, 4UL);
1696 * This is an integer logarithm so that shifts can be used later
1697 * to extract the more random high bits from the multiplicative
1698 * hash function before the remainder is taken.
1700 static inline unsigned long wait_table_bits(unsigned long size
)
1705 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1707 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1708 unsigned long *zones_size
, unsigned long *zholes_size
)
1710 unsigned long realtotalpages
, totalpages
= 0;
1713 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1714 totalpages
+= zones_size
[i
];
1715 pgdat
->node_spanned_pages
= totalpages
;
1717 realtotalpages
= totalpages
;
1719 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1720 realtotalpages
-= zholes_size
[i
];
1721 pgdat
->node_present_pages
= realtotalpages
;
1722 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1727 * Initially all pages are reserved - free ones are freed
1728 * up by free_all_bootmem() once the early boot process is
1729 * done. Non-atomic initialization, single-pass.
1731 void __devinit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1732 unsigned long start_pfn
)
1735 unsigned long end_pfn
= start_pfn
+ size
;
1738 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++, page
++) {
1739 if (!early_pfn_valid(pfn
))
1741 page
= pfn_to_page(pfn
);
1742 set_page_links(page
, zone
, nid
, pfn
);
1743 set_page_count(page
, 1);
1744 reset_page_mapcount(page
);
1745 SetPageReserved(page
);
1746 INIT_LIST_HEAD(&page
->lru
);
1747 #ifdef WANT_PAGE_VIRTUAL
1748 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1749 if (!is_highmem_idx(zone
))
1750 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1755 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1759 for (order
= 0; order
< MAX_ORDER
; order
++) {
1760 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1761 zone
->free_area
[order
].nr_free
= 0;
1765 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1766 void zonetable_add(struct zone
*zone
, int nid
, int zid
, unsigned long pfn
,
1769 unsigned long snum
= pfn_to_section_nr(pfn
);
1770 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1773 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1775 for (; snum
<= end
; snum
++)
1776 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1779 #ifndef __HAVE_ARCH_MEMMAP_INIT
1780 #define memmap_init(size, nid, zone, start_pfn) \
1781 memmap_init_zone((size), (nid), (zone), (start_pfn))
1784 static int __devinit
zone_batchsize(struct zone
*zone
)
1789 * The per-cpu-pages pools are set to around 1000th of the
1790 * size of the zone. But no more than 1/2 of a meg.
1792 * OK, so we don't know how big the cache is. So guess.
1794 batch
= zone
->present_pages
/ 1024;
1795 if (batch
* PAGE_SIZE
> 512 * 1024)
1796 batch
= (512 * 1024) / PAGE_SIZE
;
1797 batch
/= 4; /* We effectively *= 4 below */
1802 * Clamp the batch to a 2^n - 1 value. Having a power
1803 * of 2 value was found to be more likely to have
1804 * suboptimal cache aliasing properties in some cases.
1806 * For example if 2 tasks are alternately allocating
1807 * batches of pages, one task can end up with a lot
1808 * of pages of one half of the possible page colors
1809 * and the other with pages of the other colors.
1811 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1816 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1818 struct per_cpu_pages
*pcp
;
1820 memset(p
, 0, sizeof(*p
));
1822 pcp
= &p
->pcp
[0]; /* hot */
1824 pcp
->high
= 6 * batch
;
1825 pcp
->batch
= max(1UL, 1 * batch
);
1826 INIT_LIST_HEAD(&pcp
->list
);
1828 pcp
= &p
->pcp
[1]; /* cold*/
1830 pcp
->high
= 2 * batch
;
1831 pcp
->batch
= max(1UL, batch
/2);
1832 INIT_LIST_HEAD(&pcp
->list
);
1836 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1837 * to the value high for the pageset p.
1840 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
1843 struct per_cpu_pages
*pcp
;
1845 pcp
= &p
->pcp
[0]; /* hot list */
1847 pcp
->batch
= max(1UL, high
/4);
1848 if ((high
/4) > (PAGE_SHIFT
* 8))
1849 pcp
->batch
= PAGE_SHIFT
* 8;
1855 * Boot pageset table. One per cpu which is going to be used for all
1856 * zones and all nodes. The parameters will be set in such a way
1857 * that an item put on a list will immediately be handed over to
1858 * the buddy list. This is safe since pageset manipulation is done
1859 * with interrupts disabled.
1861 * Some NUMA counter updates may also be caught by the boot pagesets.
1863 * The boot_pagesets must be kept even after bootup is complete for
1864 * unused processors and/or zones. They do play a role for bootstrapping
1865 * hotplugged processors.
1867 * zoneinfo_show() and maybe other functions do
1868 * not check if the processor is online before following the pageset pointer.
1869 * Other parts of the kernel may not check if the zone is available.
1871 static struct per_cpu_pageset
1872 boot_pageset
[NR_CPUS
];
1875 * Dynamically allocate memory for the
1876 * per cpu pageset array in struct zone.
1878 static int __devinit
process_zones(int cpu
)
1880 struct zone
*zone
, *dzone
;
1882 for_each_zone(zone
) {
1884 zone
->pageset
[cpu
] = kmalloc_node(sizeof(struct per_cpu_pageset
),
1885 GFP_KERNEL
, cpu_to_node(cpu
));
1886 if (!zone
->pageset
[cpu
])
1889 setup_pageset(zone
->pageset
[cpu
], zone_batchsize(zone
));
1891 if (percpu_pagelist_fraction
)
1892 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
1893 (zone
->present_pages
/ percpu_pagelist_fraction
));
1898 for_each_zone(dzone
) {
1901 kfree(dzone
->pageset
[cpu
]);
1902 dzone
->pageset
[cpu
] = NULL
;
1907 static inline void free_zone_pagesets(int cpu
)
1911 for_each_zone(zone
) {
1912 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1914 zone_pcp(zone
, cpu
) = NULL
;
1919 static int __devinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1920 unsigned long action
,
1923 int cpu
= (long)hcpu
;
1924 int ret
= NOTIFY_OK
;
1927 case CPU_UP_PREPARE
:
1928 if (process_zones(cpu
))
1931 case CPU_UP_CANCELED
:
1933 free_zone_pagesets(cpu
);
1941 static struct notifier_block pageset_notifier
=
1942 { &pageset_cpuup_callback
, NULL
, 0 };
1944 void __init
setup_per_cpu_pageset(void)
1948 /* Initialize per_cpu_pageset for cpu 0.
1949 * A cpuup callback will do this for every cpu
1950 * as it comes online
1952 err
= process_zones(smp_processor_id());
1954 register_cpu_notifier(&pageset_notifier
);
1960 void zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1963 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1966 * The per-page waitqueue mechanism uses hashed waitqueues
1969 zone
->wait_table_size
= wait_table_size(zone_size_pages
);
1970 zone
->wait_table_bits
= wait_table_bits(zone
->wait_table_size
);
1971 zone
->wait_table
= (wait_queue_head_t
*)
1972 alloc_bootmem_node(pgdat
, zone
->wait_table_size
1973 * sizeof(wait_queue_head_t
));
1975 for(i
= 0; i
< zone
->wait_table_size
; ++i
)
1976 init_waitqueue_head(zone
->wait_table
+ i
);
1979 static __devinit
void zone_pcp_init(struct zone
*zone
)
1982 unsigned long batch
= zone_batchsize(zone
);
1984 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1986 /* Early boot. Slab allocator not functional yet */
1987 zone
->pageset
[cpu
] = &boot_pageset
[cpu
];
1988 setup_pageset(&boot_pageset
[cpu
],0);
1990 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1993 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1994 zone
->name
, zone
->present_pages
, batch
);
1997 static __devinit
void init_currently_empty_zone(struct zone
*zone
,
1998 unsigned long zone_start_pfn
, unsigned long size
)
2000 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2002 zone_wait_table_init(zone
, size
);
2003 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2005 zone
->zone_mem_map
= pfn_to_page(zone_start_pfn
);
2006 zone
->zone_start_pfn
= zone_start_pfn
;
2008 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
2010 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2014 * Set up the zone data structures:
2015 * - mark all pages reserved
2016 * - mark all memory queues empty
2017 * - clear the memory bitmaps
2019 static void __init
free_area_init_core(struct pglist_data
*pgdat
,
2020 unsigned long *zones_size
, unsigned long *zholes_size
)
2023 int nid
= pgdat
->node_id
;
2024 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
2026 pgdat_resize_init(pgdat
);
2027 pgdat
->nr_zones
= 0;
2028 init_waitqueue_head(&pgdat
->kswapd_wait
);
2029 pgdat
->kswapd_max_order
= 0;
2031 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2032 struct zone
*zone
= pgdat
->node_zones
+ j
;
2033 unsigned long size
, realsize
;
2035 realsize
= size
= zones_size
[j
];
2037 realsize
-= zholes_size
[j
];
2039 if (j
< ZONE_HIGHMEM
)
2040 nr_kernel_pages
+= realsize
;
2041 nr_all_pages
+= realsize
;
2043 zone
->spanned_pages
= size
;
2044 zone
->present_pages
= realsize
;
2045 zone
->name
= zone_names
[j
];
2046 spin_lock_init(&zone
->lock
);
2047 spin_lock_init(&zone
->lru_lock
);
2048 zone_seqlock_init(zone
);
2049 zone
->zone_pgdat
= pgdat
;
2050 zone
->free_pages
= 0;
2052 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
2054 zone_pcp_init(zone
);
2055 INIT_LIST_HEAD(&zone
->active_list
);
2056 INIT_LIST_HEAD(&zone
->inactive_list
);
2057 zone
->nr_scan_active
= 0;
2058 zone
->nr_scan_inactive
= 0;
2059 zone
->nr_active
= 0;
2060 zone
->nr_inactive
= 0;
2061 atomic_set(&zone
->reclaim_in_progress
, 0);
2065 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2066 init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2067 zone_start_pfn
+= size
;
2071 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2073 /* Skip empty nodes */
2074 if (!pgdat
->node_spanned_pages
)
2077 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2078 /* ia64 gets its own node_mem_map, before this, without bootmem */
2079 if (!pgdat
->node_mem_map
) {
2083 size
= (pgdat
->node_spanned_pages
+ 1) * sizeof(struct page
);
2084 map
= alloc_remap(pgdat
->node_id
, size
);
2086 map
= alloc_bootmem_node(pgdat
, size
);
2087 pgdat
->node_mem_map
= map
;
2089 #ifdef CONFIG_FLATMEM
2091 * With no DISCONTIG, the global mem_map is just set as node 0's
2093 if (pgdat
== NODE_DATA(0))
2094 mem_map
= NODE_DATA(0)->node_mem_map
;
2096 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2099 void __init
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2100 unsigned long *zones_size
, unsigned long node_start_pfn
,
2101 unsigned long *zholes_size
)
2103 pgdat
->node_id
= nid
;
2104 pgdat
->node_start_pfn
= node_start_pfn
;
2105 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
2107 alloc_node_mem_map(pgdat
);
2109 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2112 #ifndef CONFIG_NEED_MULTIPLE_NODES
2113 static bootmem_data_t contig_bootmem_data
;
2114 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2116 EXPORT_SYMBOL(contig_page_data
);
2119 void __init
free_area_init(unsigned long *zones_size
)
2121 free_area_init_node(0, NODE_DATA(0), zones_size
,
2122 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2125 #ifdef CONFIG_PROC_FS
2127 #include <linux/seq_file.h>
2129 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
2134 for (pgdat
= pgdat_list
; pgdat
&& node
; pgdat
= pgdat
->pgdat_next
)
2140 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2142 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2145 return pgdat
->pgdat_next
;
2148 static void frag_stop(struct seq_file
*m
, void *arg
)
2153 * This walks the free areas for each zone.
2155 static int frag_show(struct seq_file
*m
, void *arg
)
2157 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2159 struct zone
*node_zones
= pgdat
->node_zones
;
2160 unsigned long flags
;
2163 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
2164 if (!populated_zone(zone
))
2167 spin_lock_irqsave(&zone
->lock
, flags
);
2168 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
2169 for (order
= 0; order
< MAX_ORDER
; ++order
)
2170 seq_printf(m
, "%6lu ", zone
->free_area
[order
].nr_free
);
2171 spin_unlock_irqrestore(&zone
->lock
, flags
);
2177 struct seq_operations fragmentation_op
= {
2178 .start
= frag_start
,
2185 * Output information about zones in @pgdat.
2187 static int zoneinfo_show(struct seq_file
*m
, void *arg
)
2189 pg_data_t
*pgdat
= arg
;
2191 struct zone
*node_zones
= pgdat
->node_zones
;
2192 unsigned long flags
;
2194 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; zone
++) {
2197 if (!populated_zone(zone
))
2200 spin_lock_irqsave(&zone
->lock
, flags
);
2201 seq_printf(m
, "Node %d, zone %8s", pgdat
->node_id
, zone
->name
);
2209 "\n scanned %lu (a: %lu i: %lu)"
2218 zone
->pages_scanned
,
2219 zone
->nr_scan_active
, zone
->nr_scan_inactive
,
2220 zone
->spanned_pages
,
2221 zone
->present_pages
);
2223 "\n protection: (%lu",
2224 zone
->lowmem_reserve
[0]);
2225 for (i
= 1; i
< ARRAY_SIZE(zone
->lowmem_reserve
); i
++)
2226 seq_printf(m
, ", %lu", zone
->lowmem_reserve
[i
]);
2230 for (i
= 0; i
< ARRAY_SIZE(zone
->pageset
); i
++) {
2231 struct per_cpu_pageset
*pageset
;
2234 pageset
= zone_pcp(zone
, i
);
2235 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2236 if (pageset
->pcp
[j
].count
)
2239 if (j
== ARRAY_SIZE(pageset
->pcp
))
2241 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2243 "\n cpu: %i pcp: %i"
2248 pageset
->pcp
[j
].count
,
2249 pageset
->pcp
[j
].high
,
2250 pageset
->pcp
[j
].batch
);
2256 "\n numa_foreign: %lu"
2257 "\n interleave_hit: %lu"
2258 "\n local_node: %lu"
2259 "\n other_node: %lu",
2262 pageset
->numa_foreign
,
2263 pageset
->interleave_hit
,
2264 pageset
->local_node
,
2265 pageset
->other_node
);
2269 "\n all_unreclaimable: %u"
2270 "\n prev_priority: %i"
2271 "\n temp_priority: %i"
2272 "\n start_pfn: %lu",
2273 zone
->all_unreclaimable
,
2274 zone
->prev_priority
,
2275 zone
->temp_priority
,
2276 zone
->zone_start_pfn
);
2277 spin_unlock_irqrestore(&zone
->lock
, flags
);
2283 struct seq_operations zoneinfo_op
= {
2284 .start
= frag_start
, /* iterate over all zones. The same as in
2288 .show
= zoneinfo_show
,
2291 static char *vmstat_text
[] = {
2295 "nr_page_table_pages",
2326 "pgscan_kswapd_high",
2327 "pgscan_kswapd_normal",
2328 "pgscan_kswapd_dma32",
2329 "pgscan_kswapd_dma",
2331 "pgscan_direct_high",
2332 "pgscan_direct_normal",
2333 "pgscan_direct_dma32",
2334 "pgscan_direct_dma",
2339 "kswapd_inodesteal",
2347 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
2349 struct page_state
*ps
;
2351 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2354 ps
= kmalloc(sizeof(*ps
), GFP_KERNEL
);
2357 return ERR_PTR(-ENOMEM
);
2358 get_full_page_state(ps
);
2359 ps
->pgpgin
/= 2; /* sectors -> kbytes */
2361 return (unsigned long *)ps
+ *pos
;
2364 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2367 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2369 return (unsigned long *)m
->private + *pos
;
2372 static int vmstat_show(struct seq_file
*m
, void *arg
)
2374 unsigned long *l
= arg
;
2375 unsigned long off
= l
- (unsigned long *)m
->private;
2377 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
2381 static void vmstat_stop(struct seq_file
*m
, void *arg
)
2387 struct seq_operations vmstat_op
= {
2388 .start
= vmstat_start
,
2389 .next
= vmstat_next
,
2390 .stop
= vmstat_stop
,
2391 .show
= vmstat_show
,
2394 #endif /* CONFIG_PROC_FS */
2396 #ifdef CONFIG_HOTPLUG_CPU
2397 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2398 unsigned long action
, void *hcpu
)
2400 int cpu
= (unsigned long)hcpu
;
2402 unsigned long *src
, *dest
;
2404 if (action
== CPU_DEAD
) {
2407 /* Drain local pagecache count. */
2408 count
= &per_cpu(nr_pagecache_local
, cpu
);
2409 atomic_add(*count
, &nr_pagecache
);
2411 local_irq_disable();
2414 /* Add dead cpu's page_states to our own. */
2415 dest
= (unsigned long *)&__get_cpu_var(page_states
);
2416 src
= (unsigned long *)&per_cpu(page_states
, cpu
);
2418 for (i
= 0; i
< sizeof(struct page_state
)/sizeof(unsigned long);
2428 #endif /* CONFIG_HOTPLUG_CPU */
2430 void __init
page_alloc_init(void)
2432 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2436 * setup_per_zone_lowmem_reserve - called whenever
2437 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2438 * has a correct pages reserved value, so an adequate number of
2439 * pages are left in the zone after a successful __alloc_pages().
2441 static void setup_per_zone_lowmem_reserve(void)
2443 struct pglist_data
*pgdat
;
2446 for_each_pgdat(pgdat
) {
2447 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2448 struct zone
*zone
= pgdat
->node_zones
+ j
;
2449 unsigned long present_pages
= zone
->present_pages
;
2451 zone
->lowmem_reserve
[j
] = 0;
2453 for (idx
= j
-1; idx
>= 0; idx
--) {
2454 struct zone
*lower_zone
;
2456 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2457 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2459 lower_zone
= pgdat
->node_zones
+ idx
;
2460 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2461 sysctl_lowmem_reserve_ratio
[idx
];
2462 present_pages
+= lower_zone
->present_pages
;
2469 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2470 * that the pages_{min,low,high} values for each zone are set correctly
2471 * with respect to min_free_kbytes.
2473 void setup_per_zone_pages_min(void)
2475 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2476 unsigned long lowmem_pages
= 0;
2478 unsigned long flags
;
2480 /* Calculate total number of !ZONE_HIGHMEM pages */
2481 for_each_zone(zone
) {
2482 if (!is_highmem(zone
))
2483 lowmem_pages
+= zone
->present_pages
;
2486 for_each_zone(zone
) {
2488 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2489 tmp
= (pages_min
* zone
->present_pages
) / lowmem_pages
;
2490 if (is_highmem(zone
)) {
2492 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2493 * need highmem pages, so cap pages_min to a small
2496 * The (pages_high-pages_low) and (pages_low-pages_min)
2497 * deltas controls asynch page reclaim, and so should
2498 * not be capped for highmem.
2502 min_pages
= zone
->present_pages
/ 1024;
2503 if (min_pages
< SWAP_CLUSTER_MAX
)
2504 min_pages
= SWAP_CLUSTER_MAX
;
2505 if (min_pages
> 128)
2507 zone
->pages_min
= min_pages
;
2510 * If it's a lowmem zone, reserve a number of pages
2511 * proportionate to the zone's size.
2513 zone
->pages_min
= tmp
;
2516 zone
->pages_low
= zone
->pages_min
+ tmp
/ 4;
2517 zone
->pages_high
= zone
->pages_min
+ tmp
/ 2;
2518 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2523 * Initialise min_free_kbytes.
2525 * For small machines we want it small (128k min). For large machines
2526 * we want it large (64MB max). But it is not linear, because network
2527 * bandwidth does not increase linearly with machine size. We use
2529 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2530 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2546 static int __init
init_per_zone_pages_min(void)
2548 unsigned long lowmem_kbytes
;
2550 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2552 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2553 if (min_free_kbytes
< 128)
2554 min_free_kbytes
= 128;
2555 if (min_free_kbytes
> 65536)
2556 min_free_kbytes
= 65536;
2557 setup_per_zone_pages_min();
2558 setup_per_zone_lowmem_reserve();
2561 module_init(init_per_zone_pages_min
)
2564 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2565 * that we can call two helper functions whenever min_free_kbytes
2568 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2569 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2571 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2572 setup_per_zone_pages_min();
2577 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2578 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2579 * whenever sysctl_lowmem_reserve_ratio changes.
2581 * The reserve ratio obviously has absolutely no relation with the
2582 * pages_min watermarks. The lowmem reserve ratio can only make sense
2583 * if in function of the boot time zone sizes.
2585 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2586 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2588 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2589 setup_per_zone_lowmem_reserve();
2594 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2595 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2596 * can have before it gets flushed back to buddy allocator.
2599 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
2600 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2606 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2607 if (!write
|| (ret
== -EINVAL
))
2609 for_each_zone(zone
) {
2610 for_each_online_cpu(cpu
) {
2612 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
2613 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
2619 __initdata
int hashdist
= HASHDIST_DEFAULT
;
2622 static int __init
set_hashdist(char *str
)
2626 hashdist
= simple_strtoul(str
, &str
, 0);
2629 __setup("hashdist=", set_hashdist
);
2633 * allocate a large system hash table from bootmem
2634 * - it is assumed that the hash table must contain an exact power-of-2
2635 * quantity of entries
2636 * - limit is the number of hash buckets, not the total allocation size
2638 void *__init
alloc_large_system_hash(const char *tablename
,
2639 unsigned long bucketsize
,
2640 unsigned long numentries
,
2643 unsigned int *_hash_shift
,
2644 unsigned int *_hash_mask
,
2645 unsigned long limit
)
2647 unsigned long long max
= limit
;
2648 unsigned long log2qty
, size
;
2651 /* allow the kernel cmdline to have a say */
2653 /* round applicable memory size up to nearest megabyte */
2654 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2655 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2656 numentries
>>= 20 - PAGE_SHIFT
;
2657 numentries
<<= 20 - PAGE_SHIFT
;
2659 /* limit to 1 bucket per 2^scale bytes of low memory */
2660 if (scale
> PAGE_SHIFT
)
2661 numentries
>>= (scale
- PAGE_SHIFT
);
2663 numentries
<<= (PAGE_SHIFT
- scale
);
2665 /* rounded up to nearest power of 2 in size */
2666 numentries
= 1UL << (long_log2(numentries
) + 1);
2668 /* limit allocation size to 1/16 total memory by default */
2670 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2671 do_div(max
, bucketsize
);
2674 if (numentries
> max
)
2677 log2qty
= long_log2(numentries
);
2680 size
= bucketsize
<< log2qty
;
2681 if (flags
& HASH_EARLY
)
2682 table
= alloc_bootmem(size
);
2684 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
2686 unsigned long order
;
2687 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
2689 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
2691 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2694 panic("Failed to allocate %s hash table\n", tablename
);
2696 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2699 long_log2(size
) - PAGE_SHIFT
,
2703 *_hash_shift
= log2qty
;
2705 *_hash_mask
= (1 << log2qty
) - 1;