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>
40 #include <asm/tlbflush.h>
44 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
47 nodemask_t node_online_map __read_mostly
= { { [0] = 1UL } };
48 EXPORT_SYMBOL(node_online_map
);
49 nodemask_t node_possible_map __read_mostly
= NODE_MASK_ALL
;
50 EXPORT_SYMBOL(node_possible_map
);
51 struct pglist_data
*pgdat_list __read_mostly
;
52 unsigned long totalram_pages __read_mostly
;
53 unsigned long totalhigh_pages __read_mostly
;
56 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
);
59 * results with 256, 32 in the lowmem_reserve sysctl:
60 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
61 * 1G machine -> (16M dma, 784M normal, 224M high)
62 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
63 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
64 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
66 * TBD: should special case ZONE_DMA32 machines here - in those we normally
67 * don't need any ZONE_NORMAL reservation
69 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = { 256, 256, 32 };
71 EXPORT_SYMBOL(totalram_pages
);
74 * Used by page_zone() to look up the address of the struct zone whose
75 * id is encoded in the upper bits of page->flags
77 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
78 EXPORT_SYMBOL(zone_table
);
80 static char *zone_names
[MAX_NR_ZONES
] = { "DMA", "DMA32", "Normal", "HighMem" };
81 int min_free_kbytes
= 1024;
83 unsigned long __initdata nr_kernel_pages
;
84 unsigned long __initdata nr_all_pages
;
86 #ifdef CONFIG_DEBUG_VM
87 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
91 unsigned long pfn
= page_to_pfn(page
);
94 seq
= zone_span_seqbegin(zone
);
95 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
97 else if (pfn
< zone
->zone_start_pfn
)
99 } while (zone_span_seqretry(zone
, seq
));
104 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
106 #ifdef CONFIG_HOLES_IN_ZONE
107 if (!pfn_valid(page_to_pfn(page
)))
110 if (zone
!= page_zone(page
))
116 * Temporary debugging check for pages not lying within a given zone.
118 static int bad_range(struct zone
*zone
, struct page
*page
)
120 if (page_outside_zone_boundaries(zone
, page
))
122 if (!page_is_consistent(zone
, page
))
129 static inline int bad_range(struct zone
*zone
, struct page
*page
)
135 static void bad_page(const char *function
, struct page
*page
)
137 printk(KERN_EMERG
"Bad page state at %s (in process '%s', page %p)\n",
138 function
, current
->comm
, page
);
139 printk(KERN_EMERG
"flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
140 (int)(2*sizeof(unsigned long)), (unsigned long)page
->flags
,
141 page
->mapping
, page_mapcount(page
), page_count(page
));
142 printk(KERN_EMERG
"Backtrace:\n");
144 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n");
145 page
->flags
&= ~(1 << PG_lru
|
154 set_page_count(page
, 0);
155 reset_page_mapcount(page
);
156 page
->mapping
= NULL
;
157 add_taint(TAINT_BAD_PAGE
);
161 * Higher-order pages are called "compound pages". They are structured thusly:
163 * The first PAGE_SIZE page is called the "head page".
165 * The remaining PAGE_SIZE pages are called "tail pages".
167 * All pages have PG_compound set. All pages have their ->private pointing at
168 * the head page (even the head page has this).
170 * The first tail page's ->mapping, if non-zero, holds the address of the
171 * compound page's put_page() function.
173 * The order of the allocation is stored in the first tail page's ->index
174 * This is only for debug at present. This usage means that zero-order pages
175 * may not be compound.
177 static void prep_compound_page(struct page
*page
, unsigned long order
)
180 int nr_pages
= 1 << order
;
182 page
[1].mapping
= NULL
;
183 page
[1].index
= order
;
184 for (i
= 0; i
< nr_pages
; i
++) {
185 struct page
*p
= page
+ i
;
188 set_page_private(p
, (unsigned long)page
);
192 static void destroy_compound_page(struct page
*page
, unsigned long order
)
195 int nr_pages
= 1 << order
;
197 if (!PageCompound(page
))
200 if (page
[1].index
!= order
)
201 bad_page(__FUNCTION__
, page
);
203 for (i
= 0; i
< nr_pages
; i
++) {
204 struct page
*p
= page
+ i
;
206 if (!PageCompound(p
))
207 bad_page(__FUNCTION__
, page
);
208 if (page_private(p
) != (unsigned long)page
)
209 bad_page(__FUNCTION__
, page
);
210 ClearPageCompound(p
);
215 * function for dealing with page's order in buddy system.
216 * zone->lock is already acquired when we use these.
217 * So, we don't need atomic page->flags operations here.
219 static inline unsigned long page_order(struct page
*page
) {
220 return page_private(page
);
223 static inline void set_page_order(struct page
*page
, int order
) {
224 set_page_private(page
, order
);
225 __SetPagePrivate(page
);
228 static inline void rmv_page_order(struct page
*page
)
230 __ClearPagePrivate(page
);
231 set_page_private(page
, 0);
235 * Locate the struct page for both the matching buddy in our
236 * pair (buddy1) and the combined O(n+1) page they form (page).
238 * 1) Any buddy B1 will have an order O twin B2 which satisfies
239 * the following equation:
241 * For example, if the starting buddy (buddy2) is #8 its order
243 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
245 * 2) Any buddy B will have an order O+1 parent P which
246 * satisfies the following equation:
249 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
251 static inline struct page
*
252 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
254 unsigned long buddy_idx
= page_idx
^ (1 << order
);
256 return page
+ (buddy_idx
- page_idx
);
259 static inline unsigned long
260 __find_combined_index(unsigned long page_idx
, unsigned int order
)
262 return (page_idx
& ~(1 << order
));
266 * This function checks whether a page is free && is the buddy
267 * we can do coalesce a page and its buddy if
268 * (a) the buddy is not in a hole &&
269 * (b) the buddy is free &&
270 * (c) the buddy is on the buddy system &&
271 * (d) a page and its buddy have the same order.
272 * for recording page's order, we use page_private(page) and PG_private.
275 static inline int page_is_buddy(struct page
*page
, int order
)
277 #ifdef CONFIG_HOLES_IN_ZONE
278 if (!pfn_valid(page_to_pfn(page
)))
282 if (PagePrivate(page
) &&
283 (page_order(page
) == order
) &&
284 page_count(page
) == 0)
290 * Freeing function for a buddy system allocator.
292 * The concept of a buddy system is to maintain direct-mapped table
293 * (containing bit values) for memory blocks of various "orders".
294 * The bottom level table contains the map for the smallest allocatable
295 * units of memory (here, pages), and each level above it describes
296 * pairs of units from the levels below, hence, "buddies".
297 * At a high level, all that happens here is marking the table entry
298 * at the bottom level available, and propagating the changes upward
299 * as necessary, plus some accounting needed to play nicely with other
300 * parts of the VM system.
301 * At each level, we keep a list of pages, which are heads of continuous
302 * free pages of length of (1 << order) and marked with PG_Private.Page's
303 * order is recorded in page_private(page) field.
304 * So when we are allocating or freeing one, we can derive the state of the
305 * other. That is, if we allocate a small block, and both were
306 * free, the remainder of the region must be split into blocks.
307 * If a block is freed, and its buddy is also free, then this
308 * triggers coalescing into a block of larger size.
313 static inline void __free_pages_bulk (struct page
*page
,
314 struct zone
*zone
, unsigned int order
)
316 unsigned long page_idx
;
317 int order_size
= 1 << order
;
320 destroy_compound_page(page
, order
);
322 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
324 BUG_ON(page_idx
& (order_size
- 1));
325 BUG_ON(bad_range(zone
, page
));
327 zone
->free_pages
+= order_size
;
328 while (order
< MAX_ORDER
-1) {
329 unsigned long combined_idx
;
330 struct free_area
*area
;
333 buddy
= __page_find_buddy(page
, page_idx
, order
);
334 if (!page_is_buddy(buddy
, order
))
335 break; /* Move the buddy up one level. */
337 list_del(&buddy
->lru
);
338 area
= zone
->free_area
+ order
;
340 rmv_page_order(buddy
);
341 combined_idx
= __find_combined_index(page_idx
, order
);
342 page
= page
+ (combined_idx
- page_idx
);
343 page_idx
= combined_idx
;
346 set_page_order(page
, order
);
347 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
348 zone
->free_area
[order
].nr_free
++;
351 static inline int free_pages_check(const char *function
, struct page
*page
)
353 if (unlikely(page_mapcount(page
) |
354 (page
->mapping
!= NULL
) |
355 (page_count(page
) != 0) |
365 1 << PG_reserved
))))
366 bad_page(function
, page
);
368 __ClearPageDirty(page
);
370 * For now, we report if PG_reserved was found set, but do not
371 * clear it, and do not free the page. But we shall soon need
372 * to do more, for when the ZERO_PAGE count wraps negative.
374 return PageReserved(page
);
378 * Frees a list of pages.
379 * Assumes all pages on list are in same zone, and of same order.
380 * count is the number of pages to free.
382 * If the zone was previously in an "all pages pinned" state then look to
383 * see if this freeing clears that state.
385 * And clear the zone's pages_scanned counter, to hold off the "all pages are
386 * pinned" detection logic.
389 free_pages_bulk(struct zone
*zone
, int count
,
390 struct list_head
*list
, unsigned int order
)
392 struct page
*page
= NULL
;
395 spin_lock(&zone
->lock
);
396 zone
->all_unreclaimable
= 0;
397 zone
->pages_scanned
= 0;
398 while (!list_empty(list
) && count
--) {
399 page
= list_entry(list
->prev
, struct page
, lru
);
400 /* have to delete it as __free_pages_bulk list manipulates */
401 list_del(&page
->lru
);
402 __free_pages_bulk(page
, zone
, order
);
405 spin_unlock(&zone
->lock
);
409 void __free_pages_ok(struct page
*page
, unsigned int order
)
416 arch_free_page(page
, order
);
420 for (i
= 1 ; i
< (1 << order
) ; ++i
)
421 __put_page(page
+ i
);
424 for (i
= 0 ; i
< (1 << order
) ; ++i
)
425 reserved
+= free_pages_check(__FUNCTION__
, page
+ i
);
429 list_add(&page
->lru
, &list
);
430 mod_page_state(pgfree
, 1 << order
);
431 kernel_map_pages(page
, 1<<order
, 0);
432 local_irq_save(flags
);
433 free_pages_bulk(page_zone(page
), 1, &list
, order
);
434 local_irq_restore(flags
);
438 * permit the bootmem allocator to evade page validation on high-order frees
440 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
443 __ClearPageReserved(page
);
444 set_page_count(page
, 0);
446 free_hot_cold_page(page
, 0);
451 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
452 struct page
*p
= &page
[loop
];
454 if (loop
+ 16 < BITS_PER_LONG
)
456 __ClearPageReserved(p
);
457 set_page_count(p
, 0);
460 arch_free_page(page
, order
);
462 mod_page_state(pgfree
, 1 << order
);
464 list_add(&page
->lru
, &list
);
465 kernel_map_pages(page
, 1 << order
, 0);
466 free_pages_bulk(page_zone(page
), 1, &list
, order
);
472 * The order of subdivision here is critical for the IO subsystem.
473 * Please do not alter this order without good reasons and regression
474 * testing. Specifically, as large blocks of memory are subdivided,
475 * the order in which smaller blocks are delivered depends on the order
476 * they're subdivided in this function. This is the primary factor
477 * influencing the order in which pages are delivered to the IO
478 * subsystem according to empirical testing, and this is also justified
479 * by considering the behavior of a buddy system containing a single
480 * large block of memory acted on by a series of small allocations.
481 * This behavior is a critical factor in sglist merging's success.
485 static inline void expand(struct zone
*zone
, struct page
*page
,
486 int low
, int high
, struct free_area
*area
)
488 unsigned long size
= 1 << high
;
494 BUG_ON(bad_range(zone
, &page
[size
]));
495 list_add(&page
[size
].lru
, &area
->free_list
);
497 set_page_order(&page
[size
], high
);
502 * This page is about to be returned from the page allocator
504 static int prep_new_page(struct page
*page
, int order
)
506 if (unlikely(page_mapcount(page
) |
507 (page
->mapping
!= NULL
) |
508 (page_count(page
) != 0) |
519 1 << PG_reserved
))))
520 bad_page(__FUNCTION__
, page
);
523 * For now, we report if PG_reserved was found set, but do not
524 * clear it, and do not allocate the page: as a safety net.
526 if (PageReserved(page
))
529 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
530 1 << PG_referenced
| 1 << PG_arch_1
|
531 1 << PG_checked
| 1 << PG_mappedtodisk
);
532 set_page_private(page
, 0);
533 set_page_refs(page
, order
);
534 kernel_map_pages(page
, 1 << order
, 1);
539 * Do the hard work of removing an element from the buddy allocator.
540 * Call me with the zone->lock already held.
542 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
544 struct free_area
* area
;
545 unsigned int current_order
;
548 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
549 area
= zone
->free_area
+ current_order
;
550 if (list_empty(&area
->free_list
))
553 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
554 list_del(&page
->lru
);
555 rmv_page_order(page
);
557 zone
->free_pages
-= 1UL << order
;
558 expand(zone
, page
, order
, current_order
, area
);
566 * Obtain a specified number of elements from the buddy allocator, all under
567 * a single hold of the lock, for efficiency. Add them to the supplied list.
568 * Returns the number of new pages which were placed at *list.
570 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
571 unsigned long count
, struct list_head
*list
)
575 spin_lock(&zone
->lock
);
576 for (i
= 0; i
< count
; ++i
) {
577 struct page
*page
= __rmqueue(zone
, order
);
578 if (unlikely(page
== NULL
))
580 list_add_tail(&page
->lru
, list
);
582 spin_unlock(&zone
->lock
);
587 /* Called from the slab reaper to drain remote pagesets */
588 void drain_remote_pages(void)
594 local_irq_save(flags
);
595 for_each_zone(zone
) {
596 struct per_cpu_pageset
*pset
;
598 /* Do not drain local pagesets */
599 if (zone
->zone_pgdat
->node_id
== numa_node_id())
602 pset
= zone
->pageset
[smp_processor_id()];
603 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
604 struct per_cpu_pages
*pcp
;
608 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
612 local_irq_restore(flags
);
616 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
617 static void __drain_pages(unsigned int cpu
)
623 for_each_zone(zone
) {
624 struct per_cpu_pageset
*pset
;
626 pset
= zone_pcp(zone
, cpu
);
627 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
628 struct per_cpu_pages
*pcp
;
631 local_irq_save(flags
);
632 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
634 local_irq_restore(flags
);
638 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
642 void mark_free_pages(struct zone
*zone
)
644 unsigned long zone_pfn
, flags
;
646 struct list_head
*curr
;
648 if (!zone
->spanned_pages
)
651 spin_lock_irqsave(&zone
->lock
, flags
);
652 for (zone_pfn
= 0; zone_pfn
< zone
->spanned_pages
; ++zone_pfn
)
653 ClearPageNosaveFree(pfn_to_page(zone_pfn
+ zone
->zone_start_pfn
));
655 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
656 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
657 unsigned long start_pfn
, i
;
659 start_pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
661 for (i
=0; i
< (1<<order
); i
++)
662 SetPageNosaveFree(pfn_to_page(start_pfn
+i
));
664 spin_unlock_irqrestore(&zone
->lock
, flags
);
668 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
670 void drain_local_pages(void)
674 local_irq_save(flags
);
675 __drain_pages(smp_processor_id());
676 local_irq_restore(flags
);
678 #endif /* CONFIG_PM */
680 static void zone_statistics(struct zonelist
*zonelist
, struct zone
*z
)
685 pg_data_t
*pg
= z
->zone_pgdat
;
686 pg_data_t
*orig
= zonelist
->zones
[0]->zone_pgdat
;
687 struct per_cpu_pageset
*p
;
689 local_irq_save(flags
);
690 cpu
= smp_processor_id();
696 zone_pcp(zonelist
->zones
[0], cpu
)->numa_foreign
++;
698 if (pg
== NODE_DATA(numa_node_id()))
702 local_irq_restore(flags
);
707 * Free a 0-order page
709 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
711 struct zone
*zone
= page_zone(page
);
712 struct per_cpu_pages
*pcp
;
715 arch_free_page(page
, 0);
718 page
->mapping
= NULL
;
719 if (free_pages_check(__FUNCTION__
, page
))
722 inc_page_state(pgfree
);
723 kernel_map_pages(page
, 1, 0);
725 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
726 local_irq_save(flags
);
727 list_add(&page
->lru
, &pcp
->list
);
729 if (pcp
->count
>= pcp
->high
)
730 pcp
->count
-= free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
731 local_irq_restore(flags
);
735 void fastcall
free_hot_page(struct page
*page
)
737 free_hot_cold_page(page
, 0);
740 void fastcall
free_cold_page(struct page
*page
)
742 free_hot_cold_page(page
, 1);
745 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
749 BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
750 for(i
= 0; i
< (1 << order
); i
++)
751 clear_highpage(page
+ i
);
755 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
756 * we cheat by calling it from here, in the order > 0 path. Saves a branch
760 buffered_rmqueue(struct zone
*zone
, int order
, gfp_t gfp_flags
)
764 int cold
= !!(gfp_flags
& __GFP_COLD
);
768 struct per_cpu_pages
*pcp
;
771 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
772 local_irq_save(flags
);
774 pcp
->count
+= rmqueue_bulk(zone
, 0,
775 pcp
->batch
, &pcp
->list
);
776 if (likely(pcp
->count
)) {
777 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
778 list_del(&page
->lru
);
781 local_irq_restore(flags
);
784 spin_lock_irqsave(&zone
->lock
, flags
);
785 page
= __rmqueue(zone
, order
);
786 spin_unlock_irqrestore(&zone
->lock
, flags
);
790 BUG_ON(bad_range(zone
, page
));
791 mod_page_state_zone(zone
, pgalloc
, 1 << order
);
792 if (prep_new_page(page
, order
))
795 if (gfp_flags
& __GFP_ZERO
)
796 prep_zero_page(page
, order
, gfp_flags
);
798 if (order
&& (gfp_flags
& __GFP_COMP
))
799 prep_compound_page(page
, order
);
804 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
805 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
806 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
807 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
808 #define ALLOC_HARDER 0x10 /* try to alloc harder */
809 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
810 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
813 * Return 1 if free pages are above 'mark'. This takes into account the order
816 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
817 int classzone_idx
, int alloc_flags
)
819 /* free_pages my go negative - that's OK */
820 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
823 if (alloc_flags
& ALLOC_HIGH
)
825 if (alloc_flags
& ALLOC_HARDER
)
828 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
830 for (o
= 0; o
< order
; o
++) {
831 /* At the next order, this order's pages become unavailable */
832 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
834 /* Require fewer higher order pages to be free */
837 if (free_pages
<= min
)
844 * get_page_from_freeliest goes through the zonelist trying to allocate
848 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
849 struct zonelist
*zonelist
, int alloc_flags
)
851 struct zone
**z
= zonelist
->zones
;
852 struct page
*page
= NULL
;
853 int classzone_idx
= zone_idx(*z
);
856 * Go through the zonelist once, looking for a zone with enough free.
857 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
860 if ((alloc_flags
& ALLOC_CPUSET
) &&
861 !cpuset_zone_allowed(*z
, gfp_mask
))
864 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
866 if (alloc_flags
& ALLOC_WMARK_MIN
)
867 mark
= (*z
)->pages_min
;
868 else if (alloc_flags
& ALLOC_WMARK_LOW
)
869 mark
= (*z
)->pages_low
;
871 mark
= (*z
)->pages_high
;
872 if (!zone_watermark_ok(*z
, order
, mark
,
873 classzone_idx
, alloc_flags
))
877 page
= buffered_rmqueue(*z
, order
, gfp_mask
);
879 zone_statistics(zonelist
, *z
);
882 } while (*(++z
) != NULL
);
887 * This is the 'heart' of the zoned buddy allocator.
889 struct page
* fastcall
890 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
891 struct zonelist
*zonelist
)
893 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
896 struct reclaim_state reclaim_state
;
897 struct task_struct
*p
= current
;
900 int did_some_progress
;
902 might_sleep_if(wait
);
905 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
907 if (unlikely(*z
== NULL
)) {
908 /* Should this ever happen?? */
912 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
913 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
918 wakeup_kswapd(*z
, order
);
922 * OK, we're below the kswapd watermark and have kicked background
923 * reclaim. Now things get more complex, so set up alloc_flags according
924 * to how we want to proceed.
926 * The caller may dip into page reserves a bit more if the caller
927 * cannot run direct reclaim, or if the caller has realtime scheduling
930 alloc_flags
= ALLOC_WMARK_MIN
;
931 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
932 alloc_flags
|= ALLOC_HARDER
;
933 if (gfp_mask
& __GFP_HIGH
)
934 alloc_flags
|= ALLOC_HIGH
;
935 alloc_flags
|= ALLOC_CPUSET
;
938 * Go through the zonelist again. Let __GFP_HIGH and allocations
939 * coming from realtime tasks go deeper into reserves.
941 * This is the last chance, in general, before the goto nopage.
942 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
943 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
945 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
949 /* This allocation should allow future memory freeing. */
951 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
952 && !in_interrupt()) {
953 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
955 /* go through the zonelist yet again, ignoring mins */
956 page
= get_page_from_freelist(gfp_mask
, order
,
957 zonelist
, ALLOC_NO_WATERMARKS
);
960 if (gfp_mask
& __GFP_NOFAIL
) {
961 blk_congestion_wait(WRITE
, HZ
/50);
968 /* Atomic allocations - we can't balance anything */
975 /* We now go into synchronous reclaim */
976 p
->flags
|= PF_MEMALLOC
;
977 reclaim_state
.reclaimed_slab
= 0;
978 p
->reclaim_state
= &reclaim_state
;
980 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
982 p
->reclaim_state
= NULL
;
983 p
->flags
&= ~PF_MEMALLOC
;
987 if (likely(did_some_progress
)) {
988 page
= get_page_from_freelist(gfp_mask
, order
,
989 zonelist
, alloc_flags
);
992 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
994 * Go through the zonelist yet one more time, keep
995 * very high watermark here, this is only to catch
996 * a parallel oom killing, we must fail if we're still
997 * under heavy pressure.
999 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1000 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1004 out_of_memory(gfp_mask
, order
);
1009 * Don't let big-order allocations loop unless the caller explicitly
1010 * requests that. Wait for some write requests to complete then retry.
1012 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1013 * <= 3, but that may not be true in other implementations.
1016 if (!(gfp_mask
& __GFP_NORETRY
)) {
1017 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
1019 if (gfp_mask
& __GFP_NOFAIL
)
1023 blk_congestion_wait(WRITE
, HZ
/50);
1028 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1029 printk(KERN_WARNING
"%s: page allocation failure."
1030 " order:%d, mode:0x%x\n",
1031 p
->comm
, order
, gfp_mask
);
1039 EXPORT_SYMBOL(__alloc_pages
);
1042 * Common helper functions.
1044 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1047 page
= alloc_pages(gfp_mask
, order
);
1050 return (unsigned long) page_address(page
);
1053 EXPORT_SYMBOL(__get_free_pages
);
1055 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1060 * get_zeroed_page() returns a 32-bit address, which cannot represent
1063 BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1065 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1067 return (unsigned long) page_address(page
);
1071 EXPORT_SYMBOL(get_zeroed_page
);
1073 void __pagevec_free(struct pagevec
*pvec
)
1075 int i
= pagevec_count(pvec
);
1078 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1081 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1083 if (put_page_testzero(page
)) {
1085 free_hot_page(page
);
1087 __free_pages_ok(page
, order
);
1091 EXPORT_SYMBOL(__free_pages
);
1093 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1096 BUG_ON(!virt_addr_valid((void *)addr
));
1097 __free_pages(virt_to_page((void *)addr
), order
);
1101 EXPORT_SYMBOL(free_pages
);
1104 * Total amount of free (allocatable) RAM:
1106 unsigned int nr_free_pages(void)
1108 unsigned int sum
= 0;
1112 sum
+= zone
->free_pages
;
1117 EXPORT_SYMBOL(nr_free_pages
);
1120 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1122 unsigned int i
, sum
= 0;
1124 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1125 sum
+= pgdat
->node_zones
[i
].free_pages
;
1131 static unsigned int nr_free_zone_pages(int offset
)
1133 /* Just pick one node, since fallback list is circular */
1134 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1135 unsigned int sum
= 0;
1137 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1138 struct zone
**zonep
= zonelist
->zones
;
1141 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1142 unsigned long size
= zone
->present_pages
;
1143 unsigned long high
= zone
->pages_high
;
1152 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1154 unsigned int nr_free_buffer_pages(void)
1156 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1160 * Amount of free RAM allocatable within all zones
1162 unsigned int nr_free_pagecache_pages(void)
1164 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1167 #ifdef CONFIG_HIGHMEM
1168 unsigned int nr_free_highpages (void)
1171 unsigned int pages
= 0;
1173 for_each_pgdat(pgdat
)
1174 pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1181 static void show_node(struct zone
*zone
)
1183 printk("Node %d ", zone
->zone_pgdat
->node_id
);
1186 #define show_node(zone) do { } while (0)
1190 * Accumulate the page_state information across all CPUs.
1191 * The result is unavoidably approximate - it can change
1192 * during and after execution of this function.
1194 static DEFINE_PER_CPU(struct page_state
, page_states
) = {0};
1196 atomic_t nr_pagecache
= ATOMIC_INIT(0);
1197 EXPORT_SYMBOL(nr_pagecache
);
1199 DEFINE_PER_CPU(long, nr_pagecache_local
) = 0;
1202 static void __get_page_state(struct page_state
*ret
, int nr
, cpumask_t
*cpumask
)
1206 memset(ret
, 0, sizeof(*ret
));
1208 cpu
= first_cpu(*cpumask
);
1209 while (cpu
< NR_CPUS
) {
1210 unsigned long *in
, *out
, off
;
1212 in
= (unsigned long *)&per_cpu(page_states
, cpu
);
1214 cpu
= next_cpu(cpu
, *cpumask
);
1217 prefetch(&per_cpu(page_states
, cpu
));
1219 out
= (unsigned long *)ret
;
1220 for (off
= 0; off
< nr
; off
++)
1225 void get_page_state_node(struct page_state
*ret
, int node
)
1228 cpumask_t mask
= node_to_cpumask(node
);
1230 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1231 nr
/= sizeof(unsigned long);
1233 __get_page_state(ret
, nr
+1, &mask
);
1236 void get_page_state(struct page_state
*ret
)
1239 cpumask_t mask
= CPU_MASK_ALL
;
1241 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1242 nr
/= sizeof(unsigned long);
1244 __get_page_state(ret
, nr
+ 1, &mask
);
1247 void get_full_page_state(struct page_state
*ret
)
1249 cpumask_t mask
= CPU_MASK_ALL
;
1251 __get_page_state(ret
, sizeof(*ret
) / sizeof(unsigned long), &mask
);
1254 unsigned long __read_page_state(unsigned long offset
)
1256 unsigned long ret
= 0;
1262 in
= (unsigned long)&per_cpu(page_states
, cpu
) + offset
;
1263 ret
+= *((unsigned long *)in
);
1268 void __mod_page_state(unsigned long offset
, unsigned long delta
)
1270 unsigned long flags
;
1273 local_irq_save(flags
);
1274 ptr
= &__get_cpu_var(page_states
);
1275 *(unsigned long*)(ptr
+ offset
) += delta
;
1276 local_irq_restore(flags
);
1279 EXPORT_SYMBOL(__mod_page_state
);
1281 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
1282 unsigned long *free
, struct pglist_data
*pgdat
)
1284 struct zone
*zones
= pgdat
->node_zones
;
1290 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1291 *active
+= zones
[i
].nr_active
;
1292 *inactive
+= zones
[i
].nr_inactive
;
1293 *free
+= zones
[i
].free_pages
;
1297 void get_zone_counts(unsigned long *active
,
1298 unsigned long *inactive
, unsigned long *free
)
1300 struct pglist_data
*pgdat
;
1305 for_each_pgdat(pgdat
) {
1306 unsigned long l
, m
, n
;
1307 __get_zone_counts(&l
, &m
, &n
, pgdat
);
1314 void si_meminfo(struct sysinfo
*val
)
1316 val
->totalram
= totalram_pages
;
1318 val
->freeram
= nr_free_pages();
1319 val
->bufferram
= nr_blockdev_pages();
1320 #ifdef CONFIG_HIGHMEM
1321 val
->totalhigh
= totalhigh_pages
;
1322 val
->freehigh
= nr_free_highpages();
1327 val
->mem_unit
= PAGE_SIZE
;
1330 EXPORT_SYMBOL(si_meminfo
);
1333 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1335 pg_data_t
*pgdat
= NODE_DATA(nid
);
1337 val
->totalram
= pgdat
->node_present_pages
;
1338 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1339 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1340 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1341 val
->mem_unit
= PAGE_SIZE
;
1345 #define K(x) ((x) << (PAGE_SHIFT-10))
1348 * Show free area list (used inside shift_scroll-lock stuff)
1349 * We also calculate the percentage fragmentation. We do this by counting the
1350 * memory on each free list with the exception of the first item on the list.
1352 void show_free_areas(void)
1354 struct page_state ps
;
1355 int cpu
, temperature
;
1356 unsigned long active
;
1357 unsigned long inactive
;
1361 for_each_zone(zone
) {
1363 printk("%s per-cpu:", zone
->name
);
1365 if (!zone
->present_pages
) {
1371 for_each_online_cpu(cpu
) {
1372 struct per_cpu_pageset
*pageset
;
1374 pageset
= zone_pcp(zone
, cpu
);
1376 for (temperature
= 0; temperature
< 2; temperature
++)
1377 printk("cpu %d %s: high %d, batch %d used:%d\n",
1379 temperature
? "cold" : "hot",
1380 pageset
->pcp
[temperature
].high
,
1381 pageset
->pcp
[temperature
].batch
,
1382 pageset
->pcp
[temperature
].count
);
1386 get_page_state(&ps
);
1387 get_zone_counts(&active
, &inactive
, &free
);
1389 printk("Free pages: %11ukB (%ukB HighMem)\n",
1391 K(nr_free_highpages()));
1393 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1394 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1403 ps
.nr_page_table_pages
);
1405 for_each_zone(zone
) {
1417 " pages_scanned:%lu"
1418 " all_unreclaimable? %s"
1421 K(zone
->free_pages
),
1424 K(zone
->pages_high
),
1426 K(zone
->nr_inactive
),
1427 K(zone
->present_pages
),
1428 zone
->pages_scanned
,
1429 (zone
->all_unreclaimable
? "yes" : "no")
1431 printk("lowmem_reserve[]:");
1432 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1433 printk(" %lu", zone
->lowmem_reserve
[i
]);
1437 for_each_zone(zone
) {
1438 unsigned long nr
, flags
, order
, total
= 0;
1441 printk("%s: ", zone
->name
);
1442 if (!zone
->present_pages
) {
1447 spin_lock_irqsave(&zone
->lock
, flags
);
1448 for (order
= 0; order
< MAX_ORDER
; order
++) {
1449 nr
= zone
->free_area
[order
].nr_free
;
1450 total
+= nr
<< order
;
1451 printk("%lu*%lukB ", nr
, K(1UL) << order
);
1453 spin_unlock_irqrestore(&zone
->lock
, flags
);
1454 printk("= %lukB\n", K(total
));
1457 show_swap_cache_info();
1461 * Builds allocation fallback zone lists.
1463 static int __init
build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
, int j
, int k
)
1470 zone
= pgdat
->node_zones
+ ZONE_HIGHMEM
;
1471 if (zone
->present_pages
) {
1472 #ifndef CONFIG_HIGHMEM
1475 zonelist
->zones
[j
++] = zone
;
1478 zone
= pgdat
->node_zones
+ ZONE_NORMAL
;
1479 if (zone
->present_pages
)
1480 zonelist
->zones
[j
++] = zone
;
1482 zone
= pgdat
->node_zones
+ ZONE_DMA32
;
1483 if (zone
->present_pages
)
1484 zonelist
->zones
[j
++] = zone
;
1486 zone
= pgdat
->node_zones
+ ZONE_DMA
;
1487 if (zone
->present_pages
)
1488 zonelist
->zones
[j
++] = zone
;
1494 static inline int highest_zone(int zone_bits
)
1496 int res
= ZONE_NORMAL
;
1497 if (zone_bits
& (__force
int)__GFP_HIGHMEM
)
1499 if (zone_bits
& (__force
int)__GFP_DMA32
)
1501 if (zone_bits
& (__force
int)__GFP_DMA
)
1507 #define MAX_NODE_LOAD (num_online_nodes())
1508 static int __initdata node_load
[MAX_NUMNODES
];
1510 * find_next_best_node - find the next node that should appear in a given node's fallback list
1511 * @node: node whose fallback list we're appending
1512 * @used_node_mask: nodemask_t of already used nodes
1514 * We use a number of factors to determine which is the next node that should
1515 * appear on a given node's fallback list. The node should not have appeared
1516 * already in @node's fallback list, and it should be the next closest node
1517 * according to the distance array (which contains arbitrary distance values
1518 * from each node to each node in the system), and should also prefer nodes
1519 * with no CPUs, since presumably they'll have very little allocation pressure
1520 * on them otherwise.
1521 * It returns -1 if no node is found.
1523 static int __init
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1526 int min_val
= INT_MAX
;
1529 for_each_online_node(i
) {
1532 /* Start from local node */
1533 n
= (node
+i
) % num_online_nodes();
1535 /* Don't want a node to appear more than once */
1536 if (node_isset(n
, *used_node_mask
))
1539 /* Use the local node if we haven't already */
1540 if (!node_isset(node
, *used_node_mask
)) {
1545 /* Use the distance array to find the distance */
1546 val
= node_distance(node
, n
);
1548 /* Give preference to headless and unused nodes */
1549 tmp
= node_to_cpumask(n
);
1550 if (!cpus_empty(tmp
))
1551 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1553 /* Slight preference for less loaded node */
1554 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1555 val
+= node_load
[n
];
1557 if (val
< min_val
) {
1564 node_set(best_node
, *used_node_mask
);
1569 static void __init
build_zonelists(pg_data_t
*pgdat
)
1571 int i
, j
, k
, node
, local_node
;
1572 int prev_node
, load
;
1573 struct zonelist
*zonelist
;
1574 nodemask_t used_mask
;
1576 /* initialize zonelists */
1577 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1578 zonelist
= pgdat
->node_zonelists
+ i
;
1579 zonelist
->zones
[0] = NULL
;
1582 /* NUMA-aware ordering of nodes */
1583 local_node
= pgdat
->node_id
;
1584 load
= num_online_nodes();
1585 prev_node
= local_node
;
1586 nodes_clear(used_mask
);
1587 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1589 * We don't want to pressure a particular node.
1590 * So adding penalty to the first node in same
1591 * distance group to make it round-robin.
1593 if (node_distance(local_node
, node
) !=
1594 node_distance(local_node
, prev_node
))
1595 node_load
[node
] += load
;
1598 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1599 zonelist
= pgdat
->node_zonelists
+ i
;
1600 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1602 k
= highest_zone(i
);
1604 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1605 zonelist
->zones
[j
] = NULL
;
1610 #else /* CONFIG_NUMA */
1612 static void __init
build_zonelists(pg_data_t
*pgdat
)
1614 int i
, j
, k
, node
, local_node
;
1616 local_node
= pgdat
->node_id
;
1617 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1618 struct zonelist
*zonelist
;
1620 zonelist
= pgdat
->node_zonelists
+ i
;
1623 k
= highest_zone(i
);
1624 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1626 * Now we build the zonelist so that it contains the zones
1627 * of all the other nodes.
1628 * We don't want to pressure a particular node, so when
1629 * building the zones for node N, we make sure that the
1630 * zones coming right after the local ones are those from
1631 * node N+1 (modulo N)
1633 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1634 if (!node_online(node
))
1636 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1638 for (node
= 0; node
< local_node
; node
++) {
1639 if (!node_online(node
))
1641 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1644 zonelist
->zones
[j
] = NULL
;
1648 #endif /* CONFIG_NUMA */
1650 void __init
build_all_zonelists(void)
1654 for_each_online_node(i
)
1655 build_zonelists(NODE_DATA(i
));
1656 printk("Built %i zonelists\n", num_online_nodes());
1657 cpuset_init_current_mems_allowed();
1661 * Helper functions to size the waitqueue hash table.
1662 * Essentially these want to choose hash table sizes sufficiently
1663 * large so that collisions trying to wait on pages are rare.
1664 * But in fact, the number of active page waitqueues on typical
1665 * systems is ridiculously low, less than 200. So this is even
1666 * conservative, even though it seems large.
1668 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1669 * waitqueues, i.e. the size of the waitq table given the number of pages.
1671 #define PAGES_PER_WAITQUEUE 256
1673 static inline unsigned long wait_table_size(unsigned long pages
)
1675 unsigned long size
= 1;
1677 pages
/= PAGES_PER_WAITQUEUE
;
1679 while (size
< pages
)
1683 * Once we have dozens or even hundreds of threads sleeping
1684 * on IO we've got bigger problems than wait queue collision.
1685 * Limit the size of the wait table to a reasonable size.
1687 size
= min(size
, 4096UL);
1689 return max(size
, 4UL);
1693 * This is an integer logarithm so that shifts can be used later
1694 * to extract the more random high bits from the multiplicative
1695 * hash function before the remainder is taken.
1697 static inline unsigned long wait_table_bits(unsigned long size
)
1702 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1704 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1705 unsigned long *zones_size
, unsigned long *zholes_size
)
1707 unsigned long realtotalpages
, totalpages
= 0;
1710 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1711 totalpages
+= zones_size
[i
];
1712 pgdat
->node_spanned_pages
= totalpages
;
1714 realtotalpages
= totalpages
;
1716 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1717 realtotalpages
-= zholes_size
[i
];
1718 pgdat
->node_present_pages
= realtotalpages
;
1719 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1724 * Initially all pages are reserved - free ones are freed
1725 * up by free_all_bootmem() once the early boot process is
1726 * done. Non-atomic initialization, single-pass.
1728 void __devinit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1729 unsigned long start_pfn
)
1732 unsigned long end_pfn
= start_pfn
+ size
;
1735 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++, page
++) {
1736 if (!early_pfn_valid(pfn
))
1738 page
= pfn_to_page(pfn
);
1739 set_page_links(page
, zone
, nid
, pfn
);
1740 set_page_count(page
, 1);
1741 reset_page_mapcount(page
);
1742 SetPageReserved(page
);
1743 INIT_LIST_HEAD(&page
->lru
);
1744 #ifdef WANT_PAGE_VIRTUAL
1745 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1746 if (!is_highmem_idx(zone
))
1747 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1752 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1756 for (order
= 0; order
< MAX_ORDER
; order
++) {
1757 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1758 zone
->free_area
[order
].nr_free
= 0;
1762 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1763 void zonetable_add(struct zone
*zone
, int nid
, int zid
, unsigned long pfn
,
1766 unsigned long snum
= pfn_to_section_nr(pfn
);
1767 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1770 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1772 for (; snum
<= end
; snum
++)
1773 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1776 #ifndef __HAVE_ARCH_MEMMAP_INIT
1777 #define memmap_init(size, nid, zone, start_pfn) \
1778 memmap_init_zone((size), (nid), (zone), (start_pfn))
1781 static int __devinit
zone_batchsize(struct zone
*zone
)
1786 * The per-cpu-pages pools are set to around 1000th of the
1787 * size of the zone. But no more than 1/2 of a meg.
1789 * OK, so we don't know how big the cache is. So guess.
1791 batch
= zone
->present_pages
/ 1024;
1792 if (batch
* PAGE_SIZE
> 512 * 1024)
1793 batch
= (512 * 1024) / PAGE_SIZE
;
1794 batch
/= 4; /* We effectively *= 4 below */
1799 * Clamp the batch to a 2^n - 1 value. Having a power
1800 * of 2 value was found to be more likely to have
1801 * suboptimal cache aliasing properties in some cases.
1803 * For example if 2 tasks are alternately allocating
1804 * batches of pages, one task can end up with a lot
1805 * of pages of one half of the possible page colors
1806 * and the other with pages of the other colors.
1808 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1813 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1815 struct per_cpu_pages
*pcp
;
1817 memset(p
, 0, sizeof(*p
));
1819 pcp
= &p
->pcp
[0]; /* hot */
1821 pcp
->high
= 6 * batch
;
1822 pcp
->batch
= max(1UL, 1 * batch
);
1823 INIT_LIST_HEAD(&pcp
->list
);
1825 pcp
= &p
->pcp
[1]; /* cold*/
1827 pcp
->high
= 2 * batch
;
1828 pcp
->batch
= max(1UL, batch
/2);
1829 INIT_LIST_HEAD(&pcp
->list
);
1834 * Boot pageset table. One per cpu which is going to be used for all
1835 * zones and all nodes. The parameters will be set in such a way
1836 * that an item put on a list will immediately be handed over to
1837 * the buddy list. This is safe since pageset manipulation is done
1838 * with interrupts disabled.
1840 * Some NUMA counter updates may also be caught by the boot pagesets.
1842 * The boot_pagesets must be kept even after bootup is complete for
1843 * unused processors and/or zones. They do play a role for bootstrapping
1844 * hotplugged processors.
1846 * zoneinfo_show() and maybe other functions do
1847 * not check if the processor is online before following the pageset pointer.
1848 * Other parts of the kernel may not check if the zone is available.
1850 static struct per_cpu_pageset
1851 boot_pageset
[NR_CPUS
];
1854 * Dynamically allocate memory for the
1855 * per cpu pageset array in struct zone.
1857 static int __devinit
process_zones(int cpu
)
1859 struct zone
*zone
, *dzone
;
1861 for_each_zone(zone
) {
1863 zone
->pageset
[cpu
] = kmalloc_node(sizeof(struct per_cpu_pageset
),
1864 GFP_KERNEL
, cpu_to_node(cpu
));
1865 if (!zone
->pageset
[cpu
])
1868 setup_pageset(zone
->pageset
[cpu
], zone_batchsize(zone
));
1873 for_each_zone(dzone
) {
1876 kfree(dzone
->pageset
[cpu
]);
1877 dzone
->pageset
[cpu
] = NULL
;
1882 static inline void free_zone_pagesets(int cpu
)
1887 for_each_zone(zone
) {
1888 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1890 zone_pcp(zone
, cpu
) = NULL
;
1896 static int __devinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1897 unsigned long action
,
1900 int cpu
= (long)hcpu
;
1901 int ret
= NOTIFY_OK
;
1904 case CPU_UP_PREPARE
:
1905 if (process_zones(cpu
))
1908 case CPU_UP_CANCELED
:
1910 free_zone_pagesets(cpu
);
1918 static struct notifier_block pageset_notifier
=
1919 { &pageset_cpuup_callback
, NULL
, 0 };
1921 void __init
setup_per_cpu_pageset(void)
1925 /* Initialize per_cpu_pageset for cpu 0.
1926 * A cpuup callback will do this for every cpu
1927 * as it comes online
1929 err
= process_zones(smp_processor_id());
1931 register_cpu_notifier(&pageset_notifier
);
1937 void zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1940 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1943 * The per-page waitqueue mechanism uses hashed waitqueues
1946 zone
->wait_table_size
= wait_table_size(zone_size_pages
);
1947 zone
->wait_table_bits
= wait_table_bits(zone
->wait_table_size
);
1948 zone
->wait_table
= (wait_queue_head_t
*)
1949 alloc_bootmem_node(pgdat
, zone
->wait_table_size
1950 * sizeof(wait_queue_head_t
));
1952 for(i
= 0; i
< zone
->wait_table_size
; ++i
)
1953 init_waitqueue_head(zone
->wait_table
+ i
);
1956 static __devinit
void zone_pcp_init(struct zone
*zone
)
1959 unsigned long batch
= zone_batchsize(zone
);
1961 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1963 /* Early boot. Slab allocator not functional yet */
1964 zone
->pageset
[cpu
] = &boot_pageset
[cpu
];
1965 setup_pageset(&boot_pageset
[cpu
],0);
1967 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1970 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1971 zone
->name
, zone
->present_pages
, batch
);
1974 static __devinit
void init_currently_empty_zone(struct zone
*zone
,
1975 unsigned long zone_start_pfn
, unsigned long size
)
1977 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1979 zone_wait_table_init(zone
, size
);
1980 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1982 zone
->zone_mem_map
= pfn_to_page(zone_start_pfn
);
1983 zone
->zone_start_pfn
= zone_start_pfn
;
1985 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1987 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
1991 * Set up the zone data structures:
1992 * - mark all pages reserved
1993 * - mark all memory queues empty
1994 * - clear the memory bitmaps
1996 static void __init
free_area_init_core(struct pglist_data
*pgdat
,
1997 unsigned long *zones_size
, unsigned long *zholes_size
)
2000 int nid
= pgdat
->node_id
;
2001 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
2003 pgdat_resize_init(pgdat
);
2004 pgdat
->nr_zones
= 0;
2005 init_waitqueue_head(&pgdat
->kswapd_wait
);
2006 pgdat
->kswapd_max_order
= 0;
2008 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2009 struct zone
*zone
= pgdat
->node_zones
+ j
;
2010 unsigned long size
, realsize
;
2012 realsize
= size
= zones_size
[j
];
2014 realsize
-= zholes_size
[j
];
2016 if (j
< ZONE_HIGHMEM
)
2017 nr_kernel_pages
+= realsize
;
2018 nr_all_pages
+= realsize
;
2020 zone
->spanned_pages
= size
;
2021 zone
->present_pages
= realsize
;
2022 zone
->name
= zone_names
[j
];
2023 spin_lock_init(&zone
->lock
);
2024 spin_lock_init(&zone
->lru_lock
);
2025 zone_seqlock_init(zone
);
2026 zone
->zone_pgdat
= pgdat
;
2027 zone
->free_pages
= 0;
2029 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
2031 zone_pcp_init(zone
);
2032 INIT_LIST_HEAD(&zone
->active_list
);
2033 INIT_LIST_HEAD(&zone
->inactive_list
);
2034 zone
->nr_scan_active
= 0;
2035 zone
->nr_scan_inactive
= 0;
2036 zone
->nr_active
= 0;
2037 zone
->nr_inactive
= 0;
2038 atomic_set(&zone
->reclaim_in_progress
, 0);
2042 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2043 init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2044 zone_start_pfn
+= size
;
2048 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2050 /* Skip empty nodes */
2051 if (!pgdat
->node_spanned_pages
)
2054 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2055 /* ia64 gets its own node_mem_map, before this, without bootmem */
2056 if (!pgdat
->node_mem_map
) {
2060 size
= (pgdat
->node_spanned_pages
+ 1) * sizeof(struct page
);
2061 map
= alloc_remap(pgdat
->node_id
, size
);
2063 map
= alloc_bootmem_node(pgdat
, size
);
2064 pgdat
->node_mem_map
= map
;
2066 #ifdef CONFIG_FLATMEM
2068 * With no DISCONTIG, the global mem_map is just set as node 0's
2070 if (pgdat
== NODE_DATA(0))
2071 mem_map
= NODE_DATA(0)->node_mem_map
;
2073 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2076 void __init
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2077 unsigned long *zones_size
, unsigned long node_start_pfn
,
2078 unsigned long *zholes_size
)
2080 pgdat
->node_id
= nid
;
2081 pgdat
->node_start_pfn
= node_start_pfn
;
2082 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
2084 alloc_node_mem_map(pgdat
);
2086 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2089 #ifndef CONFIG_NEED_MULTIPLE_NODES
2090 static bootmem_data_t contig_bootmem_data
;
2091 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2093 EXPORT_SYMBOL(contig_page_data
);
2096 void __init
free_area_init(unsigned long *zones_size
)
2098 free_area_init_node(0, NODE_DATA(0), zones_size
,
2099 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2102 #ifdef CONFIG_PROC_FS
2104 #include <linux/seq_file.h>
2106 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
2111 for (pgdat
= pgdat_list
; pgdat
&& node
; pgdat
= pgdat
->pgdat_next
)
2117 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2119 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2122 return pgdat
->pgdat_next
;
2125 static void frag_stop(struct seq_file
*m
, void *arg
)
2130 * This walks the free areas for each zone.
2132 static int frag_show(struct seq_file
*m
, void *arg
)
2134 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2136 struct zone
*node_zones
= pgdat
->node_zones
;
2137 unsigned long flags
;
2140 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
2141 if (!zone
->present_pages
)
2144 spin_lock_irqsave(&zone
->lock
, flags
);
2145 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
2146 for (order
= 0; order
< MAX_ORDER
; ++order
)
2147 seq_printf(m
, "%6lu ", zone
->free_area
[order
].nr_free
);
2148 spin_unlock_irqrestore(&zone
->lock
, flags
);
2154 struct seq_operations fragmentation_op
= {
2155 .start
= frag_start
,
2162 * Output information about zones in @pgdat.
2164 static int zoneinfo_show(struct seq_file
*m
, void *arg
)
2166 pg_data_t
*pgdat
= arg
;
2168 struct zone
*node_zones
= pgdat
->node_zones
;
2169 unsigned long flags
;
2171 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; zone
++) {
2174 if (!zone
->present_pages
)
2177 spin_lock_irqsave(&zone
->lock
, flags
);
2178 seq_printf(m
, "Node %d, zone %8s", pgdat
->node_id
, zone
->name
);
2186 "\n scanned %lu (a: %lu i: %lu)"
2195 zone
->pages_scanned
,
2196 zone
->nr_scan_active
, zone
->nr_scan_inactive
,
2197 zone
->spanned_pages
,
2198 zone
->present_pages
);
2200 "\n protection: (%lu",
2201 zone
->lowmem_reserve
[0]);
2202 for (i
= 1; i
< ARRAY_SIZE(zone
->lowmem_reserve
); i
++)
2203 seq_printf(m
, ", %lu", zone
->lowmem_reserve
[i
]);
2207 for (i
= 0; i
< ARRAY_SIZE(zone
->pageset
); i
++) {
2208 struct per_cpu_pageset
*pageset
;
2211 pageset
= zone_pcp(zone
, i
);
2212 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2213 if (pageset
->pcp
[j
].count
)
2216 if (j
== ARRAY_SIZE(pageset
->pcp
))
2218 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2220 "\n cpu: %i pcp: %i"
2225 pageset
->pcp
[j
].count
,
2226 pageset
->pcp
[j
].high
,
2227 pageset
->pcp
[j
].batch
);
2233 "\n numa_foreign: %lu"
2234 "\n interleave_hit: %lu"
2235 "\n local_node: %lu"
2236 "\n other_node: %lu",
2239 pageset
->numa_foreign
,
2240 pageset
->interleave_hit
,
2241 pageset
->local_node
,
2242 pageset
->other_node
);
2246 "\n all_unreclaimable: %u"
2247 "\n prev_priority: %i"
2248 "\n temp_priority: %i"
2249 "\n start_pfn: %lu",
2250 zone
->all_unreclaimable
,
2251 zone
->prev_priority
,
2252 zone
->temp_priority
,
2253 zone
->zone_start_pfn
);
2254 spin_unlock_irqrestore(&zone
->lock
, flags
);
2260 struct seq_operations zoneinfo_op
= {
2261 .start
= frag_start
, /* iterate over all zones. The same as in
2265 .show
= zoneinfo_show
,
2268 static char *vmstat_text
[] = {
2272 "nr_page_table_pages",
2303 "pgscan_kswapd_high",
2304 "pgscan_kswapd_normal",
2305 "pgscan_kswapd_dma32",
2306 "pgscan_kswapd_dma",
2308 "pgscan_direct_high",
2309 "pgscan_direct_normal",
2310 "pgscan_direct_dma32",
2311 "pgscan_direct_dma",
2316 "kswapd_inodesteal",
2324 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
2326 struct page_state
*ps
;
2328 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2331 ps
= kmalloc(sizeof(*ps
), GFP_KERNEL
);
2334 return ERR_PTR(-ENOMEM
);
2335 get_full_page_state(ps
);
2336 ps
->pgpgin
/= 2; /* sectors -> kbytes */
2338 return (unsigned long *)ps
+ *pos
;
2341 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2344 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2346 return (unsigned long *)m
->private + *pos
;
2349 static int vmstat_show(struct seq_file
*m
, void *arg
)
2351 unsigned long *l
= arg
;
2352 unsigned long off
= l
- (unsigned long *)m
->private;
2354 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
2358 static void vmstat_stop(struct seq_file
*m
, void *arg
)
2364 struct seq_operations vmstat_op
= {
2365 .start
= vmstat_start
,
2366 .next
= vmstat_next
,
2367 .stop
= vmstat_stop
,
2368 .show
= vmstat_show
,
2371 #endif /* CONFIG_PROC_FS */
2373 #ifdef CONFIG_HOTPLUG_CPU
2374 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2375 unsigned long action
, void *hcpu
)
2377 int cpu
= (unsigned long)hcpu
;
2379 unsigned long *src
, *dest
;
2381 if (action
== CPU_DEAD
) {
2384 /* Drain local pagecache count. */
2385 count
= &per_cpu(nr_pagecache_local
, cpu
);
2386 atomic_add(*count
, &nr_pagecache
);
2388 local_irq_disable();
2391 /* Add dead cpu's page_states to our own. */
2392 dest
= (unsigned long *)&__get_cpu_var(page_states
);
2393 src
= (unsigned long *)&per_cpu(page_states
, cpu
);
2395 for (i
= 0; i
< sizeof(struct page_state
)/sizeof(unsigned long);
2405 #endif /* CONFIG_HOTPLUG_CPU */
2407 void __init
page_alloc_init(void)
2409 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2413 * setup_per_zone_lowmem_reserve - called whenever
2414 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2415 * has a correct pages reserved value, so an adequate number of
2416 * pages are left in the zone after a successful __alloc_pages().
2418 static void setup_per_zone_lowmem_reserve(void)
2420 struct pglist_data
*pgdat
;
2423 for_each_pgdat(pgdat
) {
2424 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2425 struct zone
*zone
= pgdat
->node_zones
+ j
;
2426 unsigned long present_pages
= zone
->present_pages
;
2428 zone
->lowmem_reserve
[j
] = 0;
2430 for (idx
= j
-1; idx
>= 0; idx
--) {
2431 struct zone
*lower_zone
;
2433 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2434 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2436 lower_zone
= pgdat
->node_zones
+ idx
;
2437 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2438 sysctl_lowmem_reserve_ratio
[idx
];
2439 present_pages
+= lower_zone
->present_pages
;
2446 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2447 * that the pages_{min,low,high} values for each zone are set correctly
2448 * with respect to min_free_kbytes.
2450 void setup_per_zone_pages_min(void)
2452 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2453 unsigned long lowmem_pages
= 0;
2455 unsigned long flags
;
2457 /* Calculate total number of !ZONE_HIGHMEM pages */
2458 for_each_zone(zone
) {
2459 if (!is_highmem(zone
))
2460 lowmem_pages
+= zone
->present_pages
;
2463 for_each_zone(zone
) {
2465 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2466 tmp
= (pages_min
* zone
->present_pages
) / lowmem_pages
;
2467 if (is_highmem(zone
)) {
2469 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2470 * need highmem pages, so cap pages_min to a small
2473 * The (pages_high-pages_low) and (pages_low-pages_min)
2474 * deltas controls asynch page reclaim, and so should
2475 * not be capped for highmem.
2479 min_pages
= zone
->present_pages
/ 1024;
2480 if (min_pages
< SWAP_CLUSTER_MAX
)
2481 min_pages
= SWAP_CLUSTER_MAX
;
2482 if (min_pages
> 128)
2484 zone
->pages_min
= min_pages
;
2487 * If it's a lowmem zone, reserve a number of pages
2488 * proportionate to the zone's size.
2490 zone
->pages_min
= tmp
;
2493 zone
->pages_low
= zone
->pages_min
+ tmp
/ 4;
2494 zone
->pages_high
= zone
->pages_min
+ tmp
/ 2;
2495 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2500 * Initialise min_free_kbytes.
2502 * For small machines we want it small (128k min). For large machines
2503 * we want it large (64MB max). But it is not linear, because network
2504 * bandwidth does not increase linearly with machine size. We use
2506 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2507 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2523 static int __init
init_per_zone_pages_min(void)
2525 unsigned long lowmem_kbytes
;
2527 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2529 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2530 if (min_free_kbytes
< 128)
2531 min_free_kbytes
= 128;
2532 if (min_free_kbytes
> 65536)
2533 min_free_kbytes
= 65536;
2534 setup_per_zone_pages_min();
2535 setup_per_zone_lowmem_reserve();
2538 module_init(init_per_zone_pages_min
)
2541 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2542 * that we can call two helper functions whenever min_free_kbytes
2545 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2546 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2548 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2549 setup_per_zone_pages_min();
2554 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2555 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2556 * whenever sysctl_lowmem_reserve_ratio changes.
2558 * The reserve ratio obviously has absolutely no relation with the
2559 * pages_min watermarks. The lowmem reserve ratio can only make sense
2560 * if in function of the boot time zone sizes.
2562 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2563 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2565 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2566 setup_per_zone_lowmem_reserve();
2570 __initdata
int hashdist
= HASHDIST_DEFAULT
;
2573 static int __init
set_hashdist(char *str
)
2577 hashdist
= simple_strtoul(str
, &str
, 0);
2580 __setup("hashdist=", set_hashdist
);
2584 * allocate a large system hash table from bootmem
2585 * - it is assumed that the hash table must contain an exact power-of-2
2586 * quantity of entries
2587 * - limit is the number of hash buckets, not the total allocation size
2589 void *__init
alloc_large_system_hash(const char *tablename
,
2590 unsigned long bucketsize
,
2591 unsigned long numentries
,
2594 unsigned int *_hash_shift
,
2595 unsigned int *_hash_mask
,
2596 unsigned long limit
)
2598 unsigned long long max
= limit
;
2599 unsigned long log2qty
, size
;
2602 /* allow the kernel cmdline to have a say */
2604 /* round applicable memory size up to nearest megabyte */
2605 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2606 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2607 numentries
>>= 20 - PAGE_SHIFT
;
2608 numentries
<<= 20 - PAGE_SHIFT
;
2610 /* limit to 1 bucket per 2^scale bytes of low memory */
2611 if (scale
> PAGE_SHIFT
)
2612 numentries
>>= (scale
- PAGE_SHIFT
);
2614 numentries
<<= (PAGE_SHIFT
- scale
);
2616 /* rounded up to nearest power of 2 in size */
2617 numentries
= 1UL << (long_log2(numentries
) + 1);
2619 /* limit allocation size to 1/16 total memory by default */
2621 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2622 do_div(max
, bucketsize
);
2625 if (numentries
> max
)
2628 log2qty
= long_log2(numentries
);
2631 size
= bucketsize
<< log2qty
;
2632 if (flags
& HASH_EARLY
)
2633 table
= alloc_bootmem(size
);
2635 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
2637 unsigned long order
;
2638 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
2640 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
2642 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2645 panic("Failed to allocate %s hash table\n", tablename
);
2647 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2650 long_log2(size
) - PAGE_SHIFT
,
2654 *_hash_shift
= log2qty
;
2656 *_hash_mask
= (1 << log2qty
) - 1;