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
;
57 * results with 256, 32 in the lowmem_reserve sysctl:
58 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
59 * 1G machine -> (16M dma, 784M normal, 224M high)
60 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
61 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
62 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
64 * TBD: should special case ZONE_DMA32 machines here - in those we normally
65 * don't need any ZONE_NORMAL reservation
67 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = { 256, 256, 32 };
69 EXPORT_SYMBOL(totalram_pages
);
72 * Used by page_zone() to look up the address of the struct zone whose
73 * id is encoded in the upper bits of page->flags
75 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
76 EXPORT_SYMBOL(zone_table
);
78 static char *zone_names
[MAX_NR_ZONES
] = { "DMA", "DMA32", "Normal", "HighMem" };
79 int min_free_kbytes
= 1024;
81 unsigned long __initdata nr_kernel_pages
;
82 unsigned long __initdata nr_all_pages
;
84 #ifdef CONFIG_DEBUG_VM
85 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
89 unsigned long pfn
= page_to_pfn(page
);
92 seq
= zone_span_seqbegin(zone
);
93 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
95 else if (pfn
< zone
->zone_start_pfn
)
97 } while (zone_span_seqretry(zone
, seq
));
102 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
104 #ifdef CONFIG_HOLES_IN_ZONE
105 if (!pfn_valid(page_to_pfn(page
)))
108 if (zone
!= page_zone(page
))
114 * Temporary debugging check for pages not lying within a given zone.
116 static int bad_range(struct zone
*zone
, struct page
*page
)
118 if (page_outside_zone_boundaries(zone
, page
))
120 if (!page_is_consistent(zone
, page
))
127 static inline int bad_range(struct zone
*zone
, struct page
*page
)
133 static void bad_page(const char *function
, struct page
*page
)
135 printk(KERN_EMERG
"Bad page state at %s (in process '%s', page %p)\n",
136 function
, current
->comm
, page
);
137 printk(KERN_EMERG
"flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
138 (int)(2*sizeof(unsigned long)), (unsigned long)page
->flags
,
139 page
->mapping
, page_mapcount(page
), page_count(page
));
140 printk(KERN_EMERG
"Backtrace:\n");
142 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n");
143 page
->flags
&= ~(1 << PG_lru
|
152 set_page_count(page
, 0);
153 reset_page_mapcount(page
);
154 page
->mapping
= NULL
;
155 add_taint(TAINT_BAD_PAGE
);
159 * Higher-order pages are called "compound pages". They are structured thusly:
161 * The first PAGE_SIZE page is called the "head page".
163 * The remaining PAGE_SIZE pages are called "tail pages".
165 * All pages have PG_compound set. All pages have their ->private pointing at
166 * the head page (even the head page has this).
168 * The first tail page's ->mapping, if non-zero, holds the address of the
169 * compound page's put_page() function.
171 * The order of the allocation is stored in the first tail page's ->index
172 * This is only for debug at present. This usage means that zero-order pages
173 * may not be compound.
175 static void prep_compound_page(struct page
*page
, unsigned long order
)
178 int nr_pages
= 1 << order
;
180 page
[1].mapping
= NULL
;
181 page
[1].index
= order
;
182 for (i
= 0; i
< nr_pages
; i
++) {
183 struct page
*p
= page
+ i
;
186 set_page_private(p
, (unsigned long)page
);
190 static void destroy_compound_page(struct page
*page
, unsigned long order
)
193 int nr_pages
= 1 << order
;
195 if (!PageCompound(page
))
198 if (page
[1].index
!= order
)
199 bad_page(__FUNCTION__
, page
);
201 for (i
= 0; i
< nr_pages
; i
++) {
202 struct page
*p
= page
+ i
;
204 if (!PageCompound(p
))
205 bad_page(__FUNCTION__
, page
);
206 if (page_private(p
) != (unsigned long)page
)
207 bad_page(__FUNCTION__
, 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
;
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(const char *function
, struct page
*page
)
351 if (unlikely(page_mapcount(page
) |
352 (page
->mapping
!= NULL
) |
353 (page_count(page
) != 0) |
363 1 << PG_reserved
))))
364 bad_page(function
, page
);
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(__FUNCTION__
, page
+ i
);
427 list_add(&page
->lru
, &list
);
428 mod_page_state(pgfree
, 1 << order
);
429 kernel_map_pages(page
, 1<<order
, 0);
430 local_irq_save(flags
);
431 free_pages_bulk(page_zone(page
), 1, &list
, order
);
432 local_irq_restore(flags
);
437 * The order of subdivision here is critical for the IO subsystem.
438 * Please do not alter this order without good reasons and regression
439 * testing. Specifically, as large blocks of memory are subdivided,
440 * the order in which smaller blocks are delivered depends on the order
441 * they're subdivided in this function. This is the primary factor
442 * influencing the order in which pages are delivered to the IO
443 * subsystem according to empirical testing, and this is also justified
444 * by considering the behavior of a buddy system containing a single
445 * large block of memory acted on by a series of small allocations.
446 * This behavior is a critical factor in sglist merging's success.
450 static inline void expand(struct zone
*zone
, struct page
*page
,
451 int low
, int high
, struct free_area
*area
)
453 unsigned long size
= 1 << high
;
459 BUG_ON(bad_range(zone
, &page
[size
]));
460 list_add(&page
[size
].lru
, &area
->free_list
);
462 set_page_order(&page
[size
], high
);
467 * This page is about to be returned from the page allocator
469 static int prep_new_page(struct page
*page
, int order
)
471 if (unlikely(page_mapcount(page
) |
472 (page
->mapping
!= NULL
) |
473 (page_count(page
) != 0) |
484 1 << PG_reserved
))))
485 bad_page(__FUNCTION__
, page
);
488 * For now, we report if PG_reserved was found set, but do not
489 * clear it, and do not allocate the page: as a safety net.
491 if (PageReserved(page
))
494 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
495 1 << PG_referenced
| 1 << PG_arch_1
|
496 1 << PG_checked
| 1 << PG_mappedtodisk
);
497 set_page_private(page
, 0);
498 set_page_refs(page
, order
);
499 kernel_map_pages(page
, 1 << order
, 1);
504 * Do the hard work of removing an element from the buddy allocator.
505 * Call me with the zone->lock already held.
507 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
509 struct free_area
* area
;
510 unsigned int current_order
;
513 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
514 area
= zone
->free_area
+ current_order
;
515 if (list_empty(&area
->free_list
))
518 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
519 list_del(&page
->lru
);
520 rmv_page_order(page
);
522 zone
->free_pages
-= 1UL << order
;
523 expand(zone
, page
, order
, current_order
, area
);
531 * Obtain a specified number of elements from the buddy allocator, all under
532 * a single hold of the lock, for efficiency. Add them to the supplied list.
533 * Returns the number of new pages which were placed at *list.
535 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
536 unsigned long count
, struct list_head
*list
)
540 spin_lock(&zone
->lock
);
541 for (i
= 0; i
< count
; ++i
) {
542 struct page
*page
= __rmqueue(zone
, order
);
543 if (unlikely(page
== NULL
))
545 list_add_tail(&page
->lru
, list
);
547 spin_unlock(&zone
->lock
);
552 /* Called from the slab reaper to drain remote pagesets */
553 void drain_remote_pages(void)
559 local_irq_save(flags
);
560 for_each_zone(zone
) {
561 struct per_cpu_pageset
*pset
;
563 /* Do not drain local pagesets */
564 if (zone
->zone_pgdat
->node_id
== numa_node_id())
567 pset
= zone
->pageset
[smp_processor_id()];
568 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
569 struct per_cpu_pages
*pcp
;
573 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
577 local_irq_restore(flags
);
581 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
582 static void __drain_pages(unsigned int cpu
)
588 for_each_zone(zone
) {
589 struct per_cpu_pageset
*pset
;
591 pset
= zone_pcp(zone
, cpu
);
592 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
593 struct per_cpu_pages
*pcp
;
596 local_irq_save(flags
);
597 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
599 local_irq_restore(flags
);
603 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
607 void mark_free_pages(struct zone
*zone
)
609 unsigned long zone_pfn
, flags
;
611 struct list_head
*curr
;
613 if (!zone
->spanned_pages
)
616 spin_lock_irqsave(&zone
->lock
, flags
);
617 for (zone_pfn
= 0; zone_pfn
< zone
->spanned_pages
; ++zone_pfn
)
618 ClearPageNosaveFree(pfn_to_page(zone_pfn
+ zone
->zone_start_pfn
));
620 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
621 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
622 unsigned long start_pfn
, i
;
624 start_pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
626 for (i
=0; i
< (1<<order
); i
++)
627 SetPageNosaveFree(pfn_to_page(start_pfn
+i
));
629 spin_unlock_irqrestore(&zone
->lock
, flags
);
633 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
635 void drain_local_pages(void)
639 local_irq_save(flags
);
640 __drain_pages(smp_processor_id());
641 local_irq_restore(flags
);
643 #endif /* CONFIG_PM */
645 static void zone_statistics(struct zonelist
*zonelist
, struct zone
*z
)
650 pg_data_t
*pg
= z
->zone_pgdat
;
651 pg_data_t
*orig
= zonelist
->zones
[0]->zone_pgdat
;
652 struct per_cpu_pageset
*p
;
654 local_irq_save(flags
);
655 cpu
= smp_processor_id();
661 zone_pcp(zonelist
->zones
[0], cpu
)->numa_foreign
++;
663 if (pg
== NODE_DATA(numa_node_id()))
667 local_irq_restore(flags
);
672 * Free a 0-order page
674 static void FASTCALL(free_hot_cold_page(struct page
*page
, int cold
));
675 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
677 struct zone
*zone
= page_zone(page
);
678 struct per_cpu_pages
*pcp
;
681 arch_free_page(page
, 0);
684 page
->mapping
= NULL
;
685 if (free_pages_check(__FUNCTION__
, page
))
688 inc_page_state(pgfree
);
689 kernel_map_pages(page
, 1, 0);
691 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
692 local_irq_save(flags
);
693 list_add(&page
->lru
, &pcp
->list
);
695 if (pcp
->count
>= pcp
->high
)
696 pcp
->count
-= free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
697 local_irq_restore(flags
);
701 void fastcall
free_hot_page(struct page
*page
)
703 free_hot_cold_page(page
, 0);
706 void fastcall
free_cold_page(struct page
*page
)
708 free_hot_cold_page(page
, 1);
711 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
715 BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
716 for(i
= 0; i
< (1 << order
); i
++)
717 clear_highpage(page
+ i
);
721 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
722 * we cheat by calling it from here, in the order > 0 path. Saves a branch
726 buffered_rmqueue(struct zone
*zone
, int order
, gfp_t gfp_flags
)
730 int cold
= !!(gfp_flags
& __GFP_COLD
);
734 struct per_cpu_pages
*pcp
;
737 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
738 local_irq_save(flags
);
740 pcp
->count
+= rmqueue_bulk(zone
, 0,
741 pcp
->batch
, &pcp
->list
);
742 if (likely(pcp
->count
)) {
743 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
744 list_del(&page
->lru
);
747 local_irq_restore(flags
);
750 spin_lock_irqsave(&zone
->lock
, flags
);
751 page
= __rmqueue(zone
, order
);
752 spin_unlock_irqrestore(&zone
->lock
, flags
);
756 BUG_ON(bad_range(zone
, page
));
757 mod_page_state_zone(zone
, pgalloc
, 1 << order
);
758 if (prep_new_page(page
, order
))
761 if (gfp_flags
& __GFP_ZERO
)
762 prep_zero_page(page
, order
, gfp_flags
);
764 if (order
&& (gfp_flags
& __GFP_COMP
))
765 prep_compound_page(page
, order
);
770 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
771 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
772 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
773 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
774 #define ALLOC_HARDER 0x10 /* try to alloc harder */
775 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
776 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
779 * Return 1 if free pages are above 'mark'. This takes into account the order
782 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
783 int classzone_idx
, int alloc_flags
)
785 /* free_pages my go negative - that's OK */
786 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
789 if (alloc_flags
& ALLOC_HIGH
)
791 if (alloc_flags
& ALLOC_HARDER
)
794 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
796 for (o
= 0; o
< order
; o
++) {
797 /* At the next order, this order's pages become unavailable */
798 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
800 /* Require fewer higher order pages to be free */
803 if (free_pages
<= min
)
810 * get_page_from_freeliest goes through the zonelist trying to allocate
814 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
815 struct zonelist
*zonelist
, int alloc_flags
)
817 struct zone
**z
= zonelist
->zones
;
818 struct page
*page
= NULL
;
819 int classzone_idx
= zone_idx(*z
);
822 * Go through the zonelist once, looking for a zone with enough free.
823 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
826 if ((alloc_flags
& ALLOC_CPUSET
) &&
827 !cpuset_zone_allowed(*z
, gfp_mask
))
830 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
832 if (alloc_flags
& ALLOC_WMARK_MIN
)
833 mark
= (*z
)->pages_min
;
834 else if (alloc_flags
& ALLOC_WMARK_LOW
)
835 mark
= (*z
)->pages_low
;
837 mark
= (*z
)->pages_high
;
838 if (!zone_watermark_ok(*z
, order
, mark
,
839 classzone_idx
, alloc_flags
))
843 page
= buffered_rmqueue(*z
, order
, gfp_mask
);
845 zone_statistics(zonelist
, *z
);
848 } while (*(++z
) != NULL
);
853 * This is the 'heart' of the zoned buddy allocator.
855 struct page
* fastcall
856 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
857 struct zonelist
*zonelist
)
859 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
862 struct reclaim_state reclaim_state
;
863 struct task_struct
*p
= current
;
866 int did_some_progress
;
868 might_sleep_if(wait
);
871 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
873 if (unlikely(*z
== NULL
)) {
874 /* Should this ever happen?? */
878 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
879 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
884 wakeup_kswapd(*z
, order
);
888 * OK, we're below the kswapd watermark and have kicked background
889 * reclaim. Now things get more complex, so set up alloc_flags according
890 * to how we want to proceed.
892 * The caller may dip into page reserves a bit more if the caller
893 * cannot run direct reclaim, or if the caller has realtime scheduling
896 alloc_flags
= ALLOC_WMARK_MIN
;
897 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
898 alloc_flags
|= ALLOC_HARDER
;
899 if (gfp_mask
& __GFP_HIGH
)
900 alloc_flags
|= ALLOC_HIGH
;
901 alloc_flags
|= ALLOC_CPUSET
;
904 * Go through the zonelist again. Let __GFP_HIGH and allocations
905 * coming from realtime tasks go deeper into reserves.
907 * This is the last chance, in general, before the goto nopage.
908 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
909 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
911 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
915 /* This allocation should allow future memory freeing. */
917 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
918 && !in_interrupt()) {
919 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
921 /* go through the zonelist yet again, ignoring mins */
922 page
= get_page_from_freelist(gfp_mask
, order
,
923 zonelist
, ALLOC_NO_WATERMARKS
);
926 if (gfp_mask
& __GFP_NOFAIL
) {
927 blk_congestion_wait(WRITE
, HZ
/50);
934 /* Atomic allocations - we can't balance anything */
941 /* We now go into synchronous reclaim */
942 p
->flags
|= PF_MEMALLOC
;
943 reclaim_state
.reclaimed_slab
= 0;
944 p
->reclaim_state
= &reclaim_state
;
946 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
948 p
->reclaim_state
= NULL
;
949 p
->flags
&= ~PF_MEMALLOC
;
953 if (likely(did_some_progress
)) {
954 page
= get_page_from_freelist(gfp_mask
, order
,
955 zonelist
, alloc_flags
);
958 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
960 * Go through the zonelist yet one more time, keep
961 * very high watermark here, this is only to catch
962 * a parallel oom killing, we must fail if we're still
963 * under heavy pressure.
965 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
966 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
970 out_of_memory(gfp_mask
, order
);
975 * Don't let big-order allocations loop unless the caller explicitly
976 * requests that. Wait for some write requests to complete then retry.
978 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
979 * <= 3, but that may not be true in other implementations.
982 if (!(gfp_mask
& __GFP_NORETRY
)) {
983 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
985 if (gfp_mask
& __GFP_NOFAIL
)
989 blk_congestion_wait(WRITE
, HZ
/50);
994 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
995 printk(KERN_WARNING
"%s: page allocation failure."
996 " order:%d, mode:0x%x\n",
997 p
->comm
, order
, gfp_mask
);
1005 EXPORT_SYMBOL(__alloc_pages
);
1008 * Common helper functions.
1010 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1013 page
= alloc_pages(gfp_mask
, order
);
1016 return (unsigned long) page_address(page
);
1019 EXPORT_SYMBOL(__get_free_pages
);
1021 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1026 * get_zeroed_page() returns a 32-bit address, which cannot represent
1029 BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1031 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1033 return (unsigned long) page_address(page
);
1037 EXPORT_SYMBOL(get_zeroed_page
);
1039 void __pagevec_free(struct pagevec
*pvec
)
1041 int i
= pagevec_count(pvec
);
1044 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1047 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1049 if (put_page_testzero(page
)) {
1051 free_hot_page(page
);
1053 __free_pages_ok(page
, order
);
1057 EXPORT_SYMBOL(__free_pages
);
1059 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1062 BUG_ON(!virt_addr_valid((void *)addr
));
1063 __free_pages(virt_to_page((void *)addr
), order
);
1067 EXPORT_SYMBOL(free_pages
);
1070 * Total amount of free (allocatable) RAM:
1072 unsigned int nr_free_pages(void)
1074 unsigned int sum
= 0;
1078 sum
+= zone
->free_pages
;
1083 EXPORT_SYMBOL(nr_free_pages
);
1086 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1088 unsigned int i
, sum
= 0;
1090 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1091 sum
+= pgdat
->node_zones
[i
].free_pages
;
1097 static unsigned int nr_free_zone_pages(int offset
)
1099 /* Just pick one node, since fallback list is circular */
1100 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1101 unsigned int sum
= 0;
1103 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1104 struct zone
**zonep
= zonelist
->zones
;
1107 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1108 unsigned long size
= zone
->present_pages
;
1109 unsigned long high
= zone
->pages_high
;
1118 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1120 unsigned int nr_free_buffer_pages(void)
1122 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1126 * Amount of free RAM allocatable within all zones
1128 unsigned int nr_free_pagecache_pages(void)
1130 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1133 #ifdef CONFIG_HIGHMEM
1134 unsigned int nr_free_highpages (void)
1137 unsigned int pages
= 0;
1139 for_each_pgdat(pgdat
)
1140 pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1147 static void show_node(struct zone
*zone
)
1149 printk("Node %d ", zone
->zone_pgdat
->node_id
);
1152 #define show_node(zone) do { } while (0)
1156 * Accumulate the page_state information across all CPUs.
1157 * The result is unavoidably approximate - it can change
1158 * during and after execution of this function.
1160 static DEFINE_PER_CPU(struct page_state
, page_states
) = {0};
1162 atomic_t nr_pagecache
= ATOMIC_INIT(0);
1163 EXPORT_SYMBOL(nr_pagecache
);
1165 DEFINE_PER_CPU(long, nr_pagecache_local
) = 0;
1168 static void __get_page_state(struct page_state
*ret
, int nr
, cpumask_t
*cpumask
)
1172 memset(ret
, 0, sizeof(*ret
));
1174 cpu
= first_cpu(*cpumask
);
1175 while (cpu
< NR_CPUS
) {
1176 unsigned long *in
, *out
, off
;
1178 in
= (unsigned long *)&per_cpu(page_states
, cpu
);
1180 cpu
= next_cpu(cpu
, *cpumask
);
1183 prefetch(&per_cpu(page_states
, cpu
));
1185 out
= (unsigned long *)ret
;
1186 for (off
= 0; off
< nr
; off
++)
1191 void get_page_state_node(struct page_state
*ret
, int node
)
1194 cpumask_t mask
= node_to_cpumask(node
);
1196 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1197 nr
/= sizeof(unsigned long);
1199 __get_page_state(ret
, nr
+1, &mask
);
1202 void get_page_state(struct page_state
*ret
)
1205 cpumask_t mask
= CPU_MASK_ALL
;
1207 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1208 nr
/= sizeof(unsigned long);
1210 __get_page_state(ret
, nr
+ 1, &mask
);
1213 void get_full_page_state(struct page_state
*ret
)
1215 cpumask_t mask
= CPU_MASK_ALL
;
1217 __get_page_state(ret
, sizeof(*ret
) / sizeof(unsigned long), &mask
);
1220 unsigned long __read_page_state(unsigned long offset
)
1222 unsigned long ret
= 0;
1228 in
= (unsigned long)&per_cpu(page_states
, cpu
) + offset
;
1229 ret
+= *((unsigned long *)in
);
1234 void __mod_page_state(unsigned long offset
, unsigned long delta
)
1236 unsigned long flags
;
1239 local_irq_save(flags
);
1240 ptr
= &__get_cpu_var(page_states
);
1241 *(unsigned long*)(ptr
+ offset
) += delta
;
1242 local_irq_restore(flags
);
1245 EXPORT_SYMBOL(__mod_page_state
);
1247 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
1248 unsigned long *free
, struct pglist_data
*pgdat
)
1250 struct zone
*zones
= pgdat
->node_zones
;
1256 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1257 *active
+= zones
[i
].nr_active
;
1258 *inactive
+= zones
[i
].nr_inactive
;
1259 *free
+= zones
[i
].free_pages
;
1263 void get_zone_counts(unsigned long *active
,
1264 unsigned long *inactive
, unsigned long *free
)
1266 struct pglist_data
*pgdat
;
1271 for_each_pgdat(pgdat
) {
1272 unsigned long l
, m
, n
;
1273 __get_zone_counts(&l
, &m
, &n
, pgdat
);
1280 void si_meminfo(struct sysinfo
*val
)
1282 val
->totalram
= totalram_pages
;
1284 val
->freeram
= nr_free_pages();
1285 val
->bufferram
= nr_blockdev_pages();
1286 #ifdef CONFIG_HIGHMEM
1287 val
->totalhigh
= totalhigh_pages
;
1288 val
->freehigh
= nr_free_highpages();
1293 val
->mem_unit
= PAGE_SIZE
;
1296 EXPORT_SYMBOL(si_meminfo
);
1299 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1301 pg_data_t
*pgdat
= NODE_DATA(nid
);
1303 val
->totalram
= pgdat
->node_present_pages
;
1304 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1305 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1306 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1307 val
->mem_unit
= PAGE_SIZE
;
1311 #define K(x) ((x) << (PAGE_SHIFT-10))
1314 * Show free area list (used inside shift_scroll-lock stuff)
1315 * We also calculate the percentage fragmentation. We do this by counting the
1316 * memory on each free list with the exception of the first item on the list.
1318 void show_free_areas(void)
1320 struct page_state ps
;
1321 int cpu
, temperature
;
1322 unsigned long active
;
1323 unsigned long inactive
;
1327 for_each_zone(zone
) {
1329 printk("%s per-cpu:", zone
->name
);
1331 if (!zone
->present_pages
) {
1337 for_each_online_cpu(cpu
) {
1338 struct per_cpu_pageset
*pageset
;
1340 pageset
= zone_pcp(zone
, cpu
);
1342 for (temperature
= 0; temperature
< 2; temperature
++)
1343 printk("cpu %d %s: high %d, batch %d used:%d\n",
1345 temperature
? "cold" : "hot",
1346 pageset
->pcp
[temperature
].high
,
1347 pageset
->pcp
[temperature
].batch
,
1348 pageset
->pcp
[temperature
].count
);
1352 get_page_state(&ps
);
1353 get_zone_counts(&active
, &inactive
, &free
);
1355 printk("Free pages: %11ukB (%ukB HighMem)\n",
1357 K(nr_free_highpages()));
1359 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1360 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1369 ps
.nr_page_table_pages
);
1371 for_each_zone(zone
) {
1383 " pages_scanned:%lu"
1384 " all_unreclaimable? %s"
1387 K(zone
->free_pages
),
1390 K(zone
->pages_high
),
1392 K(zone
->nr_inactive
),
1393 K(zone
->present_pages
),
1394 zone
->pages_scanned
,
1395 (zone
->all_unreclaimable
? "yes" : "no")
1397 printk("lowmem_reserve[]:");
1398 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1399 printk(" %lu", zone
->lowmem_reserve
[i
]);
1403 for_each_zone(zone
) {
1404 unsigned long nr
, flags
, order
, total
= 0;
1407 printk("%s: ", zone
->name
);
1408 if (!zone
->present_pages
) {
1413 spin_lock_irqsave(&zone
->lock
, flags
);
1414 for (order
= 0; order
< MAX_ORDER
; order
++) {
1415 nr
= zone
->free_area
[order
].nr_free
;
1416 total
+= nr
<< order
;
1417 printk("%lu*%lukB ", nr
, K(1UL) << order
);
1419 spin_unlock_irqrestore(&zone
->lock
, flags
);
1420 printk("= %lukB\n", K(total
));
1423 show_swap_cache_info();
1427 * Builds allocation fallback zone lists.
1429 static int __init
build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
, int j
, int k
)
1436 zone
= pgdat
->node_zones
+ ZONE_HIGHMEM
;
1437 if (zone
->present_pages
) {
1438 #ifndef CONFIG_HIGHMEM
1441 zonelist
->zones
[j
++] = zone
;
1444 zone
= pgdat
->node_zones
+ ZONE_NORMAL
;
1445 if (zone
->present_pages
)
1446 zonelist
->zones
[j
++] = zone
;
1448 zone
= pgdat
->node_zones
+ ZONE_DMA32
;
1449 if (zone
->present_pages
)
1450 zonelist
->zones
[j
++] = zone
;
1452 zone
= pgdat
->node_zones
+ ZONE_DMA
;
1453 if (zone
->present_pages
)
1454 zonelist
->zones
[j
++] = zone
;
1460 static inline int highest_zone(int zone_bits
)
1462 int res
= ZONE_NORMAL
;
1463 if (zone_bits
& (__force
int)__GFP_HIGHMEM
)
1465 if (zone_bits
& (__force
int)__GFP_DMA32
)
1467 if (zone_bits
& (__force
int)__GFP_DMA
)
1473 #define MAX_NODE_LOAD (num_online_nodes())
1474 static int __initdata node_load
[MAX_NUMNODES
];
1476 * find_next_best_node - find the next node that should appear in a given node's fallback list
1477 * @node: node whose fallback list we're appending
1478 * @used_node_mask: nodemask_t of already used nodes
1480 * We use a number of factors to determine which is the next node that should
1481 * appear on a given node's fallback list. The node should not have appeared
1482 * already in @node's fallback list, and it should be the next closest node
1483 * according to the distance array (which contains arbitrary distance values
1484 * from each node to each node in the system), and should also prefer nodes
1485 * with no CPUs, since presumably they'll have very little allocation pressure
1486 * on them otherwise.
1487 * It returns -1 if no node is found.
1489 static int __init
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1492 int min_val
= INT_MAX
;
1495 for_each_online_node(i
) {
1498 /* Start from local node */
1499 n
= (node
+i
) % num_online_nodes();
1501 /* Don't want a node to appear more than once */
1502 if (node_isset(n
, *used_node_mask
))
1505 /* Use the local node if we haven't already */
1506 if (!node_isset(node
, *used_node_mask
)) {
1511 /* Use the distance array to find the distance */
1512 val
= node_distance(node
, n
);
1514 /* Give preference to headless and unused nodes */
1515 tmp
= node_to_cpumask(n
);
1516 if (!cpus_empty(tmp
))
1517 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1519 /* Slight preference for less loaded node */
1520 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1521 val
+= node_load
[n
];
1523 if (val
< min_val
) {
1530 node_set(best_node
, *used_node_mask
);
1535 static void __init
build_zonelists(pg_data_t
*pgdat
)
1537 int i
, j
, k
, node
, local_node
;
1538 int prev_node
, load
;
1539 struct zonelist
*zonelist
;
1540 nodemask_t used_mask
;
1542 /* initialize zonelists */
1543 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1544 zonelist
= pgdat
->node_zonelists
+ i
;
1545 zonelist
->zones
[0] = NULL
;
1548 /* NUMA-aware ordering of nodes */
1549 local_node
= pgdat
->node_id
;
1550 load
= num_online_nodes();
1551 prev_node
= local_node
;
1552 nodes_clear(used_mask
);
1553 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1555 * We don't want to pressure a particular node.
1556 * So adding penalty to the first node in same
1557 * distance group to make it round-robin.
1559 if (node_distance(local_node
, node
) !=
1560 node_distance(local_node
, prev_node
))
1561 node_load
[node
] += load
;
1564 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1565 zonelist
= pgdat
->node_zonelists
+ i
;
1566 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1568 k
= highest_zone(i
);
1570 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1571 zonelist
->zones
[j
] = NULL
;
1576 #else /* CONFIG_NUMA */
1578 static void __init
build_zonelists(pg_data_t
*pgdat
)
1580 int i
, j
, k
, node
, local_node
;
1582 local_node
= pgdat
->node_id
;
1583 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1584 struct zonelist
*zonelist
;
1586 zonelist
= pgdat
->node_zonelists
+ i
;
1589 k
= highest_zone(i
);
1590 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1592 * Now we build the zonelist so that it contains the zones
1593 * of all the other nodes.
1594 * We don't want to pressure a particular node, so when
1595 * building the zones for node N, we make sure that the
1596 * zones coming right after the local ones are those from
1597 * node N+1 (modulo N)
1599 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1600 if (!node_online(node
))
1602 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1604 for (node
= 0; node
< local_node
; node
++) {
1605 if (!node_online(node
))
1607 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1610 zonelist
->zones
[j
] = NULL
;
1614 #endif /* CONFIG_NUMA */
1616 void __init
build_all_zonelists(void)
1620 for_each_online_node(i
)
1621 build_zonelists(NODE_DATA(i
));
1622 printk("Built %i zonelists\n", num_online_nodes());
1623 cpuset_init_current_mems_allowed();
1627 * Helper functions to size the waitqueue hash table.
1628 * Essentially these want to choose hash table sizes sufficiently
1629 * large so that collisions trying to wait on pages are rare.
1630 * But in fact, the number of active page waitqueues on typical
1631 * systems is ridiculously low, less than 200. So this is even
1632 * conservative, even though it seems large.
1634 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1635 * waitqueues, i.e. the size of the waitq table given the number of pages.
1637 #define PAGES_PER_WAITQUEUE 256
1639 static inline unsigned long wait_table_size(unsigned long pages
)
1641 unsigned long size
= 1;
1643 pages
/= PAGES_PER_WAITQUEUE
;
1645 while (size
< pages
)
1649 * Once we have dozens or even hundreds of threads sleeping
1650 * on IO we've got bigger problems than wait queue collision.
1651 * Limit the size of the wait table to a reasonable size.
1653 size
= min(size
, 4096UL);
1655 return max(size
, 4UL);
1659 * This is an integer logarithm so that shifts can be used later
1660 * to extract the more random high bits from the multiplicative
1661 * hash function before the remainder is taken.
1663 static inline unsigned long wait_table_bits(unsigned long size
)
1668 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1670 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1671 unsigned long *zones_size
, unsigned long *zholes_size
)
1673 unsigned long realtotalpages
, totalpages
= 0;
1676 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1677 totalpages
+= zones_size
[i
];
1678 pgdat
->node_spanned_pages
= totalpages
;
1680 realtotalpages
= totalpages
;
1682 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1683 realtotalpages
-= zholes_size
[i
];
1684 pgdat
->node_present_pages
= realtotalpages
;
1685 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1690 * Initially all pages are reserved - free ones are freed
1691 * up by free_all_bootmem() once the early boot process is
1692 * done. Non-atomic initialization, single-pass.
1694 void __devinit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1695 unsigned long start_pfn
)
1698 unsigned long end_pfn
= start_pfn
+ size
;
1701 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++, page
++) {
1702 if (!early_pfn_valid(pfn
))
1704 page
= pfn_to_page(pfn
);
1705 set_page_links(page
, zone
, nid
, pfn
);
1706 set_page_count(page
, 1);
1707 reset_page_mapcount(page
);
1708 SetPageReserved(page
);
1709 INIT_LIST_HEAD(&page
->lru
);
1710 #ifdef WANT_PAGE_VIRTUAL
1711 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1712 if (!is_highmem_idx(zone
))
1713 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1718 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1722 for (order
= 0; order
< MAX_ORDER
; order
++) {
1723 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1724 zone
->free_area
[order
].nr_free
= 0;
1728 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1729 void zonetable_add(struct zone
*zone
, int nid
, int zid
, unsigned long pfn
,
1732 unsigned long snum
= pfn_to_section_nr(pfn
);
1733 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1736 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1738 for (; snum
<= end
; snum
++)
1739 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1742 #ifndef __HAVE_ARCH_MEMMAP_INIT
1743 #define memmap_init(size, nid, zone, start_pfn) \
1744 memmap_init_zone((size), (nid), (zone), (start_pfn))
1747 static int __devinit
zone_batchsize(struct zone
*zone
)
1752 * The per-cpu-pages pools are set to around 1000th of the
1753 * size of the zone. But no more than 1/2 of a meg.
1755 * OK, so we don't know how big the cache is. So guess.
1757 batch
= zone
->present_pages
/ 1024;
1758 if (batch
* PAGE_SIZE
> 512 * 1024)
1759 batch
= (512 * 1024) / PAGE_SIZE
;
1760 batch
/= 4; /* We effectively *= 4 below */
1765 * Clamp the batch to a 2^n - 1 value. Having a power
1766 * of 2 value was found to be more likely to have
1767 * suboptimal cache aliasing properties in some cases.
1769 * For example if 2 tasks are alternately allocating
1770 * batches of pages, one task can end up with a lot
1771 * of pages of one half of the possible page colors
1772 * and the other with pages of the other colors.
1774 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1779 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1781 struct per_cpu_pages
*pcp
;
1783 memset(p
, 0, sizeof(*p
));
1785 pcp
= &p
->pcp
[0]; /* hot */
1787 pcp
->high
= 6 * batch
;
1788 pcp
->batch
= max(1UL, 1 * batch
);
1789 INIT_LIST_HEAD(&pcp
->list
);
1791 pcp
= &p
->pcp
[1]; /* cold*/
1793 pcp
->high
= 2 * batch
;
1794 pcp
->batch
= max(1UL, batch
/2);
1795 INIT_LIST_HEAD(&pcp
->list
);
1800 * Boot pageset table. One per cpu which is going to be used for all
1801 * zones and all nodes. The parameters will be set in such a way
1802 * that an item put on a list will immediately be handed over to
1803 * the buddy list. This is safe since pageset manipulation is done
1804 * with interrupts disabled.
1806 * Some NUMA counter updates may also be caught by the boot pagesets.
1808 * The boot_pagesets must be kept even after bootup is complete for
1809 * unused processors and/or zones. They do play a role for bootstrapping
1810 * hotplugged processors.
1812 * zoneinfo_show() and maybe other functions do
1813 * not check if the processor is online before following the pageset pointer.
1814 * Other parts of the kernel may not check if the zone is available.
1816 static struct per_cpu_pageset
1817 boot_pageset
[NR_CPUS
];
1820 * Dynamically allocate memory for the
1821 * per cpu pageset array in struct zone.
1823 static int __devinit
process_zones(int cpu
)
1825 struct zone
*zone
, *dzone
;
1827 for_each_zone(zone
) {
1829 zone
->pageset
[cpu
] = kmalloc_node(sizeof(struct per_cpu_pageset
),
1830 GFP_KERNEL
, cpu_to_node(cpu
));
1831 if (!zone
->pageset
[cpu
])
1834 setup_pageset(zone
->pageset
[cpu
], zone_batchsize(zone
));
1839 for_each_zone(dzone
) {
1842 kfree(dzone
->pageset
[cpu
]);
1843 dzone
->pageset
[cpu
] = NULL
;
1848 static inline void free_zone_pagesets(int cpu
)
1853 for_each_zone(zone
) {
1854 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1856 zone_pcp(zone
, cpu
) = NULL
;
1862 static int __devinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1863 unsigned long action
,
1866 int cpu
= (long)hcpu
;
1867 int ret
= NOTIFY_OK
;
1870 case CPU_UP_PREPARE
:
1871 if (process_zones(cpu
))
1874 case CPU_UP_CANCELED
:
1876 free_zone_pagesets(cpu
);
1884 static struct notifier_block pageset_notifier
=
1885 { &pageset_cpuup_callback
, NULL
, 0 };
1887 void __init
setup_per_cpu_pageset(void)
1891 /* Initialize per_cpu_pageset for cpu 0.
1892 * A cpuup callback will do this for every cpu
1893 * as it comes online
1895 err
= process_zones(smp_processor_id());
1897 register_cpu_notifier(&pageset_notifier
);
1903 void zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1906 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1909 * The per-page waitqueue mechanism uses hashed waitqueues
1912 zone
->wait_table_size
= wait_table_size(zone_size_pages
);
1913 zone
->wait_table_bits
= wait_table_bits(zone
->wait_table_size
);
1914 zone
->wait_table
= (wait_queue_head_t
*)
1915 alloc_bootmem_node(pgdat
, zone
->wait_table_size
1916 * sizeof(wait_queue_head_t
));
1918 for(i
= 0; i
< zone
->wait_table_size
; ++i
)
1919 init_waitqueue_head(zone
->wait_table
+ i
);
1922 static __devinit
void zone_pcp_init(struct zone
*zone
)
1925 unsigned long batch
= zone_batchsize(zone
);
1927 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1929 /* Early boot. Slab allocator not functional yet */
1930 zone
->pageset
[cpu
] = &boot_pageset
[cpu
];
1931 setup_pageset(&boot_pageset
[cpu
],0);
1933 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1936 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1937 zone
->name
, zone
->present_pages
, batch
);
1940 static __devinit
void init_currently_empty_zone(struct zone
*zone
,
1941 unsigned long zone_start_pfn
, unsigned long size
)
1943 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1945 zone_wait_table_init(zone
, size
);
1946 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1948 zone
->zone_mem_map
= pfn_to_page(zone_start_pfn
);
1949 zone
->zone_start_pfn
= zone_start_pfn
;
1951 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1953 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
1957 * Set up the zone data structures:
1958 * - mark all pages reserved
1959 * - mark all memory queues empty
1960 * - clear the memory bitmaps
1962 static void __init
free_area_init_core(struct pglist_data
*pgdat
,
1963 unsigned long *zones_size
, unsigned long *zholes_size
)
1966 int nid
= pgdat
->node_id
;
1967 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
1969 pgdat_resize_init(pgdat
);
1970 pgdat
->nr_zones
= 0;
1971 init_waitqueue_head(&pgdat
->kswapd_wait
);
1972 pgdat
->kswapd_max_order
= 0;
1974 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
1975 struct zone
*zone
= pgdat
->node_zones
+ j
;
1976 unsigned long size
, realsize
;
1978 realsize
= size
= zones_size
[j
];
1980 realsize
-= zholes_size
[j
];
1982 if (j
< ZONE_HIGHMEM
)
1983 nr_kernel_pages
+= realsize
;
1984 nr_all_pages
+= realsize
;
1986 zone
->spanned_pages
= size
;
1987 zone
->present_pages
= realsize
;
1988 zone
->name
= zone_names
[j
];
1989 spin_lock_init(&zone
->lock
);
1990 spin_lock_init(&zone
->lru_lock
);
1991 zone_seqlock_init(zone
);
1992 zone
->zone_pgdat
= pgdat
;
1993 zone
->free_pages
= 0;
1995 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
1997 zone_pcp_init(zone
);
1998 INIT_LIST_HEAD(&zone
->active_list
);
1999 INIT_LIST_HEAD(&zone
->inactive_list
);
2000 zone
->nr_scan_active
= 0;
2001 zone
->nr_scan_inactive
= 0;
2002 zone
->nr_active
= 0;
2003 zone
->nr_inactive
= 0;
2004 atomic_set(&zone
->reclaim_in_progress
, 0);
2008 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2009 init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2010 zone_start_pfn
+= size
;
2014 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2016 /* Skip empty nodes */
2017 if (!pgdat
->node_spanned_pages
)
2020 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2021 /* ia64 gets its own node_mem_map, before this, without bootmem */
2022 if (!pgdat
->node_mem_map
) {
2026 size
= (pgdat
->node_spanned_pages
+ 1) * sizeof(struct page
);
2027 map
= alloc_remap(pgdat
->node_id
, size
);
2029 map
= alloc_bootmem_node(pgdat
, size
);
2030 pgdat
->node_mem_map
= map
;
2032 #ifdef CONFIG_FLATMEM
2034 * With no DISCONTIG, the global mem_map is just set as node 0's
2036 if (pgdat
== NODE_DATA(0))
2037 mem_map
= NODE_DATA(0)->node_mem_map
;
2039 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2042 void __init
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2043 unsigned long *zones_size
, unsigned long node_start_pfn
,
2044 unsigned long *zholes_size
)
2046 pgdat
->node_id
= nid
;
2047 pgdat
->node_start_pfn
= node_start_pfn
;
2048 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
2050 alloc_node_mem_map(pgdat
);
2052 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2055 #ifndef CONFIG_NEED_MULTIPLE_NODES
2056 static bootmem_data_t contig_bootmem_data
;
2057 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2059 EXPORT_SYMBOL(contig_page_data
);
2062 void __init
free_area_init(unsigned long *zones_size
)
2064 free_area_init_node(0, NODE_DATA(0), zones_size
,
2065 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2068 #ifdef CONFIG_PROC_FS
2070 #include <linux/seq_file.h>
2072 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
2077 for (pgdat
= pgdat_list
; pgdat
&& node
; pgdat
= pgdat
->pgdat_next
)
2083 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2085 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2088 return pgdat
->pgdat_next
;
2091 static void frag_stop(struct seq_file
*m
, void *arg
)
2096 * This walks the free areas for each zone.
2098 static int frag_show(struct seq_file
*m
, void *arg
)
2100 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2102 struct zone
*node_zones
= pgdat
->node_zones
;
2103 unsigned long flags
;
2106 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
2107 if (!zone
->present_pages
)
2110 spin_lock_irqsave(&zone
->lock
, flags
);
2111 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
2112 for (order
= 0; order
< MAX_ORDER
; ++order
)
2113 seq_printf(m
, "%6lu ", zone
->free_area
[order
].nr_free
);
2114 spin_unlock_irqrestore(&zone
->lock
, flags
);
2120 struct seq_operations fragmentation_op
= {
2121 .start
= frag_start
,
2128 * Output information about zones in @pgdat.
2130 static int zoneinfo_show(struct seq_file
*m
, void *arg
)
2132 pg_data_t
*pgdat
= arg
;
2134 struct zone
*node_zones
= pgdat
->node_zones
;
2135 unsigned long flags
;
2137 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; zone
++) {
2140 if (!zone
->present_pages
)
2143 spin_lock_irqsave(&zone
->lock
, flags
);
2144 seq_printf(m
, "Node %d, zone %8s", pgdat
->node_id
, zone
->name
);
2152 "\n scanned %lu (a: %lu i: %lu)"
2161 zone
->pages_scanned
,
2162 zone
->nr_scan_active
, zone
->nr_scan_inactive
,
2163 zone
->spanned_pages
,
2164 zone
->present_pages
);
2166 "\n protection: (%lu",
2167 zone
->lowmem_reserve
[0]);
2168 for (i
= 1; i
< ARRAY_SIZE(zone
->lowmem_reserve
); i
++)
2169 seq_printf(m
, ", %lu", zone
->lowmem_reserve
[i
]);
2173 for (i
= 0; i
< ARRAY_SIZE(zone
->pageset
); i
++) {
2174 struct per_cpu_pageset
*pageset
;
2177 pageset
= zone_pcp(zone
, i
);
2178 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2179 if (pageset
->pcp
[j
].count
)
2182 if (j
== ARRAY_SIZE(pageset
->pcp
))
2184 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2186 "\n cpu: %i pcp: %i"
2191 pageset
->pcp
[j
].count
,
2192 pageset
->pcp
[j
].high
,
2193 pageset
->pcp
[j
].batch
);
2199 "\n numa_foreign: %lu"
2200 "\n interleave_hit: %lu"
2201 "\n local_node: %lu"
2202 "\n other_node: %lu",
2205 pageset
->numa_foreign
,
2206 pageset
->interleave_hit
,
2207 pageset
->local_node
,
2208 pageset
->other_node
);
2212 "\n all_unreclaimable: %u"
2213 "\n prev_priority: %i"
2214 "\n temp_priority: %i"
2215 "\n start_pfn: %lu",
2216 zone
->all_unreclaimable
,
2217 zone
->prev_priority
,
2218 zone
->temp_priority
,
2219 zone
->zone_start_pfn
);
2220 spin_unlock_irqrestore(&zone
->lock
, flags
);
2226 struct seq_operations zoneinfo_op
= {
2227 .start
= frag_start
, /* iterate over all zones. The same as in
2231 .show
= zoneinfo_show
,
2234 static char *vmstat_text
[] = {
2238 "nr_page_table_pages",
2263 "pgscan_kswapd_high",
2264 "pgscan_kswapd_normal",
2266 "pgscan_kswapd_dma",
2267 "pgscan_direct_high",
2268 "pgscan_direct_normal",
2269 "pgscan_direct_dma",
2274 "kswapd_inodesteal",
2282 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
2284 struct page_state
*ps
;
2286 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2289 ps
= kmalloc(sizeof(*ps
), GFP_KERNEL
);
2292 return ERR_PTR(-ENOMEM
);
2293 get_full_page_state(ps
);
2294 ps
->pgpgin
/= 2; /* sectors -> kbytes */
2296 return (unsigned long *)ps
+ *pos
;
2299 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2302 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2304 return (unsigned long *)m
->private + *pos
;
2307 static int vmstat_show(struct seq_file
*m
, void *arg
)
2309 unsigned long *l
= arg
;
2310 unsigned long off
= l
- (unsigned long *)m
->private;
2312 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
2316 static void vmstat_stop(struct seq_file
*m
, void *arg
)
2322 struct seq_operations vmstat_op
= {
2323 .start
= vmstat_start
,
2324 .next
= vmstat_next
,
2325 .stop
= vmstat_stop
,
2326 .show
= vmstat_show
,
2329 #endif /* CONFIG_PROC_FS */
2331 #ifdef CONFIG_HOTPLUG_CPU
2332 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2333 unsigned long action
, void *hcpu
)
2335 int cpu
= (unsigned long)hcpu
;
2337 unsigned long *src
, *dest
;
2339 if (action
== CPU_DEAD
) {
2342 /* Drain local pagecache count. */
2343 count
= &per_cpu(nr_pagecache_local
, cpu
);
2344 atomic_add(*count
, &nr_pagecache
);
2346 local_irq_disable();
2349 /* Add dead cpu's page_states to our own. */
2350 dest
= (unsigned long *)&__get_cpu_var(page_states
);
2351 src
= (unsigned long *)&per_cpu(page_states
, cpu
);
2353 for (i
= 0; i
< sizeof(struct page_state
)/sizeof(unsigned long);
2363 #endif /* CONFIG_HOTPLUG_CPU */
2365 void __init
page_alloc_init(void)
2367 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2371 * setup_per_zone_lowmem_reserve - called whenever
2372 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2373 * has a correct pages reserved value, so an adequate number of
2374 * pages are left in the zone after a successful __alloc_pages().
2376 static void setup_per_zone_lowmem_reserve(void)
2378 struct pglist_data
*pgdat
;
2381 for_each_pgdat(pgdat
) {
2382 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2383 struct zone
*zone
= pgdat
->node_zones
+ j
;
2384 unsigned long present_pages
= zone
->present_pages
;
2386 zone
->lowmem_reserve
[j
] = 0;
2388 for (idx
= j
-1; idx
>= 0; idx
--) {
2389 struct zone
*lower_zone
;
2391 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2392 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2394 lower_zone
= pgdat
->node_zones
+ idx
;
2395 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2396 sysctl_lowmem_reserve_ratio
[idx
];
2397 present_pages
+= lower_zone
->present_pages
;
2404 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2405 * that the pages_{min,low,high} values for each zone are set correctly
2406 * with respect to min_free_kbytes.
2408 void setup_per_zone_pages_min(void)
2410 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2411 unsigned long lowmem_pages
= 0;
2413 unsigned long flags
;
2415 /* Calculate total number of !ZONE_HIGHMEM pages */
2416 for_each_zone(zone
) {
2417 if (!is_highmem(zone
))
2418 lowmem_pages
+= zone
->present_pages
;
2421 for_each_zone(zone
) {
2423 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2424 tmp
= (pages_min
* zone
->present_pages
) / lowmem_pages
;
2425 if (is_highmem(zone
)) {
2427 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2428 * need highmem pages, so cap pages_min to a small
2431 * The (pages_high-pages_low) and (pages_low-pages_min)
2432 * deltas controls asynch page reclaim, and so should
2433 * not be capped for highmem.
2437 min_pages
= zone
->present_pages
/ 1024;
2438 if (min_pages
< SWAP_CLUSTER_MAX
)
2439 min_pages
= SWAP_CLUSTER_MAX
;
2440 if (min_pages
> 128)
2442 zone
->pages_min
= min_pages
;
2445 * If it's a lowmem zone, reserve a number of pages
2446 * proportionate to the zone's size.
2448 zone
->pages_min
= tmp
;
2451 zone
->pages_low
= zone
->pages_min
+ tmp
/ 4;
2452 zone
->pages_high
= zone
->pages_min
+ tmp
/ 2;
2453 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2458 * Initialise min_free_kbytes.
2460 * For small machines we want it small (128k min). For large machines
2461 * we want it large (64MB max). But it is not linear, because network
2462 * bandwidth does not increase linearly with machine size. We use
2464 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2465 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2481 static int __init
init_per_zone_pages_min(void)
2483 unsigned long lowmem_kbytes
;
2485 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2487 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2488 if (min_free_kbytes
< 128)
2489 min_free_kbytes
= 128;
2490 if (min_free_kbytes
> 65536)
2491 min_free_kbytes
= 65536;
2492 setup_per_zone_pages_min();
2493 setup_per_zone_lowmem_reserve();
2496 module_init(init_per_zone_pages_min
)
2499 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2500 * that we can call two helper functions whenever min_free_kbytes
2503 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2504 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2506 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2507 setup_per_zone_pages_min();
2512 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2513 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2514 * whenever sysctl_lowmem_reserve_ratio changes.
2516 * The reserve ratio obviously has absolutely no relation with the
2517 * pages_min watermarks. The lowmem reserve ratio can only make sense
2518 * if in function of the boot time zone sizes.
2520 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2521 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2523 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2524 setup_per_zone_lowmem_reserve();
2528 __initdata
int hashdist
= HASHDIST_DEFAULT
;
2531 static int __init
set_hashdist(char *str
)
2535 hashdist
= simple_strtoul(str
, &str
, 0);
2538 __setup("hashdist=", set_hashdist
);
2542 * allocate a large system hash table from bootmem
2543 * - it is assumed that the hash table must contain an exact power-of-2
2544 * quantity of entries
2545 * - limit is the number of hash buckets, not the total allocation size
2547 void *__init
alloc_large_system_hash(const char *tablename
,
2548 unsigned long bucketsize
,
2549 unsigned long numentries
,
2552 unsigned int *_hash_shift
,
2553 unsigned int *_hash_mask
,
2554 unsigned long limit
)
2556 unsigned long long max
= limit
;
2557 unsigned long log2qty
, size
;
2560 /* allow the kernel cmdline to have a say */
2562 /* round applicable memory size up to nearest megabyte */
2563 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2564 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2565 numentries
>>= 20 - PAGE_SHIFT
;
2566 numentries
<<= 20 - PAGE_SHIFT
;
2568 /* limit to 1 bucket per 2^scale bytes of low memory */
2569 if (scale
> PAGE_SHIFT
)
2570 numentries
>>= (scale
- PAGE_SHIFT
);
2572 numentries
<<= (PAGE_SHIFT
- scale
);
2574 /* rounded up to nearest power of 2 in size */
2575 numentries
= 1UL << (long_log2(numentries
) + 1);
2577 /* limit allocation size to 1/16 total memory by default */
2579 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2580 do_div(max
, bucketsize
);
2583 if (numentries
> max
)
2586 log2qty
= long_log2(numentries
);
2589 size
= bucketsize
<< log2qty
;
2590 if (flags
& HASH_EARLY
)
2591 table
= alloc_bootmem(size
);
2593 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
2595 unsigned long order
;
2596 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
2598 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
2600 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2603 panic("Failed to allocate %s hash table\n", tablename
);
2605 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2608 long_log2(size
) - PAGE_SHIFT
,
2612 *_hash_shift
= log2qty
;
2614 *_hash_mask
= (1 << log2qty
) - 1;