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 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
88 unsigned long pfn
= page_to_pfn(page
);
91 seq
= zone_span_seqbegin(zone
);
92 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
94 else if (pfn
< zone
->zone_start_pfn
)
96 } while (zone_span_seqretry(zone
, seq
));
101 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
103 #ifdef CONFIG_HOLES_IN_ZONE
104 if (!pfn_valid(page_to_pfn(page
)))
107 if (zone
!= page_zone(page
))
113 * Temporary debugging check for pages not lying within a given zone.
115 static int bad_range(struct zone
*zone
, struct page
*page
)
117 if (page_outside_zone_boundaries(zone
, page
))
119 if (!page_is_consistent(zone
, page
))
125 static void bad_page(const char *function
, struct page
*page
)
127 printk(KERN_EMERG
"Bad page state at %s (in process '%s', page %p)\n",
128 function
, current
->comm
, page
);
129 printk(KERN_EMERG
"flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
130 (int)(2*sizeof(unsigned long)), (unsigned long)page
->flags
,
131 page
->mapping
, page_mapcount(page
), page_count(page
));
132 printk(KERN_EMERG
"Backtrace:\n");
134 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n");
135 page
->flags
&= ~(1 << PG_lru
|
144 set_page_count(page
, 0);
145 reset_page_mapcount(page
);
146 page
->mapping
= NULL
;
147 add_taint(TAINT_BAD_PAGE
);
151 * Higher-order pages are called "compound pages". They are structured thusly:
153 * The first PAGE_SIZE page is called the "head page".
155 * The remaining PAGE_SIZE pages are called "tail pages".
157 * All pages have PG_compound set. All pages have their ->private pointing at
158 * the head page (even the head page has this).
160 * The first tail page's ->mapping, if non-zero, holds the address of the
161 * compound page's put_page() function.
163 * The order of the allocation is stored in the first tail page's ->index
164 * This is only for debug at present. This usage means that zero-order pages
165 * may not be compound.
167 static void prep_compound_page(struct page
*page
, unsigned long order
)
170 int nr_pages
= 1 << order
;
172 page
[1].mapping
= NULL
;
173 page
[1].index
= order
;
174 for (i
= 0; i
< nr_pages
; i
++) {
175 struct page
*p
= page
+ i
;
178 set_page_private(p
, (unsigned long)page
);
182 static void destroy_compound_page(struct page
*page
, unsigned long order
)
185 int nr_pages
= 1 << order
;
187 if (!PageCompound(page
))
190 if (page
[1].index
!= order
)
191 bad_page(__FUNCTION__
, page
);
193 for (i
= 0; i
< nr_pages
; i
++) {
194 struct page
*p
= page
+ i
;
196 if (!PageCompound(p
))
197 bad_page(__FUNCTION__
, page
);
198 if (page_private(p
) != (unsigned long)page
)
199 bad_page(__FUNCTION__
, page
);
200 ClearPageCompound(p
);
205 * function for dealing with page's order in buddy system.
206 * zone->lock is already acquired when we use these.
207 * So, we don't need atomic page->flags operations here.
209 static inline unsigned long page_order(struct page
*page
) {
210 return page_private(page
);
213 static inline void set_page_order(struct page
*page
, int order
) {
214 set_page_private(page
, order
);
215 __SetPagePrivate(page
);
218 static inline void rmv_page_order(struct page
*page
)
220 __ClearPagePrivate(page
);
221 set_page_private(page
, 0);
225 * Locate the struct page for both the matching buddy in our
226 * pair (buddy1) and the combined O(n+1) page they form (page).
228 * 1) Any buddy B1 will have an order O twin B2 which satisfies
229 * the following equation:
231 * For example, if the starting buddy (buddy2) is #8 its order
233 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
235 * 2) Any buddy B will have an order O+1 parent P which
236 * satisfies the following equation:
239 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
241 static inline struct page
*
242 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
244 unsigned long buddy_idx
= page_idx
^ (1 << order
);
246 return page
+ (buddy_idx
- page_idx
);
249 static inline unsigned long
250 __find_combined_index(unsigned long page_idx
, unsigned int order
)
252 return (page_idx
& ~(1 << order
));
256 * This function checks whether a page is free && is the buddy
257 * we can do coalesce a page and its buddy if
258 * (a) the buddy is free &&
259 * (b) the buddy is on the buddy system &&
260 * (c) a page and its buddy have the same order.
261 * for recording page's order, we use page_private(page) and PG_private.
264 static inline int page_is_buddy(struct page
*page
, int order
)
266 if (PagePrivate(page
) &&
267 (page_order(page
) == order
) &&
268 page_count(page
) == 0)
274 * Freeing function for a buddy system allocator.
276 * The concept of a buddy system is to maintain direct-mapped table
277 * (containing bit values) for memory blocks of various "orders".
278 * The bottom level table contains the map for the smallest allocatable
279 * units of memory (here, pages), and each level above it describes
280 * pairs of units from the levels below, hence, "buddies".
281 * At a high level, all that happens here is marking the table entry
282 * at the bottom level available, and propagating the changes upward
283 * as necessary, plus some accounting needed to play nicely with other
284 * parts of the VM system.
285 * At each level, we keep a list of pages, which are heads of continuous
286 * free pages of length of (1 << order) and marked with PG_Private.Page's
287 * order is recorded in page_private(page) field.
288 * So when we are allocating or freeing one, we can derive the state of the
289 * other. That is, if we allocate a small block, and both were
290 * free, the remainder of the region must be split into blocks.
291 * If a block is freed, and its buddy is also free, then this
292 * triggers coalescing into a block of larger size.
297 static inline void __free_pages_bulk (struct page
*page
,
298 struct zone
*zone
, unsigned int order
)
300 unsigned long page_idx
;
301 int order_size
= 1 << order
;
304 destroy_compound_page(page
, order
);
306 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
308 BUG_ON(page_idx
& (order_size
- 1));
309 BUG_ON(bad_range(zone
, page
));
311 zone
->free_pages
+= order_size
;
312 while (order
< MAX_ORDER
-1) {
313 unsigned long combined_idx
;
314 struct free_area
*area
;
317 combined_idx
= __find_combined_index(page_idx
, order
);
318 buddy
= __page_find_buddy(page
, page_idx
, order
);
320 if (bad_range(zone
, buddy
))
322 if (!page_is_buddy(buddy
, order
))
323 break; /* Move the buddy up one level. */
324 list_del(&buddy
->lru
);
325 area
= zone
->free_area
+ order
;
327 rmv_page_order(buddy
);
328 page
= page
+ (combined_idx
- page_idx
);
329 page_idx
= combined_idx
;
332 set_page_order(page
, order
);
333 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
334 zone
->free_area
[order
].nr_free
++;
337 static inline int free_pages_check(const char *function
, struct page
*page
)
339 if ( page_mapcount(page
) ||
340 page
->mapping
!= NULL
||
341 page_count(page
) != 0 ||
352 bad_page(function
, page
);
354 __ClearPageDirty(page
);
356 * For now, we report if PG_reserved was found set, but do not
357 * clear it, and do not free the page. But we shall soon need
358 * to do more, for when the ZERO_PAGE count wraps negative.
360 return PageReserved(page
);
364 * Frees a list of pages.
365 * Assumes all pages on list are in same zone, and of same order.
366 * count is the number of pages to free.
368 * If the zone was previously in an "all pages pinned" state then look to
369 * see if this freeing clears that state.
371 * And clear the zone's pages_scanned counter, to hold off the "all pages are
372 * pinned" detection logic.
375 free_pages_bulk(struct zone
*zone
, int count
,
376 struct list_head
*list
, unsigned int order
)
379 struct page
*page
= NULL
;
382 spin_lock_irqsave(&zone
->lock
, flags
);
383 zone
->all_unreclaimable
= 0;
384 zone
->pages_scanned
= 0;
385 while (!list_empty(list
) && count
--) {
386 page
= list_entry(list
->prev
, struct page
, lru
);
387 /* have to delete it as __free_pages_bulk list manipulates */
388 list_del(&page
->lru
);
389 __free_pages_bulk(page
, zone
, order
);
392 spin_unlock_irqrestore(&zone
->lock
, flags
);
396 void __free_pages_ok(struct page
*page
, unsigned int order
)
402 arch_free_page(page
, order
);
406 for (i
= 1 ; i
< (1 << order
) ; ++i
)
407 __put_page(page
+ i
);
410 for (i
= 0 ; i
< (1 << order
) ; ++i
)
411 reserved
+= free_pages_check(__FUNCTION__
, page
+ i
);
415 list_add(&page
->lru
, &list
);
416 mod_page_state(pgfree
, 1 << order
);
417 kernel_map_pages(page
, 1<<order
, 0);
418 free_pages_bulk(page_zone(page
), 1, &list
, order
);
423 * The order of subdivision here is critical for the IO subsystem.
424 * Please do not alter this order without good reasons and regression
425 * testing. Specifically, as large blocks of memory are subdivided,
426 * the order in which smaller blocks are delivered depends on the order
427 * they're subdivided in this function. This is the primary factor
428 * influencing the order in which pages are delivered to the IO
429 * subsystem according to empirical testing, and this is also justified
430 * by considering the behavior of a buddy system containing a single
431 * large block of memory acted on by a series of small allocations.
432 * This behavior is a critical factor in sglist merging's success.
436 static inline struct page
*
437 expand(struct zone
*zone
, struct page
*page
,
438 int low
, int high
, struct free_area
*area
)
440 unsigned long size
= 1 << high
;
446 BUG_ON(bad_range(zone
, &page
[size
]));
447 list_add(&page
[size
].lru
, &area
->free_list
);
449 set_page_order(&page
[size
], high
);
454 void set_page_refs(struct page
*page
, int order
)
457 set_page_count(page
, 1);
462 * We need to reference all the pages for this order, otherwise if
463 * anyone accesses one of the pages with (get/put) it will be freed.
464 * - eg: access_process_vm()
466 for (i
= 0; i
< (1 << order
); i
++)
467 set_page_count(page
+ i
, 1);
468 #endif /* CONFIG_MMU */
472 * This page is about to be returned from the page allocator
474 static int prep_new_page(struct page
*page
, int order
)
476 if ( page_mapcount(page
) ||
477 page
->mapping
!= NULL
||
478 page_count(page
) != 0 ||
490 bad_page(__FUNCTION__
, page
);
493 * For now, we report if PG_reserved was found set, but do not
494 * clear it, and do not allocate the page: as a safety net.
496 if (PageReserved(page
))
499 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
500 1 << PG_referenced
| 1 << PG_arch_1
|
501 1 << PG_checked
| 1 << PG_mappedtodisk
);
502 set_page_private(page
, 0);
503 set_page_refs(page
, order
);
504 kernel_map_pages(page
, 1 << order
, 1);
509 * Do the hard work of removing an element from the buddy allocator.
510 * Call me with the zone->lock already held.
512 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
514 struct free_area
* area
;
515 unsigned int current_order
;
518 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
519 area
= zone
->free_area
+ current_order
;
520 if (list_empty(&area
->free_list
))
523 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
524 list_del(&page
->lru
);
525 rmv_page_order(page
);
527 zone
->free_pages
-= 1UL << order
;
528 return expand(zone
, page
, order
, current_order
, area
);
535 * Obtain a specified number of elements from the buddy allocator, all under
536 * a single hold of the lock, for efficiency. Add them to the supplied list.
537 * Returns the number of new pages which were placed at *list.
539 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
540 unsigned long count
, struct list_head
*list
)
547 spin_lock_irqsave(&zone
->lock
, flags
);
548 for (i
= 0; i
< count
; ++i
) {
549 page
= __rmqueue(zone
, order
);
553 list_add_tail(&page
->lru
, list
);
555 spin_unlock_irqrestore(&zone
->lock
, flags
);
560 /* Called from the slab reaper to drain remote pagesets */
561 void drain_remote_pages(void)
567 local_irq_save(flags
);
568 for_each_zone(zone
) {
569 struct per_cpu_pageset
*pset
;
571 /* Do not drain local pagesets */
572 if (zone
->zone_pgdat
->node_id
== numa_node_id())
575 pset
= zone
->pageset
[smp_processor_id()];
576 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
577 struct per_cpu_pages
*pcp
;
581 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
585 local_irq_restore(flags
);
589 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
590 static void __drain_pages(unsigned int cpu
)
595 for_each_zone(zone
) {
596 struct per_cpu_pageset
*pset
;
598 pset
= zone_pcp(zone
, cpu
);
599 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
600 struct per_cpu_pages
*pcp
;
603 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
608 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
612 void mark_free_pages(struct zone
*zone
)
614 unsigned long zone_pfn
, flags
;
616 struct list_head
*curr
;
618 if (!zone
->spanned_pages
)
621 spin_lock_irqsave(&zone
->lock
, flags
);
622 for (zone_pfn
= 0; zone_pfn
< zone
->spanned_pages
; ++zone_pfn
)
623 ClearPageNosaveFree(pfn_to_page(zone_pfn
+ zone
->zone_start_pfn
));
625 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
626 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
627 unsigned long start_pfn
, i
;
629 start_pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
631 for (i
=0; i
< (1<<order
); i
++)
632 SetPageNosaveFree(pfn_to_page(start_pfn
+i
));
634 spin_unlock_irqrestore(&zone
->lock
, flags
);
638 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
640 void drain_local_pages(void)
644 local_irq_save(flags
);
645 __drain_pages(smp_processor_id());
646 local_irq_restore(flags
);
648 #endif /* CONFIG_PM */
650 static void zone_statistics(struct zonelist
*zonelist
, struct zone
*z
)
655 pg_data_t
*pg
= z
->zone_pgdat
;
656 pg_data_t
*orig
= zonelist
->zones
[0]->zone_pgdat
;
657 struct per_cpu_pageset
*p
;
659 local_irq_save(flags
);
660 cpu
= smp_processor_id();
666 zone_pcp(zonelist
->zones
[0], cpu
)->numa_foreign
++;
668 if (pg
== NODE_DATA(numa_node_id()))
672 local_irq_restore(flags
);
677 * Free a 0-order page
679 static void FASTCALL(free_hot_cold_page(struct page
*page
, int cold
));
680 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
682 struct zone
*zone
= page_zone(page
);
683 struct per_cpu_pages
*pcp
;
686 arch_free_page(page
, 0);
689 page
->mapping
= NULL
;
690 if (free_pages_check(__FUNCTION__
, page
))
693 inc_page_state(pgfree
);
694 kernel_map_pages(page
, 1, 0);
696 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
697 local_irq_save(flags
);
698 list_add(&page
->lru
, &pcp
->list
);
700 if (pcp
->count
>= pcp
->high
)
701 pcp
->count
-= free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
702 local_irq_restore(flags
);
706 void fastcall
free_hot_page(struct page
*page
)
708 free_hot_cold_page(page
, 0);
711 void fastcall
free_cold_page(struct page
*page
)
713 free_hot_cold_page(page
, 1);
716 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
720 BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
721 for(i
= 0; i
< (1 << order
); i
++)
722 clear_highpage(page
+ i
);
726 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
727 * we cheat by calling it from here, in the order > 0 path. Saves a branch
731 buffered_rmqueue(struct zone
*zone
, int order
, gfp_t gfp_flags
)
735 int cold
= !!(gfp_flags
& __GFP_COLD
);
739 struct per_cpu_pages
*pcp
;
742 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
743 local_irq_save(flags
);
744 if (pcp
->count
<= pcp
->low
)
745 pcp
->count
+= rmqueue_bulk(zone
, 0,
746 pcp
->batch
, &pcp
->list
);
748 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
749 list_del(&page
->lru
);
752 local_irq_restore(flags
);
755 spin_lock_irqsave(&zone
->lock
, flags
);
756 page
= __rmqueue(zone
, order
);
757 spin_unlock_irqrestore(&zone
->lock
, flags
);
761 BUG_ON(bad_range(zone
, page
));
762 mod_page_state_zone(zone
, pgalloc
, 1 << order
);
763 if (prep_new_page(page
, order
))
766 if (gfp_flags
& __GFP_ZERO
)
767 prep_zero_page(page
, order
, gfp_flags
);
769 if (order
&& (gfp_flags
& __GFP_COMP
))
770 prep_compound_page(page
, order
);
775 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
776 #define ALLOC_HARDER 0x02 /* try to alloc harder */
777 #define ALLOC_HIGH 0x04 /* __GFP_HIGH set */
778 #define ALLOC_CPUSET 0x08 /* check for correct cpuset */
781 * Return 1 if free pages are above 'mark'. This takes into account the order
784 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
785 int classzone_idx
, int alloc_flags
)
787 /* free_pages my go negative - that's OK */
788 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
791 if (alloc_flags
& ALLOC_HIGH
)
793 if (alloc_flags
& ALLOC_HARDER
)
796 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
798 for (o
= 0; o
< order
; o
++) {
799 /* At the next order, this order's pages become unavailable */
800 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
802 /* Require fewer higher order pages to be free */
805 if (free_pages
<= min
)
812 * get_page_from_freeliest goes through the zonelist trying to allocate
816 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
817 struct zonelist
*zonelist
, int alloc_flags
)
819 struct zone
**z
= zonelist
->zones
;
820 struct page
*page
= NULL
;
821 int classzone_idx
= zone_idx(*z
);
824 * Go through the zonelist once, looking for a zone with enough free.
825 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
828 if ((alloc_flags
& ALLOC_CPUSET
) &&
829 !cpuset_zone_allowed(*z
, gfp_mask
))
832 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
833 if (!zone_watermark_ok(*z
, order
, (*z
)->pages_low
,
834 classzone_idx
, alloc_flags
))
838 page
= buffered_rmqueue(*z
, order
, gfp_mask
);
840 zone_statistics(zonelist
, *z
);
843 } while (*(++z
) != NULL
);
848 * This is the 'heart' of the zoned buddy allocator.
850 struct page
* fastcall
851 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
852 struct zonelist
*zonelist
)
854 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
857 struct reclaim_state reclaim_state
;
858 struct task_struct
*p
= current
;
861 int did_some_progress
;
863 might_sleep_if(wait
);
866 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
868 if (unlikely(*z
== NULL
)) {
869 /* Should this ever happen?? */
873 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
874 zonelist
, ALLOC_CPUSET
);
879 wakeup_kswapd(*z
, order
);
883 * OK, we're below the kswapd watermark and have kicked background
884 * reclaim. Now things get more complex, so set up alloc_flags according
885 * to how we want to proceed.
887 * The caller may dip into page reserves a bit more if the caller
888 * cannot run direct reclaim, or if the caller has realtime scheduling
892 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
893 alloc_flags
|= ALLOC_HARDER
;
894 if (gfp_mask
& __GFP_HIGH
)
895 alloc_flags
|= ALLOC_HIGH
;
897 alloc_flags
|= ALLOC_CPUSET
;
900 * Go through the zonelist again. Let __GFP_HIGH and allocations
901 * coming from realtime tasks go deeper into reserves.
903 * This is the last chance, in general, before the goto nopage.
904 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
905 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
907 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
911 /* This allocation should allow future memory freeing. */
913 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
914 && !in_interrupt()) {
915 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
917 /* go through the zonelist yet again, ignoring mins */
918 page
= get_page_from_freelist(gfp_mask
, order
,
919 zonelist
, ALLOC_NO_WATERMARKS
|ALLOC_CPUSET
);
922 if (gfp_mask
& __GFP_NOFAIL
) {
923 blk_congestion_wait(WRITE
, HZ
/50);
930 /* Atomic allocations - we can't balance anything */
937 /* We now go into synchronous reclaim */
938 p
->flags
|= PF_MEMALLOC
;
939 reclaim_state
.reclaimed_slab
= 0;
940 p
->reclaim_state
= &reclaim_state
;
942 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
944 p
->reclaim_state
= NULL
;
945 p
->flags
&= ~PF_MEMALLOC
;
949 if (likely(did_some_progress
)) {
950 page
= get_page_from_freelist(gfp_mask
, order
,
951 zonelist
, alloc_flags
);
954 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
956 * Go through the zonelist yet one more time, keep
957 * very high watermark here, this is only to catch
958 * a parallel oom killing, we must fail if we're still
959 * under heavy pressure.
961 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
962 zonelist
, ALLOC_CPUSET
);
966 out_of_memory(gfp_mask
, order
);
971 * Don't let big-order allocations loop unless the caller explicitly
972 * requests that. Wait for some write requests to complete then retry.
974 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
975 * <= 3, but that may not be true in other implementations.
978 if (!(gfp_mask
& __GFP_NORETRY
)) {
979 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
981 if (gfp_mask
& __GFP_NOFAIL
)
985 blk_congestion_wait(WRITE
, HZ
/50);
990 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
991 printk(KERN_WARNING
"%s: page allocation failure."
992 " order:%d, mode:0x%x\n",
993 p
->comm
, order
, gfp_mask
);
1001 EXPORT_SYMBOL(__alloc_pages
);
1004 * Common helper functions.
1006 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1009 page
= alloc_pages(gfp_mask
, order
);
1012 return (unsigned long) page_address(page
);
1015 EXPORT_SYMBOL(__get_free_pages
);
1017 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1022 * get_zeroed_page() returns a 32-bit address, which cannot represent
1025 BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1027 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1029 return (unsigned long) page_address(page
);
1033 EXPORT_SYMBOL(get_zeroed_page
);
1035 void __pagevec_free(struct pagevec
*pvec
)
1037 int i
= pagevec_count(pvec
);
1040 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1043 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1045 if (put_page_testzero(page
)) {
1047 free_hot_page(page
);
1049 __free_pages_ok(page
, order
);
1053 EXPORT_SYMBOL(__free_pages
);
1055 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1058 BUG_ON(!virt_addr_valid((void *)addr
));
1059 __free_pages(virt_to_page((void *)addr
), order
);
1063 EXPORT_SYMBOL(free_pages
);
1066 * Total amount of free (allocatable) RAM:
1068 unsigned int nr_free_pages(void)
1070 unsigned int sum
= 0;
1074 sum
+= zone
->free_pages
;
1079 EXPORT_SYMBOL(nr_free_pages
);
1082 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1084 unsigned int i
, sum
= 0;
1086 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1087 sum
+= pgdat
->node_zones
[i
].free_pages
;
1093 static unsigned int nr_free_zone_pages(int offset
)
1095 /* Just pick one node, since fallback list is circular */
1096 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1097 unsigned int sum
= 0;
1099 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1100 struct zone
**zonep
= zonelist
->zones
;
1103 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1104 unsigned long size
= zone
->present_pages
;
1105 unsigned long high
= zone
->pages_high
;
1114 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1116 unsigned int nr_free_buffer_pages(void)
1118 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1122 * Amount of free RAM allocatable within all zones
1124 unsigned int nr_free_pagecache_pages(void)
1126 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1129 #ifdef CONFIG_HIGHMEM
1130 unsigned int nr_free_highpages (void)
1133 unsigned int pages
= 0;
1135 for_each_pgdat(pgdat
)
1136 pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1143 static void show_node(struct zone
*zone
)
1145 printk("Node %d ", zone
->zone_pgdat
->node_id
);
1148 #define show_node(zone) do { } while (0)
1152 * Accumulate the page_state information across all CPUs.
1153 * The result is unavoidably approximate - it can change
1154 * during and after execution of this function.
1156 static DEFINE_PER_CPU(struct page_state
, page_states
) = {0};
1158 atomic_t nr_pagecache
= ATOMIC_INIT(0);
1159 EXPORT_SYMBOL(nr_pagecache
);
1161 DEFINE_PER_CPU(long, nr_pagecache_local
) = 0;
1164 void __get_page_state(struct page_state
*ret
, int nr
, cpumask_t
*cpumask
)
1168 memset(ret
, 0, sizeof(*ret
));
1169 cpus_and(*cpumask
, *cpumask
, cpu_online_map
);
1171 cpu
= first_cpu(*cpumask
);
1172 while (cpu
< NR_CPUS
) {
1173 unsigned long *in
, *out
, off
;
1175 in
= (unsigned long *)&per_cpu(page_states
, cpu
);
1177 cpu
= next_cpu(cpu
, *cpumask
);
1180 prefetch(&per_cpu(page_states
, cpu
));
1182 out
= (unsigned long *)ret
;
1183 for (off
= 0; off
< nr
; off
++)
1188 void get_page_state_node(struct page_state
*ret
, int node
)
1191 cpumask_t mask
= node_to_cpumask(node
);
1193 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1194 nr
/= sizeof(unsigned long);
1196 __get_page_state(ret
, nr
+1, &mask
);
1199 void get_page_state(struct page_state
*ret
)
1202 cpumask_t mask
= CPU_MASK_ALL
;
1204 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1205 nr
/= sizeof(unsigned long);
1207 __get_page_state(ret
, nr
+ 1, &mask
);
1210 void get_full_page_state(struct page_state
*ret
)
1212 cpumask_t mask
= CPU_MASK_ALL
;
1214 __get_page_state(ret
, sizeof(*ret
) / sizeof(unsigned long), &mask
);
1217 unsigned long __read_page_state(unsigned long offset
)
1219 unsigned long ret
= 0;
1222 for_each_online_cpu(cpu
) {
1225 in
= (unsigned long)&per_cpu(page_states
, cpu
) + offset
;
1226 ret
+= *((unsigned long *)in
);
1231 void __mod_page_state(unsigned long offset
, unsigned long delta
)
1233 unsigned long flags
;
1236 local_irq_save(flags
);
1237 ptr
= &__get_cpu_var(page_states
);
1238 *(unsigned long*)(ptr
+ offset
) += delta
;
1239 local_irq_restore(flags
);
1242 EXPORT_SYMBOL(__mod_page_state
);
1244 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
1245 unsigned long *free
, struct pglist_data
*pgdat
)
1247 struct zone
*zones
= pgdat
->node_zones
;
1253 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1254 *active
+= zones
[i
].nr_active
;
1255 *inactive
+= zones
[i
].nr_inactive
;
1256 *free
+= zones
[i
].free_pages
;
1260 void get_zone_counts(unsigned long *active
,
1261 unsigned long *inactive
, unsigned long *free
)
1263 struct pglist_data
*pgdat
;
1268 for_each_pgdat(pgdat
) {
1269 unsigned long l
, m
, n
;
1270 __get_zone_counts(&l
, &m
, &n
, pgdat
);
1277 void si_meminfo(struct sysinfo
*val
)
1279 val
->totalram
= totalram_pages
;
1281 val
->freeram
= nr_free_pages();
1282 val
->bufferram
= nr_blockdev_pages();
1283 #ifdef CONFIG_HIGHMEM
1284 val
->totalhigh
= totalhigh_pages
;
1285 val
->freehigh
= nr_free_highpages();
1290 val
->mem_unit
= PAGE_SIZE
;
1293 EXPORT_SYMBOL(si_meminfo
);
1296 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1298 pg_data_t
*pgdat
= NODE_DATA(nid
);
1300 val
->totalram
= pgdat
->node_present_pages
;
1301 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1302 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1303 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1304 val
->mem_unit
= PAGE_SIZE
;
1308 #define K(x) ((x) << (PAGE_SHIFT-10))
1311 * Show free area list (used inside shift_scroll-lock stuff)
1312 * We also calculate the percentage fragmentation. We do this by counting the
1313 * memory on each free list with the exception of the first item on the list.
1315 void show_free_areas(void)
1317 struct page_state ps
;
1318 int cpu
, temperature
;
1319 unsigned long active
;
1320 unsigned long inactive
;
1324 for_each_zone(zone
) {
1326 printk("%s per-cpu:", zone
->name
);
1328 if (!zone
->present_pages
) {
1334 for_each_online_cpu(cpu
) {
1335 struct per_cpu_pageset
*pageset
;
1337 pageset
= zone_pcp(zone
, cpu
);
1339 for (temperature
= 0; temperature
< 2; temperature
++)
1340 printk("cpu %d %s: low %d, high %d, batch %d used:%d\n",
1342 temperature
? "cold" : "hot",
1343 pageset
->pcp
[temperature
].low
,
1344 pageset
->pcp
[temperature
].high
,
1345 pageset
->pcp
[temperature
].batch
,
1346 pageset
->pcp
[temperature
].count
);
1350 get_page_state(&ps
);
1351 get_zone_counts(&active
, &inactive
, &free
);
1353 printk("Free pages: %11ukB (%ukB HighMem)\n",
1355 K(nr_free_highpages()));
1357 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1358 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1367 ps
.nr_page_table_pages
);
1369 for_each_zone(zone
) {
1381 " pages_scanned:%lu"
1382 " all_unreclaimable? %s"
1385 K(zone
->free_pages
),
1388 K(zone
->pages_high
),
1390 K(zone
->nr_inactive
),
1391 K(zone
->present_pages
),
1392 zone
->pages_scanned
,
1393 (zone
->all_unreclaimable
? "yes" : "no")
1395 printk("lowmem_reserve[]:");
1396 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1397 printk(" %lu", zone
->lowmem_reserve
[i
]);
1401 for_each_zone(zone
) {
1402 unsigned long nr
, flags
, order
, total
= 0;
1405 printk("%s: ", zone
->name
);
1406 if (!zone
->present_pages
) {
1411 spin_lock_irqsave(&zone
->lock
, flags
);
1412 for (order
= 0; order
< MAX_ORDER
; order
++) {
1413 nr
= zone
->free_area
[order
].nr_free
;
1414 total
+= nr
<< order
;
1415 printk("%lu*%lukB ", nr
, K(1UL) << order
);
1417 spin_unlock_irqrestore(&zone
->lock
, flags
);
1418 printk("= %lukB\n", K(total
));
1421 show_swap_cache_info();
1425 * Builds allocation fallback zone lists.
1427 static int __init
build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
, int j
, int k
)
1434 zone
= pgdat
->node_zones
+ ZONE_HIGHMEM
;
1435 if (zone
->present_pages
) {
1436 #ifndef CONFIG_HIGHMEM
1439 zonelist
->zones
[j
++] = zone
;
1442 zone
= pgdat
->node_zones
+ ZONE_NORMAL
;
1443 if (zone
->present_pages
)
1444 zonelist
->zones
[j
++] = zone
;
1446 zone
= pgdat
->node_zones
+ ZONE_DMA32
;
1447 if (zone
->present_pages
)
1448 zonelist
->zones
[j
++] = zone
;
1450 zone
= pgdat
->node_zones
+ ZONE_DMA
;
1451 if (zone
->present_pages
)
1452 zonelist
->zones
[j
++] = zone
;
1458 static inline int highest_zone(int zone_bits
)
1460 int res
= ZONE_NORMAL
;
1461 if (zone_bits
& (__force
int)__GFP_HIGHMEM
)
1463 if (zone_bits
& (__force
int)__GFP_DMA32
)
1465 if (zone_bits
& (__force
int)__GFP_DMA
)
1471 #define MAX_NODE_LOAD (num_online_nodes())
1472 static int __initdata node_load
[MAX_NUMNODES
];
1474 * find_next_best_node - find the next node that should appear in a given node's fallback list
1475 * @node: node whose fallback list we're appending
1476 * @used_node_mask: nodemask_t of already used nodes
1478 * We use a number of factors to determine which is the next node that should
1479 * appear on a given node's fallback list. The node should not have appeared
1480 * already in @node's fallback list, and it should be the next closest node
1481 * according to the distance array (which contains arbitrary distance values
1482 * from each node to each node in the system), and should also prefer nodes
1483 * with no CPUs, since presumably they'll have very little allocation pressure
1484 * on them otherwise.
1485 * It returns -1 if no node is found.
1487 static int __init
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1490 int min_val
= INT_MAX
;
1493 for_each_online_node(i
) {
1496 /* Start from local node */
1497 n
= (node
+i
) % num_online_nodes();
1499 /* Don't want a node to appear more than once */
1500 if (node_isset(n
, *used_node_mask
))
1503 /* Use the local node if we haven't already */
1504 if (!node_isset(node
, *used_node_mask
)) {
1509 /* Use the distance array to find the distance */
1510 val
= node_distance(node
, n
);
1512 /* Give preference to headless and unused nodes */
1513 tmp
= node_to_cpumask(n
);
1514 if (!cpus_empty(tmp
))
1515 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1517 /* Slight preference for less loaded node */
1518 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1519 val
+= node_load
[n
];
1521 if (val
< min_val
) {
1528 node_set(best_node
, *used_node_mask
);
1533 static void __init
build_zonelists(pg_data_t
*pgdat
)
1535 int i
, j
, k
, node
, local_node
;
1536 int prev_node
, load
;
1537 struct zonelist
*zonelist
;
1538 nodemask_t used_mask
;
1540 /* initialize zonelists */
1541 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1542 zonelist
= pgdat
->node_zonelists
+ i
;
1543 zonelist
->zones
[0] = NULL
;
1546 /* NUMA-aware ordering of nodes */
1547 local_node
= pgdat
->node_id
;
1548 load
= num_online_nodes();
1549 prev_node
= local_node
;
1550 nodes_clear(used_mask
);
1551 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1553 * We don't want to pressure a particular node.
1554 * So adding penalty to the first node in same
1555 * distance group to make it round-robin.
1557 if (node_distance(local_node
, node
) !=
1558 node_distance(local_node
, prev_node
))
1559 node_load
[node
] += load
;
1562 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1563 zonelist
= pgdat
->node_zonelists
+ i
;
1564 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1566 k
= highest_zone(i
);
1568 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1569 zonelist
->zones
[j
] = NULL
;
1574 #else /* CONFIG_NUMA */
1576 static void __init
build_zonelists(pg_data_t
*pgdat
)
1578 int i
, j
, k
, node
, local_node
;
1580 local_node
= pgdat
->node_id
;
1581 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1582 struct zonelist
*zonelist
;
1584 zonelist
= pgdat
->node_zonelists
+ i
;
1587 k
= highest_zone(i
);
1588 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1590 * Now we build the zonelist so that it contains the zones
1591 * of all the other nodes.
1592 * We don't want to pressure a particular node, so when
1593 * building the zones for node N, we make sure that the
1594 * zones coming right after the local ones are those from
1595 * node N+1 (modulo N)
1597 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1598 if (!node_online(node
))
1600 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1602 for (node
= 0; node
< local_node
; node
++) {
1603 if (!node_online(node
))
1605 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1608 zonelist
->zones
[j
] = NULL
;
1612 #endif /* CONFIG_NUMA */
1614 void __init
build_all_zonelists(void)
1618 for_each_online_node(i
)
1619 build_zonelists(NODE_DATA(i
));
1620 printk("Built %i zonelists\n", num_online_nodes());
1621 cpuset_init_current_mems_allowed();
1625 * Helper functions to size the waitqueue hash table.
1626 * Essentially these want to choose hash table sizes sufficiently
1627 * large so that collisions trying to wait on pages are rare.
1628 * But in fact, the number of active page waitqueues on typical
1629 * systems is ridiculously low, less than 200. So this is even
1630 * conservative, even though it seems large.
1632 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1633 * waitqueues, i.e. the size of the waitq table given the number of pages.
1635 #define PAGES_PER_WAITQUEUE 256
1637 static inline unsigned long wait_table_size(unsigned long pages
)
1639 unsigned long size
= 1;
1641 pages
/= PAGES_PER_WAITQUEUE
;
1643 while (size
< pages
)
1647 * Once we have dozens or even hundreds of threads sleeping
1648 * on IO we've got bigger problems than wait queue collision.
1649 * Limit the size of the wait table to a reasonable size.
1651 size
= min(size
, 4096UL);
1653 return max(size
, 4UL);
1657 * This is an integer logarithm so that shifts can be used later
1658 * to extract the more random high bits from the multiplicative
1659 * hash function before the remainder is taken.
1661 static inline unsigned long wait_table_bits(unsigned long size
)
1666 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1668 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1669 unsigned long *zones_size
, unsigned long *zholes_size
)
1671 unsigned long realtotalpages
, totalpages
= 0;
1674 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1675 totalpages
+= zones_size
[i
];
1676 pgdat
->node_spanned_pages
= totalpages
;
1678 realtotalpages
= totalpages
;
1680 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1681 realtotalpages
-= zholes_size
[i
];
1682 pgdat
->node_present_pages
= realtotalpages
;
1683 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1688 * Initially all pages are reserved - free ones are freed
1689 * up by free_all_bootmem() once the early boot process is
1690 * done. Non-atomic initialization, single-pass.
1692 void __devinit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1693 unsigned long start_pfn
)
1696 unsigned long end_pfn
= start_pfn
+ size
;
1699 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++, page
++) {
1700 if (!early_pfn_valid(pfn
))
1702 if (!early_pfn_in_nid(pfn
, nid
))
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 * We will be trying to allcoate bigger chunks of contiguous
1766 * memory of the order of fls(batch). This should result in
1767 * better cache coloring.
1769 * A sanity check also to ensure that batch is still in limits.
1771 batch
= (1 << fls(batch
+ batch
/2));
1773 if (fls(batch
) >= (PAGE_SHIFT
+ MAX_ORDER
- 2))
1774 batch
= PAGE_SHIFT
+ ((MAX_ORDER
- 1 - PAGE_SHIFT
)/2);
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 */
1788 pcp
->high
= 6 * batch
;
1789 pcp
->batch
= max(1UL, 1 * batch
);
1790 INIT_LIST_HEAD(&pcp
->list
);
1792 pcp
= &p
->pcp
[1]; /* cold*/
1795 pcp
->high
= 2 * batch
;
1796 pcp
->batch
= max(1UL, batch
/2);
1797 INIT_LIST_HEAD(&pcp
->list
);
1802 * Boot pageset table. One per cpu which is going to be used for all
1803 * zones and all nodes. The parameters will be set in such a way
1804 * that an item put on a list will immediately be handed over to
1805 * the buddy list. This is safe since pageset manipulation is done
1806 * with interrupts disabled.
1808 * Some NUMA counter updates may also be caught by the boot pagesets.
1810 * The boot_pagesets must be kept even after bootup is complete for
1811 * unused processors and/or zones. They do play a role for bootstrapping
1812 * hotplugged processors.
1814 * zoneinfo_show() and maybe other functions do
1815 * not check if the processor is online before following the pageset pointer.
1816 * Other parts of the kernel may not check if the zone is available.
1818 static struct per_cpu_pageset
1819 boot_pageset
[NR_CPUS
];
1822 * Dynamically allocate memory for the
1823 * per cpu pageset array in struct zone.
1825 static int __devinit
process_zones(int cpu
)
1827 struct zone
*zone
, *dzone
;
1829 for_each_zone(zone
) {
1831 zone
->pageset
[cpu
] = kmalloc_node(sizeof(struct per_cpu_pageset
),
1832 GFP_KERNEL
, cpu_to_node(cpu
));
1833 if (!zone
->pageset
[cpu
])
1836 setup_pageset(zone
->pageset
[cpu
], zone_batchsize(zone
));
1841 for_each_zone(dzone
) {
1844 kfree(dzone
->pageset
[cpu
]);
1845 dzone
->pageset
[cpu
] = NULL
;
1850 static inline void free_zone_pagesets(int cpu
)
1855 for_each_zone(zone
) {
1856 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1858 zone_pcp(zone
, cpu
) = NULL
;
1864 static int __devinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1865 unsigned long action
,
1868 int cpu
= (long)hcpu
;
1869 int ret
= NOTIFY_OK
;
1872 case CPU_UP_PREPARE
:
1873 if (process_zones(cpu
))
1876 case CPU_UP_CANCELED
:
1878 free_zone_pagesets(cpu
);
1886 static struct notifier_block pageset_notifier
=
1887 { &pageset_cpuup_callback
, NULL
, 0 };
1889 void __init
setup_per_cpu_pageset()
1893 /* Initialize per_cpu_pageset for cpu 0.
1894 * A cpuup callback will do this for every cpu
1895 * as it comes online
1897 err
= process_zones(smp_processor_id());
1899 register_cpu_notifier(&pageset_notifier
);
1905 void zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1908 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1911 * The per-page waitqueue mechanism uses hashed waitqueues
1914 zone
->wait_table_size
= wait_table_size(zone_size_pages
);
1915 zone
->wait_table_bits
= wait_table_bits(zone
->wait_table_size
);
1916 zone
->wait_table
= (wait_queue_head_t
*)
1917 alloc_bootmem_node(pgdat
, zone
->wait_table_size
1918 * sizeof(wait_queue_head_t
));
1920 for(i
= 0; i
< zone
->wait_table_size
; ++i
)
1921 init_waitqueue_head(zone
->wait_table
+ i
);
1924 static __devinit
void zone_pcp_init(struct zone
*zone
)
1927 unsigned long batch
= zone_batchsize(zone
);
1929 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1931 /* Early boot. Slab allocator not functional yet */
1932 zone
->pageset
[cpu
] = &boot_pageset
[cpu
];
1933 setup_pageset(&boot_pageset
[cpu
],0);
1935 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1938 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1939 zone
->name
, zone
->present_pages
, batch
);
1942 static __devinit
void init_currently_empty_zone(struct zone
*zone
,
1943 unsigned long zone_start_pfn
, unsigned long size
)
1945 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1947 zone_wait_table_init(zone
, size
);
1948 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1950 zone
->zone_mem_map
= pfn_to_page(zone_start_pfn
);
1951 zone
->zone_start_pfn
= zone_start_pfn
;
1953 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1955 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
1959 * Set up the zone data structures:
1960 * - mark all pages reserved
1961 * - mark all memory queues empty
1962 * - clear the memory bitmaps
1964 static void __init
free_area_init_core(struct pglist_data
*pgdat
,
1965 unsigned long *zones_size
, unsigned long *zholes_size
)
1968 int nid
= pgdat
->node_id
;
1969 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
1971 pgdat_resize_init(pgdat
);
1972 pgdat
->nr_zones
= 0;
1973 init_waitqueue_head(&pgdat
->kswapd_wait
);
1974 pgdat
->kswapd_max_order
= 0;
1976 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
1977 struct zone
*zone
= pgdat
->node_zones
+ j
;
1978 unsigned long size
, realsize
;
1980 realsize
= size
= zones_size
[j
];
1982 realsize
-= zholes_size
[j
];
1984 if (j
< ZONE_HIGHMEM
)
1985 nr_kernel_pages
+= realsize
;
1986 nr_all_pages
+= realsize
;
1988 zone
->spanned_pages
= size
;
1989 zone
->present_pages
= realsize
;
1990 zone
->name
= zone_names
[j
];
1991 spin_lock_init(&zone
->lock
);
1992 spin_lock_init(&zone
->lru_lock
);
1993 zone_seqlock_init(zone
);
1994 zone
->zone_pgdat
= pgdat
;
1995 zone
->free_pages
= 0;
1997 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
1999 zone_pcp_init(zone
);
2000 INIT_LIST_HEAD(&zone
->active_list
);
2001 INIT_LIST_HEAD(&zone
->inactive_list
);
2002 zone
->nr_scan_active
= 0;
2003 zone
->nr_scan_inactive
= 0;
2004 zone
->nr_active
= 0;
2005 zone
->nr_inactive
= 0;
2006 atomic_set(&zone
->reclaim_in_progress
, 0);
2010 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2011 init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2012 zone_start_pfn
+= size
;
2016 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2018 /* Skip empty nodes */
2019 if (!pgdat
->node_spanned_pages
)
2022 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2023 /* ia64 gets its own node_mem_map, before this, without bootmem */
2024 if (!pgdat
->node_mem_map
) {
2028 size
= (pgdat
->node_spanned_pages
+ 1) * sizeof(struct page
);
2029 map
= alloc_remap(pgdat
->node_id
, size
);
2031 map
= alloc_bootmem_node(pgdat
, size
);
2032 pgdat
->node_mem_map
= map
;
2034 #ifdef CONFIG_FLATMEM
2036 * With no DISCONTIG, the global mem_map is just set as node 0's
2038 if (pgdat
== NODE_DATA(0))
2039 mem_map
= NODE_DATA(0)->node_mem_map
;
2041 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2044 void __init
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2045 unsigned long *zones_size
, unsigned long node_start_pfn
,
2046 unsigned long *zholes_size
)
2048 pgdat
->node_id
= nid
;
2049 pgdat
->node_start_pfn
= node_start_pfn
;
2050 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
2052 alloc_node_mem_map(pgdat
);
2054 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2057 #ifndef CONFIG_NEED_MULTIPLE_NODES
2058 static bootmem_data_t contig_bootmem_data
;
2059 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2061 EXPORT_SYMBOL(contig_page_data
);
2064 void __init
free_area_init(unsigned long *zones_size
)
2066 free_area_init_node(0, NODE_DATA(0), zones_size
,
2067 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2070 #ifdef CONFIG_PROC_FS
2072 #include <linux/seq_file.h>
2074 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
2079 for (pgdat
= pgdat_list
; pgdat
&& node
; pgdat
= pgdat
->pgdat_next
)
2085 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2087 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2090 return pgdat
->pgdat_next
;
2093 static void frag_stop(struct seq_file
*m
, void *arg
)
2098 * This walks the free areas for each zone.
2100 static int frag_show(struct seq_file
*m
, void *arg
)
2102 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2104 struct zone
*node_zones
= pgdat
->node_zones
;
2105 unsigned long flags
;
2108 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
2109 if (!zone
->present_pages
)
2112 spin_lock_irqsave(&zone
->lock
, flags
);
2113 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
2114 for (order
= 0; order
< MAX_ORDER
; ++order
)
2115 seq_printf(m
, "%6lu ", zone
->free_area
[order
].nr_free
);
2116 spin_unlock_irqrestore(&zone
->lock
, flags
);
2122 struct seq_operations fragmentation_op
= {
2123 .start
= frag_start
,
2130 * Output information about zones in @pgdat.
2132 static int zoneinfo_show(struct seq_file
*m
, void *arg
)
2134 pg_data_t
*pgdat
= arg
;
2136 struct zone
*node_zones
= pgdat
->node_zones
;
2137 unsigned long flags
;
2139 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; zone
++) {
2142 if (!zone
->present_pages
)
2145 spin_lock_irqsave(&zone
->lock
, flags
);
2146 seq_printf(m
, "Node %d, zone %8s", pgdat
->node_id
, zone
->name
);
2154 "\n scanned %lu (a: %lu i: %lu)"
2163 zone
->pages_scanned
,
2164 zone
->nr_scan_active
, zone
->nr_scan_inactive
,
2165 zone
->spanned_pages
,
2166 zone
->present_pages
);
2168 "\n protection: (%lu",
2169 zone
->lowmem_reserve
[0]);
2170 for (i
= 1; i
< ARRAY_SIZE(zone
->lowmem_reserve
); i
++)
2171 seq_printf(m
, ", %lu", zone
->lowmem_reserve
[i
]);
2175 for (i
= 0; i
< ARRAY_SIZE(zone
->pageset
); i
++) {
2176 struct per_cpu_pageset
*pageset
;
2179 pageset
= zone_pcp(zone
, i
);
2180 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2181 if (pageset
->pcp
[j
].count
)
2184 if (j
== ARRAY_SIZE(pageset
->pcp
))
2186 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2188 "\n cpu: %i pcp: %i"
2194 pageset
->pcp
[j
].count
,
2195 pageset
->pcp
[j
].low
,
2196 pageset
->pcp
[j
].high
,
2197 pageset
->pcp
[j
].batch
);
2203 "\n numa_foreign: %lu"
2204 "\n interleave_hit: %lu"
2205 "\n local_node: %lu"
2206 "\n other_node: %lu",
2209 pageset
->numa_foreign
,
2210 pageset
->interleave_hit
,
2211 pageset
->local_node
,
2212 pageset
->other_node
);
2216 "\n all_unreclaimable: %u"
2217 "\n prev_priority: %i"
2218 "\n temp_priority: %i"
2219 "\n start_pfn: %lu",
2220 zone
->all_unreclaimable
,
2221 zone
->prev_priority
,
2222 zone
->temp_priority
,
2223 zone
->zone_start_pfn
);
2224 spin_unlock_irqrestore(&zone
->lock
, flags
);
2230 struct seq_operations zoneinfo_op
= {
2231 .start
= frag_start
, /* iterate over all zones. The same as in
2235 .show
= zoneinfo_show
,
2238 static char *vmstat_text
[] = {
2242 "nr_page_table_pages",
2267 "pgscan_kswapd_high",
2268 "pgscan_kswapd_normal",
2270 "pgscan_kswapd_dma",
2271 "pgscan_direct_high",
2272 "pgscan_direct_normal",
2273 "pgscan_direct_dma",
2278 "kswapd_inodesteal",
2286 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
2288 struct page_state
*ps
;
2290 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2293 ps
= kmalloc(sizeof(*ps
), GFP_KERNEL
);
2296 return ERR_PTR(-ENOMEM
);
2297 get_full_page_state(ps
);
2298 ps
->pgpgin
/= 2; /* sectors -> kbytes */
2300 return (unsigned long *)ps
+ *pos
;
2303 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2306 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2308 return (unsigned long *)m
->private + *pos
;
2311 static int vmstat_show(struct seq_file
*m
, void *arg
)
2313 unsigned long *l
= arg
;
2314 unsigned long off
= l
- (unsigned long *)m
->private;
2316 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
2320 static void vmstat_stop(struct seq_file
*m
, void *arg
)
2326 struct seq_operations vmstat_op
= {
2327 .start
= vmstat_start
,
2328 .next
= vmstat_next
,
2329 .stop
= vmstat_stop
,
2330 .show
= vmstat_show
,
2333 #endif /* CONFIG_PROC_FS */
2335 #ifdef CONFIG_HOTPLUG_CPU
2336 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2337 unsigned long action
, void *hcpu
)
2339 int cpu
= (unsigned long)hcpu
;
2341 unsigned long *src
, *dest
;
2343 if (action
== CPU_DEAD
) {
2346 /* Drain local pagecache count. */
2347 count
= &per_cpu(nr_pagecache_local
, cpu
);
2348 atomic_add(*count
, &nr_pagecache
);
2350 local_irq_disable();
2353 /* Add dead cpu's page_states to our own. */
2354 dest
= (unsigned long *)&__get_cpu_var(page_states
);
2355 src
= (unsigned long *)&per_cpu(page_states
, cpu
);
2357 for (i
= 0; i
< sizeof(struct page_state
)/sizeof(unsigned long);
2367 #endif /* CONFIG_HOTPLUG_CPU */
2369 void __init
page_alloc_init(void)
2371 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2375 * setup_per_zone_lowmem_reserve - called whenever
2376 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2377 * has a correct pages reserved value, so an adequate number of
2378 * pages are left in the zone after a successful __alloc_pages().
2380 static void setup_per_zone_lowmem_reserve(void)
2382 struct pglist_data
*pgdat
;
2385 for_each_pgdat(pgdat
) {
2386 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2387 struct zone
*zone
= pgdat
->node_zones
+ j
;
2388 unsigned long present_pages
= zone
->present_pages
;
2390 zone
->lowmem_reserve
[j
] = 0;
2392 for (idx
= j
-1; idx
>= 0; idx
--) {
2393 struct zone
*lower_zone
;
2395 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2396 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2398 lower_zone
= pgdat
->node_zones
+ idx
;
2399 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2400 sysctl_lowmem_reserve_ratio
[idx
];
2401 present_pages
+= lower_zone
->present_pages
;
2408 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2409 * that the pages_{min,low,high} values for each zone are set correctly
2410 * with respect to min_free_kbytes.
2412 void setup_per_zone_pages_min(void)
2414 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2415 unsigned long lowmem_pages
= 0;
2417 unsigned long flags
;
2419 /* Calculate total number of !ZONE_HIGHMEM pages */
2420 for_each_zone(zone
) {
2421 if (!is_highmem(zone
))
2422 lowmem_pages
+= zone
->present_pages
;
2425 for_each_zone(zone
) {
2427 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2428 tmp
= (pages_min
* zone
->present_pages
) / lowmem_pages
;
2429 if (is_highmem(zone
)) {
2431 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2432 * need highmem pages, so cap pages_min to a small
2435 * The (pages_high-pages_low) and (pages_low-pages_min)
2436 * deltas controls asynch page reclaim, and so should
2437 * not be capped for highmem.
2441 min_pages
= zone
->present_pages
/ 1024;
2442 if (min_pages
< SWAP_CLUSTER_MAX
)
2443 min_pages
= SWAP_CLUSTER_MAX
;
2444 if (min_pages
> 128)
2446 zone
->pages_min
= min_pages
;
2449 * If it's a lowmem zone, reserve a number of pages
2450 * proportionate to the zone's size.
2452 zone
->pages_min
= tmp
;
2455 zone
->pages_low
= zone
->pages_min
+ tmp
/ 4;
2456 zone
->pages_high
= zone
->pages_min
+ tmp
/ 2;
2457 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2462 * Initialise min_free_kbytes.
2464 * For small machines we want it small (128k min). For large machines
2465 * we want it large (64MB max). But it is not linear, because network
2466 * bandwidth does not increase linearly with machine size. We use
2468 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2469 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2485 static int __init
init_per_zone_pages_min(void)
2487 unsigned long lowmem_kbytes
;
2489 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2491 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2492 if (min_free_kbytes
< 128)
2493 min_free_kbytes
= 128;
2494 if (min_free_kbytes
> 65536)
2495 min_free_kbytes
= 65536;
2496 setup_per_zone_pages_min();
2497 setup_per_zone_lowmem_reserve();
2500 module_init(init_per_zone_pages_min
)
2503 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2504 * that we can call two helper functions whenever min_free_kbytes
2507 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2508 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2510 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2511 setup_per_zone_pages_min();
2516 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2517 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2518 * whenever sysctl_lowmem_reserve_ratio changes.
2520 * The reserve ratio obviously has absolutely no relation with the
2521 * pages_min watermarks. The lowmem reserve ratio can only make sense
2522 * if in function of the boot time zone sizes.
2524 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2525 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2527 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2528 setup_per_zone_lowmem_reserve();
2532 __initdata
int hashdist
= HASHDIST_DEFAULT
;
2535 static int __init
set_hashdist(char *str
)
2539 hashdist
= simple_strtoul(str
, &str
, 0);
2542 __setup("hashdist=", set_hashdist
);
2546 * allocate a large system hash table from bootmem
2547 * - it is assumed that the hash table must contain an exact power-of-2
2548 * quantity of entries
2549 * - limit is the number of hash buckets, not the total allocation size
2551 void *__init
alloc_large_system_hash(const char *tablename
,
2552 unsigned long bucketsize
,
2553 unsigned long numentries
,
2556 unsigned int *_hash_shift
,
2557 unsigned int *_hash_mask
,
2558 unsigned long limit
)
2560 unsigned long long max
= limit
;
2561 unsigned long log2qty
, size
;
2564 /* allow the kernel cmdline to have a say */
2566 /* round applicable memory size up to nearest megabyte */
2567 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2568 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2569 numentries
>>= 20 - PAGE_SHIFT
;
2570 numentries
<<= 20 - PAGE_SHIFT
;
2572 /* limit to 1 bucket per 2^scale bytes of low memory */
2573 if (scale
> PAGE_SHIFT
)
2574 numentries
>>= (scale
- PAGE_SHIFT
);
2576 numentries
<<= (PAGE_SHIFT
- scale
);
2578 /* rounded up to nearest power of 2 in size */
2579 numentries
= 1UL << (long_log2(numentries
) + 1);
2581 /* limit allocation size to 1/16 total memory by default */
2583 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2584 do_div(max
, bucketsize
);
2587 if (numentries
> max
)
2590 log2qty
= long_log2(numentries
);
2593 size
= bucketsize
<< log2qty
;
2594 if (flags
& HASH_EARLY
)
2595 table
= alloc_bootmem(size
);
2597 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
2599 unsigned long order
;
2600 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
2602 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
2604 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2607 panic("Failed to allocate %s hash table\n", tablename
);
2609 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2612 long_log2(size
) - PAGE_SHIFT
,
2616 *_hash_shift
= log2qty
;
2618 *_hash_mask
= (1 << log2qty
) - 1;