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
|
145 set_page_count(page
, 0);
146 reset_page_mapcount(page
);
147 page
->mapping
= NULL
;
148 add_taint(TAINT_BAD_PAGE
);
152 * Higher-order pages are called "compound pages". They are structured thusly:
154 * The first PAGE_SIZE page is called the "head page".
156 * The remaining PAGE_SIZE pages are called "tail pages".
158 * All pages have PG_compound set. All pages have their ->private pointing at
159 * the head page (even the head page has this).
161 * The first tail page's ->mapping, if non-zero, holds the address of the
162 * compound page's put_page() function.
164 * The order of the allocation is stored in the first tail page's ->index
165 * This is only for debug at present. This usage means that zero-order pages
166 * may not be compound.
168 static void prep_compound_page(struct page
*page
, unsigned long order
)
171 int nr_pages
= 1 << order
;
173 page
[1].mapping
= NULL
;
174 page
[1].index
= order
;
175 for (i
= 0; i
< nr_pages
; i
++) {
176 struct page
*p
= page
+ i
;
179 set_page_private(p
, (unsigned long)page
);
183 static void destroy_compound_page(struct page
*page
, unsigned long order
)
186 int nr_pages
= 1 << order
;
188 if (!PageCompound(page
))
191 if (page
[1].index
!= order
)
192 bad_page(__FUNCTION__
, page
);
194 for (i
= 0; i
< nr_pages
; i
++) {
195 struct page
*p
= page
+ i
;
197 if (!PageCompound(p
))
198 bad_page(__FUNCTION__
, page
);
199 if (page_private(p
) != (unsigned long)page
)
200 bad_page(__FUNCTION__
, page
);
201 ClearPageCompound(p
);
206 * function for dealing with page's order in buddy system.
207 * zone->lock is already acquired when we use these.
208 * So, we don't need atomic page->flags operations here.
210 static inline unsigned long page_order(struct page
*page
) {
211 return page_private(page
);
214 static inline void set_page_order(struct page
*page
, int order
) {
215 set_page_private(page
, order
);
216 __SetPagePrivate(page
);
219 static inline void rmv_page_order(struct page
*page
)
221 __ClearPagePrivate(page
);
222 set_page_private(page
, 0);
226 * Locate the struct page for both the matching buddy in our
227 * pair (buddy1) and the combined O(n+1) page they form (page).
229 * 1) Any buddy B1 will have an order O twin B2 which satisfies
230 * the following equation:
232 * For example, if the starting buddy (buddy2) is #8 its order
234 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
236 * 2) Any buddy B will have an order O+1 parent P which
237 * satisfies the following equation:
240 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
242 static inline struct page
*
243 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
245 unsigned long buddy_idx
= page_idx
^ (1 << order
);
247 return page
+ (buddy_idx
- page_idx
);
250 static inline unsigned long
251 __find_combined_index(unsigned long page_idx
, unsigned int order
)
253 return (page_idx
& ~(1 << order
));
257 * This function checks whether a page is free && is the buddy
258 * we can do coalesce a page and its buddy if
259 * (a) the buddy is free &&
260 * (b) the buddy is on the buddy system &&
261 * (c) a page and its buddy have the same order.
262 * for recording page's order, we use page_private(page) and PG_private.
265 static inline int page_is_buddy(struct page
*page
, int order
)
267 if (PagePrivate(page
) &&
268 (page_order(page
) == order
) &&
269 page_count(page
) == 0)
275 * Freeing function for a buddy system allocator.
277 * The concept of a buddy system is to maintain direct-mapped table
278 * (containing bit values) for memory blocks of various "orders".
279 * The bottom level table contains the map for the smallest allocatable
280 * units of memory (here, pages), and each level above it describes
281 * pairs of units from the levels below, hence, "buddies".
282 * At a high level, all that happens here is marking the table entry
283 * at the bottom level available, and propagating the changes upward
284 * as necessary, plus some accounting needed to play nicely with other
285 * parts of the VM system.
286 * At each level, we keep a list of pages, which are heads of continuous
287 * free pages of length of (1 << order) and marked with PG_Private.Page's
288 * order is recorded in page_private(page) field.
289 * So when we are allocating or freeing one, we can derive the state of the
290 * other. That is, if we allocate a small block, and both were
291 * free, the remainder of the region must be split into blocks.
292 * If a block is freed, and its buddy is also free, then this
293 * triggers coalescing into a block of larger size.
298 static inline void __free_pages_bulk (struct page
*page
,
299 struct zone
*zone
, unsigned int order
)
301 unsigned long page_idx
;
302 int order_size
= 1 << order
;
305 destroy_compound_page(page
, order
);
307 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
309 BUG_ON(page_idx
& (order_size
- 1));
310 BUG_ON(bad_range(zone
, page
));
312 zone
->free_pages
+= order_size
;
313 while (order
< MAX_ORDER
-1) {
314 unsigned long combined_idx
;
315 struct free_area
*area
;
318 combined_idx
= __find_combined_index(page_idx
, order
);
319 buddy
= __page_find_buddy(page
, page_idx
, order
);
321 if (bad_range(zone
, buddy
))
323 if (!page_is_buddy(buddy
, order
))
324 break; /* Move the buddy up one level. */
325 list_del(&buddy
->lru
);
326 area
= zone
->free_area
+ order
;
328 rmv_page_order(buddy
);
329 page
= page
+ (combined_idx
- page_idx
);
330 page_idx
= combined_idx
;
333 set_page_order(page
, order
);
334 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
335 zone
->free_area
[order
].nr_free
++;
338 static inline void free_pages_check(const char *function
, struct page
*page
)
340 if ( page_mapcount(page
) ||
341 page
->mapping
!= NULL
||
342 page_count(page
) != 0 ||
353 bad_page(function
, page
);
355 __ClearPageDirty(page
);
359 * Frees a list of pages.
360 * Assumes all pages on list are in same zone, and of same order.
361 * count is the number of pages to free.
363 * If the zone was previously in an "all pages pinned" state then look to
364 * see if this freeing clears that state.
366 * And clear the zone's pages_scanned counter, to hold off the "all pages are
367 * pinned" detection logic.
370 free_pages_bulk(struct zone
*zone
, int count
,
371 struct list_head
*list
, unsigned int order
)
374 struct page
*page
= NULL
;
377 spin_lock_irqsave(&zone
->lock
, flags
);
378 zone
->all_unreclaimable
= 0;
379 zone
->pages_scanned
= 0;
380 while (!list_empty(list
) && count
--) {
381 page
= list_entry(list
->prev
, struct page
, lru
);
382 /* have to delete it as __free_pages_bulk list manipulates */
383 list_del(&page
->lru
);
384 __free_pages_bulk(page
, zone
, order
);
387 spin_unlock_irqrestore(&zone
->lock
, flags
);
391 void __free_pages_ok(struct page
*page
, unsigned int order
)
396 arch_free_page(page
, order
);
398 mod_page_state(pgfree
, 1 << order
);
402 for (i
= 1 ; i
< (1 << order
) ; ++i
)
403 __put_page(page
+ i
);
406 for (i
= 0 ; i
< (1 << order
) ; ++i
)
407 free_pages_check(__FUNCTION__
, page
+ i
);
408 list_add(&page
->lru
, &list
);
409 kernel_map_pages(page
, 1<<order
, 0);
410 free_pages_bulk(page_zone(page
), 1, &list
, order
);
415 * The order of subdivision here is critical for the IO subsystem.
416 * Please do not alter this order without good reasons and regression
417 * testing. Specifically, as large blocks of memory are subdivided,
418 * the order in which smaller blocks are delivered depends on the order
419 * they're subdivided in this function. This is the primary factor
420 * influencing the order in which pages are delivered to the IO
421 * subsystem according to empirical testing, and this is also justified
422 * by considering the behavior of a buddy system containing a single
423 * large block of memory acted on by a series of small allocations.
424 * This behavior is a critical factor in sglist merging's success.
428 static inline struct page
*
429 expand(struct zone
*zone
, struct page
*page
,
430 int low
, int high
, struct free_area
*area
)
432 unsigned long size
= 1 << high
;
438 BUG_ON(bad_range(zone
, &page
[size
]));
439 list_add(&page
[size
].lru
, &area
->free_list
);
441 set_page_order(&page
[size
], high
);
446 void set_page_refs(struct page
*page
, int order
)
449 set_page_count(page
, 1);
454 * We need to reference all the pages for this order, otherwise if
455 * anyone accesses one of the pages with (get/put) it will be freed.
456 * - eg: access_process_vm()
458 for (i
= 0; i
< (1 << order
); i
++)
459 set_page_count(page
+ i
, 1);
460 #endif /* CONFIG_MMU */
464 * This page is about to be returned from the page allocator
466 static void prep_new_page(struct page
*page
, int order
)
468 if ( page_mapcount(page
) ||
469 page
->mapping
!= NULL
||
470 page_count(page
) != 0 ||
482 bad_page(__FUNCTION__
, page
);
484 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
485 1 << PG_referenced
| 1 << PG_arch_1
|
486 1 << PG_checked
| 1 << PG_mappedtodisk
);
487 set_page_private(page
, 0);
488 set_page_refs(page
, order
);
489 kernel_map_pages(page
, 1 << order
, 1);
493 * Do the hard work of removing an element from the buddy allocator.
494 * Call me with the zone->lock already held.
496 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
498 struct free_area
* area
;
499 unsigned int current_order
;
502 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
503 area
= zone
->free_area
+ current_order
;
504 if (list_empty(&area
->free_list
))
507 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
508 list_del(&page
->lru
);
509 rmv_page_order(page
);
511 zone
->free_pages
-= 1UL << order
;
512 return expand(zone
, page
, order
, current_order
, area
);
519 * Obtain a specified number of elements from the buddy allocator, all under
520 * a single hold of the lock, for efficiency. Add them to the supplied list.
521 * Returns the number of new pages which were placed at *list.
523 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
524 unsigned long count
, struct list_head
*list
)
531 spin_lock_irqsave(&zone
->lock
, flags
);
532 for (i
= 0; i
< count
; ++i
) {
533 page
= __rmqueue(zone
, order
);
537 list_add_tail(&page
->lru
, list
);
539 spin_unlock_irqrestore(&zone
->lock
, flags
);
544 /* Called from the slab reaper to drain remote pagesets */
545 void drain_remote_pages(void)
551 local_irq_save(flags
);
552 for_each_zone(zone
) {
553 struct per_cpu_pageset
*pset
;
555 /* Do not drain local pagesets */
556 if (zone
->zone_pgdat
->node_id
== numa_node_id())
559 pset
= zone
->pageset
[smp_processor_id()];
560 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
561 struct per_cpu_pages
*pcp
;
565 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
569 local_irq_restore(flags
);
573 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
574 static void __drain_pages(unsigned int cpu
)
579 for_each_zone(zone
) {
580 struct per_cpu_pageset
*pset
;
582 pset
= zone_pcp(zone
, cpu
);
583 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
584 struct per_cpu_pages
*pcp
;
587 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
592 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
596 void mark_free_pages(struct zone
*zone
)
598 unsigned long zone_pfn
, flags
;
600 struct list_head
*curr
;
602 if (!zone
->spanned_pages
)
605 spin_lock_irqsave(&zone
->lock
, flags
);
606 for (zone_pfn
= 0; zone_pfn
< zone
->spanned_pages
; ++zone_pfn
)
607 ClearPageNosaveFree(pfn_to_page(zone_pfn
+ zone
->zone_start_pfn
));
609 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
610 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
611 unsigned long start_pfn
, i
;
613 start_pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
615 for (i
=0; i
< (1<<order
); i
++)
616 SetPageNosaveFree(pfn_to_page(start_pfn
+i
));
618 spin_unlock_irqrestore(&zone
->lock
, flags
);
622 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
624 void drain_local_pages(void)
628 local_irq_save(flags
);
629 __drain_pages(smp_processor_id());
630 local_irq_restore(flags
);
632 #endif /* CONFIG_PM */
634 static void zone_statistics(struct zonelist
*zonelist
, struct zone
*z
)
639 pg_data_t
*pg
= z
->zone_pgdat
;
640 pg_data_t
*orig
= zonelist
->zones
[0]->zone_pgdat
;
641 struct per_cpu_pageset
*p
;
643 local_irq_save(flags
);
644 cpu
= smp_processor_id();
650 zone_pcp(zonelist
->zones
[0], cpu
)->numa_foreign
++;
652 if (pg
== NODE_DATA(numa_node_id()))
656 local_irq_restore(flags
);
661 * Free a 0-order page
663 static void FASTCALL(free_hot_cold_page(struct page
*page
, int cold
));
664 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
666 struct zone
*zone
= page_zone(page
);
667 struct per_cpu_pages
*pcp
;
670 arch_free_page(page
, 0);
672 kernel_map_pages(page
, 1, 0);
673 inc_page_state(pgfree
);
675 page
->mapping
= NULL
;
676 free_pages_check(__FUNCTION__
, page
);
677 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
678 local_irq_save(flags
);
679 list_add(&page
->lru
, &pcp
->list
);
681 if (pcp
->count
>= pcp
->high
)
682 pcp
->count
-= free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
683 local_irq_restore(flags
);
687 void fastcall
free_hot_page(struct page
*page
)
689 free_hot_cold_page(page
, 0);
692 void fastcall
free_cold_page(struct page
*page
)
694 free_hot_cold_page(page
, 1);
697 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
701 BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
702 for(i
= 0; i
< (1 << order
); i
++)
703 clear_highpage(page
+ i
);
707 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
708 * we cheat by calling it from here, in the order > 0 path. Saves a branch
712 buffered_rmqueue(struct zone
*zone
, int order
, gfp_t gfp_flags
)
715 struct page
*page
= NULL
;
716 int cold
= !!(gfp_flags
& __GFP_COLD
);
719 struct per_cpu_pages
*pcp
;
721 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
722 local_irq_save(flags
);
723 if (pcp
->count
<= pcp
->low
)
724 pcp
->count
+= rmqueue_bulk(zone
, 0,
725 pcp
->batch
, &pcp
->list
);
727 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
728 list_del(&page
->lru
);
731 local_irq_restore(flags
);
734 spin_lock_irqsave(&zone
->lock
, flags
);
735 page
= __rmqueue(zone
, order
);
736 spin_unlock_irqrestore(&zone
->lock
, flags
);
740 BUG_ON(bad_range(zone
, page
));
741 mod_page_state_zone(zone
, pgalloc
, 1 << order
);
742 prep_new_page(page
, order
);
744 if (gfp_flags
& __GFP_ZERO
)
745 prep_zero_page(page
, order
, gfp_flags
);
747 if (order
&& (gfp_flags
& __GFP_COMP
))
748 prep_compound_page(page
, order
);
753 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
754 #define ALLOC_HARDER 0x02 /* try to alloc harder */
755 #define ALLOC_HIGH 0x04 /* __GFP_HIGH set */
756 #define ALLOC_CPUSET 0x08 /* check for correct cpuset */
759 * Return 1 if free pages are above 'mark'. This takes into account the order
762 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
763 int classzone_idx
, int alloc_flags
)
765 /* free_pages my go negative - that's OK */
766 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
769 if (alloc_flags
& ALLOC_HIGH
)
771 if (alloc_flags
& ALLOC_HARDER
)
774 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
776 for (o
= 0; o
< order
; o
++) {
777 /* At the next order, this order's pages become unavailable */
778 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
780 /* Require fewer higher order pages to be free */
783 if (free_pages
<= min
)
790 * get_page_from_freeliest goes through the zonelist trying to allocate
794 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
795 struct zonelist
*zonelist
, int alloc_flags
)
797 struct zone
**z
= zonelist
->zones
;
798 struct page
*page
= NULL
;
799 int classzone_idx
= zone_idx(*z
);
802 * Go through the zonelist once, looking for a zone with enough free.
803 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
806 if ((alloc_flags
& ALLOC_CPUSET
) &&
807 !cpuset_zone_allowed(*z
, gfp_mask
))
810 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
811 if (!zone_watermark_ok(*z
, order
, (*z
)->pages_low
,
812 classzone_idx
, alloc_flags
))
816 page
= buffered_rmqueue(*z
, order
, gfp_mask
);
818 zone_statistics(zonelist
, *z
);
821 } while (*(++z
) != NULL
);
826 * This is the 'heart' of the zoned buddy allocator.
828 struct page
* fastcall
829 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
830 struct zonelist
*zonelist
)
832 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
835 struct reclaim_state reclaim_state
;
836 struct task_struct
*p
= current
;
839 int did_some_progress
;
841 might_sleep_if(wait
);
844 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
846 if (unlikely(*z
== NULL
)) {
847 /* Should this ever happen?? */
851 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
852 zonelist
, ALLOC_CPUSET
);
857 wakeup_kswapd(*z
, order
);
861 * OK, we're below the kswapd watermark and have kicked background
862 * reclaim. Now things get more complex, so set up alloc_flags according
863 * to how we want to proceed.
865 * The caller may dip into page reserves a bit more if the caller
866 * cannot run direct reclaim, or if the caller has realtime scheduling
870 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
871 alloc_flags
|= ALLOC_HARDER
;
872 if (gfp_mask
& __GFP_HIGH
)
873 alloc_flags
|= ALLOC_HIGH
;
875 alloc_flags
|= ALLOC_CPUSET
;
878 * Go through the zonelist again. Let __GFP_HIGH and allocations
879 * coming from realtime tasks go deeper into reserves.
881 * This is the last chance, in general, before the goto nopage.
882 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
883 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
885 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
889 /* This allocation should allow future memory freeing. */
891 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
892 && !in_interrupt()) {
893 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
895 /* go through the zonelist yet again, ignoring mins */
896 page
= get_page_from_freelist(gfp_mask
, order
,
897 zonelist
, ALLOC_NO_WATERMARKS
|ALLOC_CPUSET
);
900 if (gfp_mask
& __GFP_NOFAIL
) {
901 blk_congestion_wait(WRITE
, HZ
/50);
908 /* Atomic allocations - we can't balance anything */
915 /* We now go into synchronous reclaim */
916 p
->flags
|= PF_MEMALLOC
;
917 reclaim_state
.reclaimed_slab
= 0;
918 p
->reclaim_state
= &reclaim_state
;
920 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
922 p
->reclaim_state
= NULL
;
923 p
->flags
&= ~PF_MEMALLOC
;
927 if (likely(did_some_progress
)) {
928 page
= get_page_from_freelist(gfp_mask
, order
,
929 zonelist
, alloc_flags
);
932 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
934 * Go through the zonelist yet one more time, keep
935 * very high watermark here, this is only to catch
936 * a parallel oom killing, we must fail if we're still
937 * under heavy pressure.
939 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
940 zonelist
, ALLOC_CPUSET
);
944 out_of_memory(gfp_mask
, order
);
949 * Don't let big-order allocations loop unless the caller explicitly
950 * requests that. Wait for some write requests to complete then retry.
952 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
953 * <= 3, but that may not be true in other implementations.
956 if (!(gfp_mask
& __GFP_NORETRY
)) {
957 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
959 if (gfp_mask
& __GFP_NOFAIL
)
963 blk_congestion_wait(WRITE
, HZ
/50);
968 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
969 printk(KERN_WARNING
"%s: page allocation failure."
970 " order:%d, mode:0x%x\n",
971 p
->comm
, order
, gfp_mask
);
979 EXPORT_SYMBOL(__alloc_pages
);
982 * Common helper functions.
984 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
987 page
= alloc_pages(gfp_mask
, order
);
990 return (unsigned long) page_address(page
);
993 EXPORT_SYMBOL(__get_free_pages
);
995 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1000 * get_zeroed_page() returns a 32-bit address, which cannot represent
1003 BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1005 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1007 return (unsigned long) page_address(page
);
1011 EXPORT_SYMBOL(get_zeroed_page
);
1013 void __pagevec_free(struct pagevec
*pvec
)
1015 int i
= pagevec_count(pvec
);
1018 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1021 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1023 if (put_page_testzero(page
)) {
1025 free_hot_page(page
);
1027 __free_pages_ok(page
, order
);
1031 EXPORT_SYMBOL(__free_pages
);
1033 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1036 BUG_ON(!virt_addr_valid((void *)addr
));
1037 __free_pages(virt_to_page((void *)addr
), order
);
1041 EXPORT_SYMBOL(free_pages
);
1044 * Total amount of free (allocatable) RAM:
1046 unsigned int nr_free_pages(void)
1048 unsigned int sum
= 0;
1052 sum
+= zone
->free_pages
;
1057 EXPORT_SYMBOL(nr_free_pages
);
1060 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1062 unsigned int i
, sum
= 0;
1064 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1065 sum
+= pgdat
->node_zones
[i
].free_pages
;
1071 static unsigned int nr_free_zone_pages(int offset
)
1073 /* Just pick one node, since fallback list is circular */
1074 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1075 unsigned int sum
= 0;
1077 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1078 struct zone
**zonep
= zonelist
->zones
;
1081 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1082 unsigned long size
= zone
->present_pages
;
1083 unsigned long high
= zone
->pages_high
;
1092 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1094 unsigned int nr_free_buffer_pages(void)
1096 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1100 * Amount of free RAM allocatable within all zones
1102 unsigned int nr_free_pagecache_pages(void)
1104 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1107 #ifdef CONFIG_HIGHMEM
1108 unsigned int nr_free_highpages (void)
1111 unsigned int pages
= 0;
1113 for_each_pgdat(pgdat
)
1114 pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1121 static void show_node(struct zone
*zone
)
1123 printk("Node %d ", zone
->zone_pgdat
->node_id
);
1126 #define show_node(zone) do { } while (0)
1130 * Accumulate the page_state information across all CPUs.
1131 * The result is unavoidably approximate - it can change
1132 * during and after execution of this function.
1134 static DEFINE_PER_CPU(struct page_state
, page_states
) = {0};
1136 atomic_t nr_pagecache
= ATOMIC_INIT(0);
1137 EXPORT_SYMBOL(nr_pagecache
);
1139 DEFINE_PER_CPU(long, nr_pagecache_local
) = 0;
1142 void __get_page_state(struct page_state
*ret
, int nr
, cpumask_t
*cpumask
)
1146 memset(ret
, 0, sizeof(*ret
));
1147 cpus_and(*cpumask
, *cpumask
, cpu_online_map
);
1149 cpu
= first_cpu(*cpumask
);
1150 while (cpu
< NR_CPUS
) {
1151 unsigned long *in
, *out
, off
;
1153 in
= (unsigned long *)&per_cpu(page_states
, cpu
);
1155 cpu
= next_cpu(cpu
, *cpumask
);
1158 prefetch(&per_cpu(page_states
, cpu
));
1160 out
= (unsigned long *)ret
;
1161 for (off
= 0; off
< nr
; off
++)
1166 void get_page_state_node(struct page_state
*ret
, int node
)
1169 cpumask_t mask
= node_to_cpumask(node
);
1171 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1172 nr
/= sizeof(unsigned long);
1174 __get_page_state(ret
, nr
+1, &mask
);
1177 void get_page_state(struct page_state
*ret
)
1180 cpumask_t mask
= CPU_MASK_ALL
;
1182 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1183 nr
/= sizeof(unsigned long);
1185 __get_page_state(ret
, nr
+ 1, &mask
);
1188 void get_full_page_state(struct page_state
*ret
)
1190 cpumask_t mask
= CPU_MASK_ALL
;
1192 __get_page_state(ret
, sizeof(*ret
) / sizeof(unsigned long), &mask
);
1195 unsigned long __read_page_state(unsigned long offset
)
1197 unsigned long ret
= 0;
1200 for_each_online_cpu(cpu
) {
1203 in
= (unsigned long)&per_cpu(page_states
, cpu
) + offset
;
1204 ret
+= *((unsigned long *)in
);
1209 void __mod_page_state(unsigned long offset
, unsigned long delta
)
1211 unsigned long flags
;
1214 local_irq_save(flags
);
1215 ptr
= &__get_cpu_var(page_states
);
1216 *(unsigned long*)(ptr
+ offset
) += delta
;
1217 local_irq_restore(flags
);
1220 EXPORT_SYMBOL(__mod_page_state
);
1222 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
1223 unsigned long *free
, struct pglist_data
*pgdat
)
1225 struct zone
*zones
= pgdat
->node_zones
;
1231 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1232 *active
+= zones
[i
].nr_active
;
1233 *inactive
+= zones
[i
].nr_inactive
;
1234 *free
+= zones
[i
].free_pages
;
1238 void get_zone_counts(unsigned long *active
,
1239 unsigned long *inactive
, unsigned long *free
)
1241 struct pglist_data
*pgdat
;
1246 for_each_pgdat(pgdat
) {
1247 unsigned long l
, m
, n
;
1248 __get_zone_counts(&l
, &m
, &n
, pgdat
);
1255 void si_meminfo(struct sysinfo
*val
)
1257 val
->totalram
= totalram_pages
;
1259 val
->freeram
= nr_free_pages();
1260 val
->bufferram
= nr_blockdev_pages();
1261 #ifdef CONFIG_HIGHMEM
1262 val
->totalhigh
= totalhigh_pages
;
1263 val
->freehigh
= nr_free_highpages();
1268 val
->mem_unit
= PAGE_SIZE
;
1271 EXPORT_SYMBOL(si_meminfo
);
1274 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1276 pg_data_t
*pgdat
= NODE_DATA(nid
);
1278 val
->totalram
= pgdat
->node_present_pages
;
1279 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1280 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1281 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1282 val
->mem_unit
= PAGE_SIZE
;
1286 #define K(x) ((x) << (PAGE_SHIFT-10))
1289 * Show free area list (used inside shift_scroll-lock stuff)
1290 * We also calculate the percentage fragmentation. We do this by counting the
1291 * memory on each free list with the exception of the first item on the list.
1293 void show_free_areas(void)
1295 struct page_state ps
;
1296 int cpu
, temperature
;
1297 unsigned long active
;
1298 unsigned long inactive
;
1302 for_each_zone(zone
) {
1304 printk("%s per-cpu:", zone
->name
);
1306 if (!zone
->present_pages
) {
1312 for_each_online_cpu(cpu
) {
1313 struct per_cpu_pageset
*pageset
;
1315 pageset
= zone_pcp(zone
, cpu
);
1317 for (temperature
= 0; temperature
< 2; temperature
++)
1318 printk("cpu %d %s: low %d, high %d, batch %d used:%d\n",
1320 temperature
? "cold" : "hot",
1321 pageset
->pcp
[temperature
].low
,
1322 pageset
->pcp
[temperature
].high
,
1323 pageset
->pcp
[temperature
].batch
,
1324 pageset
->pcp
[temperature
].count
);
1328 get_page_state(&ps
);
1329 get_zone_counts(&active
, &inactive
, &free
);
1331 printk("Free pages: %11ukB (%ukB HighMem)\n",
1333 K(nr_free_highpages()));
1335 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1336 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1345 ps
.nr_page_table_pages
);
1347 for_each_zone(zone
) {
1359 " pages_scanned:%lu"
1360 " all_unreclaimable? %s"
1363 K(zone
->free_pages
),
1366 K(zone
->pages_high
),
1368 K(zone
->nr_inactive
),
1369 K(zone
->present_pages
),
1370 zone
->pages_scanned
,
1371 (zone
->all_unreclaimable
? "yes" : "no")
1373 printk("lowmem_reserve[]:");
1374 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1375 printk(" %lu", zone
->lowmem_reserve
[i
]);
1379 for_each_zone(zone
) {
1380 unsigned long nr
, flags
, order
, total
= 0;
1383 printk("%s: ", zone
->name
);
1384 if (!zone
->present_pages
) {
1389 spin_lock_irqsave(&zone
->lock
, flags
);
1390 for (order
= 0; order
< MAX_ORDER
; order
++) {
1391 nr
= zone
->free_area
[order
].nr_free
;
1392 total
+= nr
<< order
;
1393 printk("%lu*%lukB ", nr
, K(1UL) << order
);
1395 spin_unlock_irqrestore(&zone
->lock
, flags
);
1396 printk("= %lukB\n", K(total
));
1399 show_swap_cache_info();
1403 * Builds allocation fallback zone lists.
1405 static int __init
build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
, int j
, int k
)
1412 zone
= pgdat
->node_zones
+ ZONE_HIGHMEM
;
1413 if (zone
->present_pages
) {
1414 #ifndef CONFIG_HIGHMEM
1417 zonelist
->zones
[j
++] = zone
;
1420 zone
= pgdat
->node_zones
+ ZONE_NORMAL
;
1421 if (zone
->present_pages
)
1422 zonelist
->zones
[j
++] = zone
;
1424 zone
= pgdat
->node_zones
+ ZONE_DMA32
;
1425 if (zone
->present_pages
)
1426 zonelist
->zones
[j
++] = zone
;
1428 zone
= pgdat
->node_zones
+ ZONE_DMA
;
1429 if (zone
->present_pages
)
1430 zonelist
->zones
[j
++] = zone
;
1436 static inline int highest_zone(int zone_bits
)
1438 int res
= ZONE_NORMAL
;
1439 if (zone_bits
& (__force
int)__GFP_HIGHMEM
)
1441 if (zone_bits
& (__force
int)__GFP_DMA32
)
1443 if (zone_bits
& (__force
int)__GFP_DMA
)
1449 #define MAX_NODE_LOAD (num_online_nodes())
1450 static int __initdata node_load
[MAX_NUMNODES
];
1452 * find_next_best_node - find the next node that should appear in a given node's fallback list
1453 * @node: node whose fallback list we're appending
1454 * @used_node_mask: nodemask_t of already used nodes
1456 * We use a number of factors to determine which is the next node that should
1457 * appear on a given node's fallback list. The node should not have appeared
1458 * already in @node's fallback list, and it should be the next closest node
1459 * according to the distance array (which contains arbitrary distance values
1460 * from each node to each node in the system), and should also prefer nodes
1461 * with no CPUs, since presumably they'll have very little allocation pressure
1462 * on them otherwise.
1463 * It returns -1 if no node is found.
1465 static int __init
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1468 int min_val
= INT_MAX
;
1471 for_each_online_node(i
) {
1474 /* Start from local node */
1475 n
= (node
+i
) % num_online_nodes();
1477 /* Don't want a node to appear more than once */
1478 if (node_isset(n
, *used_node_mask
))
1481 /* Use the local node if we haven't already */
1482 if (!node_isset(node
, *used_node_mask
)) {
1487 /* Use the distance array to find the distance */
1488 val
= node_distance(node
, n
);
1490 /* Give preference to headless and unused nodes */
1491 tmp
= node_to_cpumask(n
);
1492 if (!cpus_empty(tmp
))
1493 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1495 /* Slight preference for less loaded node */
1496 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1497 val
+= node_load
[n
];
1499 if (val
< min_val
) {
1506 node_set(best_node
, *used_node_mask
);
1511 static void __init
build_zonelists(pg_data_t
*pgdat
)
1513 int i
, j
, k
, node
, local_node
;
1514 int prev_node
, load
;
1515 struct zonelist
*zonelist
;
1516 nodemask_t used_mask
;
1518 /* initialize zonelists */
1519 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1520 zonelist
= pgdat
->node_zonelists
+ i
;
1521 zonelist
->zones
[0] = NULL
;
1524 /* NUMA-aware ordering of nodes */
1525 local_node
= pgdat
->node_id
;
1526 load
= num_online_nodes();
1527 prev_node
= local_node
;
1528 nodes_clear(used_mask
);
1529 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1531 * We don't want to pressure a particular node.
1532 * So adding penalty to the first node in same
1533 * distance group to make it round-robin.
1535 if (node_distance(local_node
, node
) !=
1536 node_distance(local_node
, prev_node
))
1537 node_load
[node
] += load
;
1540 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1541 zonelist
= pgdat
->node_zonelists
+ i
;
1542 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1544 k
= highest_zone(i
);
1546 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1547 zonelist
->zones
[j
] = NULL
;
1552 #else /* CONFIG_NUMA */
1554 static void __init
build_zonelists(pg_data_t
*pgdat
)
1556 int i
, j
, k
, node
, local_node
;
1558 local_node
= pgdat
->node_id
;
1559 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1560 struct zonelist
*zonelist
;
1562 zonelist
= pgdat
->node_zonelists
+ i
;
1565 k
= highest_zone(i
);
1566 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1568 * Now we build the zonelist so that it contains the zones
1569 * of all the other nodes.
1570 * We don't want to pressure a particular node, so when
1571 * building the zones for node N, we make sure that the
1572 * zones coming right after the local ones are those from
1573 * node N+1 (modulo N)
1575 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1576 if (!node_online(node
))
1578 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1580 for (node
= 0; node
< local_node
; node
++) {
1581 if (!node_online(node
))
1583 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1586 zonelist
->zones
[j
] = NULL
;
1590 #endif /* CONFIG_NUMA */
1592 void __init
build_all_zonelists(void)
1596 for_each_online_node(i
)
1597 build_zonelists(NODE_DATA(i
));
1598 printk("Built %i zonelists\n", num_online_nodes());
1599 cpuset_init_current_mems_allowed();
1603 * Helper functions to size the waitqueue hash table.
1604 * Essentially these want to choose hash table sizes sufficiently
1605 * large so that collisions trying to wait on pages are rare.
1606 * But in fact, the number of active page waitqueues on typical
1607 * systems is ridiculously low, less than 200. So this is even
1608 * conservative, even though it seems large.
1610 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1611 * waitqueues, i.e. the size of the waitq table given the number of pages.
1613 #define PAGES_PER_WAITQUEUE 256
1615 static inline unsigned long wait_table_size(unsigned long pages
)
1617 unsigned long size
= 1;
1619 pages
/= PAGES_PER_WAITQUEUE
;
1621 while (size
< pages
)
1625 * Once we have dozens or even hundreds of threads sleeping
1626 * on IO we've got bigger problems than wait queue collision.
1627 * Limit the size of the wait table to a reasonable size.
1629 size
= min(size
, 4096UL);
1631 return max(size
, 4UL);
1635 * This is an integer logarithm so that shifts can be used later
1636 * to extract the more random high bits from the multiplicative
1637 * hash function before the remainder is taken.
1639 static inline unsigned long wait_table_bits(unsigned long size
)
1644 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1646 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1647 unsigned long *zones_size
, unsigned long *zholes_size
)
1649 unsigned long realtotalpages
, totalpages
= 0;
1652 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1653 totalpages
+= zones_size
[i
];
1654 pgdat
->node_spanned_pages
= totalpages
;
1656 realtotalpages
= totalpages
;
1658 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1659 realtotalpages
-= zholes_size
[i
];
1660 pgdat
->node_present_pages
= realtotalpages
;
1661 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1666 * Initially all pages are reserved - free ones are freed
1667 * up by free_all_bootmem() once the early boot process is
1668 * done. Non-atomic initialization, single-pass.
1670 void __devinit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1671 unsigned long start_pfn
)
1674 unsigned long end_pfn
= start_pfn
+ size
;
1677 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++, page
++) {
1678 if (!early_pfn_valid(pfn
))
1680 if (!early_pfn_in_nid(pfn
, nid
))
1682 page
= pfn_to_page(pfn
);
1683 set_page_links(page
, zone
, nid
, pfn
);
1684 set_page_count(page
, 1);
1685 reset_page_mapcount(page
);
1686 SetPageReserved(page
);
1687 INIT_LIST_HEAD(&page
->lru
);
1688 #ifdef WANT_PAGE_VIRTUAL
1689 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1690 if (!is_highmem_idx(zone
))
1691 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1696 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1700 for (order
= 0; order
< MAX_ORDER
; order
++) {
1701 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1702 zone
->free_area
[order
].nr_free
= 0;
1706 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1707 void zonetable_add(struct zone
*zone
, int nid
, int zid
, unsigned long pfn
,
1710 unsigned long snum
= pfn_to_section_nr(pfn
);
1711 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1714 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1716 for (; snum
<= end
; snum
++)
1717 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1720 #ifndef __HAVE_ARCH_MEMMAP_INIT
1721 #define memmap_init(size, nid, zone, start_pfn) \
1722 memmap_init_zone((size), (nid), (zone), (start_pfn))
1725 static int __devinit
zone_batchsize(struct zone
*zone
)
1730 * The per-cpu-pages pools are set to around 1000th of the
1731 * size of the zone. But no more than 1/2 of a meg.
1733 * OK, so we don't know how big the cache is. So guess.
1735 batch
= zone
->present_pages
/ 1024;
1736 if (batch
* PAGE_SIZE
> 512 * 1024)
1737 batch
= (512 * 1024) / PAGE_SIZE
;
1738 batch
/= 4; /* We effectively *= 4 below */
1743 * We will be trying to allcoate bigger chunks of contiguous
1744 * memory of the order of fls(batch). This should result in
1745 * better cache coloring.
1747 * A sanity check also to ensure that batch is still in limits.
1749 batch
= (1 << fls(batch
+ batch
/2));
1751 if (fls(batch
) >= (PAGE_SHIFT
+ MAX_ORDER
- 2))
1752 batch
= PAGE_SHIFT
+ ((MAX_ORDER
- 1 - PAGE_SHIFT
)/2);
1757 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1759 struct per_cpu_pages
*pcp
;
1761 memset(p
, 0, sizeof(*p
));
1763 pcp
= &p
->pcp
[0]; /* hot */
1766 pcp
->high
= 6 * batch
;
1767 pcp
->batch
= max(1UL, 1 * batch
);
1768 INIT_LIST_HEAD(&pcp
->list
);
1770 pcp
= &p
->pcp
[1]; /* cold*/
1773 pcp
->high
= 2 * batch
;
1774 pcp
->batch
= max(1UL, batch
/2);
1775 INIT_LIST_HEAD(&pcp
->list
);
1780 * Boot pageset table. One per cpu which is going to be used for all
1781 * zones and all nodes. The parameters will be set in such a way
1782 * that an item put on a list will immediately be handed over to
1783 * the buddy list. This is safe since pageset manipulation is done
1784 * with interrupts disabled.
1786 * Some NUMA counter updates may also be caught by the boot pagesets.
1788 * The boot_pagesets must be kept even after bootup is complete for
1789 * unused processors and/or zones. They do play a role for bootstrapping
1790 * hotplugged processors.
1792 * zoneinfo_show() and maybe other functions do
1793 * not check if the processor is online before following the pageset pointer.
1794 * Other parts of the kernel may not check if the zone is available.
1796 static struct per_cpu_pageset
1797 boot_pageset
[NR_CPUS
];
1800 * Dynamically allocate memory for the
1801 * per cpu pageset array in struct zone.
1803 static int __devinit
process_zones(int cpu
)
1805 struct zone
*zone
, *dzone
;
1807 for_each_zone(zone
) {
1809 zone
->pageset
[cpu
] = kmalloc_node(sizeof(struct per_cpu_pageset
),
1810 GFP_KERNEL
, cpu_to_node(cpu
));
1811 if (!zone
->pageset
[cpu
])
1814 setup_pageset(zone
->pageset
[cpu
], zone_batchsize(zone
));
1819 for_each_zone(dzone
) {
1822 kfree(dzone
->pageset
[cpu
]);
1823 dzone
->pageset
[cpu
] = NULL
;
1828 static inline void free_zone_pagesets(int cpu
)
1833 for_each_zone(zone
) {
1834 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1836 zone_pcp(zone
, cpu
) = NULL
;
1842 static int __devinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1843 unsigned long action
,
1846 int cpu
= (long)hcpu
;
1847 int ret
= NOTIFY_OK
;
1850 case CPU_UP_PREPARE
:
1851 if (process_zones(cpu
))
1854 case CPU_UP_CANCELED
:
1856 free_zone_pagesets(cpu
);
1864 static struct notifier_block pageset_notifier
=
1865 { &pageset_cpuup_callback
, NULL
, 0 };
1867 void __init
setup_per_cpu_pageset()
1871 /* Initialize per_cpu_pageset for cpu 0.
1872 * A cpuup callback will do this for every cpu
1873 * as it comes online
1875 err
= process_zones(smp_processor_id());
1877 register_cpu_notifier(&pageset_notifier
);
1883 void zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1886 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1889 * The per-page waitqueue mechanism uses hashed waitqueues
1892 zone
->wait_table_size
= wait_table_size(zone_size_pages
);
1893 zone
->wait_table_bits
= wait_table_bits(zone
->wait_table_size
);
1894 zone
->wait_table
= (wait_queue_head_t
*)
1895 alloc_bootmem_node(pgdat
, zone
->wait_table_size
1896 * sizeof(wait_queue_head_t
));
1898 for(i
= 0; i
< zone
->wait_table_size
; ++i
)
1899 init_waitqueue_head(zone
->wait_table
+ i
);
1902 static __devinit
void zone_pcp_init(struct zone
*zone
)
1905 unsigned long batch
= zone_batchsize(zone
);
1907 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1909 /* Early boot. Slab allocator not functional yet */
1910 zone
->pageset
[cpu
] = &boot_pageset
[cpu
];
1911 setup_pageset(&boot_pageset
[cpu
],0);
1913 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1916 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1917 zone
->name
, zone
->present_pages
, batch
);
1920 static __devinit
void init_currently_empty_zone(struct zone
*zone
,
1921 unsigned long zone_start_pfn
, unsigned long size
)
1923 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1925 zone_wait_table_init(zone
, size
);
1926 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1928 zone
->zone_mem_map
= pfn_to_page(zone_start_pfn
);
1929 zone
->zone_start_pfn
= zone_start_pfn
;
1931 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1933 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
1937 * Set up the zone data structures:
1938 * - mark all pages reserved
1939 * - mark all memory queues empty
1940 * - clear the memory bitmaps
1942 static void __init
free_area_init_core(struct pglist_data
*pgdat
,
1943 unsigned long *zones_size
, unsigned long *zholes_size
)
1946 int nid
= pgdat
->node_id
;
1947 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
1949 pgdat_resize_init(pgdat
);
1950 pgdat
->nr_zones
= 0;
1951 init_waitqueue_head(&pgdat
->kswapd_wait
);
1952 pgdat
->kswapd_max_order
= 0;
1954 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
1955 struct zone
*zone
= pgdat
->node_zones
+ j
;
1956 unsigned long size
, realsize
;
1958 realsize
= size
= zones_size
[j
];
1960 realsize
-= zholes_size
[j
];
1962 if (j
< ZONE_HIGHMEM
)
1963 nr_kernel_pages
+= realsize
;
1964 nr_all_pages
+= realsize
;
1966 zone
->spanned_pages
= size
;
1967 zone
->present_pages
= realsize
;
1968 zone
->name
= zone_names
[j
];
1969 spin_lock_init(&zone
->lock
);
1970 spin_lock_init(&zone
->lru_lock
);
1971 zone_seqlock_init(zone
);
1972 zone
->zone_pgdat
= pgdat
;
1973 zone
->free_pages
= 0;
1975 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
1977 zone_pcp_init(zone
);
1978 INIT_LIST_HEAD(&zone
->active_list
);
1979 INIT_LIST_HEAD(&zone
->inactive_list
);
1980 zone
->nr_scan_active
= 0;
1981 zone
->nr_scan_inactive
= 0;
1982 zone
->nr_active
= 0;
1983 zone
->nr_inactive
= 0;
1984 atomic_set(&zone
->reclaim_in_progress
, 0);
1988 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
1989 init_currently_empty_zone(zone
, zone_start_pfn
, size
);
1990 zone_start_pfn
+= size
;
1994 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
1996 /* Skip empty nodes */
1997 if (!pgdat
->node_spanned_pages
)
2000 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2001 /* ia64 gets its own node_mem_map, before this, without bootmem */
2002 if (!pgdat
->node_mem_map
) {
2006 size
= (pgdat
->node_spanned_pages
+ 1) * sizeof(struct page
);
2007 map
= alloc_remap(pgdat
->node_id
, size
);
2009 map
= alloc_bootmem_node(pgdat
, size
);
2010 pgdat
->node_mem_map
= map
;
2012 #ifdef CONFIG_FLATMEM
2014 * With no DISCONTIG, the global mem_map is just set as node 0's
2016 if (pgdat
== NODE_DATA(0))
2017 mem_map
= NODE_DATA(0)->node_mem_map
;
2019 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2022 void __init
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2023 unsigned long *zones_size
, unsigned long node_start_pfn
,
2024 unsigned long *zholes_size
)
2026 pgdat
->node_id
= nid
;
2027 pgdat
->node_start_pfn
= node_start_pfn
;
2028 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
2030 alloc_node_mem_map(pgdat
);
2032 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2035 #ifndef CONFIG_NEED_MULTIPLE_NODES
2036 static bootmem_data_t contig_bootmem_data
;
2037 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2039 EXPORT_SYMBOL(contig_page_data
);
2042 void __init
free_area_init(unsigned long *zones_size
)
2044 free_area_init_node(0, NODE_DATA(0), zones_size
,
2045 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2048 #ifdef CONFIG_PROC_FS
2050 #include <linux/seq_file.h>
2052 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
2057 for (pgdat
= pgdat_list
; pgdat
&& node
; pgdat
= pgdat
->pgdat_next
)
2063 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2065 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2068 return pgdat
->pgdat_next
;
2071 static void frag_stop(struct seq_file
*m
, void *arg
)
2076 * This walks the free areas for each zone.
2078 static int frag_show(struct seq_file
*m
, void *arg
)
2080 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2082 struct zone
*node_zones
= pgdat
->node_zones
;
2083 unsigned long flags
;
2086 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
2087 if (!zone
->present_pages
)
2090 spin_lock_irqsave(&zone
->lock
, flags
);
2091 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
2092 for (order
= 0; order
< MAX_ORDER
; ++order
)
2093 seq_printf(m
, "%6lu ", zone
->free_area
[order
].nr_free
);
2094 spin_unlock_irqrestore(&zone
->lock
, flags
);
2100 struct seq_operations fragmentation_op
= {
2101 .start
= frag_start
,
2108 * Output information about zones in @pgdat.
2110 static int zoneinfo_show(struct seq_file
*m
, void *arg
)
2112 pg_data_t
*pgdat
= arg
;
2114 struct zone
*node_zones
= pgdat
->node_zones
;
2115 unsigned long flags
;
2117 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; zone
++) {
2120 if (!zone
->present_pages
)
2123 spin_lock_irqsave(&zone
->lock
, flags
);
2124 seq_printf(m
, "Node %d, zone %8s", pgdat
->node_id
, zone
->name
);
2132 "\n scanned %lu (a: %lu i: %lu)"
2141 zone
->pages_scanned
,
2142 zone
->nr_scan_active
, zone
->nr_scan_inactive
,
2143 zone
->spanned_pages
,
2144 zone
->present_pages
);
2146 "\n protection: (%lu",
2147 zone
->lowmem_reserve
[0]);
2148 for (i
= 1; i
< ARRAY_SIZE(zone
->lowmem_reserve
); i
++)
2149 seq_printf(m
, ", %lu", zone
->lowmem_reserve
[i
]);
2153 for (i
= 0; i
< ARRAY_SIZE(zone
->pageset
); i
++) {
2154 struct per_cpu_pageset
*pageset
;
2157 pageset
= zone_pcp(zone
, i
);
2158 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2159 if (pageset
->pcp
[j
].count
)
2162 if (j
== ARRAY_SIZE(pageset
->pcp
))
2164 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2166 "\n cpu: %i pcp: %i"
2172 pageset
->pcp
[j
].count
,
2173 pageset
->pcp
[j
].low
,
2174 pageset
->pcp
[j
].high
,
2175 pageset
->pcp
[j
].batch
);
2181 "\n numa_foreign: %lu"
2182 "\n interleave_hit: %lu"
2183 "\n local_node: %lu"
2184 "\n other_node: %lu",
2187 pageset
->numa_foreign
,
2188 pageset
->interleave_hit
,
2189 pageset
->local_node
,
2190 pageset
->other_node
);
2194 "\n all_unreclaimable: %u"
2195 "\n prev_priority: %i"
2196 "\n temp_priority: %i"
2197 "\n start_pfn: %lu",
2198 zone
->all_unreclaimable
,
2199 zone
->prev_priority
,
2200 zone
->temp_priority
,
2201 zone
->zone_start_pfn
);
2202 spin_unlock_irqrestore(&zone
->lock
, flags
);
2208 struct seq_operations zoneinfo_op
= {
2209 .start
= frag_start
, /* iterate over all zones. The same as in
2213 .show
= zoneinfo_show
,
2216 static char *vmstat_text
[] = {
2220 "nr_page_table_pages",
2245 "pgscan_kswapd_high",
2246 "pgscan_kswapd_normal",
2248 "pgscan_kswapd_dma",
2249 "pgscan_direct_high",
2250 "pgscan_direct_normal",
2251 "pgscan_direct_dma",
2256 "kswapd_inodesteal",
2264 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
2266 struct page_state
*ps
;
2268 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2271 ps
= kmalloc(sizeof(*ps
), GFP_KERNEL
);
2274 return ERR_PTR(-ENOMEM
);
2275 get_full_page_state(ps
);
2276 ps
->pgpgin
/= 2; /* sectors -> kbytes */
2278 return (unsigned long *)ps
+ *pos
;
2281 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2284 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2286 return (unsigned long *)m
->private + *pos
;
2289 static int vmstat_show(struct seq_file
*m
, void *arg
)
2291 unsigned long *l
= arg
;
2292 unsigned long off
= l
- (unsigned long *)m
->private;
2294 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
2298 static void vmstat_stop(struct seq_file
*m
, void *arg
)
2304 struct seq_operations vmstat_op
= {
2305 .start
= vmstat_start
,
2306 .next
= vmstat_next
,
2307 .stop
= vmstat_stop
,
2308 .show
= vmstat_show
,
2311 #endif /* CONFIG_PROC_FS */
2313 #ifdef CONFIG_HOTPLUG_CPU
2314 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2315 unsigned long action
, void *hcpu
)
2317 int cpu
= (unsigned long)hcpu
;
2319 unsigned long *src
, *dest
;
2321 if (action
== CPU_DEAD
) {
2324 /* Drain local pagecache count. */
2325 count
= &per_cpu(nr_pagecache_local
, cpu
);
2326 atomic_add(*count
, &nr_pagecache
);
2328 local_irq_disable();
2331 /* Add dead cpu's page_states to our own. */
2332 dest
= (unsigned long *)&__get_cpu_var(page_states
);
2333 src
= (unsigned long *)&per_cpu(page_states
, cpu
);
2335 for (i
= 0; i
< sizeof(struct page_state
)/sizeof(unsigned long);
2345 #endif /* CONFIG_HOTPLUG_CPU */
2347 void __init
page_alloc_init(void)
2349 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2353 * setup_per_zone_lowmem_reserve - called whenever
2354 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2355 * has a correct pages reserved value, so an adequate number of
2356 * pages are left in the zone after a successful __alloc_pages().
2358 static void setup_per_zone_lowmem_reserve(void)
2360 struct pglist_data
*pgdat
;
2363 for_each_pgdat(pgdat
) {
2364 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2365 struct zone
*zone
= pgdat
->node_zones
+ j
;
2366 unsigned long present_pages
= zone
->present_pages
;
2368 zone
->lowmem_reserve
[j
] = 0;
2370 for (idx
= j
-1; idx
>= 0; idx
--) {
2371 struct zone
*lower_zone
;
2373 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2374 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2376 lower_zone
= pgdat
->node_zones
+ idx
;
2377 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2378 sysctl_lowmem_reserve_ratio
[idx
];
2379 present_pages
+= lower_zone
->present_pages
;
2386 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2387 * that the pages_{min,low,high} values for each zone are set correctly
2388 * with respect to min_free_kbytes.
2390 void setup_per_zone_pages_min(void)
2392 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2393 unsigned long lowmem_pages
= 0;
2395 unsigned long flags
;
2397 /* Calculate total number of !ZONE_HIGHMEM pages */
2398 for_each_zone(zone
) {
2399 if (!is_highmem(zone
))
2400 lowmem_pages
+= zone
->present_pages
;
2403 for_each_zone(zone
) {
2405 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2406 tmp
= (pages_min
* zone
->present_pages
) / lowmem_pages
;
2407 if (is_highmem(zone
)) {
2409 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2410 * need highmem pages, so cap pages_min to a small
2413 * The (pages_high-pages_low) and (pages_low-pages_min)
2414 * deltas controls asynch page reclaim, and so should
2415 * not be capped for highmem.
2419 min_pages
= zone
->present_pages
/ 1024;
2420 if (min_pages
< SWAP_CLUSTER_MAX
)
2421 min_pages
= SWAP_CLUSTER_MAX
;
2422 if (min_pages
> 128)
2424 zone
->pages_min
= min_pages
;
2427 * If it's a lowmem zone, reserve a number of pages
2428 * proportionate to the zone's size.
2430 zone
->pages_min
= tmp
;
2433 zone
->pages_low
= zone
->pages_min
+ tmp
/ 4;
2434 zone
->pages_high
= zone
->pages_min
+ tmp
/ 2;
2435 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2440 * Initialise min_free_kbytes.
2442 * For small machines we want it small (128k min). For large machines
2443 * we want it large (64MB max). But it is not linear, because network
2444 * bandwidth does not increase linearly with machine size. We use
2446 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2447 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2463 static int __init
init_per_zone_pages_min(void)
2465 unsigned long lowmem_kbytes
;
2467 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2469 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2470 if (min_free_kbytes
< 128)
2471 min_free_kbytes
= 128;
2472 if (min_free_kbytes
> 65536)
2473 min_free_kbytes
= 65536;
2474 setup_per_zone_pages_min();
2475 setup_per_zone_lowmem_reserve();
2478 module_init(init_per_zone_pages_min
)
2481 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2482 * that we can call two helper functions whenever min_free_kbytes
2485 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2486 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2488 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2489 setup_per_zone_pages_min();
2494 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2495 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2496 * whenever sysctl_lowmem_reserve_ratio changes.
2498 * The reserve ratio obviously has absolutely no relation with the
2499 * pages_min watermarks. The lowmem reserve ratio can only make sense
2500 * if in function of the boot time zone sizes.
2502 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2503 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2505 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2506 setup_per_zone_lowmem_reserve();
2510 __initdata
int hashdist
= HASHDIST_DEFAULT
;
2513 static int __init
set_hashdist(char *str
)
2517 hashdist
= simple_strtoul(str
, &str
, 0);
2520 __setup("hashdist=", set_hashdist
);
2524 * allocate a large system hash table from bootmem
2525 * - it is assumed that the hash table must contain an exact power-of-2
2526 * quantity of entries
2527 * - limit is the number of hash buckets, not the total allocation size
2529 void *__init
alloc_large_system_hash(const char *tablename
,
2530 unsigned long bucketsize
,
2531 unsigned long numentries
,
2534 unsigned int *_hash_shift
,
2535 unsigned int *_hash_mask
,
2536 unsigned long limit
)
2538 unsigned long long max
= limit
;
2539 unsigned long log2qty
, size
;
2542 /* allow the kernel cmdline to have a say */
2544 /* round applicable memory size up to nearest megabyte */
2545 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2546 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2547 numentries
>>= 20 - PAGE_SHIFT
;
2548 numentries
<<= 20 - PAGE_SHIFT
;
2550 /* limit to 1 bucket per 2^scale bytes of low memory */
2551 if (scale
> PAGE_SHIFT
)
2552 numentries
>>= (scale
- PAGE_SHIFT
);
2554 numentries
<<= (PAGE_SHIFT
- scale
);
2556 /* rounded up to nearest power of 2 in size */
2557 numentries
= 1UL << (long_log2(numentries
) + 1);
2559 /* limit allocation size to 1/16 total memory by default */
2561 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2562 do_div(max
, bucketsize
);
2565 if (numentries
> max
)
2568 log2qty
= long_log2(numentries
);
2571 size
= bucketsize
<< log2qty
;
2572 if (flags
& HASH_EARLY
)
2573 table
= alloc_bootmem(size
);
2575 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
2577 unsigned long order
;
2578 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
2580 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
2582 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2585 panic("Failed to allocate %s hash table\n", tablename
);
2587 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2590 long_log2(size
) - PAGE_SHIFT
,
2594 *_hash_shift
= log2qty
;
2596 *_hash_mask
= (1 << log2qty
) - 1;