2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/config.h>
18 #include <linux/stddef.h>
20 #include <linux/swap.h>
21 #include <linux/interrupt.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/notifier.h>
32 #include <linux/topology.h>
33 #include <linux/sysctl.h>
34 #include <linux/cpu.h>
35 #include <linux/cpuset.h>
36 #include <linux/memory_hotplug.h>
37 #include <linux/nodemask.h>
38 #include <linux/vmalloc.h>
40 #include <asm/tlbflush.h>
44 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
47 nodemask_t node_online_map __read_mostly
= { { [0] = 1UL } };
48 EXPORT_SYMBOL(node_online_map
);
49 nodemask_t node_possible_map __read_mostly
= NODE_MASK_ALL
;
50 EXPORT_SYMBOL(node_possible_map
);
51 struct pglist_data
*pgdat_list __read_mostly
;
52 unsigned long totalram_pages __read_mostly
;
53 unsigned long totalhigh_pages __read_mostly
;
57 * results with 256, 32 in the lowmem_reserve sysctl:
58 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
59 * 1G machine -> (16M dma, 784M normal, 224M high)
60 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
61 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
62 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
64 * TBD: should special case ZONE_DMA32 machines here - in those we normally
65 * don't need any ZONE_NORMAL reservation
67 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = { 256, 256, 32 };
69 EXPORT_SYMBOL(totalram_pages
);
72 * Used by page_zone() to look up the address of the struct zone whose
73 * id is encoded in the upper bits of page->flags
75 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
76 EXPORT_SYMBOL(zone_table
);
78 static char *zone_names
[MAX_NR_ZONES
] = { "DMA", "DMA32", "Normal", "HighMem" };
79 int min_free_kbytes
= 1024;
81 unsigned long __initdata nr_kernel_pages
;
82 unsigned long __initdata nr_all_pages
;
84 #ifdef CONFIG_DEBUG_VM
85 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
89 unsigned long pfn
= page_to_pfn(page
);
92 seq
= zone_span_seqbegin(zone
);
93 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
95 else if (pfn
< zone
->zone_start_pfn
)
97 } while (zone_span_seqretry(zone
, seq
));
102 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
104 #ifdef CONFIG_HOLES_IN_ZONE
105 if (!pfn_valid(page_to_pfn(page
)))
108 if (zone
!= page_zone(page
))
114 * Temporary debugging check for pages not lying within a given zone.
116 static int bad_range(struct zone
*zone
, struct page
*page
)
118 if (page_outside_zone_boundaries(zone
, page
))
120 if (!page_is_consistent(zone
, page
))
127 static inline int bad_range(struct zone
*zone
, struct page
*page
)
133 static void bad_page(const char *function
, struct page
*page
)
135 printk(KERN_EMERG
"Bad page state at %s (in process '%s', page %p)\n",
136 function
, current
->comm
, page
);
137 printk(KERN_EMERG
"flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
138 (int)(2*sizeof(unsigned long)), (unsigned long)page
->flags
,
139 page
->mapping
, page_mapcount(page
), page_count(page
));
140 printk(KERN_EMERG
"Backtrace:\n");
142 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n");
143 page
->flags
&= ~(1 << PG_lru
|
152 set_page_count(page
, 0);
153 reset_page_mapcount(page
);
154 page
->mapping
= NULL
;
155 add_taint(TAINT_BAD_PAGE
);
159 * Higher-order pages are called "compound pages". They are structured thusly:
161 * The first PAGE_SIZE page is called the "head page".
163 * The remaining PAGE_SIZE pages are called "tail pages".
165 * All pages have PG_compound set. All pages have their ->private pointing at
166 * the head page (even the head page has this).
168 * The first tail page's ->mapping, if non-zero, holds the address of the
169 * compound page's put_page() function.
171 * The order of the allocation is stored in the first tail page's ->index
172 * This is only for debug at present. This usage means that zero-order pages
173 * may not be compound.
175 static void prep_compound_page(struct page
*page
, unsigned long order
)
178 int nr_pages
= 1 << order
;
180 page
[1].mapping
= NULL
;
181 page
[1].index
= order
;
182 for (i
= 0; i
< nr_pages
; i
++) {
183 struct page
*p
= page
+ i
;
186 set_page_private(p
, (unsigned long)page
);
190 static void destroy_compound_page(struct page
*page
, unsigned long order
)
193 int nr_pages
= 1 << order
;
195 if (!PageCompound(page
))
198 if (page
[1].index
!= order
)
199 bad_page(__FUNCTION__
, page
);
201 for (i
= 0; i
< nr_pages
; i
++) {
202 struct page
*p
= page
+ i
;
204 if (!PageCompound(p
))
205 bad_page(__FUNCTION__
, page
);
206 if (page_private(p
) != (unsigned long)page
)
207 bad_page(__FUNCTION__
, page
);
208 ClearPageCompound(p
);
213 * function for dealing with page's order in buddy system.
214 * zone->lock is already acquired when we use these.
215 * So, we don't need atomic page->flags operations here.
217 static inline unsigned long page_order(struct page
*page
) {
218 return page_private(page
);
221 static inline void set_page_order(struct page
*page
, int order
) {
222 set_page_private(page
, order
);
223 __SetPagePrivate(page
);
226 static inline void rmv_page_order(struct page
*page
)
228 __ClearPagePrivate(page
);
229 set_page_private(page
, 0);
233 * Locate the struct page for both the matching buddy in our
234 * pair (buddy1) and the combined O(n+1) page they form (page).
236 * 1) Any buddy B1 will have an order O twin B2 which satisfies
237 * the following equation:
239 * For example, if the starting buddy (buddy2) is #8 its order
241 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
243 * 2) Any buddy B will have an order O+1 parent P which
244 * satisfies the following equation:
247 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
249 static inline struct page
*
250 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
252 unsigned long buddy_idx
= page_idx
^ (1 << order
);
254 return page
+ (buddy_idx
- page_idx
);
257 static inline unsigned long
258 __find_combined_index(unsigned long page_idx
, unsigned int order
)
260 return (page_idx
& ~(1 << order
));
264 * This function checks whether a page is free && is the buddy
265 * we can do coalesce a page and its buddy if
266 * (a) the buddy is not in a hole &&
267 * (b) the buddy is free &&
268 * (c) the buddy is on the buddy system &&
269 * (d) a page and its buddy have the same order.
270 * for recording page's order, we use page_private(page) and PG_private.
273 static inline int page_is_buddy(struct page
*page
, int order
)
275 #ifdef CONFIG_HOLES_IN_ZONE
276 if (!pfn_valid(page_to_pfn(page
)))
280 if (PagePrivate(page
) &&
281 (page_order(page
) == order
) &&
282 page_count(page
) == 0)
288 * Freeing function for a buddy system allocator.
290 * The concept of a buddy system is to maintain direct-mapped table
291 * (containing bit values) for memory blocks of various "orders".
292 * The bottom level table contains the map for the smallest allocatable
293 * units of memory (here, pages), and each level above it describes
294 * pairs of units from the levels below, hence, "buddies".
295 * At a high level, all that happens here is marking the table entry
296 * at the bottom level available, and propagating the changes upward
297 * as necessary, plus some accounting needed to play nicely with other
298 * parts of the VM system.
299 * At each level, we keep a list of pages, which are heads of continuous
300 * free pages of length of (1 << order) and marked with PG_Private.Page's
301 * order is recorded in page_private(page) field.
302 * So when we are allocating or freeing one, we can derive the state of the
303 * other. That is, if we allocate a small block, and both were
304 * free, the remainder of the region must be split into blocks.
305 * If a block is freed, and its buddy is also free, then this
306 * triggers coalescing into a block of larger size.
311 static inline void __free_pages_bulk (struct page
*page
,
312 struct zone
*zone
, unsigned int order
)
314 unsigned long page_idx
;
315 int order_size
= 1 << order
;
318 destroy_compound_page(page
, order
);
320 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
322 BUG_ON(page_idx
& (order_size
- 1));
323 BUG_ON(bad_range(zone
, page
));
325 zone
->free_pages
+= order_size
;
326 while (order
< MAX_ORDER
-1) {
327 unsigned long combined_idx
;
328 struct free_area
*area
;
331 buddy
= __page_find_buddy(page
, page_idx
, order
);
332 if (!page_is_buddy(buddy
, order
))
333 break; /* Move the buddy up one level. */
335 list_del(&buddy
->lru
);
336 area
= zone
->free_area
+ order
;
338 rmv_page_order(buddy
);
339 combined_idx
= __find_combined_index(page_idx
, order
);
340 page
= page
+ (combined_idx
- page_idx
);
341 page_idx
= combined_idx
;
344 set_page_order(page
, order
);
345 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
346 zone
->free_area
[order
].nr_free
++;
349 static inline int free_pages_check(const char *function
, struct page
*page
)
351 if (unlikely(page_mapcount(page
) |
352 (page
->mapping
!= NULL
) |
353 (page_count(page
) != 0) |
363 1 << PG_reserved
))))
364 bad_page(function
, page
);
366 __ClearPageDirty(page
);
368 * For now, we report if PG_reserved was found set, but do not
369 * clear it, and do not free the page. But we shall soon need
370 * to do more, for when the ZERO_PAGE count wraps negative.
372 return PageReserved(page
);
376 * Frees a list of pages.
377 * Assumes all pages on list are in same zone, and of same order.
378 * count is the number of pages to free.
380 * If the zone was previously in an "all pages pinned" state then look to
381 * see if this freeing clears that state.
383 * And clear the zone's pages_scanned counter, to hold off the "all pages are
384 * pinned" detection logic.
387 free_pages_bulk(struct zone
*zone
, int count
,
388 struct list_head
*list
, unsigned int order
)
390 struct page
*page
= NULL
;
393 spin_lock(&zone
->lock
);
394 zone
->all_unreclaimable
= 0;
395 zone
->pages_scanned
= 0;
396 while (!list_empty(list
) && count
--) {
397 page
= list_entry(list
->prev
, struct page
, lru
);
398 /* have to delete it as __free_pages_bulk list manipulates */
399 list_del(&page
->lru
);
400 __free_pages_bulk(page
, zone
, order
);
403 spin_unlock(&zone
->lock
);
407 void __free_pages_ok(struct page
*page
, unsigned int order
)
414 arch_free_page(page
, order
);
418 for (i
= 1 ; i
< (1 << order
) ; ++i
)
419 __put_page(page
+ i
);
422 for (i
= 0 ; i
< (1 << order
) ; ++i
)
423 reserved
+= free_pages_check(__FUNCTION__
, page
+ i
);
427 list_add(&page
->lru
, &list
);
428 mod_page_state(pgfree
, 1 << order
);
429 kernel_map_pages(page
, 1<<order
, 0);
430 local_irq_save(flags
);
431 free_pages_bulk(page_zone(page
), 1, &list
, order
);
432 local_irq_restore(flags
);
437 * The order of subdivision here is critical for the IO subsystem.
438 * Please do not alter this order without good reasons and regression
439 * testing. Specifically, as large blocks of memory are subdivided,
440 * the order in which smaller blocks are delivered depends on the order
441 * they're subdivided in this function. This is the primary factor
442 * influencing the order in which pages are delivered to the IO
443 * subsystem according to empirical testing, and this is also justified
444 * by considering the behavior of a buddy system containing a single
445 * large block of memory acted on by a series of small allocations.
446 * This behavior is a critical factor in sglist merging's success.
450 static inline struct page
*
451 expand(struct zone
*zone
, struct page
*page
,
452 int low
, int high
, struct free_area
*area
)
454 unsigned long size
= 1 << high
;
460 BUG_ON(bad_range(zone
, &page
[size
]));
461 list_add(&page
[size
].lru
, &area
->free_list
);
463 set_page_order(&page
[size
], high
);
469 * This page is about to be returned from the page allocator
471 static int prep_new_page(struct page
*page
, int order
)
473 if (unlikely(page_mapcount(page
) |
474 (page
->mapping
!= NULL
) |
475 (page_count(page
) != 0) |
486 1 << PG_reserved
))))
487 bad_page(__FUNCTION__
, page
);
490 * For now, we report if PG_reserved was found set, but do not
491 * clear it, and do not allocate the page: as a safety net.
493 if (PageReserved(page
))
496 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
497 1 << PG_referenced
| 1 << PG_arch_1
|
498 1 << PG_checked
| 1 << PG_mappedtodisk
);
499 set_page_private(page
, 0);
500 set_page_refs(page
, order
);
501 kernel_map_pages(page
, 1 << order
, 1);
506 * Do the hard work of removing an element from the buddy allocator.
507 * Call me with the zone->lock already held.
509 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
511 struct free_area
* area
;
512 unsigned int current_order
;
515 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
516 area
= zone
->free_area
+ current_order
;
517 if (list_empty(&area
->free_list
))
520 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
521 list_del(&page
->lru
);
522 rmv_page_order(page
);
524 zone
->free_pages
-= 1UL << order
;
525 return expand(zone
, page
, order
, current_order
, area
);
532 * Obtain a specified number of elements from the buddy allocator, all under
533 * a single hold of the lock, for efficiency. Add them to the supplied list.
534 * Returns the number of new pages which were placed at *list.
536 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
537 unsigned long count
, struct list_head
*list
)
543 spin_lock(&zone
->lock
);
544 for (i
= 0; i
< count
; ++i
) {
545 page
= __rmqueue(zone
, order
);
549 list_add_tail(&page
->lru
, list
);
551 spin_unlock(&zone
->lock
);
556 /* Called from the slab reaper to drain remote pagesets */
557 void drain_remote_pages(void)
563 local_irq_save(flags
);
564 for_each_zone(zone
) {
565 struct per_cpu_pageset
*pset
;
567 /* Do not drain local pagesets */
568 if (zone
->zone_pgdat
->node_id
== numa_node_id())
571 pset
= zone
->pageset
[smp_processor_id()];
572 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
573 struct per_cpu_pages
*pcp
;
577 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
581 local_irq_restore(flags
);
585 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
586 static void __drain_pages(unsigned int cpu
)
592 for_each_zone(zone
) {
593 struct per_cpu_pageset
*pset
;
595 pset
= zone_pcp(zone
, cpu
);
596 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
597 struct per_cpu_pages
*pcp
;
600 local_irq_save(flags
);
601 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
603 local_irq_restore(flags
);
607 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
611 void mark_free_pages(struct zone
*zone
)
613 unsigned long zone_pfn
, flags
;
615 struct list_head
*curr
;
617 if (!zone
->spanned_pages
)
620 spin_lock_irqsave(&zone
->lock
, flags
);
621 for (zone_pfn
= 0; zone_pfn
< zone
->spanned_pages
; ++zone_pfn
)
622 ClearPageNosaveFree(pfn_to_page(zone_pfn
+ zone
->zone_start_pfn
));
624 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
625 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
626 unsigned long start_pfn
, i
;
628 start_pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
630 for (i
=0; i
< (1<<order
); i
++)
631 SetPageNosaveFree(pfn_to_page(start_pfn
+i
));
633 spin_unlock_irqrestore(&zone
->lock
, flags
);
637 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
639 void drain_local_pages(void)
643 local_irq_save(flags
);
644 __drain_pages(smp_processor_id());
645 local_irq_restore(flags
);
647 #endif /* CONFIG_PM */
649 static void zone_statistics(struct zonelist
*zonelist
, struct zone
*z
)
654 pg_data_t
*pg
= z
->zone_pgdat
;
655 pg_data_t
*orig
= zonelist
->zones
[0]->zone_pgdat
;
656 struct per_cpu_pageset
*p
;
658 local_irq_save(flags
);
659 cpu
= smp_processor_id();
665 zone_pcp(zonelist
->zones
[0], cpu
)->numa_foreign
++;
667 if (pg
== NODE_DATA(numa_node_id()))
671 local_irq_restore(flags
);
676 * Free a 0-order page
678 static void FASTCALL(free_hot_cold_page(struct page
*page
, int cold
));
679 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
681 struct zone
*zone
= page_zone(page
);
682 struct per_cpu_pages
*pcp
;
685 arch_free_page(page
, 0);
688 page
->mapping
= NULL
;
689 if (free_pages_check(__FUNCTION__
, page
))
692 inc_page_state(pgfree
);
693 kernel_map_pages(page
, 1, 0);
695 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
696 local_irq_save(flags
);
697 list_add(&page
->lru
, &pcp
->list
);
699 if (pcp
->count
>= pcp
->high
)
700 pcp
->count
-= free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
701 local_irq_restore(flags
);
705 void fastcall
free_hot_page(struct page
*page
)
707 free_hot_cold_page(page
, 0);
710 void fastcall
free_cold_page(struct page
*page
)
712 free_hot_cold_page(page
, 1);
715 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
719 BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
720 for(i
= 0; i
< (1 << order
); i
++)
721 clear_highpage(page
+ i
);
725 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
726 * we cheat by calling it from here, in the order > 0 path. Saves a branch
730 buffered_rmqueue(struct zone
*zone
, int order
, gfp_t gfp_flags
)
734 int cold
= !!(gfp_flags
& __GFP_COLD
);
738 struct per_cpu_pages
*pcp
;
741 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
742 local_irq_save(flags
);
743 if (pcp
->count
<= pcp
->low
)
744 pcp
->count
+= rmqueue_bulk(zone
, 0,
745 pcp
->batch
, &pcp
->list
);
746 if (likely(pcp
->count
)) {
747 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
748 list_del(&page
->lru
);
751 local_irq_restore(flags
);
754 spin_lock_irqsave(&zone
->lock
, flags
);
755 page
= __rmqueue(zone
, order
);
756 spin_unlock_irqrestore(&zone
->lock
, flags
);
760 BUG_ON(bad_range(zone
, page
));
761 mod_page_state_zone(zone
, pgalloc
, 1 << order
);
762 if (prep_new_page(page
, order
))
765 if (gfp_flags
& __GFP_ZERO
)
766 prep_zero_page(page
, order
, gfp_flags
);
768 if (order
&& (gfp_flags
& __GFP_COMP
))
769 prep_compound_page(page
, order
);
774 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
775 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
776 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
777 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
778 #define ALLOC_HARDER 0x10 /* try to alloc harder */
779 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
780 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
783 * Return 1 if free pages are above 'mark'. This takes into account the order
786 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
787 int classzone_idx
, int alloc_flags
)
789 /* free_pages my go negative - that's OK */
790 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
793 if (alloc_flags
& ALLOC_HIGH
)
795 if (alloc_flags
& ALLOC_HARDER
)
798 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
800 for (o
= 0; o
< order
; o
++) {
801 /* At the next order, this order's pages become unavailable */
802 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
804 /* Require fewer higher order pages to be free */
807 if (free_pages
<= min
)
814 * get_page_from_freeliest goes through the zonelist trying to allocate
818 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
819 struct zonelist
*zonelist
, int alloc_flags
)
821 struct zone
**z
= zonelist
->zones
;
822 struct page
*page
= NULL
;
823 int classzone_idx
= zone_idx(*z
);
826 * Go through the zonelist once, looking for a zone with enough free.
827 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
830 if ((alloc_flags
& ALLOC_CPUSET
) &&
831 !cpuset_zone_allowed(*z
, gfp_mask
))
834 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
836 if (alloc_flags
& ALLOC_WMARK_MIN
)
837 mark
= (*z
)->pages_min
;
838 else if (alloc_flags
& ALLOC_WMARK_LOW
)
839 mark
= (*z
)->pages_low
;
841 mark
= (*z
)->pages_high
;
842 if (!zone_watermark_ok(*z
, order
, mark
,
843 classzone_idx
, alloc_flags
))
847 page
= buffered_rmqueue(*z
, order
, gfp_mask
);
849 zone_statistics(zonelist
, *z
);
852 } while (*(++z
) != NULL
);
857 * This is the 'heart' of the zoned buddy allocator.
859 struct page
* fastcall
860 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
861 struct zonelist
*zonelist
)
863 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
866 struct reclaim_state reclaim_state
;
867 struct task_struct
*p
= current
;
870 int did_some_progress
;
872 might_sleep_if(wait
);
875 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
877 if (unlikely(*z
== NULL
)) {
878 /* Should this ever happen?? */
882 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
883 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
888 wakeup_kswapd(*z
, order
);
892 * OK, we're below the kswapd watermark and have kicked background
893 * reclaim. Now things get more complex, so set up alloc_flags according
894 * to how we want to proceed.
896 * The caller may dip into page reserves a bit more if the caller
897 * cannot run direct reclaim, or if the caller has realtime scheduling
900 alloc_flags
= ALLOC_WMARK_MIN
;
901 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
902 alloc_flags
|= ALLOC_HARDER
;
903 if (gfp_mask
& __GFP_HIGH
)
904 alloc_flags
|= ALLOC_HIGH
;
905 alloc_flags
|= ALLOC_CPUSET
;
908 * Go through the zonelist again. Let __GFP_HIGH and allocations
909 * coming from realtime tasks go deeper into reserves.
911 * This is the last chance, in general, before the goto nopage.
912 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
913 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
915 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
919 /* This allocation should allow future memory freeing. */
921 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
922 && !in_interrupt()) {
923 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
925 /* go through the zonelist yet again, ignoring mins */
926 page
= get_page_from_freelist(gfp_mask
, order
,
927 zonelist
, ALLOC_NO_WATERMARKS
);
930 if (gfp_mask
& __GFP_NOFAIL
) {
931 blk_congestion_wait(WRITE
, HZ
/50);
938 /* Atomic allocations - we can't balance anything */
945 /* We now go into synchronous reclaim */
946 p
->flags
|= PF_MEMALLOC
;
947 reclaim_state
.reclaimed_slab
= 0;
948 p
->reclaim_state
= &reclaim_state
;
950 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
952 p
->reclaim_state
= NULL
;
953 p
->flags
&= ~PF_MEMALLOC
;
957 if (likely(did_some_progress
)) {
958 page
= get_page_from_freelist(gfp_mask
, order
,
959 zonelist
, alloc_flags
);
962 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
964 * Go through the zonelist yet one more time, keep
965 * very high watermark here, this is only to catch
966 * a parallel oom killing, we must fail if we're still
967 * under heavy pressure.
969 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
970 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
974 out_of_memory(gfp_mask
, order
);
979 * Don't let big-order allocations loop unless the caller explicitly
980 * requests that. Wait for some write requests to complete then retry.
982 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
983 * <= 3, but that may not be true in other implementations.
986 if (!(gfp_mask
& __GFP_NORETRY
)) {
987 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
989 if (gfp_mask
& __GFP_NOFAIL
)
993 blk_congestion_wait(WRITE
, HZ
/50);
998 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
999 printk(KERN_WARNING
"%s: page allocation failure."
1000 " order:%d, mode:0x%x\n",
1001 p
->comm
, order
, gfp_mask
);
1009 EXPORT_SYMBOL(__alloc_pages
);
1012 * Common helper functions.
1014 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1017 page
= alloc_pages(gfp_mask
, order
);
1020 return (unsigned long) page_address(page
);
1023 EXPORT_SYMBOL(__get_free_pages
);
1025 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1030 * get_zeroed_page() returns a 32-bit address, which cannot represent
1033 BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1035 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1037 return (unsigned long) page_address(page
);
1041 EXPORT_SYMBOL(get_zeroed_page
);
1043 void __pagevec_free(struct pagevec
*pvec
)
1045 int i
= pagevec_count(pvec
);
1048 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1051 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1053 if (put_page_testzero(page
)) {
1055 free_hot_page(page
);
1057 __free_pages_ok(page
, order
);
1061 EXPORT_SYMBOL(__free_pages
);
1063 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1066 BUG_ON(!virt_addr_valid((void *)addr
));
1067 __free_pages(virt_to_page((void *)addr
), order
);
1071 EXPORT_SYMBOL(free_pages
);
1074 * Total amount of free (allocatable) RAM:
1076 unsigned int nr_free_pages(void)
1078 unsigned int sum
= 0;
1082 sum
+= zone
->free_pages
;
1087 EXPORT_SYMBOL(nr_free_pages
);
1090 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1092 unsigned int i
, sum
= 0;
1094 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1095 sum
+= pgdat
->node_zones
[i
].free_pages
;
1101 static unsigned int nr_free_zone_pages(int offset
)
1103 /* Just pick one node, since fallback list is circular */
1104 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1105 unsigned int sum
= 0;
1107 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1108 struct zone
**zonep
= zonelist
->zones
;
1111 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1112 unsigned long size
= zone
->present_pages
;
1113 unsigned long high
= zone
->pages_high
;
1122 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1124 unsigned int nr_free_buffer_pages(void)
1126 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1130 * Amount of free RAM allocatable within all zones
1132 unsigned int nr_free_pagecache_pages(void)
1134 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1137 #ifdef CONFIG_HIGHMEM
1138 unsigned int nr_free_highpages (void)
1141 unsigned int pages
= 0;
1143 for_each_pgdat(pgdat
)
1144 pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1151 static void show_node(struct zone
*zone
)
1153 printk("Node %d ", zone
->zone_pgdat
->node_id
);
1156 #define show_node(zone) do { } while (0)
1160 * Accumulate the page_state information across all CPUs.
1161 * The result is unavoidably approximate - it can change
1162 * during and after execution of this function.
1164 static DEFINE_PER_CPU(struct page_state
, page_states
) = {0};
1166 atomic_t nr_pagecache
= ATOMIC_INIT(0);
1167 EXPORT_SYMBOL(nr_pagecache
);
1169 DEFINE_PER_CPU(long, nr_pagecache_local
) = 0;
1172 void __get_page_state(struct page_state
*ret
, int nr
, cpumask_t
*cpumask
)
1176 memset(ret
, 0, sizeof(*ret
));
1177 cpus_and(*cpumask
, *cpumask
, cpu_online_map
);
1179 cpu
= first_cpu(*cpumask
);
1180 while (cpu
< NR_CPUS
) {
1181 unsigned long *in
, *out
, off
;
1183 in
= (unsigned long *)&per_cpu(page_states
, cpu
);
1185 cpu
= next_cpu(cpu
, *cpumask
);
1188 prefetch(&per_cpu(page_states
, cpu
));
1190 out
= (unsigned long *)ret
;
1191 for (off
= 0; off
< nr
; off
++)
1196 void get_page_state_node(struct page_state
*ret
, int node
)
1199 cpumask_t mask
= node_to_cpumask(node
);
1201 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1202 nr
/= sizeof(unsigned long);
1204 __get_page_state(ret
, nr
+1, &mask
);
1207 void get_page_state(struct page_state
*ret
)
1210 cpumask_t mask
= CPU_MASK_ALL
;
1212 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1213 nr
/= sizeof(unsigned long);
1215 __get_page_state(ret
, nr
+ 1, &mask
);
1218 void get_full_page_state(struct page_state
*ret
)
1220 cpumask_t mask
= CPU_MASK_ALL
;
1222 __get_page_state(ret
, sizeof(*ret
) / sizeof(unsigned long), &mask
);
1225 unsigned long __read_page_state(unsigned long offset
)
1227 unsigned long ret
= 0;
1230 for_each_online_cpu(cpu
) {
1233 in
= (unsigned long)&per_cpu(page_states
, cpu
) + offset
;
1234 ret
+= *((unsigned long *)in
);
1239 void __mod_page_state(unsigned long offset
, unsigned long delta
)
1241 unsigned long flags
;
1244 local_irq_save(flags
);
1245 ptr
= &__get_cpu_var(page_states
);
1246 *(unsigned long*)(ptr
+ offset
) += delta
;
1247 local_irq_restore(flags
);
1250 EXPORT_SYMBOL(__mod_page_state
);
1252 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
1253 unsigned long *free
, struct pglist_data
*pgdat
)
1255 struct zone
*zones
= pgdat
->node_zones
;
1261 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1262 *active
+= zones
[i
].nr_active
;
1263 *inactive
+= zones
[i
].nr_inactive
;
1264 *free
+= zones
[i
].free_pages
;
1268 void get_zone_counts(unsigned long *active
,
1269 unsigned long *inactive
, unsigned long *free
)
1271 struct pglist_data
*pgdat
;
1276 for_each_pgdat(pgdat
) {
1277 unsigned long l
, m
, n
;
1278 __get_zone_counts(&l
, &m
, &n
, pgdat
);
1285 void si_meminfo(struct sysinfo
*val
)
1287 val
->totalram
= totalram_pages
;
1289 val
->freeram
= nr_free_pages();
1290 val
->bufferram
= nr_blockdev_pages();
1291 #ifdef CONFIG_HIGHMEM
1292 val
->totalhigh
= totalhigh_pages
;
1293 val
->freehigh
= nr_free_highpages();
1298 val
->mem_unit
= PAGE_SIZE
;
1301 EXPORT_SYMBOL(si_meminfo
);
1304 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1306 pg_data_t
*pgdat
= NODE_DATA(nid
);
1308 val
->totalram
= pgdat
->node_present_pages
;
1309 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1310 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1311 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1312 val
->mem_unit
= PAGE_SIZE
;
1316 #define K(x) ((x) << (PAGE_SHIFT-10))
1319 * Show free area list (used inside shift_scroll-lock stuff)
1320 * We also calculate the percentage fragmentation. We do this by counting the
1321 * memory on each free list with the exception of the first item on the list.
1323 void show_free_areas(void)
1325 struct page_state ps
;
1326 int cpu
, temperature
;
1327 unsigned long active
;
1328 unsigned long inactive
;
1332 for_each_zone(zone
) {
1334 printk("%s per-cpu:", zone
->name
);
1336 if (!zone
->present_pages
) {
1342 for_each_online_cpu(cpu
) {
1343 struct per_cpu_pageset
*pageset
;
1345 pageset
= zone_pcp(zone
, cpu
);
1347 for (temperature
= 0; temperature
< 2; temperature
++)
1348 printk("cpu %d %s: low %d, high %d, batch %d used:%d\n",
1350 temperature
? "cold" : "hot",
1351 pageset
->pcp
[temperature
].low
,
1352 pageset
->pcp
[temperature
].high
,
1353 pageset
->pcp
[temperature
].batch
,
1354 pageset
->pcp
[temperature
].count
);
1358 get_page_state(&ps
);
1359 get_zone_counts(&active
, &inactive
, &free
);
1361 printk("Free pages: %11ukB (%ukB HighMem)\n",
1363 K(nr_free_highpages()));
1365 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1366 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1375 ps
.nr_page_table_pages
);
1377 for_each_zone(zone
) {
1389 " pages_scanned:%lu"
1390 " all_unreclaimable? %s"
1393 K(zone
->free_pages
),
1396 K(zone
->pages_high
),
1398 K(zone
->nr_inactive
),
1399 K(zone
->present_pages
),
1400 zone
->pages_scanned
,
1401 (zone
->all_unreclaimable
? "yes" : "no")
1403 printk("lowmem_reserve[]:");
1404 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1405 printk(" %lu", zone
->lowmem_reserve
[i
]);
1409 for_each_zone(zone
) {
1410 unsigned long nr
, flags
, order
, total
= 0;
1413 printk("%s: ", zone
->name
);
1414 if (!zone
->present_pages
) {
1419 spin_lock_irqsave(&zone
->lock
, flags
);
1420 for (order
= 0; order
< MAX_ORDER
; order
++) {
1421 nr
= zone
->free_area
[order
].nr_free
;
1422 total
+= nr
<< order
;
1423 printk("%lu*%lukB ", nr
, K(1UL) << order
);
1425 spin_unlock_irqrestore(&zone
->lock
, flags
);
1426 printk("= %lukB\n", K(total
));
1429 show_swap_cache_info();
1433 * Builds allocation fallback zone lists.
1435 static int __init
build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
, int j
, int k
)
1442 zone
= pgdat
->node_zones
+ ZONE_HIGHMEM
;
1443 if (zone
->present_pages
) {
1444 #ifndef CONFIG_HIGHMEM
1447 zonelist
->zones
[j
++] = zone
;
1450 zone
= pgdat
->node_zones
+ ZONE_NORMAL
;
1451 if (zone
->present_pages
)
1452 zonelist
->zones
[j
++] = zone
;
1454 zone
= pgdat
->node_zones
+ ZONE_DMA32
;
1455 if (zone
->present_pages
)
1456 zonelist
->zones
[j
++] = zone
;
1458 zone
= pgdat
->node_zones
+ ZONE_DMA
;
1459 if (zone
->present_pages
)
1460 zonelist
->zones
[j
++] = zone
;
1466 static inline int highest_zone(int zone_bits
)
1468 int res
= ZONE_NORMAL
;
1469 if (zone_bits
& (__force
int)__GFP_HIGHMEM
)
1471 if (zone_bits
& (__force
int)__GFP_DMA32
)
1473 if (zone_bits
& (__force
int)__GFP_DMA
)
1479 #define MAX_NODE_LOAD (num_online_nodes())
1480 static int __initdata node_load
[MAX_NUMNODES
];
1482 * find_next_best_node - find the next node that should appear in a given node's fallback list
1483 * @node: node whose fallback list we're appending
1484 * @used_node_mask: nodemask_t of already used nodes
1486 * We use a number of factors to determine which is the next node that should
1487 * appear on a given node's fallback list. The node should not have appeared
1488 * already in @node's fallback list, and it should be the next closest node
1489 * according to the distance array (which contains arbitrary distance values
1490 * from each node to each node in the system), and should also prefer nodes
1491 * with no CPUs, since presumably they'll have very little allocation pressure
1492 * on them otherwise.
1493 * It returns -1 if no node is found.
1495 static int __init
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1498 int min_val
= INT_MAX
;
1501 for_each_online_node(i
) {
1504 /* Start from local node */
1505 n
= (node
+i
) % num_online_nodes();
1507 /* Don't want a node to appear more than once */
1508 if (node_isset(n
, *used_node_mask
))
1511 /* Use the local node if we haven't already */
1512 if (!node_isset(node
, *used_node_mask
)) {
1517 /* Use the distance array to find the distance */
1518 val
= node_distance(node
, n
);
1520 /* Give preference to headless and unused nodes */
1521 tmp
= node_to_cpumask(n
);
1522 if (!cpus_empty(tmp
))
1523 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1525 /* Slight preference for less loaded node */
1526 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1527 val
+= node_load
[n
];
1529 if (val
< min_val
) {
1536 node_set(best_node
, *used_node_mask
);
1541 static void __init
build_zonelists(pg_data_t
*pgdat
)
1543 int i
, j
, k
, node
, local_node
;
1544 int prev_node
, load
;
1545 struct zonelist
*zonelist
;
1546 nodemask_t used_mask
;
1548 /* initialize zonelists */
1549 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1550 zonelist
= pgdat
->node_zonelists
+ i
;
1551 zonelist
->zones
[0] = NULL
;
1554 /* NUMA-aware ordering of nodes */
1555 local_node
= pgdat
->node_id
;
1556 load
= num_online_nodes();
1557 prev_node
= local_node
;
1558 nodes_clear(used_mask
);
1559 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1561 * We don't want to pressure a particular node.
1562 * So adding penalty to the first node in same
1563 * distance group to make it round-robin.
1565 if (node_distance(local_node
, node
) !=
1566 node_distance(local_node
, prev_node
))
1567 node_load
[node
] += load
;
1570 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1571 zonelist
= pgdat
->node_zonelists
+ i
;
1572 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1574 k
= highest_zone(i
);
1576 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1577 zonelist
->zones
[j
] = NULL
;
1582 #else /* CONFIG_NUMA */
1584 static void __init
build_zonelists(pg_data_t
*pgdat
)
1586 int i
, j
, k
, node
, local_node
;
1588 local_node
= pgdat
->node_id
;
1589 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1590 struct zonelist
*zonelist
;
1592 zonelist
= pgdat
->node_zonelists
+ i
;
1595 k
= highest_zone(i
);
1596 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1598 * Now we build the zonelist so that it contains the zones
1599 * of all the other nodes.
1600 * We don't want to pressure a particular node, so when
1601 * building the zones for node N, we make sure that the
1602 * zones coming right after the local ones are those from
1603 * node N+1 (modulo N)
1605 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1606 if (!node_online(node
))
1608 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1610 for (node
= 0; node
< local_node
; node
++) {
1611 if (!node_online(node
))
1613 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1616 zonelist
->zones
[j
] = NULL
;
1620 #endif /* CONFIG_NUMA */
1622 void __init
build_all_zonelists(void)
1626 for_each_online_node(i
)
1627 build_zonelists(NODE_DATA(i
));
1628 printk("Built %i zonelists\n", num_online_nodes());
1629 cpuset_init_current_mems_allowed();
1633 * Helper functions to size the waitqueue hash table.
1634 * Essentially these want to choose hash table sizes sufficiently
1635 * large so that collisions trying to wait on pages are rare.
1636 * But in fact, the number of active page waitqueues on typical
1637 * systems is ridiculously low, less than 200. So this is even
1638 * conservative, even though it seems large.
1640 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1641 * waitqueues, i.e. the size of the waitq table given the number of pages.
1643 #define PAGES_PER_WAITQUEUE 256
1645 static inline unsigned long wait_table_size(unsigned long pages
)
1647 unsigned long size
= 1;
1649 pages
/= PAGES_PER_WAITQUEUE
;
1651 while (size
< pages
)
1655 * Once we have dozens or even hundreds of threads sleeping
1656 * on IO we've got bigger problems than wait queue collision.
1657 * Limit the size of the wait table to a reasonable size.
1659 size
= min(size
, 4096UL);
1661 return max(size
, 4UL);
1665 * This is an integer logarithm so that shifts can be used later
1666 * to extract the more random high bits from the multiplicative
1667 * hash function before the remainder is taken.
1669 static inline unsigned long wait_table_bits(unsigned long size
)
1674 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1676 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1677 unsigned long *zones_size
, unsigned long *zholes_size
)
1679 unsigned long realtotalpages
, totalpages
= 0;
1682 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1683 totalpages
+= zones_size
[i
];
1684 pgdat
->node_spanned_pages
= totalpages
;
1686 realtotalpages
= totalpages
;
1688 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1689 realtotalpages
-= zholes_size
[i
];
1690 pgdat
->node_present_pages
= realtotalpages
;
1691 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1696 * Initially all pages are reserved - free ones are freed
1697 * up by free_all_bootmem() once the early boot process is
1698 * done. Non-atomic initialization, single-pass.
1700 void __devinit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1701 unsigned long start_pfn
)
1704 unsigned long end_pfn
= start_pfn
+ size
;
1707 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++, page
++) {
1708 if (!early_pfn_valid(pfn
))
1710 page
= pfn_to_page(pfn
);
1711 set_page_links(page
, zone
, nid
, pfn
);
1712 set_page_count(page
, 1);
1713 reset_page_mapcount(page
);
1714 SetPageReserved(page
);
1715 INIT_LIST_HEAD(&page
->lru
);
1716 #ifdef WANT_PAGE_VIRTUAL
1717 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1718 if (!is_highmem_idx(zone
))
1719 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1724 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1728 for (order
= 0; order
< MAX_ORDER
; order
++) {
1729 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1730 zone
->free_area
[order
].nr_free
= 0;
1734 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1735 void zonetable_add(struct zone
*zone
, int nid
, int zid
, unsigned long pfn
,
1738 unsigned long snum
= pfn_to_section_nr(pfn
);
1739 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1742 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1744 for (; snum
<= end
; snum
++)
1745 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1748 #ifndef __HAVE_ARCH_MEMMAP_INIT
1749 #define memmap_init(size, nid, zone, start_pfn) \
1750 memmap_init_zone((size), (nid), (zone), (start_pfn))
1753 static int __devinit
zone_batchsize(struct zone
*zone
)
1758 * The per-cpu-pages pools are set to around 1000th of the
1759 * size of the zone. But no more than 1/2 of a meg.
1761 * OK, so we don't know how big the cache is. So guess.
1763 batch
= zone
->present_pages
/ 1024;
1764 if (batch
* PAGE_SIZE
> 512 * 1024)
1765 batch
= (512 * 1024) / PAGE_SIZE
;
1766 batch
/= 4; /* We effectively *= 4 below */
1771 * Clamp the batch to a 2^n - 1 value. Having a power
1772 * of 2 value was found to be more likely to have
1773 * suboptimal cache aliasing properties in some cases.
1775 * For example if 2 tasks are alternately allocating
1776 * batches of pages, one task can end up with a lot
1777 * of pages of one half of the possible page colors
1778 * and the other with pages of the other colors.
1780 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1785 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1787 struct per_cpu_pages
*pcp
;
1789 memset(p
, 0, sizeof(*p
));
1791 pcp
= &p
->pcp
[0]; /* hot */
1794 pcp
->high
= 6 * batch
;
1795 pcp
->batch
= max(1UL, 1 * batch
);
1796 INIT_LIST_HEAD(&pcp
->list
);
1798 pcp
= &p
->pcp
[1]; /* cold*/
1801 pcp
->high
= 2 * batch
;
1802 pcp
->batch
= max(1UL, batch
/2);
1803 INIT_LIST_HEAD(&pcp
->list
);
1808 * Boot pageset table. One per cpu which is going to be used for all
1809 * zones and all nodes. The parameters will be set in such a way
1810 * that an item put on a list will immediately be handed over to
1811 * the buddy list. This is safe since pageset manipulation is done
1812 * with interrupts disabled.
1814 * Some NUMA counter updates may also be caught by the boot pagesets.
1816 * The boot_pagesets must be kept even after bootup is complete for
1817 * unused processors and/or zones. They do play a role for bootstrapping
1818 * hotplugged processors.
1820 * zoneinfo_show() and maybe other functions do
1821 * not check if the processor is online before following the pageset pointer.
1822 * Other parts of the kernel may not check if the zone is available.
1824 static struct per_cpu_pageset
1825 boot_pageset
[NR_CPUS
];
1828 * Dynamically allocate memory for the
1829 * per cpu pageset array in struct zone.
1831 static int __devinit
process_zones(int cpu
)
1833 struct zone
*zone
, *dzone
;
1835 for_each_zone(zone
) {
1837 zone
->pageset
[cpu
] = kmalloc_node(sizeof(struct per_cpu_pageset
),
1838 GFP_KERNEL
, cpu_to_node(cpu
));
1839 if (!zone
->pageset
[cpu
])
1842 setup_pageset(zone
->pageset
[cpu
], zone_batchsize(zone
));
1847 for_each_zone(dzone
) {
1850 kfree(dzone
->pageset
[cpu
]);
1851 dzone
->pageset
[cpu
] = NULL
;
1856 static inline void free_zone_pagesets(int cpu
)
1861 for_each_zone(zone
) {
1862 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1864 zone_pcp(zone
, cpu
) = NULL
;
1870 static int __devinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1871 unsigned long action
,
1874 int cpu
= (long)hcpu
;
1875 int ret
= NOTIFY_OK
;
1878 case CPU_UP_PREPARE
:
1879 if (process_zones(cpu
))
1882 case CPU_UP_CANCELED
:
1884 free_zone_pagesets(cpu
);
1892 static struct notifier_block pageset_notifier
=
1893 { &pageset_cpuup_callback
, NULL
, 0 };
1895 void __init
setup_per_cpu_pageset(void)
1899 /* Initialize per_cpu_pageset for cpu 0.
1900 * A cpuup callback will do this for every cpu
1901 * as it comes online
1903 err
= process_zones(smp_processor_id());
1905 register_cpu_notifier(&pageset_notifier
);
1911 void zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1914 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1917 * The per-page waitqueue mechanism uses hashed waitqueues
1920 zone
->wait_table_size
= wait_table_size(zone_size_pages
);
1921 zone
->wait_table_bits
= wait_table_bits(zone
->wait_table_size
);
1922 zone
->wait_table
= (wait_queue_head_t
*)
1923 alloc_bootmem_node(pgdat
, zone
->wait_table_size
1924 * sizeof(wait_queue_head_t
));
1926 for(i
= 0; i
< zone
->wait_table_size
; ++i
)
1927 init_waitqueue_head(zone
->wait_table
+ i
);
1930 static __devinit
void zone_pcp_init(struct zone
*zone
)
1933 unsigned long batch
= zone_batchsize(zone
);
1935 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1937 /* Early boot. Slab allocator not functional yet */
1938 zone
->pageset
[cpu
] = &boot_pageset
[cpu
];
1939 setup_pageset(&boot_pageset
[cpu
],0);
1941 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1944 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1945 zone
->name
, zone
->present_pages
, batch
);
1948 static __devinit
void init_currently_empty_zone(struct zone
*zone
,
1949 unsigned long zone_start_pfn
, unsigned long size
)
1951 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1953 zone_wait_table_init(zone
, size
);
1954 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1956 zone
->zone_mem_map
= pfn_to_page(zone_start_pfn
);
1957 zone
->zone_start_pfn
= zone_start_pfn
;
1959 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1961 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
1965 * Set up the zone data structures:
1966 * - mark all pages reserved
1967 * - mark all memory queues empty
1968 * - clear the memory bitmaps
1970 static void __init
free_area_init_core(struct pglist_data
*pgdat
,
1971 unsigned long *zones_size
, unsigned long *zholes_size
)
1974 int nid
= pgdat
->node_id
;
1975 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
1977 pgdat_resize_init(pgdat
);
1978 pgdat
->nr_zones
= 0;
1979 init_waitqueue_head(&pgdat
->kswapd_wait
);
1980 pgdat
->kswapd_max_order
= 0;
1982 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
1983 struct zone
*zone
= pgdat
->node_zones
+ j
;
1984 unsigned long size
, realsize
;
1986 realsize
= size
= zones_size
[j
];
1988 realsize
-= zholes_size
[j
];
1990 if (j
< ZONE_HIGHMEM
)
1991 nr_kernel_pages
+= realsize
;
1992 nr_all_pages
+= realsize
;
1994 zone
->spanned_pages
= size
;
1995 zone
->present_pages
= realsize
;
1996 zone
->name
= zone_names
[j
];
1997 spin_lock_init(&zone
->lock
);
1998 spin_lock_init(&zone
->lru_lock
);
1999 zone_seqlock_init(zone
);
2000 zone
->zone_pgdat
= pgdat
;
2001 zone
->free_pages
= 0;
2003 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
2005 zone_pcp_init(zone
);
2006 INIT_LIST_HEAD(&zone
->active_list
);
2007 INIT_LIST_HEAD(&zone
->inactive_list
);
2008 zone
->nr_scan_active
= 0;
2009 zone
->nr_scan_inactive
= 0;
2010 zone
->nr_active
= 0;
2011 zone
->nr_inactive
= 0;
2012 atomic_set(&zone
->reclaim_in_progress
, 0);
2016 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2017 init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2018 zone_start_pfn
+= size
;
2022 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2024 /* Skip empty nodes */
2025 if (!pgdat
->node_spanned_pages
)
2028 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2029 /* ia64 gets its own node_mem_map, before this, without bootmem */
2030 if (!pgdat
->node_mem_map
) {
2034 size
= (pgdat
->node_spanned_pages
+ 1) * sizeof(struct page
);
2035 map
= alloc_remap(pgdat
->node_id
, size
);
2037 map
= alloc_bootmem_node(pgdat
, size
);
2038 pgdat
->node_mem_map
= map
;
2040 #ifdef CONFIG_FLATMEM
2042 * With no DISCONTIG, the global mem_map is just set as node 0's
2044 if (pgdat
== NODE_DATA(0))
2045 mem_map
= NODE_DATA(0)->node_mem_map
;
2047 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2050 void __init
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2051 unsigned long *zones_size
, unsigned long node_start_pfn
,
2052 unsigned long *zholes_size
)
2054 pgdat
->node_id
= nid
;
2055 pgdat
->node_start_pfn
= node_start_pfn
;
2056 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
2058 alloc_node_mem_map(pgdat
);
2060 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2063 #ifndef CONFIG_NEED_MULTIPLE_NODES
2064 static bootmem_data_t contig_bootmem_data
;
2065 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2067 EXPORT_SYMBOL(contig_page_data
);
2070 void __init
free_area_init(unsigned long *zones_size
)
2072 free_area_init_node(0, NODE_DATA(0), zones_size
,
2073 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2076 #ifdef CONFIG_PROC_FS
2078 #include <linux/seq_file.h>
2080 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
2085 for (pgdat
= pgdat_list
; pgdat
&& node
; pgdat
= pgdat
->pgdat_next
)
2091 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2093 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2096 return pgdat
->pgdat_next
;
2099 static void frag_stop(struct seq_file
*m
, void *arg
)
2104 * This walks the free areas for each zone.
2106 static int frag_show(struct seq_file
*m
, void *arg
)
2108 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2110 struct zone
*node_zones
= pgdat
->node_zones
;
2111 unsigned long flags
;
2114 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
2115 if (!zone
->present_pages
)
2118 spin_lock_irqsave(&zone
->lock
, flags
);
2119 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
2120 for (order
= 0; order
< MAX_ORDER
; ++order
)
2121 seq_printf(m
, "%6lu ", zone
->free_area
[order
].nr_free
);
2122 spin_unlock_irqrestore(&zone
->lock
, flags
);
2128 struct seq_operations fragmentation_op
= {
2129 .start
= frag_start
,
2136 * Output information about zones in @pgdat.
2138 static int zoneinfo_show(struct seq_file
*m
, void *arg
)
2140 pg_data_t
*pgdat
= arg
;
2142 struct zone
*node_zones
= pgdat
->node_zones
;
2143 unsigned long flags
;
2145 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; zone
++) {
2148 if (!zone
->present_pages
)
2151 spin_lock_irqsave(&zone
->lock
, flags
);
2152 seq_printf(m
, "Node %d, zone %8s", pgdat
->node_id
, zone
->name
);
2160 "\n scanned %lu (a: %lu i: %lu)"
2169 zone
->pages_scanned
,
2170 zone
->nr_scan_active
, zone
->nr_scan_inactive
,
2171 zone
->spanned_pages
,
2172 zone
->present_pages
);
2174 "\n protection: (%lu",
2175 zone
->lowmem_reserve
[0]);
2176 for (i
= 1; i
< ARRAY_SIZE(zone
->lowmem_reserve
); i
++)
2177 seq_printf(m
, ", %lu", zone
->lowmem_reserve
[i
]);
2181 for (i
= 0; i
< ARRAY_SIZE(zone
->pageset
); i
++) {
2182 struct per_cpu_pageset
*pageset
;
2185 pageset
= zone_pcp(zone
, i
);
2186 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2187 if (pageset
->pcp
[j
].count
)
2190 if (j
== ARRAY_SIZE(pageset
->pcp
))
2192 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2194 "\n cpu: %i pcp: %i"
2200 pageset
->pcp
[j
].count
,
2201 pageset
->pcp
[j
].low
,
2202 pageset
->pcp
[j
].high
,
2203 pageset
->pcp
[j
].batch
);
2209 "\n numa_foreign: %lu"
2210 "\n interleave_hit: %lu"
2211 "\n local_node: %lu"
2212 "\n other_node: %lu",
2215 pageset
->numa_foreign
,
2216 pageset
->interleave_hit
,
2217 pageset
->local_node
,
2218 pageset
->other_node
);
2222 "\n all_unreclaimable: %u"
2223 "\n prev_priority: %i"
2224 "\n temp_priority: %i"
2225 "\n start_pfn: %lu",
2226 zone
->all_unreclaimable
,
2227 zone
->prev_priority
,
2228 zone
->temp_priority
,
2229 zone
->zone_start_pfn
);
2230 spin_unlock_irqrestore(&zone
->lock
, flags
);
2236 struct seq_operations zoneinfo_op
= {
2237 .start
= frag_start
, /* iterate over all zones. The same as in
2241 .show
= zoneinfo_show
,
2244 static char *vmstat_text
[] = {
2248 "nr_page_table_pages",
2273 "pgscan_kswapd_high",
2274 "pgscan_kswapd_normal",
2276 "pgscan_kswapd_dma",
2277 "pgscan_direct_high",
2278 "pgscan_direct_normal",
2279 "pgscan_direct_dma",
2284 "kswapd_inodesteal",
2292 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
2294 struct page_state
*ps
;
2296 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2299 ps
= kmalloc(sizeof(*ps
), GFP_KERNEL
);
2302 return ERR_PTR(-ENOMEM
);
2303 get_full_page_state(ps
);
2304 ps
->pgpgin
/= 2; /* sectors -> kbytes */
2306 return (unsigned long *)ps
+ *pos
;
2309 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2312 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2314 return (unsigned long *)m
->private + *pos
;
2317 static int vmstat_show(struct seq_file
*m
, void *arg
)
2319 unsigned long *l
= arg
;
2320 unsigned long off
= l
- (unsigned long *)m
->private;
2322 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
2326 static void vmstat_stop(struct seq_file
*m
, void *arg
)
2332 struct seq_operations vmstat_op
= {
2333 .start
= vmstat_start
,
2334 .next
= vmstat_next
,
2335 .stop
= vmstat_stop
,
2336 .show
= vmstat_show
,
2339 #endif /* CONFIG_PROC_FS */
2341 #ifdef CONFIG_HOTPLUG_CPU
2342 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2343 unsigned long action
, void *hcpu
)
2345 int cpu
= (unsigned long)hcpu
;
2347 unsigned long *src
, *dest
;
2349 if (action
== CPU_DEAD
) {
2352 /* Drain local pagecache count. */
2353 count
= &per_cpu(nr_pagecache_local
, cpu
);
2354 atomic_add(*count
, &nr_pagecache
);
2356 local_irq_disable();
2359 /* Add dead cpu's page_states to our own. */
2360 dest
= (unsigned long *)&__get_cpu_var(page_states
);
2361 src
= (unsigned long *)&per_cpu(page_states
, cpu
);
2363 for (i
= 0; i
< sizeof(struct page_state
)/sizeof(unsigned long);
2373 #endif /* CONFIG_HOTPLUG_CPU */
2375 void __init
page_alloc_init(void)
2377 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2381 * setup_per_zone_lowmem_reserve - called whenever
2382 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2383 * has a correct pages reserved value, so an adequate number of
2384 * pages are left in the zone after a successful __alloc_pages().
2386 static void setup_per_zone_lowmem_reserve(void)
2388 struct pglist_data
*pgdat
;
2391 for_each_pgdat(pgdat
) {
2392 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2393 struct zone
*zone
= pgdat
->node_zones
+ j
;
2394 unsigned long present_pages
= zone
->present_pages
;
2396 zone
->lowmem_reserve
[j
] = 0;
2398 for (idx
= j
-1; idx
>= 0; idx
--) {
2399 struct zone
*lower_zone
;
2401 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2402 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2404 lower_zone
= pgdat
->node_zones
+ idx
;
2405 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2406 sysctl_lowmem_reserve_ratio
[idx
];
2407 present_pages
+= lower_zone
->present_pages
;
2414 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2415 * that the pages_{min,low,high} values for each zone are set correctly
2416 * with respect to min_free_kbytes.
2418 void setup_per_zone_pages_min(void)
2420 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2421 unsigned long lowmem_pages
= 0;
2423 unsigned long flags
;
2425 /* Calculate total number of !ZONE_HIGHMEM pages */
2426 for_each_zone(zone
) {
2427 if (!is_highmem(zone
))
2428 lowmem_pages
+= zone
->present_pages
;
2431 for_each_zone(zone
) {
2433 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2434 tmp
= (pages_min
* zone
->present_pages
) / lowmem_pages
;
2435 if (is_highmem(zone
)) {
2437 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2438 * need highmem pages, so cap pages_min to a small
2441 * The (pages_high-pages_low) and (pages_low-pages_min)
2442 * deltas controls asynch page reclaim, and so should
2443 * not be capped for highmem.
2447 min_pages
= zone
->present_pages
/ 1024;
2448 if (min_pages
< SWAP_CLUSTER_MAX
)
2449 min_pages
= SWAP_CLUSTER_MAX
;
2450 if (min_pages
> 128)
2452 zone
->pages_min
= min_pages
;
2455 * If it's a lowmem zone, reserve a number of pages
2456 * proportionate to the zone's size.
2458 zone
->pages_min
= tmp
;
2461 zone
->pages_low
= zone
->pages_min
+ tmp
/ 4;
2462 zone
->pages_high
= zone
->pages_min
+ tmp
/ 2;
2463 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2468 * Initialise min_free_kbytes.
2470 * For small machines we want it small (128k min). For large machines
2471 * we want it large (64MB max). But it is not linear, because network
2472 * bandwidth does not increase linearly with machine size. We use
2474 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2475 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2491 static int __init
init_per_zone_pages_min(void)
2493 unsigned long lowmem_kbytes
;
2495 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2497 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2498 if (min_free_kbytes
< 128)
2499 min_free_kbytes
= 128;
2500 if (min_free_kbytes
> 65536)
2501 min_free_kbytes
= 65536;
2502 setup_per_zone_pages_min();
2503 setup_per_zone_lowmem_reserve();
2506 module_init(init_per_zone_pages_min
)
2509 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2510 * that we can call two helper functions whenever min_free_kbytes
2513 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2514 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2516 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2517 setup_per_zone_pages_min();
2522 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2523 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2524 * whenever sysctl_lowmem_reserve_ratio changes.
2526 * The reserve ratio obviously has absolutely no relation with the
2527 * pages_min watermarks. The lowmem reserve ratio can only make sense
2528 * if in function of the boot time zone sizes.
2530 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2531 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2533 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2534 setup_per_zone_lowmem_reserve();
2538 __initdata
int hashdist
= HASHDIST_DEFAULT
;
2541 static int __init
set_hashdist(char *str
)
2545 hashdist
= simple_strtoul(str
, &str
, 0);
2548 __setup("hashdist=", set_hashdist
);
2552 * allocate a large system hash table from bootmem
2553 * - it is assumed that the hash table must contain an exact power-of-2
2554 * quantity of entries
2555 * - limit is the number of hash buckets, not the total allocation size
2557 void *__init
alloc_large_system_hash(const char *tablename
,
2558 unsigned long bucketsize
,
2559 unsigned long numentries
,
2562 unsigned int *_hash_shift
,
2563 unsigned int *_hash_mask
,
2564 unsigned long limit
)
2566 unsigned long long max
= limit
;
2567 unsigned long log2qty
, size
;
2570 /* allow the kernel cmdline to have a say */
2572 /* round applicable memory size up to nearest megabyte */
2573 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2574 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2575 numentries
>>= 20 - PAGE_SHIFT
;
2576 numentries
<<= 20 - PAGE_SHIFT
;
2578 /* limit to 1 bucket per 2^scale bytes of low memory */
2579 if (scale
> PAGE_SHIFT
)
2580 numentries
>>= (scale
- PAGE_SHIFT
);
2582 numentries
<<= (PAGE_SHIFT
- scale
);
2584 /* rounded up to nearest power of 2 in size */
2585 numentries
= 1UL << (long_log2(numentries
) + 1);
2587 /* limit allocation size to 1/16 total memory by default */
2589 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2590 do_div(max
, bucketsize
);
2593 if (numentries
> max
)
2596 log2qty
= long_log2(numentries
);
2599 size
= bucketsize
<< log2qty
;
2600 if (flags
& HASH_EARLY
)
2601 table
= alloc_bootmem(size
);
2603 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
2605 unsigned long order
;
2606 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
2608 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
2610 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2613 panic("Failed to allocate %s hash table\n", tablename
);
2615 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2618 long_log2(size
) - PAGE_SHIFT
,
2622 *_hash_shift
= log2qty
;
2624 *_hash_mask
= (1 << log2qty
) - 1;