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
);
70 EXPORT_SYMBOL(nr_swap_pages
);
73 * Used by page_zone() to look up the address of the struct zone whose
74 * id is encoded in the upper bits of page->flags
76 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
77 EXPORT_SYMBOL(zone_table
);
79 static char *zone_names
[MAX_NR_ZONES
] = { "DMA", "DMA32", "Normal", "HighMem" };
80 int min_free_kbytes
= 1024;
82 unsigned long __initdata nr_kernel_pages
;
83 unsigned long __initdata nr_all_pages
;
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
))
126 static void bad_page(const char *function
, struct page
*page
)
128 printk(KERN_EMERG
"Bad page state at %s (in process '%s', page %p)\n",
129 function
, current
->comm
, page
);
130 printk(KERN_EMERG
"flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
131 (int)(2*sizeof(page_flags_t
)), (unsigned long)page
->flags
,
132 page
->mapping
, page_mapcount(page
), page_count(page
));
133 printk(KERN_EMERG
"Backtrace:\n");
135 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n");
136 page
->flags
&= ~(1 << PG_lru
|
146 set_page_count(page
, 0);
147 reset_page_mapcount(page
);
148 page
->mapping
= NULL
;
149 add_taint(TAINT_BAD_PAGE
);
152 #ifndef CONFIG_HUGETLB_PAGE
153 #define prep_compound_page(page, order) do { } while (0)
154 #define destroy_compound_page(page, order) do { } while (0)
157 * Higher-order pages are called "compound pages". They are structured thusly:
159 * The first PAGE_SIZE page is called the "head page".
161 * The remaining PAGE_SIZE pages are called "tail pages".
163 * All pages have PG_compound set. All pages have their ->private pointing at
164 * the head page (even the head page has this).
166 * The first tail page's ->mapping, if non-zero, holds the address of the
167 * compound page's put_page() function.
169 * The order of the allocation is stored in the first tail page's ->index
170 * This is only for debug at present. This usage means that zero-order pages
171 * may not be compound.
173 static void prep_compound_page(struct page
*page
, unsigned long order
)
176 int nr_pages
= 1 << order
;
178 page
[1].mapping
= NULL
;
179 page
[1].index
= order
;
180 for (i
= 0; i
< nr_pages
; i
++) {
181 struct page
*p
= page
+ i
;
184 set_page_private(p
, (unsigned long)page
);
188 static void destroy_compound_page(struct page
*page
, unsigned long order
)
191 int nr_pages
= 1 << order
;
193 if (!PageCompound(page
))
196 if (page
[1].index
!= order
)
197 bad_page(__FUNCTION__
, page
);
199 for (i
= 0; i
< nr_pages
; i
++) {
200 struct page
*p
= page
+ i
;
202 if (!PageCompound(p
))
203 bad_page(__FUNCTION__
, page
);
204 if (page_private(p
) != (unsigned long)page
)
205 bad_page(__FUNCTION__
, page
);
206 ClearPageCompound(p
);
209 #endif /* CONFIG_HUGETLB_PAGE */
212 * function for dealing with page's order in buddy system.
213 * zone->lock is already acquired when we use these.
214 * So, we don't need atomic page->flags operations here.
216 static inline unsigned long page_order(struct page
*page
) {
217 return page_private(page
);
220 static inline void set_page_order(struct page
*page
, int order
) {
221 set_page_private(page
, order
);
222 __SetPagePrivate(page
);
225 static inline void rmv_page_order(struct page
*page
)
227 __ClearPagePrivate(page
);
228 set_page_private(page
, 0);
232 * Locate the struct page for both the matching buddy in our
233 * pair (buddy1) and the combined O(n+1) page they form (page).
235 * 1) Any buddy B1 will have an order O twin B2 which satisfies
236 * the following equation:
238 * For example, if the starting buddy (buddy2) is #8 its order
240 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
242 * 2) Any buddy B will have an order O+1 parent P which
243 * satisfies the following equation:
246 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
248 static inline struct page
*
249 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
251 unsigned long buddy_idx
= page_idx
^ (1 << order
);
253 return page
+ (buddy_idx
- page_idx
);
256 static inline unsigned long
257 __find_combined_index(unsigned long page_idx
, unsigned int order
)
259 return (page_idx
& ~(1 << order
));
263 * This function checks whether a page is free && is the buddy
264 * we can do coalesce a page and its buddy if
265 * (a) the buddy is free &&
266 * (b) the buddy is on the buddy system &&
267 * (c) a page and its buddy have the same order.
268 * for recording page's order, we use page_private(page) and PG_private.
271 static inline int page_is_buddy(struct page
*page
, int order
)
273 if (PagePrivate(page
) &&
274 (page_order(page
) == order
) &&
275 page_count(page
) == 0)
281 * Freeing function for a buddy system allocator.
283 * The concept of a buddy system is to maintain direct-mapped table
284 * (containing bit values) for memory blocks of various "orders".
285 * The bottom level table contains the map for the smallest allocatable
286 * units of memory (here, pages), and each level above it describes
287 * pairs of units from the levels below, hence, "buddies".
288 * At a high level, all that happens here is marking the table entry
289 * at the bottom level available, and propagating the changes upward
290 * as necessary, plus some accounting needed to play nicely with other
291 * parts of the VM system.
292 * At each level, we keep a list of pages, which are heads of continuous
293 * free pages of length of (1 << order) and marked with PG_Private.Page's
294 * order is recorded in page_private(page) field.
295 * So when we are allocating or freeing one, we can derive the state of the
296 * other. That is, if we allocate a small block, and both were
297 * free, the remainder of the region must be split into blocks.
298 * If a block is freed, and its buddy is also free, then this
299 * triggers coalescing into a block of larger size.
304 static inline void __free_pages_bulk (struct page
*page
,
305 struct zone
*zone
, unsigned int order
)
307 unsigned long page_idx
;
308 int order_size
= 1 << order
;
311 destroy_compound_page(page
, order
);
313 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
315 BUG_ON(page_idx
& (order_size
- 1));
316 BUG_ON(bad_range(zone
, page
));
318 zone
->free_pages
+= order_size
;
319 while (order
< MAX_ORDER
-1) {
320 unsigned long combined_idx
;
321 struct free_area
*area
;
324 combined_idx
= __find_combined_index(page_idx
, order
);
325 buddy
= __page_find_buddy(page
, page_idx
, order
);
327 if (bad_range(zone
, buddy
))
329 if (!page_is_buddy(buddy
, order
))
330 break; /* Move the buddy up one level. */
331 list_del(&buddy
->lru
);
332 area
= zone
->free_area
+ order
;
334 rmv_page_order(buddy
);
335 page
= page
+ (combined_idx
- page_idx
);
336 page_idx
= combined_idx
;
339 set_page_order(page
, order
);
340 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
341 zone
->free_area
[order
].nr_free
++;
344 static inline void free_pages_check(const char *function
, struct page
*page
)
346 if ( page_mapcount(page
) ||
347 page
->mapping
!= NULL
||
348 page_count(page
) != 0 ||
359 bad_page(function
, page
);
361 __ClearPageDirty(page
);
365 * Frees a list of pages.
366 * Assumes all pages on list are in same zone, and of same order.
367 * count is the number of pages to free.
369 * If the zone was previously in an "all pages pinned" state then look to
370 * see if this freeing clears that state.
372 * And clear the zone's pages_scanned counter, to hold off the "all pages are
373 * pinned" detection logic.
376 free_pages_bulk(struct zone
*zone
, int count
,
377 struct list_head
*list
, unsigned int order
)
380 struct page
*page
= NULL
;
383 spin_lock_irqsave(&zone
->lock
, flags
);
384 zone
->all_unreclaimable
= 0;
385 zone
->pages_scanned
= 0;
386 while (!list_empty(list
) && count
--) {
387 page
= list_entry(list
->prev
, struct page
, lru
);
388 /* have to delete it as __free_pages_bulk list manipulates */
389 list_del(&page
->lru
);
390 __free_pages_bulk(page
, zone
, order
);
393 spin_unlock_irqrestore(&zone
->lock
, flags
);
397 void __free_pages_ok(struct page
*page
, unsigned int order
)
402 arch_free_page(page
, order
);
404 mod_page_state(pgfree
, 1 << order
);
408 for (i
= 1 ; i
< (1 << order
) ; ++i
)
409 __put_page(page
+ i
);
412 for (i
= 0 ; i
< (1 << order
) ; ++i
)
413 free_pages_check(__FUNCTION__
, page
+ i
);
414 list_add(&page
->lru
, &list
);
415 kernel_map_pages(page
, 1<<order
, 0);
416 free_pages_bulk(page_zone(page
), 1, &list
, order
);
421 * The order of subdivision here is critical for the IO subsystem.
422 * Please do not alter this order without good reasons and regression
423 * testing. Specifically, as large blocks of memory are subdivided,
424 * the order in which smaller blocks are delivered depends on the order
425 * they're subdivided in this function. This is the primary factor
426 * influencing the order in which pages are delivered to the IO
427 * subsystem according to empirical testing, and this is also justified
428 * by considering the behavior of a buddy system containing a single
429 * large block of memory acted on by a series of small allocations.
430 * This behavior is a critical factor in sglist merging's success.
434 static inline struct page
*
435 expand(struct zone
*zone
, struct page
*page
,
436 int low
, int high
, struct free_area
*area
)
438 unsigned long size
= 1 << high
;
444 BUG_ON(bad_range(zone
, &page
[size
]));
445 list_add(&page
[size
].lru
, &area
->free_list
);
447 set_page_order(&page
[size
], high
);
452 void set_page_refs(struct page
*page
, int order
)
455 set_page_count(page
, 1);
460 * We need to reference all the pages for this order, otherwise if
461 * anyone accesses one of the pages with (get/put) it will be freed.
462 * - eg: access_process_vm()
464 for (i
= 0; i
< (1 << order
); i
++)
465 set_page_count(page
+ i
, 1);
466 #endif /* CONFIG_MMU */
470 * This page is about to be returned from the page allocator
472 static void prep_new_page(struct page
*page
, int order
)
474 if ( page_mapcount(page
) ||
475 page
->mapping
!= NULL
||
476 page_count(page
) != 0 ||
488 bad_page(__FUNCTION__
, page
);
490 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
491 1 << PG_referenced
| 1 << PG_arch_1
|
492 1 << PG_checked
| 1 << PG_mappedtodisk
);
493 set_page_private(page
, 0);
494 set_page_refs(page
, order
);
495 kernel_map_pages(page
, 1 << order
, 1);
499 * Do the hard work of removing an element from the buddy allocator.
500 * Call me with the zone->lock already held.
502 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
504 struct free_area
* area
;
505 unsigned int current_order
;
508 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
509 area
= zone
->free_area
+ current_order
;
510 if (list_empty(&area
->free_list
))
513 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
514 list_del(&page
->lru
);
515 rmv_page_order(page
);
517 zone
->free_pages
-= 1UL << order
;
518 return expand(zone
, page
, order
, current_order
, area
);
525 * Obtain a specified number of elements from the buddy allocator, all under
526 * a single hold of the lock, for efficiency. Add them to the supplied list.
527 * Returns the number of new pages which were placed at *list.
529 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
530 unsigned long count
, struct list_head
*list
)
537 spin_lock_irqsave(&zone
->lock
, flags
);
538 for (i
= 0; i
< count
; ++i
) {
539 page
= __rmqueue(zone
, order
);
543 list_add_tail(&page
->lru
, list
);
545 spin_unlock_irqrestore(&zone
->lock
, flags
);
550 /* Called from the slab reaper to drain remote pagesets */
551 void drain_remote_pages(void)
557 local_irq_save(flags
);
558 for_each_zone(zone
) {
559 struct per_cpu_pageset
*pset
;
561 /* Do not drain local pagesets */
562 if (zone
->zone_pgdat
->node_id
== numa_node_id())
565 pset
= zone
->pageset
[smp_processor_id()];
566 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
567 struct per_cpu_pages
*pcp
;
571 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
575 local_irq_restore(flags
);
579 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
580 static void __drain_pages(unsigned int cpu
)
585 for_each_zone(zone
) {
586 struct per_cpu_pageset
*pset
;
588 pset
= zone_pcp(zone
, cpu
);
589 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
590 struct per_cpu_pages
*pcp
;
593 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
598 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
602 void mark_free_pages(struct zone
*zone
)
604 unsigned long zone_pfn
, flags
;
606 struct list_head
*curr
;
608 if (!zone
->spanned_pages
)
611 spin_lock_irqsave(&zone
->lock
, flags
);
612 for (zone_pfn
= 0; zone_pfn
< zone
->spanned_pages
; ++zone_pfn
)
613 ClearPageNosaveFree(pfn_to_page(zone_pfn
+ zone
->zone_start_pfn
));
615 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
616 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
617 unsigned long start_pfn
, i
;
619 start_pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
621 for (i
=0; i
< (1<<order
); i
++)
622 SetPageNosaveFree(pfn_to_page(start_pfn
+i
));
624 spin_unlock_irqrestore(&zone
->lock
, flags
);
628 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
630 void drain_local_pages(void)
634 local_irq_save(flags
);
635 __drain_pages(smp_processor_id());
636 local_irq_restore(flags
);
638 #endif /* CONFIG_PM */
640 static void zone_statistics(struct zonelist
*zonelist
, struct zone
*z
)
645 pg_data_t
*pg
= z
->zone_pgdat
;
646 pg_data_t
*orig
= zonelist
->zones
[0]->zone_pgdat
;
647 struct per_cpu_pageset
*p
;
649 local_irq_save(flags
);
650 cpu
= smp_processor_id();
656 zone_pcp(zonelist
->zones
[0], cpu
)->numa_foreign
++;
658 if (pg
== NODE_DATA(numa_node_id()))
662 local_irq_restore(flags
);
667 * Free a 0-order page
669 static void FASTCALL(free_hot_cold_page(struct page
*page
, int cold
));
670 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
672 struct zone
*zone
= page_zone(page
);
673 struct per_cpu_pages
*pcp
;
676 arch_free_page(page
, 0);
678 kernel_map_pages(page
, 1, 0);
679 inc_page_state(pgfree
);
681 page
->mapping
= NULL
;
682 free_pages_check(__FUNCTION__
, page
);
683 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
684 local_irq_save(flags
);
685 list_add(&page
->lru
, &pcp
->list
);
687 if (pcp
->count
>= pcp
->high
)
688 pcp
->count
-= free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
689 local_irq_restore(flags
);
693 void fastcall
free_hot_page(struct page
*page
)
695 free_hot_cold_page(page
, 0);
698 void fastcall
free_cold_page(struct page
*page
)
700 free_hot_cold_page(page
, 1);
703 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
707 BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
708 for(i
= 0; i
< (1 << order
); i
++)
709 clear_highpage(page
+ i
);
713 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
714 * we cheat by calling it from here, in the order > 0 path. Saves a branch
718 buffered_rmqueue(struct zone
*zone
, int order
, gfp_t gfp_flags
)
721 struct page
*page
= NULL
;
722 int cold
= !!(gfp_flags
& __GFP_COLD
);
725 struct per_cpu_pages
*pcp
;
727 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
728 local_irq_save(flags
);
729 if (pcp
->count
<= pcp
->low
)
730 pcp
->count
+= rmqueue_bulk(zone
, 0,
731 pcp
->batch
, &pcp
->list
);
733 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
734 list_del(&page
->lru
);
737 local_irq_restore(flags
);
742 spin_lock_irqsave(&zone
->lock
, flags
);
743 page
= __rmqueue(zone
, order
);
744 spin_unlock_irqrestore(&zone
->lock
, flags
);
748 BUG_ON(bad_range(zone
, page
));
749 mod_page_state_zone(zone
, pgalloc
, 1 << order
);
750 prep_new_page(page
, order
);
752 if (gfp_flags
& __GFP_ZERO
)
753 prep_zero_page(page
, order
, gfp_flags
);
755 if (order
&& (gfp_flags
& __GFP_COMP
))
756 prep_compound_page(page
, order
);
762 * Return 1 if free pages are above 'mark'. This takes into account the order
765 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
766 int classzone_idx
, int can_try_harder
, gfp_t gfp_high
)
768 /* free_pages my go negative - that's OK */
769 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
777 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
779 for (o
= 0; o
< order
; o
++) {
780 /* At the next order, this order's pages become unavailable */
781 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
783 /* Require fewer higher order pages to be free */
786 if (free_pages
<= min
)
793 should_reclaim_zone(struct zone
*z
, gfp_t gfp_mask
)
795 if (!z
->reclaim_pages
)
797 if (gfp_mask
& __GFP_NORECLAIM
)
803 * This is the 'heart' of the zoned buddy allocator.
805 struct page
* fastcall
806 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
807 struct zonelist
*zonelist
)
809 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
810 struct zone
**zones
, *z
;
812 struct reclaim_state reclaim_state
;
813 struct task_struct
*p
= current
;
818 int did_some_progress
;
820 might_sleep_if(wait
);
823 * The caller may dip into page reserves a bit more if the caller
824 * cannot run direct reclaim, or is the caller has realtime scheduling
827 can_try_harder
= (unlikely(rt_task(p
)) && !in_interrupt()) || !wait
;
829 zones
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
831 if (unlikely(zones
[0] == NULL
)) {
832 /* Should this ever happen?? */
836 classzone_idx
= zone_idx(zones
[0]);
840 * Go through the zonelist once, looking for a zone with enough free.
841 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
843 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++) {
844 int do_reclaim
= should_reclaim_zone(z
, gfp_mask
);
846 if (!cpuset_zone_allowed(z
, __GFP_HARDWALL
))
850 * If the zone is to attempt early page reclaim then this loop
851 * will try to reclaim pages and check the watermark a second
852 * time before giving up and falling back to the next zone.
855 if (!zone_watermark_ok(z
, order
, z
->pages_low
,
856 classzone_idx
, 0, 0)) {
860 zone_reclaim(z
, gfp_mask
, order
);
861 /* Only try reclaim once */
863 goto zone_reclaim_retry
;
867 page
= buffered_rmqueue(z
, order
, gfp_mask
);
872 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++)
873 wakeup_kswapd(z
, order
);
876 * Go through the zonelist again. Let __GFP_HIGH and allocations
877 * coming from realtime tasks to go deeper into reserves
879 * This is the last chance, in general, before the goto nopage.
880 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
881 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
883 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++) {
884 if (!zone_watermark_ok(z
, order
, z
->pages_min
,
885 classzone_idx
, can_try_harder
,
886 gfp_mask
& __GFP_HIGH
))
889 if (wait
&& !cpuset_zone_allowed(z
, gfp_mask
))
892 page
= buffered_rmqueue(z
, order
, gfp_mask
);
897 /* This allocation should allow future memory freeing. */
899 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
900 && !in_interrupt()) {
901 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
902 /* go through the zonelist yet again, ignoring mins */
903 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++) {
904 if (!cpuset_zone_allowed(z
, gfp_mask
))
906 page
= buffered_rmqueue(z
, order
, gfp_mask
);
914 /* Atomic allocations - we can't balance anything */
921 /* We now go into synchronous reclaim */
922 p
->flags
|= PF_MEMALLOC
;
923 reclaim_state
.reclaimed_slab
= 0;
924 p
->reclaim_state
= &reclaim_state
;
926 did_some_progress
= try_to_free_pages(zones
, gfp_mask
);
928 p
->reclaim_state
= NULL
;
929 p
->flags
&= ~PF_MEMALLOC
;
933 if (likely(did_some_progress
)) {
934 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++) {
935 if (!zone_watermark_ok(z
, order
, z
->pages_min
,
936 classzone_idx
, can_try_harder
,
937 gfp_mask
& __GFP_HIGH
))
940 if (!cpuset_zone_allowed(z
, gfp_mask
))
943 page
= buffered_rmqueue(z
, order
, gfp_mask
);
947 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
949 * Go through the zonelist yet one more time, keep
950 * very high watermark here, this is only to catch
951 * a parallel oom killing, we must fail if we're still
952 * under heavy pressure.
954 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++) {
955 if (!zone_watermark_ok(z
, order
, z
->pages_high
,
956 classzone_idx
, 0, 0))
959 if (!cpuset_zone_allowed(z
, __GFP_HARDWALL
))
962 page
= buffered_rmqueue(z
, order
, gfp_mask
);
967 out_of_memory(gfp_mask
, order
);
972 * Don't let big-order allocations loop unless the caller explicitly
973 * requests that. Wait for some write requests to complete then retry.
975 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
976 * <= 3, but that may not be true in other implementations.
979 if (!(gfp_mask
& __GFP_NORETRY
)) {
980 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
982 if (gfp_mask
& __GFP_NOFAIL
)
986 blk_congestion_wait(WRITE
, HZ
/50);
991 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
992 printk(KERN_WARNING
"%s: page allocation failure."
993 " order:%d, mode:0x%x\n",
994 p
->comm
, order
, gfp_mask
);
1000 zone_statistics(zonelist
, z
);
1004 EXPORT_SYMBOL(__alloc_pages
);
1007 * Common helper functions.
1009 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1012 page
= alloc_pages(gfp_mask
, order
);
1015 return (unsigned long) page_address(page
);
1018 EXPORT_SYMBOL(__get_free_pages
);
1020 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1025 * get_zeroed_page() returns a 32-bit address, which cannot represent
1028 BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1030 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1032 return (unsigned long) page_address(page
);
1036 EXPORT_SYMBOL(get_zeroed_page
);
1038 void __pagevec_free(struct pagevec
*pvec
)
1040 int i
= pagevec_count(pvec
);
1043 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1046 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1048 if (put_page_testzero(page
)) {
1050 free_hot_page(page
);
1052 __free_pages_ok(page
, order
);
1056 EXPORT_SYMBOL(__free_pages
);
1058 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1061 BUG_ON(!virt_addr_valid((void *)addr
));
1062 __free_pages(virt_to_page((void *)addr
), order
);
1066 EXPORT_SYMBOL(free_pages
);
1069 * Total amount of free (allocatable) RAM:
1071 unsigned int nr_free_pages(void)
1073 unsigned int sum
= 0;
1077 sum
+= zone
->free_pages
;
1082 EXPORT_SYMBOL(nr_free_pages
);
1085 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1087 unsigned int i
, sum
= 0;
1089 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1090 sum
+= pgdat
->node_zones
[i
].free_pages
;
1096 static unsigned int nr_free_zone_pages(int offset
)
1098 /* Just pick one node, since fallback list is circular */
1099 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1100 unsigned int sum
= 0;
1102 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1103 struct zone
**zonep
= zonelist
->zones
;
1106 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1107 unsigned long size
= zone
->present_pages
;
1108 unsigned long high
= zone
->pages_high
;
1117 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1119 unsigned int nr_free_buffer_pages(void)
1121 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1125 * Amount of free RAM allocatable within all zones
1127 unsigned int nr_free_pagecache_pages(void)
1129 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1132 #ifdef CONFIG_HIGHMEM
1133 unsigned int nr_free_highpages (void)
1136 unsigned int pages
= 0;
1138 for_each_pgdat(pgdat
)
1139 pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1146 static void show_node(struct zone
*zone
)
1148 printk("Node %d ", zone
->zone_pgdat
->node_id
);
1151 #define show_node(zone) do { } while (0)
1155 * Accumulate the page_state information across all CPUs.
1156 * The result is unavoidably approximate - it can change
1157 * during and after execution of this function.
1159 static DEFINE_PER_CPU(struct page_state
, page_states
) = {0};
1161 atomic_t nr_pagecache
= ATOMIC_INIT(0);
1162 EXPORT_SYMBOL(nr_pagecache
);
1164 DEFINE_PER_CPU(long, nr_pagecache_local
) = 0;
1167 void __get_page_state(struct page_state
*ret
, int nr
, cpumask_t
*cpumask
)
1171 memset(ret
, 0, sizeof(*ret
));
1172 cpus_and(*cpumask
, *cpumask
, cpu_online_map
);
1174 cpu
= first_cpu(*cpumask
);
1175 while (cpu
< NR_CPUS
) {
1176 unsigned long *in
, *out
, off
;
1178 in
= (unsigned long *)&per_cpu(page_states
, cpu
);
1180 cpu
= next_cpu(cpu
, *cpumask
);
1183 prefetch(&per_cpu(page_states
, cpu
));
1185 out
= (unsigned long *)ret
;
1186 for (off
= 0; off
< nr
; off
++)
1191 void get_page_state_node(struct page_state
*ret
, int node
)
1194 cpumask_t mask
= node_to_cpumask(node
);
1196 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1197 nr
/= sizeof(unsigned long);
1199 __get_page_state(ret
, nr
+1, &mask
);
1202 void get_page_state(struct page_state
*ret
)
1205 cpumask_t mask
= CPU_MASK_ALL
;
1207 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
1208 nr
/= sizeof(unsigned long);
1210 __get_page_state(ret
, nr
+ 1, &mask
);
1213 void get_full_page_state(struct page_state
*ret
)
1215 cpumask_t mask
= CPU_MASK_ALL
;
1217 __get_page_state(ret
, sizeof(*ret
) / sizeof(unsigned long), &mask
);
1220 unsigned long __read_page_state(unsigned long offset
)
1222 unsigned long ret
= 0;
1225 for_each_online_cpu(cpu
) {
1228 in
= (unsigned long)&per_cpu(page_states
, cpu
) + offset
;
1229 ret
+= *((unsigned long *)in
);
1234 void __mod_page_state(unsigned long offset
, unsigned long delta
)
1236 unsigned long flags
;
1239 local_irq_save(flags
);
1240 ptr
= &__get_cpu_var(page_states
);
1241 *(unsigned long*)(ptr
+ offset
) += delta
;
1242 local_irq_restore(flags
);
1245 EXPORT_SYMBOL(__mod_page_state
);
1247 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
1248 unsigned long *free
, struct pglist_data
*pgdat
)
1250 struct zone
*zones
= pgdat
->node_zones
;
1256 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1257 *active
+= zones
[i
].nr_active
;
1258 *inactive
+= zones
[i
].nr_inactive
;
1259 *free
+= zones
[i
].free_pages
;
1263 void get_zone_counts(unsigned long *active
,
1264 unsigned long *inactive
, unsigned long *free
)
1266 struct pglist_data
*pgdat
;
1271 for_each_pgdat(pgdat
) {
1272 unsigned long l
, m
, n
;
1273 __get_zone_counts(&l
, &m
, &n
, pgdat
);
1280 void si_meminfo(struct sysinfo
*val
)
1282 val
->totalram
= totalram_pages
;
1284 val
->freeram
= nr_free_pages();
1285 val
->bufferram
= nr_blockdev_pages();
1286 #ifdef CONFIG_HIGHMEM
1287 val
->totalhigh
= totalhigh_pages
;
1288 val
->freehigh
= nr_free_highpages();
1293 val
->mem_unit
= PAGE_SIZE
;
1296 EXPORT_SYMBOL(si_meminfo
);
1299 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1301 pg_data_t
*pgdat
= NODE_DATA(nid
);
1303 val
->totalram
= pgdat
->node_present_pages
;
1304 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1305 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1306 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1307 val
->mem_unit
= PAGE_SIZE
;
1311 #define K(x) ((x) << (PAGE_SHIFT-10))
1314 * Show free area list (used inside shift_scroll-lock stuff)
1315 * We also calculate the percentage fragmentation. We do this by counting the
1316 * memory on each free list with the exception of the first item on the list.
1318 void show_free_areas(void)
1320 struct page_state ps
;
1321 int cpu
, temperature
;
1322 unsigned long active
;
1323 unsigned long inactive
;
1327 for_each_zone(zone
) {
1329 printk("%s per-cpu:", zone
->name
);
1331 if (!zone
->present_pages
) {
1338 struct per_cpu_pageset
*pageset
;
1340 pageset
= zone_pcp(zone
, cpu
);
1342 for (temperature
= 0; temperature
< 2; temperature
++)
1343 printk("cpu %d %s: low %d, high %d, batch %d used:%d\n",
1345 temperature
? "cold" : "hot",
1346 pageset
->pcp
[temperature
].low
,
1347 pageset
->pcp
[temperature
].high
,
1348 pageset
->pcp
[temperature
].batch
,
1349 pageset
->pcp
[temperature
].count
);
1353 get_page_state(&ps
);
1354 get_zone_counts(&active
, &inactive
, &free
);
1356 printk("Free pages: %11ukB (%ukB HighMem)\n",
1358 K(nr_free_highpages()));
1360 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1361 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1370 ps
.nr_page_table_pages
);
1372 for_each_zone(zone
) {
1384 " pages_scanned:%lu"
1385 " all_unreclaimable? %s"
1388 K(zone
->free_pages
),
1391 K(zone
->pages_high
),
1393 K(zone
->nr_inactive
),
1394 K(zone
->present_pages
),
1395 zone
->pages_scanned
,
1396 (zone
->all_unreclaimable
? "yes" : "no")
1398 printk("lowmem_reserve[]:");
1399 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1400 printk(" %lu", zone
->lowmem_reserve
[i
]);
1404 for_each_zone(zone
) {
1405 unsigned long nr
, flags
, order
, total
= 0;
1408 printk("%s: ", zone
->name
);
1409 if (!zone
->present_pages
) {
1414 spin_lock_irqsave(&zone
->lock
, flags
);
1415 for (order
= 0; order
< MAX_ORDER
; order
++) {
1416 nr
= zone
->free_area
[order
].nr_free
;
1417 total
+= nr
<< order
;
1418 printk("%lu*%lukB ", nr
, K(1UL) << order
);
1420 spin_unlock_irqrestore(&zone
->lock
, flags
);
1421 printk("= %lukB\n", K(total
));
1424 show_swap_cache_info();
1428 * Builds allocation fallback zone lists.
1430 static int __init
build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
, int j
, int k
)
1437 zone
= pgdat
->node_zones
+ ZONE_HIGHMEM
;
1438 if (zone
->present_pages
) {
1439 #ifndef CONFIG_HIGHMEM
1442 zonelist
->zones
[j
++] = zone
;
1445 zone
= pgdat
->node_zones
+ ZONE_NORMAL
;
1446 if (zone
->present_pages
)
1447 zonelist
->zones
[j
++] = zone
;
1449 zone
= pgdat
->node_zones
+ ZONE_DMA32
;
1450 if (zone
->present_pages
)
1451 zonelist
->zones
[j
++] = zone
;
1453 zone
= pgdat
->node_zones
+ ZONE_DMA
;
1454 if (zone
->present_pages
)
1455 zonelist
->zones
[j
++] = zone
;
1461 static inline int highest_zone(int zone_bits
)
1463 int res
= ZONE_NORMAL
;
1464 if (zone_bits
& (__force
int)__GFP_HIGHMEM
)
1466 if (zone_bits
& (__force
int)__GFP_DMA32
)
1468 if (zone_bits
& (__force
int)__GFP_DMA
)
1474 #define MAX_NODE_LOAD (num_online_nodes())
1475 static int __initdata node_load
[MAX_NUMNODES
];
1477 * find_next_best_node - find the next node that should appear in a given node's fallback list
1478 * @node: node whose fallback list we're appending
1479 * @used_node_mask: nodemask_t of already used nodes
1481 * We use a number of factors to determine which is the next node that should
1482 * appear on a given node's fallback list. The node should not have appeared
1483 * already in @node's fallback list, and it should be the next closest node
1484 * according to the distance array (which contains arbitrary distance values
1485 * from each node to each node in the system), and should also prefer nodes
1486 * with no CPUs, since presumably they'll have very little allocation pressure
1487 * on them otherwise.
1488 * It returns -1 if no node is found.
1490 static int __init
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1493 int min_val
= INT_MAX
;
1496 for_each_online_node(i
) {
1499 /* Start from local node */
1500 n
= (node
+i
) % num_online_nodes();
1502 /* Don't want a node to appear more than once */
1503 if (node_isset(n
, *used_node_mask
))
1506 /* Use the local node if we haven't already */
1507 if (!node_isset(node
, *used_node_mask
)) {
1512 /* Use the distance array to find the distance */
1513 val
= node_distance(node
, n
);
1515 /* Give preference to headless and unused nodes */
1516 tmp
= node_to_cpumask(n
);
1517 if (!cpus_empty(tmp
))
1518 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1520 /* Slight preference for less loaded node */
1521 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1522 val
+= node_load
[n
];
1524 if (val
< min_val
) {
1531 node_set(best_node
, *used_node_mask
);
1536 static void __init
build_zonelists(pg_data_t
*pgdat
)
1538 int i
, j
, k
, node
, local_node
;
1539 int prev_node
, load
;
1540 struct zonelist
*zonelist
;
1541 nodemask_t used_mask
;
1543 /* initialize zonelists */
1544 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1545 zonelist
= pgdat
->node_zonelists
+ i
;
1546 zonelist
->zones
[0] = NULL
;
1549 /* NUMA-aware ordering of nodes */
1550 local_node
= pgdat
->node_id
;
1551 load
= num_online_nodes();
1552 prev_node
= local_node
;
1553 nodes_clear(used_mask
);
1554 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1556 * We don't want to pressure a particular node.
1557 * So adding penalty to the first node in same
1558 * distance group to make it round-robin.
1560 if (node_distance(local_node
, node
) !=
1561 node_distance(local_node
, prev_node
))
1562 node_load
[node
] += load
;
1565 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1566 zonelist
= pgdat
->node_zonelists
+ i
;
1567 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1569 k
= highest_zone(i
);
1571 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1572 zonelist
->zones
[j
] = NULL
;
1577 #else /* CONFIG_NUMA */
1579 static void __init
build_zonelists(pg_data_t
*pgdat
)
1581 int i
, j
, k
, node
, local_node
;
1583 local_node
= pgdat
->node_id
;
1584 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1585 struct zonelist
*zonelist
;
1587 zonelist
= pgdat
->node_zonelists
+ i
;
1590 k
= highest_zone(i
);
1591 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1593 * Now we build the zonelist so that it contains the zones
1594 * of all the other nodes.
1595 * We don't want to pressure a particular node, so when
1596 * building the zones for node N, we make sure that the
1597 * zones coming right after the local ones are those from
1598 * node N+1 (modulo N)
1600 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1601 if (!node_online(node
))
1603 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1605 for (node
= 0; node
< local_node
; node
++) {
1606 if (!node_online(node
))
1608 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1611 zonelist
->zones
[j
] = NULL
;
1615 #endif /* CONFIG_NUMA */
1617 void __init
build_all_zonelists(void)
1621 for_each_online_node(i
)
1622 build_zonelists(NODE_DATA(i
));
1623 printk("Built %i zonelists\n", num_online_nodes());
1624 cpuset_init_current_mems_allowed();
1628 * Helper functions to size the waitqueue hash table.
1629 * Essentially these want to choose hash table sizes sufficiently
1630 * large so that collisions trying to wait on pages are rare.
1631 * But in fact, the number of active page waitqueues on typical
1632 * systems is ridiculously low, less than 200. So this is even
1633 * conservative, even though it seems large.
1635 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1636 * waitqueues, i.e. the size of the waitq table given the number of pages.
1638 #define PAGES_PER_WAITQUEUE 256
1640 static inline unsigned long wait_table_size(unsigned long pages
)
1642 unsigned long size
= 1;
1644 pages
/= PAGES_PER_WAITQUEUE
;
1646 while (size
< pages
)
1650 * Once we have dozens or even hundreds of threads sleeping
1651 * on IO we've got bigger problems than wait queue collision.
1652 * Limit the size of the wait table to a reasonable size.
1654 size
= min(size
, 4096UL);
1656 return max(size
, 4UL);
1660 * This is an integer logarithm so that shifts can be used later
1661 * to extract the more random high bits from the multiplicative
1662 * hash function before the remainder is taken.
1664 static inline unsigned long wait_table_bits(unsigned long size
)
1669 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1671 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1672 unsigned long *zones_size
, unsigned long *zholes_size
)
1674 unsigned long realtotalpages
, totalpages
= 0;
1677 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1678 totalpages
+= zones_size
[i
];
1679 pgdat
->node_spanned_pages
= totalpages
;
1681 realtotalpages
= totalpages
;
1683 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1684 realtotalpages
-= zholes_size
[i
];
1685 pgdat
->node_present_pages
= realtotalpages
;
1686 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1691 * Initially all pages are reserved - free ones are freed
1692 * up by free_all_bootmem() once the early boot process is
1693 * done. Non-atomic initialization, single-pass.
1695 void __devinit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1696 unsigned long start_pfn
)
1699 unsigned long end_pfn
= start_pfn
+ size
;
1702 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++, page
++) {
1703 if (!early_pfn_valid(pfn
))
1705 if (!early_pfn_in_nid(pfn
, nid
))
1707 page
= pfn_to_page(pfn
);
1708 set_page_links(page
, zone
, nid
, pfn
);
1709 set_page_count(page
, 1);
1710 reset_page_mapcount(page
);
1711 SetPageReserved(page
);
1712 INIT_LIST_HEAD(&page
->lru
);
1713 #ifdef WANT_PAGE_VIRTUAL
1714 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1715 if (!is_highmem_idx(zone
))
1716 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1721 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1725 for (order
= 0; order
< MAX_ORDER
; order
++) {
1726 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1727 zone
->free_area
[order
].nr_free
= 0;
1731 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1732 void zonetable_add(struct zone
*zone
, int nid
, int zid
, unsigned long pfn
,
1735 unsigned long snum
= pfn_to_section_nr(pfn
);
1736 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1739 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1741 for (; snum
<= end
; snum
++)
1742 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1745 #ifndef __HAVE_ARCH_MEMMAP_INIT
1746 #define memmap_init(size, nid, zone, start_pfn) \
1747 memmap_init_zone((size), (nid), (zone), (start_pfn))
1750 static int __devinit
zone_batchsize(struct zone
*zone
)
1755 * The per-cpu-pages pools are set to around 1000th of the
1756 * size of the zone. But no more than 1/2 of a meg.
1758 * OK, so we don't know how big the cache is. So guess.
1760 batch
= zone
->present_pages
/ 1024;
1761 if (batch
* PAGE_SIZE
> 512 * 1024)
1762 batch
= (512 * 1024) / PAGE_SIZE
;
1763 batch
/= 4; /* We effectively *= 4 below */
1768 * We will be trying to allcoate bigger chunks of contiguous
1769 * memory of the order of fls(batch). This should result in
1770 * better cache coloring.
1772 * A sanity check also to ensure that batch is still in limits.
1774 batch
= (1 << fls(batch
+ batch
/2));
1776 if (fls(batch
) >= (PAGE_SHIFT
+ MAX_ORDER
- 2))
1777 batch
= PAGE_SHIFT
+ ((MAX_ORDER
- 1 - PAGE_SHIFT
)/2);
1782 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1784 struct per_cpu_pages
*pcp
;
1786 memset(p
, 0, sizeof(*p
));
1788 pcp
= &p
->pcp
[0]; /* hot */
1791 pcp
->high
= 6 * batch
;
1792 pcp
->batch
= max(1UL, 1 * batch
);
1793 INIT_LIST_HEAD(&pcp
->list
);
1795 pcp
= &p
->pcp
[1]; /* cold*/
1798 pcp
->high
= 2 * batch
;
1799 pcp
->batch
= max(1UL, batch
/2);
1800 INIT_LIST_HEAD(&pcp
->list
);
1805 * Boot pageset table. One per cpu which is going to be used for all
1806 * zones and all nodes. The parameters will be set in such a way
1807 * that an item put on a list will immediately be handed over to
1808 * the buddy list. This is safe since pageset manipulation is done
1809 * with interrupts disabled.
1811 * Some NUMA counter updates may also be caught by the boot pagesets.
1813 * The boot_pagesets must be kept even after bootup is complete for
1814 * unused processors and/or zones. They do play a role for bootstrapping
1815 * hotplugged processors.
1817 * zoneinfo_show() and maybe other functions do
1818 * not check if the processor is online before following the pageset pointer.
1819 * Other parts of the kernel may not check if the zone is available.
1821 static struct per_cpu_pageset
1822 boot_pageset
[NR_CPUS
];
1825 * Dynamically allocate memory for the
1826 * per cpu pageset array in struct zone.
1828 static int __devinit
process_zones(int cpu
)
1830 struct zone
*zone
, *dzone
;
1832 for_each_zone(zone
) {
1834 zone
->pageset
[cpu
] = kmalloc_node(sizeof(struct per_cpu_pageset
),
1835 GFP_KERNEL
, cpu_to_node(cpu
));
1836 if (!zone
->pageset
[cpu
])
1839 setup_pageset(zone
->pageset
[cpu
], zone_batchsize(zone
));
1844 for_each_zone(dzone
) {
1847 kfree(dzone
->pageset
[cpu
]);
1848 dzone
->pageset
[cpu
] = NULL
;
1853 static inline void free_zone_pagesets(int cpu
)
1858 for_each_zone(zone
) {
1859 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1861 zone_pcp(zone
, cpu
) = NULL
;
1867 static int __devinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1868 unsigned long action
,
1871 int cpu
= (long)hcpu
;
1872 int ret
= NOTIFY_OK
;
1875 case CPU_UP_PREPARE
:
1876 if (process_zones(cpu
))
1879 #ifdef CONFIG_HOTPLUG_CPU
1881 free_zone_pagesets(cpu
);
1890 static struct notifier_block pageset_notifier
=
1891 { &pageset_cpuup_callback
, NULL
, 0 };
1893 void __init
setup_per_cpu_pageset()
1897 /* Initialize per_cpu_pageset for cpu 0.
1898 * A cpuup callback will do this for every cpu
1899 * as it comes online
1901 err
= process_zones(smp_processor_id());
1903 register_cpu_notifier(&pageset_notifier
);
1909 void zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1912 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1915 * The per-page waitqueue mechanism uses hashed waitqueues
1918 zone
->wait_table_size
= wait_table_size(zone_size_pages
);
1919 zone
->wait_table_bits
= wait_table_bits(zone
->wait_table_size
);
1920 zone
->wait_table
= (wait_queue_head_t
*)
1921 alloc_bootmem_node(pgdat
, zone
->wait_table_size
1922 * sizeof(wait_queue_head_t
));
1924 for(i
= 0; i
< zone
->wait_table_size
; ++i
)
1925 init_waitqueue_head(zone
->wait_table
+ i
);
1928 static __devinit
void zone_pcp_init(struct zone
*zone
)
1931 unsigned long batch
= zone_batchsize(zone
);
1933 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1935 /* Early boot. Slab allocator not functional yet */
1936 zone
->pageset
[cpu
] = &boot_pageset
[cpu
];
1937 setup_pageset(&boot_pageset
[cpu
],0);
1939 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1942 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1943 zone
->name
, zone
->present_pages
, batch
);
1946 static __devinit
void init_currently_empty_zone(struct zone
*zone
,
1947 unsigned long zone_start_pfn
, unsigned long size
)
1949 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1951 zone_wait_table_init(zone
, size
);
1952 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1954 zone
->zone_mem_map
= pfn_to_page(zone_start_pfn
);
1955 zone
->zone_start_pfn
= zone_start_pfn
;
1957 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1959 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
1963 * Set up the zone data structures:
1964 * - mark all pages reserved
1965 * - mark all memory queues empty
1966 * - clear the memory bitmaps
1968 static void __init
free_area_init_core(struct pglist_data
*pgdat
,
1969 unsigned long *zones_size
, unsigned long *zholes_size
)
1972 int nid
= pgdat
->node_id
;
1973 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
1975 pgdat_resize_init(pgdat
);
1976 pgdat
->nr_zones
= 0;
1977 init_waitqueue_head(&pgdat
->kswapd_wait
);
1978 pgdat
->kswapd_max_order
= 0;
1980 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
1981 struct zone
*zone
= pgdat
->node_zones
+ j
;
1982 unsigned long size
, realsize
;
1984 realsize
= size
= zones_size
[j
];
1986 realsize
-= zholes_size
[j
];
1988 if (j
< ZONE_HIGHMEM
)
1989 nr_kernel_pages
+= realsize
;
1990 nr_all_pages
+= realsize
;
1992 zone
->spanned_pages
= size
;
1993 zone
->present_pages
= realsize
;
1994 zone
->name
= zone_names
[j
];
1995 spin_lock_init(&zone
->lock
);
1996 spin_lock_init(&zone
->lru_lock
);
1997 zone_seqlock_init(zone
);
1998 zone
->zone_pgdat
= pgdat
;
1999 zone
->free_pages
= 0;
2001 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
2003 zone_pcp_init(zone
);
2004 INIT_LIST_HEAD(&zone
->active_list
);
2005 INIT_LIST_HEAD(&zone
->inactive_list
);
2006 zone
->nr_scan_active
= 0;
2007 zone
->nr_scan_inactive
= 0;
2008 zone
->nr_active
= 0;
2009 zone
->nr_inactive
= 0;
2010 atomic_set(&zone
->reclaim_in_progress
, 0);
2014 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2015 init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2016 zone_start_pfn
+= size
;
2020 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2022 /* Skip empty nodes */
2023 if (!pgdat
->node_spanned_pages
)
2026 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2027 /* ia64 gets its own node_mem_map, before this, without bootmem */
2028 if (!pgdat
->node_mem_map
) {
2032 size
= (pgdat
->node_spanned_pages
+ 1) * sizeof(struct page
);
2033 map
= alloc_remap(pgdat
->node_id
, size
);
2035 map
= alloc_bootmem_node(pgdat
, size
);
2036 pgdat
->node_mem_map
= map
;
2038 #ifdef CONFIG_FLATMEM
2040 * With no DISCONTIG, the global mem_map is just set as node 0's
2042 if (pgdat
== NODE_DATA(0))
2043 mem_map
= NODE_DATA(0)->node_mem_map
;
2045 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2048 void __init
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2049 unsigned long *zones_size
, unsigned long node_start_pfn
,
2050 unsigned long *zholes_size
)
2052 pgdat
->node_id
= nid
;
2053 pgdat
->node_start_pfn
= node_start_pfn
;
2054 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
2056 alloc_node_mem_map(pgdat
);
2058 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2061 #ifndef CONFIG_NEED_MULTIPLE_NODES
2062 static bootmem_data_t contig_bootmem_data
;
2063 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2065 EXPORT_SYMBOL(contig_page_data
);
2068 void __init
free_area_init(unsigned long *zones_size
)
2070 free_area_init_node(0, NODE_DATA(0), zones_size
,
2071 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2074 #ifdef CONFIG_PROC_FS
2076 #include <linux/seq_file.h>
2078 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
2083 for (pgdat
= pgdat_list
; pgdat
&& node
; pgdat
= pgdat
->pgdat_next
)
2089 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2091 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2094 return pgdat
->pgdat_next
;
2097 static void frag_stop(struct seq_file
*m
, void *arg
)
2102 * This walks the free areas for each zone.
2104 static int frag_show(struct seq_file
*m
, void *arg
)
2106 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
2108 struct zone
*node_zones
= pgdat
->node_zones
;
2109 unsigned long flags
;
2112 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
2113 if (!zone
->present_pages
)
2116 spin_lock_irqsave(&zone
->lock
, flags
);
2117 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
2118 for (order
= 0; order
< MAX_ORDER
; ++order
)
2119 seq_printf(m
, "%6lu ", zone
->free_area
[order
].nr_free
);
2120 spin_unlock_irqrestore(&zone
->lock
, flags
);
2126 struct seq_operations fragmentation_op
= {
2127 .start
= frag_start
,
2134 * Output information about zones in @pgdat.
2136 static int zoneinfo_show(struct seq_file
*m
, void *arg
)
2138 pg_data_t
*pgdat
= arg
;
2140 struct zone
*node_zones
= pgdat
->node_zones
;
2141 unsigned long flags
;
2143 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; zone
++) {
2146 if (!zone
->present_pages
)
2149 spin_lock_irqsave(&zone
->lock
, flags
);
2150 seq_printf(m
, "Node %d, zone %8s", pgdat
->node_id
, zone
->name
);
2158 "\n scanned %lu (a: %lu i: %lu)"
2167 zone
->pages_scanned
,
2168 zone
->nr_scan_active
, zone
->nr_scan_inactive
,
2169 zone
->spanned_pages
,
2170 zone
->present_pages
);
2172 "\n protection: (%lu",
2173 zone
->lowmem_reserve
[0]);
2174 for (i
= 1; i
< ARRAY_SIZE(zone
->lowmem_reserve
); i
++)
2175 seq_printf(m
, ", %lu", zone
->lowmem_reserve
[i
]);
2179 for (i
= 0; i
< ARRAY_SIZE(zone
->pageset
); i
++) {
2180 struct per_cpu_pageset
*pageset
;
2183 pageset
= zone_pcp(zone
, i
);
2184 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2185 if (pageset
->pcp
[j
].count
)
2188 if (j
== ARRAY_SIZE(pageset
->pcp
))
2190 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
2192 "\n cpu: %i pcp: %i"
2198 pageset
->pcp
[j
].count
,
2199 pageset
->pcp
[j
].low
,
2200 pageset
->pcp
[j
].high
,
2201 pageset
->pcp
[j
].batch
);
2207 "\n numa_foreign: %lu"
2208 "\n interleave_hit: %lu"
2209 "\n local_node: %lu"
2210 "\n other_node: %lu",
2213 pageset
->numa_foreign
,
2214 pageset
->interleave_hit
,
2215 pageset
->local_node
,
2216 pageset
->other_node
);
2220 "\n all_unreclaimable: %u"
2221 "\n prev_priority: %i"
2222 "\n temp_priority: %i"
2223 "\n start_pfn: %lu",
2224 zone
->all_unreclaimable
,
2225 zone
->prev_priority
,
2226 zone
->temp_priority
,
2227 zone
->zone_start_pfn
);
2228 spin_unlock_irqrestore(&zone
->lock
, flags
);
2234 struct seq_operations zoneinfo_op
= {
2235 .start
= frag_start
, /* iterate over all zones. The same as in
2239 .show
= zoneinfo_show
,
2242 static char *vmstat_text
[] = {
2246 "nr_page_table_pages",
2271 "pgscan_kswapd_high",
2272 "pgscan_kswapd_normal",
2274 "pgscan_kswapd_dma",
2275 "pgscan_direct_high",
2276 "pgscan_direct_normal",
2277 "pgscan_direct_dma",
2282 "kswapd_inodesteal",
2290 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
2292 struct page_state
*ps
;
2294 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2297 ps
= kmalloc(sizeof(*ps
), GFP_KERNEL
);
2300 return ERR_PTR(-ENOMEM
);
2301 get_full_page_state(ps
);
2302 ps
->pgpgin
/= 2; /* sectors -> kbytes */
2304 return (unsigned long *)ps
+ *pos
;
2307 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
2310 if (*pos
>= ARRAY_SIZE(vmstat_text
))
2312 return (unsigned long *)m
->private + *pos
;
2315 static int vmstat_show(struct seq_file
*m
, void *arg
)
2317 unsigned long *l
= arg
;
2318 unsigned long off
= l
- (unsigned long *)m
->private;
2320 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
2324 static void vmstat_stop(struct seq_file
*m
, void *arg
)
2330 struct seq_operations vmstat_op
= {
2331 .start
= vmstat_start
,
2332 .next
= vmstat_next
,
2333 .stop
= vmstat_stop
,
2334 .show
= vmstat_show
,
2337 #endif /* CONFIG_PROC_FS */
2339 #ifdef CONFIG_HOTPLUG_CPU
2340 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2341 unsigned long action
, void *hcpu
)
2343 int cpu
= (unsigned long)hcpu
;
2345 unsigned long *src
, *dest
;
2347 if (action
== CPU_DEAD
) {
2350 /* Drain local pagecache count. */
2351 count
= &per_cpu(nr_pagecache_local
, cpu
);
2352 atomic_add(*count
, &nr_pagecache
);
2354 local_irq_disable();
2357 /* Add dead cpu's page_states to our own. */
2358 dest
= (unsigned long *)&__get_cpu_var(page_states
);
2359 src
= (unsigned long *)&per_cpu(page_states
, cpu
);
2361 for (i
= 0; i
< sizeof(struct page_state
)/sizeof(unsigned long);
2371 #endif /* CONFIG_HOTPLUG_CPU */
2373 void __init
page_alloc_init(void)
2375 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2379 * setup_per_zone_lowmem_reserve - called whenever
2380 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2381 * has a correct pages reserved value, so an adequate number of
2382 * pages are left in the zone after a successful __alloc_pages().
2384 static void setup_per_zone_lowmem_reserve(void)
2386 struct pglist_data
*pgdat
;
2389 for_each_pgdat(pgdat
) {
2390 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2391 struct zone
*zone
= pgdat
->node_zones
+ j
;
2392 unsigned long present_pages
= zone
->present_pages
;
2394 zone
->lowmem_reserve
[j
] = 0;
2396 for (idx
= j
-1; idx
>= 0; idx
--) {
2397 struct zone
*lower_zone
;
2399 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2400 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2402 lower_zone
= pgdat
->node_zones
+ idx
;
2403 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2404 sysctl_lowmem_reserve_ratio
[idx
];
2405 present_pages
+= lower_zone
->present_pages
;
2412 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2413 * that the pages_{min,low,high} values for each zone are set correctly
2414 * with respect to min_free_kbytes.
2416 void setup_per_zone_pages_min(void)
2418 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2419 unsigned long lowmem_pages
= 0;
2421 unsigned long flags
;
2423 /* Calculate total number of !ZONE_HIGHMEM pages */
2424 for_each_zone(zone
) {
2425 if (!is_highmem(zone
))
2426 lowmem_pages
+= zone
->present_pages
;
2429 for_each_zone(zone
) {
2430 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2431 if (is_highmem(zone
)) {
2433 * Often, highmem doesn't need to reserve any pages.
2434 * But the pages_min/low/high values are also used for
2435 * batching up page reclaim activity so we need a
2436 * decent value here.
2440 min_pages
= zone
->present_pages
/ 1024;
2441 if (min_pages
< SWAP_CLUSTER_MAX
)
2442 min_pages
= SWAP_CLUSTER_MAX
;
2443 if (min_pages
> 128)
2445 zone
->pages_min
= min_pages
;
2447 /* if it's a lowmem zone, reserve a number of pages
2448 * proportionate to the zone's size.
2450 zone
->pages_min
= (pages_min
* zone
->present_pages
) /
2455 * When interpreting these watermarks, just keep in mind that:
2456 * zone->pages_min == (zone->pages_min * 4) / 4;
2458 zone
->pages_low
= (zone
->pages_min
* 5) / 4;
2459 zone
->pages_high
= (zone
->pages_min
* 6) / 4;
2460 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2465 * Initialise min_free_kbytes.
2467 * For small machines we want it small (128k min). For large machines
2468 * we want it large (64MB max). But it is not linear, because network
2469 * bandwidth does not increase linearly with machine size. We use
2471 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2472 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2488 static int __init
init_per_zone_pages_min(void)
2490 unsigned long lowmem_kbytes
;
2492 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2494 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2495 if (min_free_kbytes
< 128)
2496 min_free_kbytes
= 128;
2497 if (min_free_kbytes
> 65536)
2498 min_free_kbytes
= 65536;
2499 setup_per_zone_pages_min();
2500 setup_per_zone_lowmem_reserve();
2503 module_init(init_per_zone_pages_min
)
2506 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2507 * that we can call two helper functions whenever min_free_kbytes
2510 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2511 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2513 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2514 setup_per_zone_pages_min();
2519 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2520 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2521 * whenever sysctl_lowmem_reserve_ratio changes.
2523 * The reserve ratio obviously has absolutely no relation with the
2524 * pages_min watermarks. The lowmem reserve ratio can only make sense
2525 * if in function of the boot time zone sizes.
2527 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2528 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2530 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2531 setup_per_zone_lowmem_reserve();
2535 __initdata
int hashdist
= HASHDIST_DEFAULT
;
2538 static int __init
set_hashdist(char *str
)
2542 hashdist
= simple_strtoul(str
, &str
, 0);
2545 __setup("hashdist=", set_hashdist
);
2549 * allocate a large system hash table from bootmem
2550 * - it is assumed that the hash table must contain an exact power-of-2
2551 * quantity of entries
2552 * - limit is the number of hash buckets, not the total allocation size
2554 void *__init
alloc_large_system_hash(const char *tablename
,
2555 unsigned long bucketsize
,
2556 unsigned long numentries
,
2559 unsigned int *_hash_shift
,
2560 unsigned int *_hash_mask
,
2561 unsigned long limit
)
2563 unsigned long long max
= limit
;
2564 unsigned long log2qty
, size
;
2567 /* allow the kernel cmdline to have a say */
2569 /* round applicable memory size up to nearest megabyte */
2570 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2571 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2572 numentries
>>= 20 - PAGE_SHIFT
;
2573 numentries
<<= 20 - PAGE_SHIFT
;
2575 /* limit to 1 bucket per 2^scale bytes of low memory */
2576 if (scale
> PAGE_SHIFT
)
2577 numentries
>>= (scale
- PAGE_SHIFT
);
2579 numentries
<<= (PAGE_SHIFT
- scale
);
2581 /* rounded up to nearest power of 2 in size */
2582 numentries
= 1UL << (long_log2(numentries
) + 1);
2584 /* limit allocation size to 1/16 total memory by default */
2586 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2587 do_div(max
, bucketsize
);
2590 if (numentries
> max
)
2593 log2qty
= long_log2(numentries
);
2596 size
= bucketsize
<< log2qty
;
2597 if (flags
& HASH_EARLY
)
2598 table
= alloc_bootmem(size
);
2600 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
2602 unsigned long order
;
2603 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
2605 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
2607 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2610 panic("Failed to allocate %s hash table\n", tablename
);
2612 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2615 long_log2(size
) - PAGE_SHIFT
,
2619 *_hash_shift
= log2qty
;
2621 *_hash_mask
= (1 << log2qty
) - 1;