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/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
40 #include <linux/sort.h>
41 #include <linux/pfn.h>
42 #include <linux/backing-dev.h>
44 #include <asm/tlbflush.h>
45 #include <asm/div64.h>
49 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
52 nodemask_t node_online_map __read_mostly
= { { [0] = 1UL } };
53 EXPORT_SYMBOL(node_online_map
);
54 nodemask_t node_possible_map __read_mostly
= NODE_MASK_ALL
;
55 EXPORT_SYMBOL(node_possible_map
);
56 unsigned long totalram_pages __read_mostly
;
57 unsigned long totalreserve_pages __read_mostly
;
59 int percpu_pagelist_fraction
;
61 static void __free_pages_ok(struct page
*page
, unsigned int order
);
64 * results with 256, 32 in the lowmem_reserve sysctl:
65 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
66 * 1G machine -> (16M dma, 784M normal, 224M high)
67 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
68 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
69 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
71 * TBD: should special case ZONE_DMA32 machines here - in those we normally
72 * don't need any ZONE_NORMAL reservation
74 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
76 #ifdef CONFIG_ZONE_DMA32
84 EXPORT_SYMBOL(totalram_pages
);
87 * Used by page_zone() to look up the address of the struct zone whose
88 * id is encoded in the upper bits of page->flags
90 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
91 EXPORT_SYMBOL(zone_table
);
93 static char *zone_names
[MAX_NR_ZONES
] = {
95 #ifdef CONFIG_ZONE_DMA32
104 int min_free_kbytes
= 1024;
106 unsigned long __meminitdata nr_kernel_pages
;
107 unsigned long __meminitdata nr_all_pages
;
108 static unsigned long __initdata dma_reserve
;
110 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
112 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
113 * ranges of memory (RAM) that may be registered with add_active_range().
114 * Ranges passed to add_active_range() will be merged if possible
115 * so the number of times add_active_range() can be called is
116 * related to the number of nodes and the number of holes
118 #ifdef CONFIG_MAX_ACTIVE_REGIONS
119 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
120 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
122 #if MAX_NUMNODES >= 32
123 /* If there can be many nodes, allow up to 50 holes per node */
124 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
126 /* By default, allow up to 256 distinct regions */
127 #define MAX_ACTIVE_REGIONS 256
131 struct node_active_region __initdata early_node_map
[MAX_ACTIVE_REGIONS
];
132 int __initdata nr_nodemap_entries
;
133 unsigned long __initdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
134 unsigned long __initdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
135 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
136 unsigned long __initdata node_boundary_start_pfn
[MAX_NUMNODES
];
137 unsigned long __initdata node_boundary_end_pfn
[MAX_NUMNODES
];
138 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
139 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
141 #ifdef CONFIG_DEBUG_VM
142 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
146 unsigned long pfn
= page_to_pfn(page
);
149 seq
= zone_span_seqbegin(zone
);
150 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
152 else if (pfn
< zone
->zone_start_pfn
)
154 } while (zone_span_seqretry(zone
, seq
));
159 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
161 #ifdef CONFIG_HOLES_IN_ZONE
162 if (!pfn_valid(page_to_pfn(page
)))
165 if (zone
!= page_zone(page
))
171 * Temporary debugging check for pages not lying within a given zone.
173 static int bad_range(struct zone
*zone
, struct page
*page
)
175 if (page_outside_zone_boundaries(zone
, page
))
177 if (!page_is_consistent(zone
, page
))
183 static inline int bad_range(struct zone
*zone
, struct page
*page
)
189 static void bad_page(struct page
*page
)
191 #if 0 // mask by Victor Yu. 02-12-2007
192 printk(KERN_EMERG
"Bad page state in process '%s'\n"
193 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
194 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
195 KERN_EMERG
"Backtrace:\n",
196 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
197 (unsigned long)page
->flags
, page
->mapping
,
198 page_mapcount(page
), page_count(page
));
200 printk(KERN_EMERG
"Bad page state in process '%s'\n"
201 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
202 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
203 KERN_EMERG
"Backtrace:\n",
204 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
205 (unsigned long)page
->flags
, page
->u
.xx
.mapping
,
206 page_mapcount(page
), page_count(page
));
209 page
->flags
&= ~(1 << PG_lru
|
219 set_page_count(page
, 0);
220 reset_page_mapcount(page
);
221 #if 0 // mask by Victor Yu. 02-12-2007
222 page
->mapping
= NULL
;
224 page
->u
.xx
.mapping
= NULL
;
226 add_taint(TAINT_BAD_PAGE
);
230 * Higher-order pages are called "compound pages". They are structured thusly:
232 * The first PAGE_SIZE page is called the "head page".
234 * The remaining PAGE_SIZE pages are called "tail pages".
236 * All pages have PG_compound set. All pages have their ->private pointing at
237 * the head page (even the head page has this).
239 * The first tail page's ->lru.next holds the address of the compound page's
240 * put_page() function. Its ->lru.prev holds the order of allocation.
241 * This usage means that zero-order pages may not be compound.
244 static void free_compound_page(struct page
*page
)
246 __free_pages_ok(page
, (unsigned long)page
[1].lru
.prev
);
249 static void prep_compound_page(struct page
*page
, unsigned long order
)
252 int nr_pages
= 1 << order
;
254 page
[1].lru
.next
= (void *)free_compound_page
; /* set dtor */
255 page
[1].lru
.prev
= (void *)order
;
256 for (i
= 0; i
< nr_pages
; i
++) {
257 struct page
*p
= page
+ i
;
259 __SetPageCompound(p
);
260 set_page_private(p
, (unsigned long)page
);
264 static void destroy_compound_page(struct page
*page
, unsigned long order
)
267 int nr_pages
= 1 << order
;
269 if (unlikely((unsigned long)page
[1].lru
.prev
!= order
))
272 for (i
= 0; i
< nr_pages
; i
++) {
273 struct page
*p
= page
+ i
;
275 if (unlikely(!PageCompound(p
) |
276 (page_private(p
) != (unsigned long)page
)))
278 __ClearPageCompound(p
);
282 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
286 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
288 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
289 * and __GFP_HIGHMEM from hard or soft interrupt context.
291 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
292 for (i
= 0; i
< (1 << order
); i
++)
293 clear_highpage(page
+ i
);
297 * function for dealing with page's order in buddy system.
298 * zone->lock is already acquired when we use these.
299 * So, we don't need atomic page->flags operations here.
301 static inline unsigned long page_order(struct page
*page
)
303 return page_private(page
);
306 static inline void set_page_order(struct page
*page
, int order
)
308 set_page_private(page
, order
);
309 __SetPageBuddy(page
);
312 static inline void rmv_page_order(struct page
*page
)
314 __ClearPageBuddy(page
);
315 set_page_private(page
, 0);
319 * Locate the struct page for both the matching buddy in our
320 * pair (buddy1) and the combined O(n+1) page they form (page).
322 * 1) Any buddy B1 will have an order O twin B2 which satisfies
323 * the following equation:
325 * For example, if the starting buddy (buddy2) is #8 its order
327 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
329 * 2) Any buddy B will have an order O+1 parent P which
330 * satisfies the following equation:
333 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
335 static inline struct page
*
336 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
338 unsigned long buddy_idx
= page_idx
^ (1 << order
);
340 return page
+ (buddy_idx
- page_idx
);
343 static inline unsigned long
344 __find_combined_index(unsigned long page_idx
, unsigned int order
)
346 return (page_idx
& ~(1 << order
));
350 * This function checks whether a page is free && is the buddy
351 * we can do coalesce a page and its buddy if
352 * (a) the buddy is not in a hole &&
353 * (b) the buddy is in the buddy system &&
354 * (c) a page and its buddy have the same order &&
355 * (d) a page and its buddy are in the same zone.
357 * For recording whether a page is in the buddy system, we use PG_buddy.
358 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
360 * For recording page's order, we use page_private(page).
362 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
365 #ifdef CONFIG_HOLES_IN_ZONE
366 if (!pfn_valid(page_to_pfn(buddy
)))
370 if (page_zone_id(page
) != page_zone_id(buddy
))
373 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
374 BUG_ON(page_count(buddy
) != 0);
381 * Freeing function for a buddy system allocator.
383 * The concept of a buddy system is to maintain direct-mapped table
384 * (containing bit values) for memory blocks of various "orders".
385 * The bottom level table contains the map for the smallest allocatable
386 * units of memory (here, pages), and each level above it describes
387 * pairs of units from the levels below, hence, "buddies".
388 * At a high level, all that happens here is marking the table entry
389 * at the bottom level available, and propagating the changes upward
390 * as necessary, plus some accounting needed to play nicely with other
391 * parts of the VM system.
392 * At each level, we keep a list of pages, which are heads of continuous
393 * free pages of length of (1 << order) and marked with PG_buddy. Page's
394 * order is recorded in page_private(page) field.
395 * So when we are allocating or freeing one, we can derive the state of the
396 * other. That is, if we allocate a small block, and both were
397 * free, the remainder of the region must be split into blocks.
398 * If a block is freed, and its buddy is also free, then this
399 * triggers coalescing into a block of larger size.
404 static inline void __free_one_page(struct page
*page
,
405 struct zone
*zone
, unsigned int order
)
407 unsigned long page_idx
;
408 int order_size
= 1 << order
;
410 if (unlikely(PageCompound(page
)))
411 destroy_compound_page(page
, order
);
413 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
415 VM_BUG_ON(page_idx
& (order_size
- 1));
416 VM_BUG_ON(bad_range(zone
, page
));
418 zone
->free_pages
+= order_size
;
419 while (order
< MAX_ORDER
-1) {
420 unsigned long combined_idx
;
421 struct free_area
*area
;
424 buddy
= __page_find_buddy(page
, page_idx
, order
);
425 if (!page_is_buddy(page
, buddy
, order
))
426 break; /* Move the buddy up one level. */
428 list_del(&buddy
->lru
);
429 area
= zone
->free_area
+ order
;
431 rmv_page_order(buddy
);
432 combined_idx
= __find_combined_index(page_idx
, order
);
433 page
= page
+ (combined_idx
- page_idx
);
434 page_idx
= combined_idx
;
437 set_page_order(page
, order
);
438 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
439 zone
->free_area
[order
].nr_free
++;
442 static inline int free_pages_check(struct page
*page
)
444 #if 0 // mask by Victor Yu. 02-12-2007
445 if (unlikely(page_mapcount(page
) |
446 (page
->mapping
!= NULL
) |
447 (page_count(page
) != 0) |
460 if (unlikely(page_mapcount(page
) |
461 (page
->u
.xx
.mapping
!= NULL
) |
462 (page_count(page
) != 0) |
477 __ClearPageDirty(page
);
479 * For now, we report if PG_reserved was found set, but do not
480 * clear it, and do not free the page. But we shall soon need
481 * to do more, for when the ZERO_PAGE count wraps negative.
483 return PageReserved(page
);
487 * Frees a list of pages.
488 * Assumes all pages on list are in same zone, and of same order.
489 * count is the number of pages to free.
491 * If the zone was previously in an "all pages pinned" state then look to
492 * see if this freeing clears that state.
494 * And clear the zone's pages_scanned counter, to hold off the "all pages are
495 * pinned" detection logic.
497 static void free_pages_bulk(struct zone
*zone
, int count
,
498 struct list_head
*list
, int order
)
500 spin_lock(&zone
->lock
);
501 zone
->all_unreclaimable
= 0;
502 zone
->pages_scanned
= 0;
506 VM_BUG_ON(list_empty(list
));
507 page
= list_entry(list
->prev
, struct page
, lru
);
508 /* have to delete it as __free_one_page list manipulates */
509 list_del(&page
->lru
);
510 __free_one_page(page
, zone
, order
);
512 spin_unlock(&zone
->lock
);
515 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
517 spin_lock(&zone
->lock
);
518 zone
->all_unreclaimable
= 0;
519 zone
->pages_scanned
= 0;
520 __free_one_page(page
, zone
,order
);
521 spin_unlock(&zone
->lock
);
524 static void __free_pages_ok(struct page
*page
, unsigned int order
)
530 for (i
= 0 ; i
< (1 << order
) ; ++i
)
531 reserved
+= free_pages_check(page
+ i
);
535 if (!PageHighMem(page
))
536 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
537 arch_free_page(page
, order
);
538 kernel_map_pages(page
, 1 << order
, 0);
540 local_irq_save(flags
);
541 __count_vm_events(PGFREE
, 1 << order
);
542 free_one_page(page_zone(page
), page
, order
);
543 local_irq_restore(flags
);
547 * permit the bootmem allocator to evade page validation on high-order frees
549 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
552 __ClearPageReserved(page
);
553 set_page_count(page
, 0);
554 set_page_refcounted(page
);
560 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
561 struct page
*p
= &page
[loop
];
563 if (loop
+ 1 < BITS_PER_LONG
)
565 __ClearPageReserved(p
);
566 set_page_count(p
, 0);
569 set_page_refcounted(page
);
570 __free_pages(page
, order
);
576 * The order of subdivision here is critical for the IO subsystem.
577 * Please do not alter this order without good reasons and regression
578 * testing. Specifically, as large blocks of memory are subdivided,
579 * the order in which smaller blocks are delivered depends on the order
580 * they're subdivided in this function. This is the primary factor
581 * influencing the order in which pages are delivered to the IO
582 * subsystem according to empirical testing, and this is also justified
583 * by considering the behavior of a buddy system containing a single
584 * large block of memory acted on by a series of small allocations.
585 * This behavior is a critical factor in sglist merging's success.
589 static inline void expand(struct zone
*zone
, struct page
*page
,
590 int low
, int high
, struct free_area
*area
)
592 unsigned long size
= 1 << high
;
598 VM_BUG_ON(bad_range(zone
, &page
[size
]));
599 list_add(&page
[size
].lru
, &area
->free_list
);
601 set_page_order(&page
[size
], high
);
606 * This page is about to be returned from the page allocator
608 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
610 #if 0 // mask by Victor Yu. 02-12-2007
611 if (unlikely(page_mapcount(page
) |
612 (page
->mapping
!= NULL
) |
613 (page_count(page
) != 0) |
627 if (unlikely(page_mapcount(page
) |
628 (page
->u
.xx
.mapping
!= NULL
) |
629 (page_count(page
) != 0) |
646 * For now, we report if PG_reserved was found set, but do not
647 * clear it, and do not allocate the page: as a safety net.
649 if (PageReserved(page
))
652 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
653 1 << PG_referenced
| 1 << PG_arch_1
|
654 1 << PG_checked
| 1 << PG_mappedtodisk
);
655 set_page_private(page
, 0);
656 set_page_refcounted(page
);
657 kernel_map_pages(page
, 1 << order
, 1);
659 if (gfp_flags
& __GFP_ZERO
)
660 prep_zero_page(page
, order
, gfp_flags
);
662 if (order
&& (gfp_flags
& __GFP_COMP
))
663 prep_compound_page(page
, order
);
669 * Do the hard work of removing an element from the buddy allocator.
670 * Call me with the zone->lock already held.
672 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
674 struct free_area
* area
;
675 unsigned int current_order
;
678 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
679 area
= zone
->free_area
+ current_order
;
680 if (list_empty(&area
->free_list
))
683 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
684 list_del(&page
->lru
);
685 rmv_page_order(page
);
687 zone
->free_pages
-= 1UL << order
;
688 expand(zone
, page
, order
, current_order
, area
);
696 * Obtain a specified number of elements from the buddy allocator, all under
697 * a single hold of the lock, for efficiency. Add them to the supplied list.
698 * Returns the number of new pages which were placed at *list.
700 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
701 unsigned long count
, struct list_head
*list
)
705 spin_lock(&zone
->lock
);
706 for (i
= 0; i
< count
; ++i
) {
707 struct page
*page
= __rmqueue(zone
, order
);
708 if (unlikely(page
== NULL
))
710 list_add_tail(&page
->lru
, list
);
712 spin_unlock(&zone
->lock
);
718 * Called from the slab reaper to drain pagesets on a particular node that
719 * belongs to the currently executing processor.
720 * Note that this function must be called with the thread pinned to
721 * a single processor.
723 void drain_node_pages(int nodeid
)
729 for (z
= 0; z
< MAX_NR_ZONES
; z
++) {
730 struct zone
*zone
= NODE_DATA(nodeid
)->node_zones
+ z
;
731 struct per_cpu_pageset
*pset
;
733 if (!populated_zone(zone
))
736 pset
= zone_pcp(zone
, smp_processor_id());
737 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
738 struct per_cpu_pages
*pcp
;
742 local_irq_save(flags
);
743 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
745 local_irq_restore(flags
);
752 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
753 static void __drain_pages(unsigned int cpu
)
759 for_each_zone(zone
) {
760 struct per_cpu_pageset
*pset
;
762 pset
= zone_pcp(zone
, cpu
);
763 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
764 struct per_cpu_pages
*pcp
;
767 local_irq_save(flags
);
768 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
770 local_irq_restore(flags
);
774 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
778 void mark_free_pages(struct zone
*zone
)
780 unsigned long pfn
, max_zone_pfn
;
783 struct list_head
*curr
;
785 if (!zone
->spanned_pages
)
788 spin_lock_irqsave(&zone
->lock
, flags
);
790 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
791 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
792 if (pfn_valid(pfn
)) {
793 struct page
*page
= pfn_to_page(pfn
);
795 if (!PageNosave(page
))
796 ClearPageNosaveFree(page
);
799 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
800 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
803 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
804 for (i
= 0; i
< (1UL << order
); i
++)
805 SetPageNosaveFree(pfn_to_page(pfn
+ i
));
808 spin_unlock_irqrestore(&zone
->lock
, flags
);
812 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
814 void drain_local_pages(void)
818 local_irq_save(flags
);
819 __drain_pages(smp_processor_id());
820 local_irq_restore(flags
);
822 #endif /* CONFIG_PM */
825 * Free a 0-order page
827 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
829 struct zone
*zone
= page_zone(page
);
830 struct per_cpu_pages
*pcp
;
834 #if 0 // mask by Victor Yu. 02-12-2007
835 page
->mapping
= NULL
;
837 page
->u
.xx
.mapping
= NULL
;
839 if (free_pages_check(page
))
842 if (!PageHighMem(page
))
843 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
844 arch_free_page(page
, 0);
845 kernel_map_pages(page
, 1, 0);
847 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
848 local_irq_save(flags
);
849 __count_vm_event(PGFREE
);
850 list_add(&page
->lru
, &pcp
->list
);
852 if (pcp
->count
>= pcp
->high
) {
853 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
854 pcp
->count
-= pcp
->batch
;
856 local_irq_restore(flags
);
860 void fastcall
free_hot_page(struct page
*page
)
862 free_hot_cold_page(page
, 0);
865 void fastcall
free_cold_page(struct page
*page
)
867 free_hot_cold_page(page
, 1);
871 * split_page takes a non-compound higher-order page, and splits it into
872 * n (1<<order) sub-pages: page[0..n]
873 * Each sub-page must be freed individually.
875 * Note: this is probably too low level an operation for use in drivers.
876 * Please consult with lkml before using this in your driver.
878 void split_page(struct page
*page
, unsigned int order
)
882 VM_BUG_ON(PageCompound(page
));
883 VM_BUG_ON(!page_count(page
));
884 for (i
= 1; i
< (1 << order
); i
++)
885 set_page_refcounted(page
+ i
);
889 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
890 * we cheat by calling it from here, in the order > 0 path. Saves a branch
893 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
894 struct zone
*zone
, int order
, gfp_t gfp_flags
)
898 int cold
= !!(gfp_flags
& __GFP_COLD
);
903 if (likely(order
== 0)) {
904 struct per_cpu_pages
*pcp
;
906 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
907 local_irq_save(flags
);
909 pcp
->count
= rmqueue_bulk(zone
, 0,
910 pcp
->batch
, &pcp
->list
);
911 if (unlikely(!pcp
->count
))
914 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
915 list_del(&page
->lru
);
918 spin_lock_irqsave(&zone
->lock
, flags
);
919 page
= __rmqueue(zone
, order
);
920 spin_unlock(&zone
->lock
);
925 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
926 zone_statistics(zonelist
, zone
);
927 local_irq_restore(flags
);
930 VM_BUG_ON(bad_range(zone
, page
));
931 if (prep_new_page(page
, order
, gfp_flags
))
936 local_irq_restore(flags
);
941 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
942 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
943 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
944 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
945 #define ALLOC_HARDER 0x10 /* try to alloc harder */
946 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
947 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
950 * Return 1 if free pages are above 'mark'. This takes into account the order
953 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
954 int classzone_idx
, int alloc_flags
)
956 /* free_pages my go negative - that's OK */
957 unsigned long min
= mark
;
958 long free_pages
= z
->free_pages
- (1 << order
) + 1;
961 if (alloc_flags
& ALLOC_HIGH
)
963 if (alloc_flags
& ALLOC_HARDER
)
966 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
968 for (o
= 0; o
< order
; o
++) {
969 /* At the next order, this order's pages become unavailable */
970 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
972 /* Require fewer higher order pages to be free */
975 if (free_pages
<= min
)
982 * get_page_from_freeliest goes through the zonelist trying to allocate
986 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
987 struct zonelist
*zonelist
, int alloc_flags
)
989 struct zone
**z
= zonelist
->zones
;
990 struct page
*page
= NULL
;
991 int classzone_idx
= zone_idx(*z
);
995 * Go through the zonelist once, looking for a zone with enough free.
996 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1000 if (unlikely(NUMA_BUILD
&& (gfp_mask
& __GFP_THISNODE
) &&
1001 zone
->zone_pgdat
!= zonelist
->zones
[0]->zone_pgdat
))
1003 if ((alloc_flags
& ALLOC_CPUSET
) &&
1004 !cpuset_zone_allowed(zone
, gfp_mask
))
1007 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1009 if (alloc_flags
& ALLOC_WMARK_MIN
)
1010 mark
= zone
->pages_min
;
1011 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1012 mark
= zone
->pages_low
;
1014 mark
= zone
->pages_high
;
1015 if (!zone_watermark_ok(zone
, order
, mark
,
1016 classzone_idx
, alloc_flags
))
1017 if (!zone_reclaim_mode
||
1018 !zone_reclaim(zone
, gfp_mask
, order
))
1022 page
= buffered_rmqueue(zonelist
, zone
, order
, gfp_mask
);
1026 } while (*(++z
) != NULL
);
1031 * This is the 'heart' of the zoned buddy allocator.
1033 struct page
* fastcall
1034 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
1035 struct zonelist
*zonelist
)
1037 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1040 struct reclaim_state reclaim_state
;
1041 struct task_struct
*p
= current
;
1044 int did_some_progress
;
1046 might_sleep_if(wait
);
1049 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
1051 if (unlikely(*z
== NULL
)) {
1052 /* Should this ever happen?? */
1056 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1057 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1062 wakeup_kswapd(*z
, order
);
1066 * OK, we're below the kswapd watermark and have kicked background
1067 * reclaim. Now things get more complex, so set up alloc_flags according
1068 * to how we want to proceed.
1070 * The caller may dip into page reserves a bit more if the caller
1071 * cannot run direct reclaim, or if the caller has realtime scheduling
1072 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1073 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1075 alloc_flags
= ALLOC_WMARK_MIN
;
1076 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1077 alloc_flags
|= ALLOC_HARDER
;
1078 if (gfp_mask
& __GFP_HIGH
)
1079 alloc_flags
|= ALLOC_HIGH
;
1081 alloc_flags
|= ALLOC_CPUSET
;
1084 * Go through the zonelist again. Let __GFP_HIGH and allocations
1085 * coming from realtime tasks go deeper into reserves.
1087 * This is the last chance, in general, before the goto nopage.
1088 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1089 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1091 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1095 /* This allocation should allow future memory freeing. */
1097 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1098 && !in_interrupt()) {
1099 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1101 /* go through the zonelist yet again, ignoring mins */
1102 page
= get_page_from_freelist(gfp_mask
, order
,
1103 zonelist
, ALLOC_NO_WATERMARKS
);
1106 if (gfp_mask
& __GFP_NOFAIL
) {
1107 congestion_wait(WRITE
, HZ
/50);
1114 /* Atomic allocations - we can't balance anything */
1121 /* We now go into synchronous reclaim */
1122 cpuset_memory_pressure_bump();
1123 p
->flags
|= PF_MEMALLOC
;
1124 reclaim_state
.reclaimed_slab
= 0;
1125 p
->reclaim_state
= &reclaim_state
;
1127 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
1129 p
->reclaim_state
= NULL
;
1130 p
->flags
&= ~PF_MEMALLOC
;
1134 if (likely(did_some_progress
)) {
1135 page
= get_page_from_freelist(gfp_mask
, order
,
1136 zonelist
, alloc_flags
);
1139 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1141 * Go through the zonelist yet one more time, keep
1142 * very high watermark here, this is only to catch
1143 * a parallel oom killing, we must fail if we're still
1144 * under heavy pressure.
1146 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1147 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1151 #if 1 // mask by Victor Yu. 03-15-2007, I don't know why it will let system dead.
1152 out_of_memory(zonelist
, gfp_mask
, order
);
1158 * Don't let big-order allocations loop unless the caller explicitly
1159 * requests that. Wait for some write requests to complete then retry.
1161 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1162 * <= 3, but that may not be true in other implementations.
1165 if (!(gfp_mask
& __GFP_NORETRY
)) {
1166 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
1168 if (gfp_mask
& __GFP_NOFAIL
)
1172 congestion_wait(WRITE
, HZ
/50);
1177 #if 0 // mask by Victor Yu. 03-14-2007
1178 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1179 printk(KERN_WARNING
"%s: page allocation failure."
1180 " order:%d, mode:0x%x\n",
1181 p
->comm
, order
, gfp_mask
);
1191 EXPORT_SYMBOL(__alloc_pages
);
1194 * Common helper functions.
1196 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1199 page
= alloc_pages(gfp_mask
, order
);
1202 return (unsigned long) page_address(page
);
1205 EXPORT_SYMBOL(__get_free_pages
);
1207 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1212 * get_zeroed_page() returns a 32-bit address, which cannot represent
1215 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1217 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1219 return (unsigned long) page_address(page
);
1223 EXPORT_SYMBOL(get_zeroed_page
);
1225 void __pagevec_free(struct pagevec
*pvec
)
1227 int i
= pagevec_count(pvec
);
1230 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1233 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1235 if (put_page_testzero(page
)) {
1237 free_hot_page(page
);
1239 __free_pages_ok(page
, order
);
1243 EXPORT_SYMBOL(__free_pages
);
1245 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1248 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1249 __free_pages(virt_to_page((void *)addr
), order
);
1253 EXPORT_SYMBOL(free_pages
);
1256 * Total amount of free (allocatable) RAM:
1258 unsigned int nr_free_pages(void)
1260 unsigned int sum
= 0;
1264 sum
+= zone
->free_pages
;
1269 EXPORT_SYMBOL(nr_free_pages
);
1272 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1274 unsigned int sum
= 0;
1277 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1278 sum
+= pgdat
->node_zones
[i
].free_pages
;
1284 static unsigned int nr_free_zone_pages(int offset
)
1286 /* Just pick one node, since fallback list is circular */
1287 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1288 unsigned int sum
= 0;
1290 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1291 struct zone
**zonep
= zonelist
->zones
;
1294 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1295 unsigned long size
= zone
->present_pages
;
1296 unsigned long high
= zone
->pages_high
;
1305 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1307 unsigned int nr_free_buffer_pages(void)
1309 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1313 * Amount of free RAM allocatable within all zones
1315 unsigned int nr_free_pagecache_pages(void)
1317 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1320 static inline void show_node(struct zone
*zone
)
1323 printk("Node %ld ", zone_to_nid(zone
));
1326 void si_meminfo(struct sysinfo
*val
)
1328 val
->totalram
= totalram_pages
;
1330 val
->freeram
= nr_free_pages();
1331 val
->bufferram
= nr_blockdev_pages();
1332 val
->totalhigh
= totalhigh_pages
;
1333 val
->freehigh
= nr_free_highpages();
1334 val
->mem_unit
= PAGE_SIZE
;
1337 EXPORT_SYMBOL(si_meminfo
);
1340 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1342 pg_data_t
*pgdat
= NODE_DATA(nid
);
1344 val
->totalram
= pgdat
->node_present_pages
;
1345 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1346 #ifdef CONFIG_HIGHMEM
1347 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1348 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1353 val
->mem_unit
= PAGE_SIZE
;
1357 #define K(x) ((x) << (PAGE_SHIFT-10))
1360 * Show free area list (used inside shift_scroll-lock stuff)
1361 * We also calculate the percentage fragmentation. We do this by counting the
1362 * memory on each free list with the exception of the first item on the list.
1364 void show_free_areas(void)
1367 unsigned long active
;
1368 unsigned long inactive
;
1372 for_each_zone(zone
) {
1373 if (!populated_zone(zone
))
1377 printk("%s per-cpu:\n", zone
->name
);
1379 for_each_online_cpu(cpu
) {
1380 struct per_cpu_pageset
*pageset
;
1382 pageset
= zone_pcp(zone
, cpu
);
1384 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1385 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1386 cpu
, pageset
->pcp
[0].high
,
1387 pageset
->pcp
[0].batch
, pageset
->pcp
[0].count
,
1388 pageset
->pcp
[1].high
, pageset
->pcp
[1].batch
,
1389 pageset
->pcp
[1].count
);
1393 get_zone_counts(&active
, &inactive
, &free
);
1395 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1396 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1399 global_page_state(NR_FILE_DIRTY
),
1400 global_page_state(NR_WRITEBACK
),
1401 global_page_state(NR_UNSTABLE_NFS
),
1403 global_page_state(NR_SLAB_RECLAIMABLE
) +
1404 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1405 global_page_state(NR_FILE_MAPPED
),
1406 global_page_state(NR_PAGETABLE
));
1408 for_each_zone(zone
) {
1411 if (!populated_zone(zone
))
1423 " pages_scanned:%lu"
1424 " all_unreclaimable? %s"
1427 K(zone
->free_pages
),
1430 K(zone
->pages_high
),
1432 K(zone
->nr_inactive
),
1433 K(zone
->present_pages
),
1434 zone
->pages_scanned
,
1435 (zone
->all_unreclaimable
? "yes" : "no")
1437 printk("lowmem_reserve[]:");
1438 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1439 printk(" %lu", zone
->lowmem_reserve
[i
]);
1443 for_each_zone(zone
) {
1444 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1446 if (!populated_zone(zone
))
1450 printk("%s: ", zone
->name
);
1452 spin_lock_irqsave(&zone
->lock
, flags
);
1453 for (order
= 0; order
< MAX_ORDER
; order
++) {
1454 nr
[order
] = zone
->free_area
[order
].nr_free
;
1455 total
+= nr
[order
] << order
;
1457 spin_unlock_irqrestore(&zone
->lock
, flags
);
1458 for (order
= 0; order
< MAX_ORDER
; order
++)
1459 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1460 printk("= %lukB\n", K(total
));
1463 show_swap_cache_info();
1467 * Builds allocation fallback zone lists.
1469 * Add all populated zones of a node to the zonelist.
1471 static int __meminit
build_zonelists_node(pg_data_t
*pgdat
,
1472 struct zonelist
*zonelist
, int nr_zones
, enum zone_type zone_type
)
1476 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1481 zone
= pgdat
->node_zones
+ zone_type
;
1482 if (populated_zone(zone
)) {
1483 zonelist
->zones
[nr_zones
++] = zone
;
1484 check_highest_zone(zone_type
);
1487 } while (zone_type
);
1492 #define MAX_NODE_LOAD (num_online_nodes())
1493 static int __meminitdata node_load
[MAX_NUMNODES
];
1495 * find_next_best_node - find the next node that should appear in a given node's fallback list
1496 * @node: node whose fallback list we're appending
1497 * @used_node_mask: nodemask_t of already used nodes
1499 * We use a number of factors to determine which is the next node that should
1500 * appear on a given node's fallback list. The node should not have appeared
1501 * already in @node's fallback list, and it should be the next closest node
1502 * according to the distance array (which contains arbitrary distance values
1503 * from each node to each node in the system), and should also prefer nodes
1504 * with no CPUs, since presumably they'll have very little allocation pressure
1505 * on them otherwise.
1506 * It returns -1 if no node is found.
1508 static int __meminit
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1511 int min_val
= INT_MAX
;
1514 /* Use the local node if we haven't already */
1515 if (!node_isset(node
, *used_node_mask
)) {
1516 node_set(node
, *used_node_mask
);
1520 for_each_online_node(n
) {
1523 /* Don't want a node to appear more than once */
1524 if (node_isset(n
, *used_node_mask
))
1527 /* Use the distance array to find the distance */
1528 val
= node_distance(node
, n
);
1530 /* Penalize nodes under us ("prefer the next node") */
1533 /* Give preference to headless and unused nodes */
1534 tmp
= node_to_cpumask(n
);
1535 if (!cpus_empty(tmp
))
1536 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1538 /* Slight preference for less loaded node */
1539 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1540 val
+= node_load
[n
];
1542 if (val
< min_val
) {
1549 node_set(best_node
, *used_node_mask
);
1554 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1556 int j
, node
, local_node
;
1558 int prev_node
, load
;
1559 struct zonelist
*zonelist
;
1560 nodemask_t used_mask
;
1562 /* initialize zonelists */
1563 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1564 zonelist
= pgdat
->node_zonelists
+ i
;
1565 zonelist
->zones
[0] = NULL
;
1568 /* NUMA-aware ordering of nodes */
1569 local_node
= pgdat
->node_id
;
1570 load
= num_online_nodes();
1571 prev_node
= local_node
;
1572 nodes_clear(used_mask
);
1573 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1574 int distance
= node_distance(local_node
, node
);
1577 * If another node is sufficiently far away then it is better
1578 * to reclaim pages in a zone before going off node.
1580 if (distance
> RECLAIM_DISTANCE
)
1581 zone_reclaim_mode
= 1;
1584 * We don't want to pressure a particular node.
1585 * So adding penalty to the first node in same
1586 * distance group to make it round-robin.
1589 if (distance
!= node_distance(local_node
, prev_node
))
1590 node_load
[node
] += load
;
1593 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1594 zonelist
= pgdat
->node_zonelists
+ i
;
1595 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1597 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1598 zonelist
->zones
[j
] = NULL
;
1603 #else /* CONFIG_NUMA */
1605 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1607 int node
, local_node
;
1610 local_node
= pgdat
->node_id
;
1611 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1612 struct zonelist
*zonelist
;
1614 zonelist
= pgdat
->node_zonelists
+ i
;
1616 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
1618 * Now we build the zonelist so that it contains the zones
1619 * of all the other nodes.
1620 * We don't want to pressure a particular node, so when
1621 * building the zones for node N, we make sure that the
1622 * zones coming right after the local ones are those from
1623 * node N+1 (modulo N)
1625 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1626 if (!node_online(node
))
1628 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1630 for (node
= 0; node
< local_node
; node
++) {
1631 if (!node_online(node
))
1633 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1636 zonelist
->zones
[j
] = NULL
;
1640 #endif /* CONFIG_NUMA */
1642 /* return values int ....just for stop_machine_run() */
1643 static int __meminit
__build_all_zonelists(void *dummy
)
1646 for_each_online_node(nid
)
1647 build_zonelists(NODE_DATA(nid
));
1651 void __meminit
build_all_zonelists(void)
1653 if (system_state
== SYSTEM_BOOTING
) {
1654 __build_all_zonelists(NULL
);
1655 cpuset_init_current_mems_allowed();
1657 /* we have to stop all cpus to guaranntee there is no user
1659 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
1660 /* cpuset refresh routine should be here */
1662 vm_total_pages
= nr_free_pagecache_pages();
1663 printk("Built %i zonelists. Total pages: %ld\n",
1664 num_online_nodes(), vm_total_pages
);
1668 * Helper functions to size the waitqueue hash table.
1669 * Essentially these want to choose hash table sizes sufficiently
1670 * large so that collisions trying to wait on pages are rare.
1671 * But in fact, the number of active page waitqueues on typical
1672 * systems is ridiculously low, less than 200. So this is even
1673 * conservative, even though it seems large.
1675 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1676 * waitqueues, i.e. the size of the waitq table given the number of pages.
1678 #define PAGES_PER_WAITQUEUE 256
1680 #ifndef CONFIG_MEMORY_HOTPLUG
1681 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1683 unsigned long size
= 1;
1685 pages
/= PAGES_PER_WAITQUEUE
;
1687 while (size
< pages
)
1691 * Once we have dozens or even hundreds of threads sleeping
1692 * on IO we've got bigger problems than wait queue collision.
1693 * Limit the size of the wait table to a reasonable size.
1695 size
= min(size
, 4096UL);
1697 return max(size
, 4UL);
1701 * A zone's size might be changed by hot-add, so it is not possible to determine
1702 * a suitable size for its wait_table. So we use the maximum size now.
1704 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1706 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1707 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1708 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1710 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1711 * or more by the traditional way. (See above). It equals:
1713 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1714 * ia64(16K page size) : = ( 8G + 4M)byte.
1715 * powerpc (64K page size) : = (32G +16M)byte.
1717 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1724 * This is an integer logarithm so that shifts can be used later
1725 * to extract the more random high bits from the multiplicative
1726 * hash function before the remainder is taken.
1728 static inline unsigned long wait_table_bits(unsigned long size
)
1733 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1736 * Initially all pages are reserved - free ones are freed
1737 * up by free_all_bootmem() once the early boot process is
1738 * done. Non-atomic initialization, single-pass.
1740 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1741 unsigned long start_pfn
)
1744 unsigned long end_pfn
= start_pfn
+ size
;
1747 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1748 if (!early_pfn_valid(pfn
))
1750 if (!early_pfn_in_nid(pfn
, nid
))
1752 page
= pfn_to_page(pfn
);
1753 set_page_links(page
, zone
, nid
, pfn
);
1754 init_page_count(page
);
1755 reset_page_mapcount(page
);
1756 SetPageReserved(page
);
1757 INIT_LIST_HEAD(&page
->lru
);
1758 #ifdef WANT_PAGE_VIRTUAL
1759 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1760 if (!is_highmem_idx(zone
))
1761 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1766 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1770 for (order
= 0; order
< MAX_ORDER
; order
++) {
1771 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1772 zone
->free_area
[order
].nr_free
= 0;
1776 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1777 void zonetable_add(struct zone
*zone
, int nid
, enum zone_type zid
,
1778 unsigned long pfn
, unsigned long size
)
1780 unsigned long snum
= pfn_to_section_nr(pfn
);
1781 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1784 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1786 for (; snum
<= end
; snum
++)
1787 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1790 #ifndef __HAVE_ARCH_MEMMAP_INIT
1791 #define memmap_init(size, nid, zone, start_pfn) \
1792 memmap_init_zone((size), (nid), (zone), (start_pfn))
1795 static int __cpuinit
zone_batchsize(struct zone
*zone
)
1800 * The per-cpu-pages pools are set to around 1000th of the
1801 * size of the zone. But no more than 1/2 of a meg.
1803 * OK, so we don't know how big the cache is. So guess.
1805 batch
= zone
->present_pages
/ 1024;
1806 if (batch
* PAGE_SIZE
> 512 * 1024)
1807 batch
= (512 * 1024) / PAGE_SIZE
;
1808 batch
/= 4; /* We effectively *= 4 below */
1813 * Clamp the batch to a 2^n - 1 value. Having a power
1814 * of 2 value was found to be more likely to have
1815 * suboptimal cache aliasing properties in some cases.
1817 * For example if 2 tasks are alternately allocating
1818 * batches of pages, one task can end up with a lot
1819 * of pages of one half of the possible page colors
1820 * and the other with pages of the other colors.
1822 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1827 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1829 struct per_cpu_pages
*pcp
;
1831 memset(p
, 0, sizeof(*p
));
1833 pcp
= &p
->pcp
[0]; /* hot */
1835 pcp
->high
= 6 * batch
;
1836 pcp
->batch
= max(1UL, 1 * batch
);
1837 INIT_LIST_HEAD(&pcp
->list
);
1839 pcp
= &p
->pcp
[1]; /* cold*/
1841 pcp
->high
= 2 * batch
;
1842 pcp
->batch
= max(1UL, batch
/2);
1843 INIT_LIST_HEAD(&pcp
->list
);
1847 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1848 * to the value high for the pageset p.
1851 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
1854 struct per_cpu_pages
*pcp
;
1856 pcp
= &p
->pcp
[0]; /* hot list */
1858 pcp
->batch
= max(1UL, high
/4);
1859 if ((high
/4) > (PAGE_SHIFT
* 8))
1860 pcp
->batch
= PAGE_SHIFT
* 8;
1866 * Boot pageset table. One per cpu which is going to be used for all
1867 * zones and all nodes. The parameters will be set in such a way
1868 * that an item put on a list will immediately be handed over to
1869 * the buddy list. This is safe since pageset manipulation is done
1870 * with interrupts disabled.
1872 * Some NUMA counter updates may also be caught by the boot pagesets.
1874 * The boot_pagesets must be kept even after bootup is complete for
1875 * unused processors and/or zones. They do play a role for bootstrapping
1876 * hotplugged processors.
1878 * zoneinfo_show() and maybe other functions do
1879 * not check if the processor is online before following the pageset pointer.
1880 * Other parts of the kernel may not check if the zone is available.
1882 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
1885 * Dynamically allocate memory for the
1886 * per cpu pageset array in struct zone.
1888 static int __cpuinit
process_zones(int cpu
)
1890 struct zone
*zone
, *dzone
;
1892 for_each_zone(zone
) {
1894 if (!populated_zone(zone
))
1897 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
1898 GFP_KERNEL
, cpu_to_node(cpu
));
1899 if (!zone_pcp(zone
, cpu
))
1902 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
1904 if (percpu_pagelist_fraction
)
1905 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
1906 (zone
->present_pages
/ percpu_pagelist_fraction
));
1911 for_each_zone(dzone
) {
1914 kfree(zone_pcp(dzone
, cpu
));
1915 zone_pcp(dzone
, cpu
) = NULL
;
1920 static inline void free_zone_pagesets(int cpu
)
1924 for_each_zone(zone
) {
1925 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1927 /* Free per_cpu_pageset if it is slab allocated */
1928 if (pset
!= &boot_pageset
[cpu
])
1930 zone_pcp(zone
, cpu
) = NULL
;
1934 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1935 unsigned long action
,
1938 int cpu
= (long)hcpu
;
1939 int ret
= NOTIFY_OK
;
1942 case CPU_UP_PREPARE
:
1943 if (process_zones(cpu
))
1946 case CPU_UP_CANCELED
:
1948 free_zone_pagesets(cpu
);
1956 static struct notifier_block __cpuinitdata pageset_notifier
=
1957 { &pageset_cpuup_callback
, NULL
, 0 };
1959 void __init
setup_per_cpu_pageset(void)
1963 /* Initialize per_cpu_pageset for cpu 0.
1964 * A cpuup callback will do this for every cpu
1965 * as it comes online
1967 err
= process_zones(smp_processor_id());
1969 register_cpu_notifier(&pageset_notifier
);
1975 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1978 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1982 * The per-page waitqueue mechanism uses hashed waitqueues
1985 zone
->wait_table_hash_nr_entries
=
1986 wait_table_hash_nr_entries(zone_size_pages
);
1987 zone
->wait_table_bits
=
1988 wait_table_bits(zone
->wait_table_hash_nr_entries
);
1989 alloc_size
= zone
->wait_table_hash_nr_entries
1990 * sizeof(wait_queue_head_t
);
1992 if (system_state
== SYSTEM_BOOTING
) {
1993 zone
->wait_table
= (wait_queue_head_t
*)
1994 alloc_bootmem_node(pgdat
, alloc_size
);
1997 * This case means that a zone whose size was 0 gets new memory
1998 * via memory hot-add.
1999 * But it may be the case that a new node was hot-added. In
2000 * this case vmalloc() will not be able to use this new node's
2001 * memory - this wait_table must be initialized to use this new
2002 * node itself as well.
2003 * To use this new node's memory, further consideration will be
2006 zone
->wait_table
= (wait_queue_head_t
*)vmalloc(alloc_size
);
2008 if (!zone
->wait_table
)
2011 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2012 init_waitqueue_head(zone
->wait_table
+ i
);
2017 static __meminit
void zone_pcp_init(struct zone
*zone
)
2020 unsigned long batch
= zone_batchsize(zone
);
2022 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2024 /* Early boot. Slab allocator not functional yet */
2025 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2026 setup_pageset(&boot_pageset
[cpu
],0);
2028 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2031 if (zone
->present_pages
)
2032 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2033 zone
->name
, zone
->present_pages
, batch
);
2036 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2037 unsigned long zone_start_pfn
,
2040 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2042 ret
= zone_wait_table_init(zone
, size
);
2045 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2047 zone
->zone_start_pfn
= zone_start_pfn
;
2049 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
2051 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2056 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2058 * Basic iterator support. Return the first range of PFNs for a node
2059 * Note: nid == MAX_NUMNODES returns first region regardless of node
2061 static int __init
first_active_region_index_in_nid(int nid
)
2065 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2066 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2073 * Basic iterator support. Return the next active range of PFNs for a node
2074 * Note: nid == MAX_NUMNODES returns next region regardles of node
2076 static int __init
next_active_region_index_in_nid(int index
, int nid
)
2078 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2079 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2085 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2087 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2088 * Architectures may implement their own version but if add_active_range()
2089 * was used and there are no special requirements, this is a convenient
2092 int __init
early_pfn_to_nid(unsigned long pfn
)
2096 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2097 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2098 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2100 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2101 return early_node_map
[i
].nid
;
2106 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2108 /* Basic iterator support to walk early_node_map[] */
2109 #define for_each_active_range_index_in_nid(i, nid) \
2110 for (i = first_active_region_index_in_nid(nid); i != -1; \
2111 i = next_active_region_index_in_nid(i, nid))
2114 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2115 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2116 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2118 * If an architecture guarantees that all ranges registered with
2119 * add_active_ranges() contain no holes and may be freed, this
2120 * this function may be used instead of calling free_bootmem() manually.
2122 void __init
free_bootmem_with_active_regions(int nid
,
2123 unsigned long max_low_pfn
)
2127 for_each_active_range_index_in_nid(i
, nid
) {
2128 unsigned long size_pages
= 0;
2129 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2131 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2134 if (end_pfn
> max_low_pfn
)
2135 end_pfn
= max_low_pfn
;
2137 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2138 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2139 PFN_PHYS(early_node_map
[i
].start_pfn
),
2140 size_pages
<< PAGE_SHIFT
);
2145 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2146 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2148 * If an architecture guarantees that all ranges registered with
2149 * add_active_ranges() contain no holes and may be freed, this
2150 * function may be used instead of calling memory_present() manually.
2152 void __init
sparse_memory_present_with_active_regions(int nid
)
2156 for_each_active_range_index_in_nid(i
, nid
)
2157 memory_present(early_node_map
[i
].nid
,
2158 early_node_map
[i
].start_pfn
,
2159 early_node_map
[i
].end_pfn
);
2163 * push_node_boundaries - Push node boundaries to at least the requested boundary
2164 * @nid: The nid of the node to push the boundary for
2165 * @start_pfn: The start pfn of the node
2166 * @end_pfn: The end pfn of the node
2168 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2169 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2170 * be hotplugged even though no physical memory exists. This function allows
2171 * an arch to push out the node boundaries so mem_map is allocated that can
2174 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2175 void __init
push_node_boundaries(unsigned int nid
,
2176 unsigned long start_pfn
, unsigned long end_pfn
)
2178 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2179 nid
, start_pfn
, end_pfn
);
2181 /* Initialise the boundary for this node if necessary */
2182 if (node_boundary_end_pfn
[nid
] == 0)
2183 node_boundary_start_pfn
[nid
] = -1UL;
2185 /* Update the boundaries */
2186 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2187 node_boundary_start_pfn
[nid
] = start_pfn
;
2188 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2189 node_boundary_end_pfn
[nid
] = end_pfn
;
2192 /* If necessary, push the node boundary out for reserve hotadd */
2193 static void __init
account_node_boundary(unsigned int nid
,
2194 unsigned long *start_pfn
, unsigned long *end_pfn
)
2196 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2197 nid
, *start_pfn
, *end_pfn
);
2199 /* Return if boundary information has not been provided */
2200 if (node_boundary_end_pfn
[nid
] == 0)
2203 /* Check the boundaries and update if necessary */
2204 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
2205 *start_pfn
= node_boundary_start_pfn
[nid
];
2206 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
2207 *end_pfn
= node_boundary_end_pfn
[nid
];
2210 void __init
push_node_boundaries(unsigned int nid
,
2211 unsigned long start_pfn
, unsigned long end_pfn
) {}
2213 static void __init
account_node_boundary(unsigned int nid
,
2214 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
2219 * get_pfn_range_for_nid - Return the start and end page frames for a node
2220 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2221 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2222 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2224 * It returns the start and end page frame of a node based on information
2225 * provided by an arch calling add_active_range(). If called for a node
2226 * with no available memory, a warning is printed and the start and end
2229 void __init
get_pfn_range_for_nid(unsigned int nid
,
2230 unsigned long *start_pfn
, unsigned long *end_pfn
)
2236 for_each_active_range_index_in_nid(i
, nid
) {
2237 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
2238 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
2241 if (*start_pfn
== -1UL) {
2242 printk(KERN_WARNING
"Node %u active with no memory\n", nid
);
2246 /* Push the node boundaries out if requested */
2247 account_node_boundary(nid
, start_pfn
, end_pfn
);
2251 * Return the number of pages a zone spans in a node, including holes
2252 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2254 unsigned long __init
zone_spanned_pages_in_node(int nid
,
2255 unsigned long zone_type
,
2256 unsigned long *ignored
)
2258 unsigned long node_start_pfn
, node_end_pfn
;
2259 unsigned long zone_start_pfn
, zone_end_pfn
;
2261 /* Get the start and end of the node and zone */
2262 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
2263 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
2264 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
2266 /* Check that this node has pages within the zone's required range */
2267 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
2270 /* Move the zone boundaries inside the node if necessary */
2271 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
2272 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
2274 /* Return the spanned pages */
2275 return zone_end_pfn
- zone_start_pfn
;
2279 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
2280 * then all holes in the requested range will be accounted for.
2282 unsigned long __init
__absent_pages_in_range(int nid
,
2283 unsigned long range_start_pfn
,
2284 unsigned long range_end_pfn
)
2287 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
2288 unsigned long start_pfn
;
2290 /* Find the end_pfn of the first active range of pfns in the node */
2291 i
= first_active_region_index_in_nid(nid
);
2295 /* Account for ranges before physical memory on this node */
2296 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
2297 hole_pages
= early_node_map
[i
].start_pfn
- range_start_pfn
;
2299 prev_end_pfn
= early_node_map
[i
].start_pfn
;
2301 /* Find all holes for the zone within the node */
2302 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
2304 /* No need to continue if prev_end_pfn is outside the zone */
2305 if (prev_end_pfn
>= range_end_pfn
)
2308 /* Make sure the end of the zone is not within the hole */
2309 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
2310 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
2312 /* Update the hole size cound and move on */
2313 if (start_pfn
> range_start_pfn
) {
2314 BUG_ON(prev_end_pfn
> start_pfn
);
2315 hole_pages
+= start_pfn
- prev_end_pfn
;
2317 prev_end_pfn
= early_node_map
[i
].end_pfn
;
2320 /* Account for ranges past physical memory on this node */
2321 if (range_end_pfn
> prev_end_pfn
)
2322 hole_pages
+= range_end_pfn
-
2323 max(range_start_pfn
, prev_end_pfn
);
2329 * absent_pages_in_range - Return number of page frames in holes within a range
2330 * @start_pfn: The start PFN to start searching for holes
2331 * @end_pfn: The end PFN to stop searching for holes
2333 * It returns the number of pages frames in memory holes within a range.
2335 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
2336 unsigned long end_pfn
)
2338 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
2341 /* Return the number of page frames in holes in a zone on a node */
2342 unsigned long __init
zone_absent_pages_in_node(int nid
,
2343 unsigned long zone_type
,
2344 unsigned long *ignored
)
2346 unsigned long node_start_pfn
, node_end_pfn
;
2347 unsigned long zone_start_pfn
, zone_end_pfn
;
2349 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
2350 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
2352 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
2355 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
2359 static inline unsigned long zone_spanned_pages_in_node(int nid
,
2360 unsigned long zone_type
,
2361 unsigned long *zones_size
)
2363 return zones_size
[zone_type
];
2366 static inline unsigned long zone_absent_pages_in_node(int nid
,
2367 unsigned long zone_type
,
2368 unsigned long *zholes_size
)
2373 return zholes_size
[zone_type
];
2378 static void __init
calculate_node_totalpages(struct pglist_data
*pgdat
,
2379 unsigned long *zones_size
, unsigned long *zholes_size
)
2381 unsigned long realtotalpages
, totalpages
= 0;
2384 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2385 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
2387 pgdat
->node_spanned_pages
= totalpages
;
2389 realtotalpages
= totalpages
;
2390 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2392 zone_absent_pages_in_node(pgdat
->node_id
, i
,
2394 pgdat
->node_present_pages
= realtotalpages
;
2395 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
2400 * Set up the zone data structures:
2401 * - mark all pages reserved
2402 * - mark all memory queues empty
2403 * - clear the memory bitmaps
2405 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
2406 unsigned long *zones_size
, unsigned long *zholes_size
)
2409 int nid
= pgdat
->node_id
;
2410 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
2413 pgdat_resize_init(pgdat
);
2414 pgdat
->nr_zones
= 0;
2415 init_waitqueue_head(&pgdat
->kswapd_wait
);
2416 pgdat
->kswapd_max_order
= 0;
2418 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2419 struct zone
*zone
= pgdat
->node_zones
+ j
;
2420 unsigned long size
, realsize
, memmap_pages
;
2422 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
2423 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
2427 * Adjust realsize so that it accounts for how much memory
2428 * is used by this zone for memmap. This affects the watermark
2429 * and per-cpu initialisations
2431 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
2432 if (realsize
>= memmap_pages
) {
2433 realsize
-= memmap_pages
;
2435 " %s zone: %lu pages used for memmap\n",
2436 zone_names
[j
], memmap_pages
);
2439 " %s zone: %lu pages exceeds realsize %lu\n",
2440 zone_names
[j
], memmap_pages
, realsize
);
2442 /* Account for reserved DMA pages */
2443 if (j
== ZONE_DMA
&& realsize
> dma_reserve
) {
2444 realsize
-= dma_reserve
;
2445 printk(KERN_DEBUG
" DMA zone: %lu pages reserved\n",
2449 if (!is_highmem_idx(j
))
2450 nr_kernel_pages
+= realsize
;
2451 nr_all_pages
+= realsize
;
2453 zone
->spanned_pages
= size
;
2454 zone
->present_pages
= realsize
;
2457 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
2459 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
2461 zone
->name
= zone_names
[j
];
2462 spin_lock_init(&zone
->lock
);
2463 spin_lock_init(&zone
->lru_lock
);
2464 zone_seqlock_init(zone
);
2465 zone
->zone_pgdat
= pgdat
;
2466 zone
->free_pages
= 0;
2468 zone
->prev_priority
= DEF_PRIORITY
;
2470 zone_pcp_init(zone
);
2471 INIT_LIST_HEAD(&zone
->active_list
);
2472 INIT_LIST_HEAD(&zone
->inactive_list
);
2473 zone
->nr_scan_active
= 0;
2474 zone
->nr_scan_inactive
= 0;
2475 zone
->nr_active
= 0;
2476 zone
->nr_inactive
= 0;
2477 zap_zone_vm_stats(zone
);
2478 atomic_set(&zone
->reclaim_in_progress
, 0);
2482 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2483 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2485 zone_start_pfn
+= size
;
2489 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2491 /* Skip empty nodes */
2492 if (!pgdat
->node_spanned_pages
)
2495 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2496 /* ia64 gets its own node_mem_map, before this, without bootmem */
2497 if (!pgdat
->node_mem_map
) {
2498 unsigned long size
, start
, end
;
2502 * The zone's endpoints aren't required to be MAX_ORDER
2503 * aligned but the node_mem_map endpoints must be in order
2504 * for the buddy allocator to function correctly.
2506 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
2507 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
2508 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
2509 size
= (end
- start
) * sizeof(struct page
);
2510 map
= alloc_remap(pgdat
->node_id
, size
);
2512 map
= alloc_bootmem_node(pgdat
, size
);
2513 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
2515 #ifdef CONFIG_FLATMEM
2517 * With no DISCONTIG, the global mem_map is just set as node 0's
2519 if (pgdat
== NODE_DATA(0)) {
2520 mem_map
= NODE_DATA(0)->node_mem_map
;
2521 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2522 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
2523 mem_map
-= pgdat
->node_start_pfn
;
2524 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2527 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2530 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2531 unsigned long *zones_size
, unsigned long node_start_pfn
,
2532 unsigned long *zholes_size
)
2534 pgdat
->node_id
= nid
;
2535 pgdat
->node_start_pfn
= node_start_pfn
;
2536 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
2538 alloc_node_mem_map(pgdat
);
2540 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2543 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2545 * add_active_range - Register a range of PFNs backed by physical memory
2546 * @nid: The node ID the range resides on
2547 * @start_pfn: The start PFN of the available physical memory
2548 * @end_pfn: The end PFN of the available physical memory
2550 * These ranges are stored in an early_node_map[] and later used by
2551 * free_area_init_nodes() to calculate zone sizes and holes. If the
2552 * range spans a memory hole, it is up to the architecture to ensure
2553 * the memory is not freed by the bootmem allocator. If possible
2554 * the range being registered will be merged with existing ranges.
2556 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
2557 unsigned long end_pfn
)
2561 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
2562 "%d entries of %d used\n",
2563 nid
, start_pfn
, end_pfn
,
2564 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
2566 /* Merge with existing active regions if possible */
2567 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2568 if (early_node_map
[i
].nid
!= nid
)
2571 /* Skip if an existing region covers this new one */
2572 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
2573 end_pfn
<= early_node_map
[i
].end_pfn
)
2576 /* Merge forward if suitable */
2577 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
2578 end_pfn
> early_node_map
[i
].end_pfn
) {
2579 early_node_map
[i
].end_pfn
= end_pfn
;
2583 /* Merge backward if suitable */
2584 if (start_pfn
< early_node_map
[i
].end_pfn
&&
2585 end_pfn
>= early_node_map
[i
].start_pfn
) {
2586 early_node_map
[i
].start_pfn
= start_pfn
;
2591 /* Check that early_node_map is large enough */
2592 if (i
>= MAX_ACTIVE_REGIONS
) {
2593 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
2594 MAX_ACTIVE_REGIONS
);
2598 early_node_map
[i
].nid
= nid
;
2599 early_node_map
[i
].start_pfn
= start_pfn
;
2600 early_node_map
[i
].end_pfn
= end_pfn
;
2601 nr_nodemap_entries
= i
+ 1;
2605 * shrink_active_range - Shrink an existing registered range of PFNs
2606 * @nid: The node id the range is on that should be shrunk
2607 * @old_end_pfn: The old end PFN of the range
2608 * @new_end_pfn: The new PFN of the range
2610 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2611 * The map is kept at the end physical page range that has already been
2612 * registered with add_active_range(). This function allows an arch to shrink
2613 * an existing registered range.
2615 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
2616 unsigned long new_end_pfn
)
2620 /* Find the old active region end and shrink */
2621 for_each_active_range_index_in_nid(i
, nid
)
2622 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
2623 early_node_map
[i
].end_pfn
= new_end_pfn
;
2629 * remove_all_active_ranges - Remove all currently registered regions
2631 * During discovery, it may be found that a table like SRAT is invalid
2632 * and an alternative discovery method must be used. This function removes
2633 * all currently registered regions.
2635 void __init
remove_all_active_ranges(void)
2637 memset(early_node_map
, 0, sizeof(early_node_map
));
2638 nr_nodemap_entries
= 0;
2639 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2640 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
2641 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
2642 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
2645 /* Compare two active node_active_regions */
2646 static int __init
cmp_node_active_region(const void *a
, const void *b
)
2648 struct node_active_region
*arange
= (struct node_active_region
*)a
;
2649 struct node_active_region
*brange
= (struct node_active_region
*)b
;
2651 /* Done this way to avoid overflows */
2652 if (arange
->start_pfn
> brange
->start_pfn
)
2654 if (arange
->start_pfn
< brange
->start_pfn
)
2660 /* sort the node_map by start_pfn */
2661 static void __init
sort_node_map(void)
2663 sort(early_node_map
, (size_t)nr_nodemap_entries
,
2664 sizeof(struct node_active_region
),
2665 cmp_node_active_region
, NULL
);
2668 /* Find the lowest pfn for a node. This depends on a sorted early_node_map */
2669 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
2673 /* Regions in the early_node_map can be in any order */
2676 /* Assuming a sorted map, the first range found has the starting pfn */
2677 for_each_active_range_index_in_nid(i
, nid
)
2678 return early_node_map
[i
].start_pfn
;
2680 printk(KERN_WARNING
"Could not find start_pfn for node %lu\n", nid
);
2685 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2687 * It returns the minimum PFN based on information provided via
2688 * add_active_range().
2690 unsigned long __init
find_min_pfn_with_active_regions(void)
2692 return find_min_pfn_for_node(MAX_NUMNODES
);
2696 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2698 * It returns the maximum PFN based on information provided via
2699 * add_active_range().
2701 unsigned long __init
find_max_pfn_with_active_regions(void)
2704 unsigned long max_pfn
= 0;
2706 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2707 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
2713 * free_area_init_nodes - Initialise all pg_data_t and zone data
2714 * @max_zone_pfn: an array of max PFNs for each zone
2716 * This will call free_area_init_node() for each active node in the system.
2717 * Using the page ranges provided by add_active_range(), the size of each
2718 * zone in each node and their holes is calculated. If the maximum PFN
2719 * between two adjacent zones match, it is assumed that the zone is empty.
2720 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2721 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2722 * starts where the previous one ended. For example, ZONE_DMA32 starts
2723 * at arch_max_dma_pfn.
2725 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
2730 /* Record where the zone boundaries are */
2731 memset(arch_zone_lowest_possible_pfn
, 0,
2732 sizeof(arch_zone_lowest_possible_pfn
));
2733 memset(arch_zone_highest_possible_pfn
, 0,
2734 sizeof(arch_zone_highest_possible_pfn
));
2735 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
2736 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
2737 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
2738 arch_zone_lowest_possible_pfn
[i
] =
2739 arch_zone_highest_possible_pfn
[i
-1];
2740 arch_zone_highest_possible_pfn
[i
] =
2741 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
2744 /* Print out the zone ranges */
2745 printk("Zone PFN ranges:\n");
2746 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2747 printk(" %-8s %8lu -> %8lu\n",
2749 arch_zone_lowest_possible_pfn
[i
],
2750 arch_zone_highest_possible_pfn
[i
]);
2752 /* Print out the early_node_map[] */
2753 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
2754 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2755 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
2756 early_node_map
[i
].start_pfn
,
2757 early_node_map
[i
].end_pfn
);
2759 /* Initialise every node */
2760 for_each_online_node(nid
) {
2761 pg_data_t
*pgdat
= NODE_DATA(nid
);
2762 free_area_init_node(nid
, pgdat
, NULL
,
2763 find_min_pfn_for_node(nid
), NULL
);
2766 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2769 * set_dma_reserve - set the specified number of pages reserved in the first zone
2770 * @new_dma_reserve: The number of pages to mark reserved
2772 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2773 * In the DMA zone, a significant percentage may be consumed by kernel image
2774 * and other unfreeable allocations which can skew the watermarks badly. This
2775 * function may optionally be used to account for unfreeable pages in the
2776 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2777 * smaller per-cpu batchsize.
2779 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
2781 dma_reserve
= new_dma_reserve
;
2784 #ifndef CONFIG_NEED_MULTIPLE_NODES
2785 static bootmem_data_t contig_bootmem_data
;
2786 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2788 EXPORT_SYMBOL(contig_page_data
);
2791 void __init
free_area_init(unsigned long *zones_size
)
2793 free_area_init_node(0, NODE_DATA(0), zones_size
,
2794 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2797 #ifdef CONFIG_HOTPLUG_CPU
2798 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2799 unsigned long action
, void *hcpu
)
2801 int cpu
= (unsigned long)hcpu
;
2803 if (action
== CPU_DEAD
) {
2804 local_irq_disable();
2806 vm_events_fold_cpu(cpu
);
2808 refresh_cpu_vm_stats(cpu
);
2812 #endif /* CONFIG_HOTPLUG_CPU */
2814 void __init
page_alloc_init(void)
2816 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2820 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2821 * or min_free_kbytes changes.
2823 static void calculate_totalreserve_pages(void)
2825 struct pglist_data
*pgdat
;
2826 unsigned long reserve_pages
= 0;
2827 enum zone_type i
, j
;
2829 for_each_online_pgdat(pgdat
) {
2830 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2831 struct zone
*zone
= pgdat
->node_zones
+ i
;
2832 unsigned long max
= 0;
2834 /* Find valid and maximum lowmem_reserve in the zone */
2835 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
2836 if (zone
->lowmem_reserve
[j
] > max
)
2837 max
= zone
->lowmem_reserve
[j
];
2840 /* we treat pages_high as reserved pages. */
2841 max
+= zone
->pages_high
;
2843 if (max
> zone
->present_pages
)
2844 max
= zone
->present_pages
;
2845 reserve_pages
+= max
;
2848 totalreserve_pages
= reserve_pages
;
2852 * setup_per_zone_lowmem_reserve - called whenever
2853 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2854 * has a correct pages reserved value, so an adequate number of
2855 * pages are left in the zone after a successful __alloc_pages().
2857 static void setup_per_zone_lowmem_reserve(void)
2859 struct pglist_data
*pgdat
;
2860 enum zone_type j
, idx
;
2862 for_each_online_pgdat(pgdat
) {
2863 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2864 struct zone
*zone
= pgdat
->node_zones
+ j
;
2865 unsigned long present_pages
= zone
->present_pages
;
2867 zone
->lowmem_reserve
[j
] = 0;
2871 struct zone
*lower_zone
;
2875 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2876 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2878 lower_zone
= pgdat
->node_zones
+ idx
;
2879 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2880 sysctl_lowmem_reserve_ratio
[idx
];
2881 present_pages
+= lower_zone
->present_pages
;
2886 /* update totalreserve_pages */
2887 calculate_totalreserve_pages();
2890 #if 1 // add by Victor Yu. 02-08-2007
2891 static u64
victor_do_div(u64 n
, u32 base
)
2898 * setup_per_zone_pages_min - called when min_free_kbytes changes.
2900 * Ensures that the pages_{min,low,high} values for each zone are set correctly
2901 * with respect to min_free_kbytes.
2903 void setup_per_zone_pages_min(void)
2905 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2906 unsigned long lowmem_pages
= 0;
2908 unsigned long flags
;
2910 /* Calculate total number of !ZONE_HIGHMEM pages */
2911 for_each_zone(zone
) {
2912 if (!is_highmem(zone
))
2913 lowmem_pages
+= zone
->present_pages
;
2917 for_each_zone(zone
) {
2920 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2921 tmp
= (u64
)pages_min
* zone
->present_pages
;
2922 #if 0 // mask by Victor Yu. 02-08-2007
2923 do_div(tmp
, lowmem_pages
);
2925 tmp
= victor_do_div(tmp
, lowmem_pages
);
2927 if (is_highmem(zone
)) {
2929 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2930 * need highmem pages, so cap pages_min to a small
2933 * The (pages_high-pages_low) and (pages_low-pages_min)
2934 * deltas controls asynch page reclaim, and so should
2935 * not be capped for highmem.
2939 min_pages
= zone
->present_pages
/ 1024;
2940 if (min_pages
< SWAP_CLUSTER_MAX
)
2941 min_pages
= SWAP_CLUSTER_MAX
;
2942 if (min_pages
> 128)
2944 zone
->pages_min
= min_pages
;
2947 * If it's a lowmem zone, reserve a number of pages
2948 * proportionate to the zone's size.
2950 zone
->pages_min
= tmp
;
2953 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
2954 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
2955 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2958 /* update totalreserve_pages */
2959 calculate_totalreserve_pages();
2963 * Initialise min_free_kbytes.
2965 * For small machines we want it small (128k min). For large machines
2966 * we want it large (64MB max). But it is not linear, because network
2967 * bandwidth does not increase linearly with machine size. We use
2969 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2970 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2986 static int __init
init_per_zone_pages_min(void)
2988 unsigned long lowmem_kbytes
;
2990 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2992 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2993 if (min_free_kbytes
< 128)
2994 min_free_kbytes
= 128;
2995 if (min_free_kbytes
> 65536)
2996 min_free_kbytes
= 65536;
2997 setup_per_zone_pages_min();
2998 setup_per_zone_lowmem_reserve();
3001 module_init(init_per_zone_pages_min
)
3004 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
3005 * that we can call two helper functions whenever min_free_kbytes
3008 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
3009 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3011 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
3012 setup_per_zone_pages_min();
3017 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
3018 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3023 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3028 zone
->min_unmapped_pages
= (zone
->present_pages
*
3029 sysctl_min_unmapped_ratio
) / 100;
3033 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
3034 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3039 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3044 zone
->min_slab_pages
= (zone
->present_pages
*
3045 sysctl_min_slab_ratio
) / 100;
3051 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
3052 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
3053 * whenever sysctl_lowmem_reserve_ratio changes.
3055 * The reserve ratio obviously has absolutely no relation with the
3056 * pages_min watermarks. The lowmem reserve ratio can only make sense
3057 * if in function of the boot time zone sizes.
3059 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
3060 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3062 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3063 setup_per_zone_lowmem_reserve();
3068 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3069 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3070 * can have before it gets flushed back to buddy allocator.
3073 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
3074 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3080 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3081 if (!write
|| (ret
== -EINVAL
))
3083 for_each_zone(zone
) {
3084 for_each_online_cpu(cpu
) {
3086 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
3087 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
3093 int hashdist
= HASHDIST_DEFAULT
;
3096 static int __init
set_hashdist(char *str
)
3100 hashdist
= simple_strtoul(str
, &str
, 0);
3103 __setup("hashdist=", set_hashdist
);
3107 * allocate a large system hash table from bootmem
3108 * - it is assumed that the hash table must contain an exact power-of-2
3109 * quantity of entries
3110 * - limit is the number of hash buckets, not the total allocation size
3112 void *__init
alloc_large_system_hash(const char *tablename
,
3113 unsigned long bucketsize
,
3114 unsigned long numentries
,
3117 unsigned int *_hash_shift
,
3118 unsigned int *_hash_mask
,
3119 unsigned long limit
)
3121 unsigned long long max
= limit
;
3122 unsigned long log2qty
, size
;
3125 /* allow the kernel cmdline to have a say */
3127 /* round applicable memory size up to nearest megabyte */
3128 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
3129 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
3130 numentries
>>= 20 - PAGE_SHIFT
;
3131 numentries
<<= 20 - PAGE_SHIFT
;
3133 /* limit to 1 bucket per 2^scale bytes of low memory */
3134 if (scale
> PAGE_SHIFT
)
3135 numentries
>>= (scale
- PAGE_SHIFT
);
3137 numentries
<<= (PAGE_SHIFT
- scale
);
3139 numentries
= roundup_pow_of_two(numentries
);
3141 /* limit allocation size to 1/16 total memory by default */
3143 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
3144 #if 0 // mask by Victor Yu. 03-16-2007
3145 do_div(max
, bucketsize
);
3147 max
= victor_do_div(max
, bucketsize
);
3151 if (numentries
> max
)
3155 * we will allocate at least a page (even on low memory systems)
3156 * so do a fixup here to ensure we utilise the space that will be
3157 * allocated, this also prevents us reporting -ve orders
3159 if (bucketsize
* numentries
< PAGE_SIZE
)
3160 numentries
= (PAGE_SIZE
+ bucketsize
- 1) / bucketsize
;
3162 log2qty
= long_log2(numentries
);
3165 size
= bucketsize
<< log2qty
;
3166 if (flags
& HASH_EARLY
)
3167 table
= alloc_bootmem(size
);
3169 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
3171 unsigned long order
;
3172 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
3174 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
3176 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
3179 panic("Failed to allocate %s hash table\n", tablename
);
3181 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3184 long_log2(size
) - PAGE_SHIFT
,
3188 *_hash_shift
= log2qty
;
3190 *_hash_mask
= (1 << log2qty
) - 1;
3195 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
3196 struct page
*pfn_to_page(unsigned long pfn
)
3198 return __pfn_to_page(pfn
);
3200 unsigned long page_to_pfn(struct page
*page
)
3202 return __page_to_pfn(page
);
3204 EXPORT_SYMBOL(pfn_to_page
);
3205 EXPORT_SYMBOL(page_to_pfn
);
3206 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
3208 #if MAX_NUMNODES > 1
3210 * Find the highest possible node id.
3212 int highest_possible_node_id(void)
3215 unsigned int highest
= 0;
3217 for_each_node_mask(node
, node_possible_map
)
3221 EXPORT_SYMBOL(highest_possible_node_id
);