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>
43 #include <asm/tlbflush.h>
44 #include <asm/div64.h>
48 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
51 nodemask_t node_online_map __read_mostly
= { { [0] = 1UL } };
52 EXPORT_SYMBOL(node_online_map
);
53 nodemask_t node_possible_map __read_mostly
= NODE_MASK_ALL
;
54 EXPORT_SYMBOL(node_possible_map
);
55 unsigned long totalram_pages __read_mostly
;
56 unsigned long totalreserve_pages __read_mostly
;
58 int percpu_pagelist_fraction
;
60 static void __free_pages_ok(struct page
*page
, unsigned int order
);
63 * results with 256, 32 in the lowmem_reserve sysctl:
64 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
65 * 1G machine -> (16M dma, 784M normal, 224M high)
66 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
67 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
68 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
70 * TBD: should special case ZONE_DMA32 machines here - in those we normally
71 * don't need any ZONE_NORMAL reservation
73 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
75 #ifdef CONFIG_ZONE_DMA32
83 EXPORT_SYMBOL(totalram_pages
);
86 * Used by page_zone() to look up the address of the struct zone whose
87 * id is encoded in the upper bits of page->flags
89 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
90 EXPORT_SYMBOL(zone_table
);
92 static char *zone_names
[MAX_NR_ZONES
] = {
94 #ifdef CONFIG_ZONE_DMA32
103 int min_free_kbytes
= 1024;
105 unsigned long __meminitdata nr_kernel_pages
;
106 unsigned long __meminitdata nr_all_pages
;
107 static unsigned long __initdata dma_reserve
;
109 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
111 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
112 * ranges of memory (RAM) that may be registered with add_active_range().
113 * Ranges passed to add_active_range() will be merged if possible
114 * so the number of times add_active_range() can be called is
115 * related to the number of nodes and the number of holes
117 #ifdef CONFIG_MAX_ACTIVE_REGIONS
118 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
119 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
121 #if MAX_NUMNODES >= 32
122 /* If there can be many nodes, allow up to 50 holes per node */
123 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
125 /* By default, allow up to 256 distinct regions */
126 #define MAX_ACTIVE_REGIONS 256
130 struct node_active_region __initdata early_node_map
[MAX_ACTIVE_REGIONS
];
131 int __initdata nr_nodemap_entries
;
132 unsigned long __initdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
133 unsigned long __initdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
134 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
135 unsigned long __initdata node_boundary_start_pfn
[MAX_NUMNODES
];
136 unsigned long __initdata node_boundary_end_pfn
[MAX_NUMNODES
];
137 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
138 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
140 #ifdef CONFIG_DEBUG_VM
141 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
145 unsigned long pfn
= page_to_pfn(page
);
148 seq
= zone_span_seqbegin(zone
);
149 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
151 else if (pfn
< zone
->zone_start_pfn
)
153 } while (zone_span_seqretry(zone
, seq
));
158 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
160 #ifdef CONFIG_HOLES_IN_ZONE
161 if (!pfn_valid(page_to_pfn(page
)))
164 if (zone
!= page_zone(page
))
170 * Temporary debugging check for pages not lying within a given zone.
172 static int bad_range(struct zone
*zone
, struct page
*page
)
174 if (page_outside_zone_boundaries(zone
, page
))
176 if (!page_is_consistent(zone
, page
))
182 static inline int bad_range(struct zone
*zone
, struct page
*page
)
188 static void bad_page(struct page
*page
)
190 printk(KERN_EMERG
"Bad page state in process '%s'\n"
191 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
192 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
193 KERN_EMERG
"Backtrace:\n",
194 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
195 (unsigned long)page
->flags
, page
->mapping
,
196 page_mapcount(page
), page_count(page
));
198 page
->flags
&= ~(1 << PG_lru
|
208 set_page_count(page
, 0);
209 reset_page_mapcount(page
);
210 page
->mapping
= NULL
;
211 add_taint(TAINT_BAD_PAGE
);
215 * Higher-order pages are called "compound pages". They are structured thusly:
217 * The first PAGE_SIZE page is called the "head page".
219 * The remaining PAGE_SIZE pages are called "tail pages".
221 * All pages have PG_compound set. All pages have their ->private pointing at
222 * the head page (even the head page has this).
224 * The first tail page's ->lru.next holds the address of the compound page's
225 * put_page() function. Its ->lru.prev holds the order of allocation.
226 * This usage means that zero-order pages may not be compound.
229 static void free_compound_page(struct page
*page
)
231 __free_pages_ok(page
, (unsigned long)page
[1].lru
.prev
);
234 static void prep_compound_page(struct page
*page
, unsigned long order
)
237 int nr_pages
= 1 << order
;
239 page
[1].lru
.next
= (void *)free_compound_page
; /* set dtor */
240 page
[1].lru
.prev
= (void *)order
;
241 for (i
= 0; i
< nr_pages
; i
++) {
242 struct page
*p
= page
+ i
;
244 __SetPageCompound(p
);
245 set_page_private(p
, (unsigned long)page
);
249 static void destroy_compound_page(struct page
*page
, unsigned long order
)
252 int nr_pages
= 1 << order
;
254 if (unlikely((unsigned long)page
[1].lru
.prev
!= order
))
257 for (i
= 0; i
< nr_pages
; i
++) {
258 struct page
*p
= page
+ i
;
260 if (unlikely(!PageCompound(p
) |
261 (page_private(p
) != (unsigned long)page
)))
263 __ClearPageCompound(p
);
267 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
271 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
273 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
274 * and __GFP_HIGHMEM from hard or soft interrupt context.
276 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
277 for (i
= 0; i
< (1 << order
); i
++)
278 clear_highpage(page
+ i
);
282 * function for dealing with page's order in buddy system.
283 * zone->lock is already acquired when we use these.
284 * So, we don't need atomic page->flags operations here.
286 static inline unsigned long page_order(struct page
*page
)
288 return page_private(page
);
291 static inline void set_page_order(struct page
*page
, int order
)
293 set_page_private(page
, order
);
294 __SetPageBuddy(page
);
297 static inline void rmv_page_order(struct page
*page
)
299 __ClearPageBuddy(page
);
300 set_page_private(page
, 0);
304 * Locate the struct page for both the matching buddy in our
305 * pair (buddy1) and the combined O(n+1) page they form (page).
307 * 1) Any buddy B1 will have an order O twin B2 which satisfies
308 * the following equation:
310 * For example, if the starting buddy (buddy2) is #8 its order
312 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
314 * 2) Any buddy B will have an order O+1 parent P which
315 * satisfies the following equation:
318 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
320 static inline struct page
*
321 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
323 unsigned long buddy_idx
= page_idx
^ (1 << order
);
325 return page
+ (buddy_idx
- page_idx
);
328 static inline unsigned long
329 __find_combined_index(unsigned long page_idx
, unsigned int order
)
331 return (page_idx
& ~(1 << order
));
335 * This function checks whether a page is free && is the buddy
336 * we can do coalesce a page and its buddy if
337 * (a) the buddy is not in a hole &&
338 * (b) the buddy is in the buddy system &&
339 * (c) a page and its buddy have the same order &&
340 * (d) a page and its buddy are in the same zone.
342 * For recording whether a page is in the buddy system, we use PG_buddy.
343 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
345 * For recording page's order, we use page_private(page).
347 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
350 #ifdef CONFIG_HOLES_IN_ZONE
351 if (!pfn_valid(page_to_pfn(buddy
)))
355 if (page_zone_id(page
) != page_zone_id(buddy
))
358 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
359 BUG_ON(page_count(buddy
) != 0);
366 * Freeing function for a buddy system allocator.
368 * The concept of a buddy system is to maintain direct-mapped table
369 * (containing bit values) for memory blocks of various "orders".
370 * The bottom level table contains the map for the smallest allocatable
371 * units of memory (here, pages), and each level above it describes
372 * pairs of units from the levels below, hence, "buddies".
373 * At a high level, all that happens here is marking the table entry
374 * at the bottom level available, and propagating the changes upward
375 * as necessary, plus some accounting needed to play nicely with other
376 * parts of the VM system.
377 * At each level, we keep a list of pages, which are heads of continuous
378 * free pages of length of (1 << order) and marked with PG_buddy. Page's
379 * order is recorded in page_private(page) field.
380 * So when we are allocating or freeing one, we can derive the state of the
381 * other. That is, if we allocate a small block, and both were
382 * free, the remainder of the region must be split into blocks.
383 * If a block is freed, and its buddy is also free, then this
384 * triggers coalescing into a block of larger size.
389 static inline void __free_one_page(struct page
*page
,
390 struct zone
*zone
, unsigned int order
)
392 unsigned long page_idx
;
393 int order_size
= 1 << order
;
395 if (unlikely(PageCompound(page
)))
396 destroy_compound_page(page
, order
);
398 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
400 VM_BUG_ON(page_idx
& (order_size
- 1));
401 VM_BUG_ON(bad_range(zone
, page
));
403 zone
->free_pages
+= order_size
;
404 while (order
< MAX_ORDER
-1) {
405 unsigned long combined_idx
;
406 struct free_area
*area
;
409 buddy
= __page_find_buddy(page
, page_idx
, order
);
410 if (!page_is_buddy(page
, buddy
, order
))
411 break; /* Move the buddy up one level. */
413 list_del(&buddy
->lru
);
414 area
= zone
->free_area
+ order
;
416 rmv_page_order(buddy
);
417 combined_idx
= __find_combined_index(page_idx
, order
);
418 page
= page
+ (combined_idx
- page_idx
);
419 page_idx
= combined_idx
;
422 set_page_order(page
, order
);
423 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
424 zone
->free_area
[order
].nr_free
++;
427 static inline int free_pages_check(struct page
*page
)
429 if (unlikely(page_mapcount(page
) |
430 (page
->mapping
!= NULL
) |
431 (page_count(page
) != 0) |
445 __ClearPageDirty(page
);
447 * For now, we report if PG_reserved was found set, but do not
448 * clear it, and do not free the page. But we shall soon need
449 * to do more, for when the ZERO_PAGE count wraps negative.
451 return PageReserved(page
);
455 * Frees a list of pages.
456 * Assumes all pages on list are in same zone, and of same order.
457 * count is the number of pages to free.
459 * If the zone was previously in an "all pages pinned" state then look to
460 * see if this freeing clears that state.
462 * And clear the zone's pages_scanned counter, to hold off the "all pages are
463 * pinned" detection logic.
465 static void free_pages_bulk(struct zone
*zone
, int count
,
466 struct list_head
*list
, int order
)
468 spin_lock(&zone
->lock
);
469 zone
->all_unreclaimable
= 0;
470 zone
->pages_scanned
= 0;
474 VM_BUG_ON(list_empty(list
));
475 page
= list_entry(list
->prev
, struct page
, lru
);
476 /* have to delete it as __free_one_page list manipulates */
477 list_del(&page
->lru
);
478 __free_one_page(page
, zone
, order
);
480 spin_unlock(&zone
->lock
);
483 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
485 spin_lock(&zone
->lock
);
486 zone
->all_unreclaimable
= 0;
487 zone
->pages_scanned
= 0;
488 __free_one_page(page
, zone
,order
);
489 spin_unlock(&zone
->lock
);
492 static void __free_pages_ok(struct page
*page
, unsigned int order
)
498 arch_free_page(page
, order
);
499 if (!PageHighMem(page
))
500 debug_check_no_locks_freed(page_address(page
),
503 for (i
= 0 ; i
< (1 << order
) ; ++i
)
504 reserved
+= free_pages_check(page
+ i
);
508 kernel_map_pages(page
, 1 << order
, 0);
509 local_irq_save(flags
);
510 __count_vm_events(PGFREE
, 1 << order
);
511 free_one_page(page_zone(page
), page
, order
);
512 local_irq_restore(flags
);
516 * permit the bootmem allocator to evade page validation on high-order frees
518 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
521 __ClearPageReserved(page
);
522 set_page_count(page
, 0);
523 set_page_refcounted(page
);
529 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
530 struct page
*p
= &page
[loop
];
532 if (loop
+ 1 < BITS_PER_LONG
)
534 __ClearPageReserved(p
);
535 set_page_count(p
, 0);
538 set_page_refcounted(page
);
539 __free_pages(page
, order
);
545 * The order of subdivision here is critical for the IO subsystem.
546 * Please do not alter this order without good reasons and regression
547 * testing. Specifically, as large blocks of memory are subdivided,
548 * the order in which smaller blocks are delivered depends on the order
549 * they're subdivided in this function. This is the primary factor
550 * influencing the order in which pages are delivered to the IO
551 * subsystem according to empirical testing, and this is also justified
552 * by considering the behavior of a buddy system containing a single
553 * large block of memory acted on by a series of small allocations.
554 * This behavior is a critical factor in sglist merging's success.
558 static inline void expand(struct zone
*zone
, struct page
*page
,
559 int low
, int high
, struct free_area
*area
)
561 unsigned long size
= 1 << high
;
567 VM_BUG_ON(bad_range(zone
, &page
[size
]));
568 list_add(&page
[size
].lru
, &area
->free_list
);
570 set_page_order(&page
[size
], high
);
575 * This page is about to be returned from the page allocator
577 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
579 if (unlikely(page_mapcount(page
) |
580 (page
->mapping
!= NULL
) |
581 (page_count(page
) != 0) |
597 * For now, we report if PG_reserved was found set, but do not
598 * clear it, and do not allocate the page: as a safety net.
600 if (PageReserved(page
))
603 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
604 1 << PG_referenced
| 1 << PG_arch_1
|
605 1 << PG_checked
| 1 << PG_mappedtodisk
);
606 set_page_private(page
, 0);
607 set_page_refcounted(page
);
608 kernel_map_pages(page
, 1 << order
, 1);
610 if (gfp_flags
& __GFP_ZERO
)
611 prep_zero_page(page
, order
, gfp_flags
);
613 if (order
&& (gfp_flags
& __GFP_COMP
))
614 prep_compound_page(page
, order
);
620 * Do the hard work of removing an element from the buddy allocator.
621 * Call me with the zone->lock already held.
623 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
625 struct free_area
* area
;
626 unsigned int current_order
;
629 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
630 area
= zone
->free_area
+ current_order
;
631 if (list_empty(&area
->free_list
))
634 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
635 list_del(&page
->lru
);
636 rmv_page_order(page
);
638 zone
->free_pages
-= 1UL << order
;
639 expand(zone
, page
, order
, current_order
, area
);
647 * Obtain a specified number of elements from the buddy allocator, all under
648 * a single hold of the lock, for efficiency. Add them to the supplied list.
649 * Returns the number of new pages which were placed at *list.
651 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
652 unsigned long count
, struct list_head
*list
)
656 spin_lock(&zone
->lock
);
657 for (i
= 0; i
< count
; ++i
) {
658 struct page
*page
= __rmqueue(zone
, order
);
659 if (unlikely(page
== NULL
))
661 list_add_tail(&page
->lru
, list
);
663 spin_unlock(&zone
->lock
);
669 * Called from the slab reaper to drain pagesets on a particular node that
670 * belongs to the currently executing processor.
671 * Note that this function must be called with the thread pinned to
672 * a single processor.
674 void drain_node_pages(int nodeid
)
680 for (z
= 0; z
< MAX_NR_ZONES
; z
++) {
681 struct zone
*zone
= NODE_DATA(nodeid
)->node_zones
+ z
;
682 struct per_cpu_pageset
*pset
;
684 if (!populated_zone(zone
))
687 pset
= zone_pcp(zone
, smp_processor_id());
688 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
689 struct per_cpu_pages
*pcp
;
693 local_irq_save(flags
);
694 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
696 local_irq_restore(flags
);
703 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
704 static void __drain_pages(unsigned int cpu
)
710 for_each_zone(zone
) {
711 struct per_cpu_pageset
*pset
;
713 pset
= zone_pcp(zone
, cpu
);
714 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
715 struct per_cpu_pages
*pcp
;
718 local_irq_save(flags
);
719 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
721 local_irq_restore(flags
);
725 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
729 void mark_free_pages(struct zone
*zone
)
731 unsigned long pfn
, max_zone_pfn
;
734 struct list_head
*curr
;
736 if (!zone
->spanned_pages
)
739 spin_lock_irqsave(&zone
->lock
, flags
);
741 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
742 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
743 if (pfn_valid(pfn
)) {
744 struct page
*page
= pfn_to_page(pfn
);
746 if (!PageNosave(page
))
747 ClearPageNosaveFree(page
);
750 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
751 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
754 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
755 for (i
= 0; i
< (1UL << order
); i
++)
756 SetPageNosaveFree(pfn_to_page(pfn
+ i
));
759 spin_unlock_irqrestore(&zone
->lock
, flags
);
763 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
765 void drain_local_pages(void)
769 local_irq_save(flags
);
770 __drain_pages(smp_processor_id());
771 local_irq_restore(flags
);
773 #endif /* CONFIG_PM */
776 * Free a 0-order page
778 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
780 struct zone
*zone
= page_zone(page
);
781 struct per_cpu_pages
*pcp
;
784 arch_free_page(page
, 0);
787 page
->mapping
= NULL
;
788 if (free_pages_check(page
))
791 kernel_map_pages(page
, 1, 0);
793 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
794 local_irq_save(flags
);
795 __count_vm_event(PGFREE
);
796 list_add(&page
->lru
, &pcp
->list
);
798 if (pcp
->count
>= pcp
->high
) {
799 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
800 pcp
->count
-= pcp
->batch
;
802 local_irq_restore(flags
);
806 void fastcall
free_hot_page(struct page
*page
)
808 free_hot_cold_page(page
, 0);
811 void fastcall
free_cold_page(struct page
*page
)
813 free_hot_cold_page(page
, 1);
817 * split_page takes a non-compound higher-order page, and splits it into
818 * n (1<<order) sub-pages: page[0..n]
819 * Each sub-page must be freed individually.
821 * Note: this is probably too low level an operation for use in drivers.
822 * Please consult with lkml before using this in your driver.
824 void split_page(struct page
*page
, unsigned int order
)
828 VM_BUG_ON(PageCompound(page
));
829 VM_BUG_ON(!page_count(page
));
830 for (i
= 1; i
< (1 << order
); i
++)
831 set_page_refcounted(page
+ i
);
835 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
836 * we cheat by calling it from here, in the order > 0 path. Saves a branch
839 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
840 struct zone
*zone
, int order
, gfp_t gfp_flags
)
844 int cold
= !!(gfp_flags
& __GFP_COLD
);
849 if (likely(order
== 0)) {
850 struct per_cpu_pages
*pcp
;
852 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
853 local_irq_save(flags
);
855 pcp
->count
+= rmqueue_bulk(zone
, 0,
856 pcp
->batch
, &pcp
->list
);
857 if (unlikely(!pcp
->count
))
860 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
861 list_del(&page
->lru
);
864 spin_lock_irqsave(&zone
->lock
, flags
);
865 page
= __rmqueue(zone
, order
);
866 spin_unlock(&zone
->lock
);
871 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
872 zone_statistics(zonelist
, zone
);
873 local_irq_restore(flags
);
876 VM_BUG_ON(bad_range(zone
, page
));
877 if (prep_new_page(page
, order
, gfp_flags
))
882 local_irq_restore(flags
);
887 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
888 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
889 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
890 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
891 #define ALLOC_HARDER 0x10 /* try to alloc harder */
892 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
893 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
896 * Return 1 if free pages are above 'mark'. This takes into account the order
899 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
900 int classzone_idx
, int alloc_flags
)
902 /* free_pages my go negative - that's OK */
903 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
906 if (alloc_flags
& ALLOC_HIGH
)
908 if (alloc_flags
& ALLOC_HARDER
)
911 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
913 for (o
= 0; o
< order
; o
++) {
914 /* At the next order, this order's pages become unavailable */
915 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
917 /* Require fewer higher order pages to be free */
920 if (free_pages
<= min
)
927 * get_page_from_freeliest goes through the zonelist trying to allocate
931 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
932 struct zonelist
*zonelist
, int alloc_flags
)
934 struct zone
**z
= zonelist
->zones
;
935 struct page
*page
= NULL
;
936 int classzone_idx
= zone_idx(*z
);
940 * Go through the zonelist once, looking for a zone with enough free.
941 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
945 if (unlikely((gfp_mask
& __GFP_THISNODE
) &&
946 zone
->zone_pgdat
!= zonelist
->zones
[0]->zone_pgdat
))
948 if ((alloc_flags
& ALLOC_CPUSET
) &&
949 !cpuset_zone_allowed(zone
, gfp_mask
))
952 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
954 if (alloc_flags
& ALLOC_WMARK_MIN
)
955 mark
= zone
->pages_min
;
956 else if (alloc_flags
& ALLOC_WMARK_LOW
)
957 mark
= zone
->pages_low
;
959 mark
= zone
->pages_high
;
960 if (!zone_watermark_ok(zone
, order
, mark
,
961 classzone_idx
, alloc_flags
))
962 if (!zone_reclaim_mode
||
963 !zone_reclaim(zone
, gfp_mask
, order
))
967 page
= buffered_rmqueue(zonelist
, zone
, order
, gfp_mask
);
971 } while (*(++z
) != NULL
);
976 * This is the 'heart' of the zoned buddy allocator.
978 struct page
* fastcall
979 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
980 struct zonelist
*zonelist
)
982 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
985 struct reclaim_state reclaim_state
;
986 struct task_struct
*p
= current
;
989 int did_some_progress
;
991 might_sleep_if(wait
);
994 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
996 if (unlikely(*z
== NULL
)) {
997 /* Should this ever happen?? */
1001 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1002 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1007 wakeup_kswapd(*z
, order
);
1011 * OK, we're below the kswapd watermark and have kicked background
1012 * reclaim. Now things get more complex, so set up alloc_flags according
1013 * to how we want to proceed.
1015 * The caller may dip into page reserves a bit more if the caller
1016 * cannot run direct reclaim, or if the caller has realtime scheduling
1017 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1018 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1020 alloc_flags
= ALLOC_WMARK_MIN
;
1021 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1022 alloc_flags
|= ALLOC_HARDER
;
1023 if (gfp_mask
& __GFP_HIGH
)
1024 alloc_flags
|= ALLOC_HIGH
;
1026 alloc_flags
|= ALLOC_CPUSET
;
1029 * Go through the zonelist again. Let __GFP_HIGH and allocations
1030 * coming from realtime tasks go deeper into reserves.
1032 * This is the last chance, in general, before the goto nopage.
1033 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1034 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1036 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1040 /* This allocation should allow future memory freeing. */
1042 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1043 && !in_interrupt()) {
1044 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1046 /* go through the zonelist yet again, ignoring mins */
1047 page
= get_page_from_freelist(gfp_mask
, order
,
1048 zonelist
, ALLOC_NO_WATERMARKS
);
1051 if (gfp_mask
& __GFP_NOFAIL
) {
1052 blk_congestion_wait(WRITE
, HZ
/50);
1059 /* Atomic allocations - we can't balance anything */
1066 /* We now go into synchronous reclaim */
1067 cpuset_memory_pressure_bump();
1068 p
->flags
|= PF_MEMALLOC
;
1069 reclaim_state
.reclaimed_slab
= 0;
1070 p
->reclaim_state
= &reclaim_state
;
1072 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
1074 p
->reclaim_state
= NULL
;
1075 p
->flags
&= ~PF_MEMALLOC
;
1079 if (likely(did_some_progress
)) {
1080 page
= get_page_from_freelist(gfp_mask
, order
,
1081 zonelist
, alloc_flags
);
1084 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1086 * Go through the zonelist yet one more time, keep
1087 * very high watermark here, this is only to catch
1088 * a parallel oom killing, we must fail if we're still
1089 * under heavy pressure.
1091 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1092 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1096 out_of_memory(zonelist
, gfp_mask
, order
);
1101 * Don't let big-order allocations loop unless the caller explicitly
1102 * requests that. Wait for some write requests to complete then retry.
1104 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1105 * <= 3, but that may not be true in other implementations.
1108 if (!(gfp_mask
& __GFP_NORETRY
)) {
1109 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
1111 if (gfp_mask
& __GFP_NOFAIL
)
1115 blk_congestion_wait(WRITE
, HZ
/50);
1120 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1121 printk(KERN_WARNING
"%s: page allocation failure."
1122 " order:%d, mode:0x%x\n",
1123 p
->comm
, order
, gfp_mask
);
1131 EXPORT_SYMBOL(__alloc_pages
);
1134 * Common helper functions.
1136 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1139 page
= alloc_pages(gfp_mask
, order
);
1142 return (unsigned long) page_address(page
);
1145 EXPORT_SYMBOL(__get_free_pages
);
1147 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1152 * get_zeroed_page() returns a 32-bit address, which cannot represent
1155 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1157 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1159 return (unsigned long) page_address(page
);
1163 EXPORT_SYMBOL(get_zeroed_page
);
1165 void __pagevec_free(struct pagevec
*pvec
)
1167 int i
= pagevec_count(pvec
);
1170 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1173 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1175 if (put_page_testzero(page
)) {
1177 free_hot_page(page
);
1179 __free_pages_ok(page
, order
);
1183 EXPORT_SYMBOL(__free_pages
);
1185 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1188 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1189 __free_pages(virt_to_page((void *)addr
), order
);
1193 EXPORT_SYMBOL(free_pages
);
1196 * Total amount of free (allocatable) RAM:
1198 unsigned int nr_free_pages(void)
1200 unsigned int sum
= 0;
1204 sum
+= zone
->free_pages
;
1209 EXPORT_SYMBOL(nr_free_pages
);
1212 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1214 unsigned int sum
= 0;
1217 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1218 sum
+= pgdat
->node_zones
[i
].free_pages
;
1224 static unsigned int nr_free_zone_pages(int offset
)
1226 /* Just pick one node, since fallback list is circular */
1227 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1228 unsigned int sum
= 0;
1230 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1231 struct zone
**zonep
= zonelist
->zones
;
1234 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1235 unsigned long size
= zone
->present_pages
;
1236 unsigned long high
= zone
->pages_high
;
1245 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1247 unsigned int nr_free_buffer_pages(void)
1249 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1253 * Amount of free RAM allocatable within all zones
1255 unsigned int nr_free_pagecache_pages(void)
1257 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1260 static void show_node(struct zone
*zone
)
1262 printk("Node %ld ", zone_to_nid(zone
));
1265 #define show_node(zone) do { } while (0)
1268 void si_meminfo(struct sysinfo
*val
)
1270 val
->totalram
= totalram_pages
;
1272 val
->freeram
= nr_free_pages();
1273 val
->bufferram
= nr_blockdev_pages();
1274 val
->totalhigh
= totalhigh_pages
;
1275 val
->freehigh
= nr_free_highpages();
1276 val
->mem_unit
= PAGE_SIZE
;
1279 EXPORT_SYMBOL(si_meminfo
);
1282 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1284 pg_data_t
*pgdat
= NODE_DATA(nid
);
1286 val
->totalram
= pgdat
->node_present_pages
;
1287 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1288 #ifdef CONFIG_HIGHMEM
1289 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1290 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1295 val
->mem_unit
= PAGE_SIZE
;
1299 #define K(x) ((x) << (PAGE_SHIFT-10))
1302 * Show free area list (used inside shift_scroll-lock stuff)
1303 * We also calculate the percentage fragmentation. We do this by counting the
1304 * memory on each free list with the exception of the first item on the list.
1306 void show_free_areas(void)
1308 int cpu
, temperature
;
1309 unsigned long active
;
1310 unsigned long inactive
;
1314 for_each_zone(zone
) {
1316 printk("%s per-cpu:", zone
->name
);
1318 if (!populated_zone(zone
)) {
1324 for_each_online_cpu(cpu
) {
1325 struct per_cpu_pageset
*pageset
;
1327 pageset
= zone_pcp(zone
, cpu
);
1329 for (temperature
= 0; temperature
< 2; temperature
++)
1330 printk("cpu %d %s: high %d, batch %d used:%d\n",
1332 temperature
? "cold" : "hot",
1333 pageset
->pcp
[temperature
].high
,
1334 pageset
->pcp
[temperature
].batch
,
1335 pageset
->pcp
[temperature
].count
);
1339 get_zone_counts(&active
, &inactive
, &free
);
1341 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1342 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1345 global_page_state(NR_FILE_DIRTY
),
1346 global_page_state(NR_WRITEBACK
),
1347 global_page_state(NR_UNSTABLE_NFS
),
1349 global_page_state(NR_SLAB_RECLAIMABLE
) +
1350 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1351 global_page_state(NR_FILE_MAPPED
),
1352 global_page_state(NR_PAGETABLE
));
1354 for_each_zone(zone
) {
1366 " pages_scanned:%lu"
1367 " all_unreclaimable? %s"
1370 K(zone
->free_pages
),
1373 K(zone
->pages_high
),
1375 K(zone
->nr_inactive
),
1376 K(zone
->present_pages
),
1377 zone
->pages_scanned
,
1378 (zone
->all_unreclaimable
? "yes" : "no")
1380 printk("lowmem_reserve[]:");
1381 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1382 printk(" %lu", zone
->lowmem_reserve
[i
]);
1386 for_each_zone(zone
) {
1387 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1390 printk("%s: ", zone
->name
);
1391 if (!populated_zone(zone
)) {
1396 spin_lock_irqsave(&zone
->lock
, flags
);
1397 for (order
= 0; order
< MAX_ORDER
; order
++) {
1398 nr
[order
] = zone
->free_area
[order
].nr_free
;
1399 total
+= nr
[order
] << order
;
1401 spin_unlock_irqrestore(&zone
->lock
, flags
);
1402 for (order
= 0; order
< MAX_ORDER
; order
++)
1403 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1404 printk("= %lukB\n", K(total
));
1407 show_swap_cache_info();
1411 * Builds allocation fallback zone lists.
1413 * Add all populated zones of a node to the zonelist.
1415 static int __meminit
build_zonelists_node(pg_data_t
*pgdat
,
1416 struct zonelist
*zonelist
, int nr_zones
, enum zone_type zone_type
)
1420 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1425 zone
= pgdat
->node_zones
+ zone_type
;
1426 if (populated_zone(zone
)) {
1427 zonelist
->zones
[nr_zones
++] = zone
;
1428 check_highest_zone(zone_type
);
1431 } while (zone_type
);
1436 #define MAX_NODE_LOAD (num_online_nodes())
1437 static int __meminitdata node_load
[MAX_NUMNODES
];
1439 * find_next_best_node - find the next node that should appear in a given node's fallback list
1440 * @node: node whose fallback list we're appending
1441 * @used_node_mask: nodemask_t of already used nodes
1443 * We use a number of factors to determine which is the next node that should
1444 * appear on a given node's fallback list. The node should not have appeared
1445 * already in @node's fallback list, and it should be the next closest node
1446 * according to the distance array (which contains arbitrary distance values
1447 * from each node to each node in the system), and should also prefer nodes
1448 * with no CPUs, since presumably they'll have very little allocation pressure
1449 * on them otherwise.
1450 * It returns -1 if no node is found.
1452 static int __meminit
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1455 int min_val
= INT_MAX
;
1458 /* Use the local node if we haven't already */
1459 if (!node_isset(node
, *used_node_mask
)) {
1460 node_set(node
, *used_node_mask
);
1464 for_each_online_node(n
) {
1467 /* Don't want a node to appear more than once */
1468 if (node_isset(n
, *used_node_mask
))
1471 /* Use the distance array to find the distance */
1472 val
= node_distance(node
, n
);
1474 /* Penalize nodes under us ("prefer the next node") */
1477 /* Give preference to headless and unused nodes */
1478 tmp
= node_to_cpumask(n
);
1479 if (!cpus_empty(tmp
))
1480 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1482 /* Slight preference for less loaded node */
1483 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1484 val
+= node_load
[n
];
1486 if (val
< min_val
) {
1493 node_set(best_node
, *used_node_mask
);
1498 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1500 int j
, node
, local_node
;
1502 int prev_node
, load
;
1503 struct zonelist
*zonelist
;
1504 nodemask_t used_mask
;
1506 /* initialize zonelists */
1507 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1508 zonelist
= pgdat
->node_zonelists
+ i
;
1509 zonelist
->zones
[0] = NULL
;
1512 /* NUMA-aware ordering of nodes */
1513 local_node
= pgdat
->node_id
;
1514 load
= num_online_nodes();
1515 prev_node
= local_node
;
1516 nodes_clear(used_mask
);
1517 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1518 int distance
= node_distance(local_node
, node
);
1521 * If another node is sufficiently far away then it is better
1522 * to reclaim pages in a zone before going off node.
1524 if (distance
> RECLAIM_DISTANCE
)
1525 zone_reclaim_mode
= 1;
1528 * We don't want to pressure a particular node.
1529 * So adding penalty to the first node in same
1530 * distance group to make it round-robin.
1533 if (distance
!= node_distance(local_node
, prev_node
))
1534 node_load
[node
] += load
;
1537 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1538 zonelist
= pgdat
->node_zonelists
+ i
;
1539 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1541 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1542 zonelist
->zones
[j
] = NULL
;
1547 #else /* CONFIG_NUMA */
1549 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1551 int node
, local_node
;
1554 local_node
= pgdat
->node_id
;
1555 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1556 struct zonelist
*zonelist
;
1558 zonelist
= pgdat
->node_zonelists
+ i
;
1560 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
1562 * Now we build the zonelist so that it contains the zones
1563 * of all the other nodes.
1564 * We don't want to pressure a particular node, so when
1565 * building the zones for node N, we make sure that the
1566 * zones coming right after the local ones are those from
1567 * node N+1 (modulo N)
1569 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1570 if (!node_online(node
))
1572 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1574 for (node
= 0; node
< local_node
; node
++) {
1575 if (!node_online(node
))
1577 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
1580 zonelist
->zones
[j
] = NULL
;
1584 #endif /* CONFIG_NUMA */
1586 /* return values int ....just for stop_machine_run() */
1587 static int __meminit
__build_all_zonelists(void *dummy
)
1590 for_each_online_node(nid
)
1591 build_zonelists(NODE_DATA(nid
));
1595 void __meminit
build_all_zonelists(void)
1597 if (system_state
== SYSTEM_BOOTING
) {
1598 __build_all_zonelists(0);
1599 cpuset_init_current_mems_allowed();
1601 /* we have to stop all cpus to guaranntee there is no user
1603 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
1604 /* cpuset refresh routine should be here */
1606 vm_total_pages
= nr_free_pagecache_pages();
1607 printk("Built %i zonelists. Total pages: %ld\n",
1608 num_online_nodes(), vm_total_pages
);
1612 * Helper functions to size the waitqueue hash table.
1613 * Essentially these want to choose hash table sizes sufficiently
1614 * large so that collisions trying to wait on pages are rare.
1615 * But in fact, the number of active page waitqueues on typical
1616 * systems is ridiculously low, less than 200. So this is even
1617 * conservative, even though it seems large.
1619 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1620 * waitqueues, i.e. the size of the waitq table given the number of pages.
1622 #define PAGES_PER_WAITQUEUE 256
1624 #ifndef CONFIG_MEMORY_HOTPLUG
1625 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1627 unsigned long size
= 1;
1629 pages
/= PAGES_PER_WAITQUEUE
;
1631 while (size
< pages
)
1635 * Once we have dozens or even hundreds of threads sleeping
1636 * on IO we've got bigger problems than wait queue collision.
1637 * Limit the size of the wait table to a reasonable size.
1639 size
= min(size
, 4096UL);
1641 return max(size
, 4UL);
1645 * A zone's size might be changed by hot-add, so it is not possible to determine
1646 * a suitable size for its wait_table. So we use the maximum size now.
1648 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1650 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1651 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1652 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1654 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1655 * or more by the traditional way. (See above). It equals:
1657 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1658 * ia64(16K page size) : = ( 8G + 4M)byte.
1659 * powerpc (64K page size) : = (32G +16M)byte.
1661 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1668 * This is an integer logarithm so that shifts can be used later
1669 * to extract the more random high bits from the multiplicative
1670 * hash function before the remainder is taken.
1672 static inline unsigned long wait_table_bits(unsigned long size
)
1677 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1680 * Initially all pages are reserved - free ones are freed
1681 * up by free_all_bootmem() once the early boot process is
1682 * done. Non-atomic initialization, single-pass.
1684 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1685 unsigned long start_pfn
)
1688 unsigned long end_pfn
= start_pfn
+ size
;
1691 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1692 if (!early_pfn_valid(pfn
))
1694 page
= pfn_to_page(pfn
);
1695 set_page_links(page
, zone
, nid
, pfn
);
1696 init_page_count(page
);
1697 reset_page_mapcount(page
);
1698 SetPageReserved(page
);
1699 INIT_LIST_HEAD(&page
->lru
);
1700 #ifdef WANT_PAGE_VIRTUAL
1701 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1702 if (!is_highmem_idx(zone
))
1703 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1708 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1712 for (order
= 0; order
< MAX_ORDER
; order
++) {
1713 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1714 zone
->free_area
[order
].nr_free
= 0;
1718 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1719 void zonetable_add(struct zone
*zone
, int nid
, enum zone_type zid
,
1720 unsigned long pfn
, unsigned long size
)
1722 unsigned long snum
= pfn_to_section_nr(pfn
);
1723 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1726 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1728 for (; snum
<= end
; snum
++)
1729 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1732 #ifndef __HAVE_ARCH_MEMMAP_INIT
1733 #define memmap_init(size, nid, zone, start_pfn) \
1734 memmap_init_zone((size), (nid), (zone), (start_pfn))
1737 static int __cpuinit
zone_batchsize(struct zone
*zone
)
1742 * The per-cpu-pages pools are set to around 1000th of the
1743 * size of the zone. But no more than 1/2 of a meg.
1745 * OK, so we don't know how big the cache is. So guess.
1747 batch
= zone
->present_pages
/ 1024;
1748 if (batch
* PAGE_SIZE
> 512 * 1024)
1749 batch
= (512 * 1024) / PAGE_SIZE
;
1750 batch
/= 4; /* We effectively *= 4 below */
1755 * Clamp the batch to a 2^n - 1 value. Having a power
1756 * of 2 value was found to be more likely to have
1757 * suboptimal cache aliasing properties in some cases.
1759 * For example if 2 tasks are alternately allocating
1760 * batches of pages, one task can end up with a lot
1761 * of pages of one half of the possible page colors
1762 * and the other with pages of the other colors.
1764 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1769 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1771 struct per_cpu_pages
*pcp
;
1773 memset(p
, 0, sizeof(*p
));
1775 pcp
= &p
->pcp
[0]; /* hot */
1777 pcp
->high
= 6 * batch
;
1778 pcp
->batch
= max(1UL, 1 * batch
);
1779 INIT_LIST_HEAD(&pcp
->list
);
1781 pcp
= &p
->pcp
[1]; /* cold*/
1783 pcp
->high
= 2 * batch
;
1784 pcp
->batch
= max(1UL, batch
/2);
1785 INIT_LIST_HEAD(&pcp
->list
);
1789 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1790 * to the value high for the pageset p.
1793 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
1796 struct per_cpu_pages
*pcp
;
1798 pcp
= &p
->pcp
[0]; /* hot list */
1800 pcp
->batch
= max(1UL, high
/4);
1801 if ((high
/4) > (PAGE_SHIFT
* 8))
1802 pcp
->batch
= PAGE_SHIFT
* 8;
1808 * Boot pageset table. One per cpu which is going to be used for all
1809 * zones and all nodes. The parameters will be set in such a way
1810 * that an item put on a list will immediately be handed over to
1811 * the buddy list. This is safe since pageset manipulation is done
1812 * with interrupts disabled.
1814 * Some NUMA counter updates may also be caught by the boot pagesets.
1816 * The boot_pagesets must be kept even after bootup is complete for
1817 * unused processors and/or zones. They do play a role for bootstrapping
1818 * hotplugged processors.
1820 * zoneinfo_show() and maybe other functions do
1821 * not check if the processor is online before following the pageset pointer.
1822 * Other parts of the kernel may not check if the zone is available.
1824 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
1827 * Dynamically allocate memory for the
1828 * per cpu pageset array in struct zone.
1830 static int __cpuinit
process_zones(int cpu
)
1832 struct zone
*zone
, *dzone
;
1834 for_each_zone(zone
) {
1836 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
1837 GFP_KERNEL
, cpu_to_node(cpu
));
1838 if (!zone_pcp(zone
, cpu
))
1841 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
1843 if (percpu_pagelist_fraction
)
1844 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
1845 (zone
->present_pages
/ percpu_pagelist_fraction
));
1850 for_each_zone(dzone
) {
1853 kfree(zone_pcp(dzone
, cpu
));
1854 zone_pcp(dzone
, cpu
) = NULL
;
1859 static inline void free_zone_pagesets(int cpu
)
1863 for_each_zone(zone
) {
1864 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1866 /* Free per_cpu_pageset if it is slab allocated */
1867 if (pset
!= &boot_pageset
[cpu
])
1869 zone_pcp(zone
, cpu
) = NULL
;
1873 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1874 unsigned long action
,
1877 int cpu
= (long)hcpu
;
1878 int ret
= NOTIFY_OK
;
1881 case CPU_UP_PREPARE
:
1882 if (process_zones(cpu
))
1885 case CPU_UP_CANCELED
:
1887 free_zone_pagesets(cpu
);
1895 static struct notifier_block __cpuinitdata pageset_notifier
=
1896 { &pageset_cpuup_callback
, NULL
, 0 };
1898 void __init
setup_per_cpu_pageset(void)
1902 /* Initialize per_cpu_pageset for cpu 0.
1903 * A cpuup callback will do this for every cpu
1904 * as it comes online
1906 err
= process_zones(smp_processor_id());
1908 register_cpu_notifier(&pageset_notifier
);
1914 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1917 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1921 * The per-page waitqueue mechanism uses hashed waitqueues
1924 zone
->wait_table_hash_nr_entries
=
1925 wait_table_hash_nr_entries(zone_size_pages
);
1926 zone
->wait_table_bits
=
1927 wait_table_bits(zone
->wait_table_hash_nr_entries
);
1928 alloc_size
= zone
->wait_table_hash_nr_entries
1929 * sizeof(wait_queue_head_t
);
1931 if (system_state
== SYSTEM_BOOTING
) {
1932 zone
->wait_table
= (wait_queue_head_t
*)
1933 alloc_bootmem_node(pgdat
, alloc_size
);
1936 * This case means that a zone whose size was 0 gets new memory
1937 * via memory hot-add.
1938 * But it may be the case that a new node was hot-added. In
1939 * this case vmalloc() will not be able to use this new node's
1940 * memory - this wait_table must be initialized to use this new
1941 * node itself as well.
1942 * To use this new node's memory, further consideration will be
1945 zone
->wait_table
= (wait_queue_head_t
*)vmalloc(alloc_size
);
1947 if (!zone
->wait_table
)
1950 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
1951 init_waitqueue_head(zone
->wait_table
+ i
);
1956 static __meminit
void zone_pcp_init(struct zone
*zone
)
1959 unsigned long batch
= zone_batchsize(zone
);
1961 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1963 /* Early boot. Slab allocator not functional yet */
1964 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
1965 setup_pageset(&boot_pageset
[cpu
],0);
1967 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1970 if (zone
->present_pages
)
1971 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1972 zone
->name
, zone
->present_pages
, batch
);
1975 __meminit
int init_currently_empty_zone(struct zone
*zone
,
1976 unsigned long zone_start_pfn
,
1979 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1981 ret
= zone_wait_table_init(zone
, size
);
1984 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1986 zone
->zone_start_pfn
= zone_start_pfn
;
1988 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1990 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
1995 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
1997 * Basic iterator support. Return the first range of PFNs for a node
1998 * Note: nid == MAX_NUMNODES returns first region regardless of node
2000 static int __init
first_active_region_index_in_nid(int nid
)
2004 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2005 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2012 * Basic iterator support. Return the next active range of PFNs for a node
2013 * Note: nid == MAX_NUMNODES returns next region regardles of node
2015 static int __init
next_active_region_index_in_nid(int index
, int nid
)
2017 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2018 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2024 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2026 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2027 * Architectures may implement their own version but if add_active_range()
2028 * was used and there are no special requirements, this is a convenient
2031 int __init
early_pfn_to_nid(unsigned long pfn
)
2035 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2036 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2037 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2039 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2040 return early_node_map
[i
].nid
;
2045 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2047 /* Basic iterator support to walk early_node_map[] */
2048 #define for_each_active_range_index_in_nid(i, nid) \
2049 for (i = first_active_region_index_in_nid(nid); i != -1; \
2050 i = next_active_region_index_in_nid(i, nid))
2053 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2054 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed
2055 * @max_low_pfn: The highest PFN that till be passed to free_bootmem_node
2057 * If an architecture guarantees that all ranges registered with
2058 * add_active_ranges() contain no holes and may be freed, this
2059 * this function may be used instead of calling free_bootmem() manually.
2061 void __init
free_bootmem_with_active_regions(int nid
,
2062 unsigned long max_low_pfn
)
2066 for_each_active_range_index_in_nid(i
, nid
) {
2067 unsigned long size_pages
= 0;
2068 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2070 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2073 if (end_pfn
> max_low_pfn
)
2074 end_pfn
= max_low_pfn
;
2076 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2077 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2078 PFN_PHYS(early_node_map
[i
].start_pfn
),
2079 size_pages
<< PAGE_SHIFT
);
2084 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2085 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used
2087 * If an architecture guarantees that all ranges registered with
2088 * add_active_ranges() contain no holes and may be freed, this
2089 * this function may be used instead of calling memory_present() manually.
2091 void __init
sparse_memory_present_with_active_regions(int nid
)
2095 for_each_active_range_index_in_nid(i
, nid
)
2096 memory_present(early_node_map
[i
].nid
,
2097 early_node_map
[i
].start_pfn
,
2098 early_node_map
[i
].end_pfn
);
2102 * push_node_boundaries - Push node boundaries to at least the requested boundary
2103 * @nid: The nid of the node to push the boundary for
2104 * @start_pfn: The start pfn of the node
2105 * @end_pfn: The end pfn of the node
2107 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2108 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2109 * be hotplugged even though no physical memory exists. This function allows
2110 * an arch to push out the node boundaries so mem_map is allocated that can
2113 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2114 void __init
push_node_boundaries(unsigned int nid
,
2115 unsigned long start_pfn
, unsigned long end_pfn
)
2117 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2118 nid
, start_pfn
, end_pfn
);
2120 /* Initialise the boundary for this node if necessary */
2121 if (node_boundary_end_pfn
[nid
] == 0)
2122 node_boundary_start_pfn
[nid
] = -1UL;
2124 /* Update the boundaries */
2125 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2126 node_boundary_start_pfn
[nid
] = start_pfn
;
2127 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2128 node_boundary_end_pfn
[nid
] = end_pfn
;
2131 /* If necessary, push the node boundary out for reserve hotadd */
2132 static void __init
account_node_boundary(unsigned int nid
,
2133 unsigned long *start_pfn
, unsigned long *end_pfn
)
2135 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2136 nid
, *start_pfn
, *end_pfn
);
2138 /* Return if boundary information has not been provided */
2139 if (node_boundary_end_pfn
[nid
] == 0)
2142 /* Check the boundaries and update if necessary */
2143 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
2144 *start_pfn
= node_boundary_start_pfn
[nid
];
2145 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
2146 *end_pfn
= node_boundary_end_pfn
[nid
];
2149 void __init
push_node_boundaries(unsigned int nid
,
2150 unsigned long start_pfn
, unsigned long end_pfn
) {}
2152 static void __init
account_node_boundary(unsigned int nid
,
2153 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
2158 * get_pfn_range_for_nid - Return the start and end page frames for a node
2159 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned
2160 * @start_pfn: Passed by reference. On return, it will have the node start_pfn
2161 * @end_pfn: Passed by reference. On return, it will have the node end_pfn
2163 * It returns the start and end page frame of a node based on information
2164 * provided by an arch calling add_active_range(). If called for a node
2165 * with no available memory, a warning is printed and the start and end
2168 void __init
get_pfn_range_for_nid(unsigned int nid
,
2169 unsigned long *start_pfn
, unsigned long *end_pfn
)
2175 for_each_active_range_index_in_nid(i
, nid
) {
2176 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
2177 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
2180 if (*start_pfn
== -1UL) {
2181 printk(KERN_WARNING
"Node %u active with no memory\n", nid
);
2185 /* Push the node boundaries out if requested */
2186 account_node_boundary(nid
, start_pfn
, end_pfn
);
2190 * Return the number of pages a zone spans in a node, including holes
2191 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2193 unsigned long __init
zone_spanned_pages_in_node(int nid
,
2194 unsigned long zone_type
,
2195 unsigned long *ignored
)
2197 unsigned long node_start_pfn
, node_end_pfn
;
2198 unsigned long zone_start_pfn
, zone_end_pfn
;
2200 /* Get the start and end of the node and zone */
2201 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
2202 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
2203 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
2205 /* Check that this node has pages within the zone's required range */
2206 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
2209 /* Move the zone boundaries inside the node if necessary */
2210 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
2211 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
2213 /* Return the spanned pages */
2214 return zone_end_pfn
- zone_start_pfn
;
2218 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
2219 * then all holes in the requested range will be accounted for
2221 unsigned long __init
__absent_pages_in_range(int nid
,
2222 unsigned long range_start_pfn
,
2223 unsigned long range_end_pfn
)
2226 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
2227 unsigned long start_pfn
;
2229 /* Find the end_pfn of the first active range of pfns in the node */
2230 i
= first_active_region_index_in_nid(nid
);
2234 /* Account for ranges before physical memory on this node */
2235 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
2236 hole_pages
= early_node_map
[i
].start_pfn
- range_start_pfn
;
2238 prev_end_pfn
= early_node_map
[i
].start_pfn
;
2240 /* Find all holes for the zone within the node */
2241 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
2243 /* No need to continue if prev_end_pfn is outside the zone */
2244 if (prev_end_pfn
>= range_end_pfn
)
2247 /* Make sure the end of the zone is not within the hole */
2248 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
2249 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
2251 /* Update the hole size cound and move on */
2252 if (start_pfn
> range_start_pfn
) {
2253 BUG_ON(prev_end_pfn
> start_pfn
);
2254 hole_pages
+= start_pfn
- prev_end_pfn
;
2256 prev_end_pfn
= early_node_map
[i
].end_pfn
;
2259 /* Account for ranges past physical memory on this node */
2260 if (range_end_pfn
> prev_end_pfn
)
2261 hole_pages
= range_end_pfn
-
2262 max(range_start_pfn
, prev_end_pfn
);
2268 * absent_pages_in_range - Return number of page frames in holes within a range
2269 * @start_pfn: The start PFN to start searching for holes
2270 * @end_pfn: The end PFN to stop searching for holes
2272 * It returns the number of pages frames in memory holes within a range
2274 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
2275 unsigned long end_pfn
)
2277 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
2280 /* Return the number of page frames in holes in a zone on a node */
2281 unsigned long __init
zone_absent_pages_in_node(int nid
,
2282 unsigned long zone_type
,
2283 unsigned long *ignored
)
2285 unsigned long node_start_pfn
, node_end_pfn
;
2286 unsigned long zone_start_pfn
, zone_end_pfn
;
2288 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
2289 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
2291 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
2294 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
2297 /* Return the zone index a PFN is in */
2298 int memmap_zone_idx(struct page
*lmem_map
)
2301 unsigned long phys_addr
= virt_to_phys(lmem_map
);
2302 unsigned long pfn
= phys_addr
>> PAGE_SHIFT
;
2304 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2305 if (pfn
< arch_zone_highest_possible_pfn
[i
])
2311 static inline unsigned long zone_spanned_pages_in_node(int nid
,
2312 unsigned long zone_type
,
2313 unsigned long *zones_size
)
2315 return zones_size
[zone_type
];
2318 static inline unsigned long zone_absent_pages_in_node(int nid
,
2319 unsigned long zone_type
,
2320 unsigned long *zholes_size
)
2325 return zholes_size
[zone_type
];
2328 static inline int memmap_zone_idx(struct page
*lmem_map
)
2330 return MAX_NR_ZONES
;
2334 static void __init
calculate_node_totalpages(struct pglist_data
*pgdat
,
2335 unsigned long *zones_size
, unsigned long *zholes_size
)
2337 unsigned long realtotalpages
, totalpages
= 0;
2340 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2341 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
2343 pgdat
->node_spanned_pages
= totalpages
;
2345 realtotalpages
= totalpages
;
2346 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2348 zone_absent_pages_in_node(pgdat
->node_id
, i
,
2350 pgdat
->node_present_pages
= realtotalpages
;
2351 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
2356 * Set up the zone data structures:
2357 * - mark all pages reserved
2358 * - mark all memory queues empty
2359 * - clear the memory bitmaps
2361 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
2362 unsigned long *zones_size
, unsigned long *zholes_size
)
2365 int nid
= pgdat
->node_id
;
2366 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
2369 pgdat_resize_init(pgdat
);
2370 pgdat
->nr_zones
= 0;
2371 init_waitqueue_head(&pgdat
->kswapd_wait
);
2372 pgdat
->kswapd_max_order
= 0;
2374 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2375 struct zone
*zone
= pgdat
->node_zones
+ j
;
2376 unsigned long size
, realsize
, memmap_pages
;
2378 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
2379 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
2383 * Adjust realsize so that it accounts for how much memory
2384 * is used by this zone for memmap. This affects the watermark
2385 * and per-cpu initialisations
2387 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
2388 if (realsize
>= memmap_pages
) {
2389 realsize
-= memmap_pages
;
2391 " %s zone: %lu pages used for memmap\n",
2392 zone_names
[j
], memmap_pages
);
2395 " %s zone: %lu pages exceeds realsize %lu\n",
2396 zone_names
[j
], memmap_pages
, realsize
);
2398 /* Account for reserved DMA pages */
2399 if (j
== ZONE_DMA
&& realsize
> dma_reserve
) {
2400 realsize
-= dma_reserve
;
2401 printk(KERN_DEBUG
" DMA zone: %lu pages reserved\n",
2405 if (!is_highmem_idx(j
))
2406 nr_kernel_pages
+= realsize
;
2407 nr_all_pages
+= realsize
;
2409 zone
->spanned_pages
= size
;
2410 zone
->present_pages
= realsize
;
2412 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
2414 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
2416 zone
->name
= zone_names
[j
];
2417 spin_lock_init(&zone
->lock
);
2418 spin_lock_init(&zone
->lru_lock
);
2419 zone_seqlock_init(zone
);
2420 zone
->zone_pgdat
= pgdat
;
2421 zone
->free_pages
= 0;
2423 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
2425 zone_pcp_init(zone
);
2426 INIT_LIST_HEAD(&zone
->active_list
);
2427 INIT_LIST_HEAD(&zone
->inactive_list
);
2428 zone
->nr_scan_active
= 0;
2429 zone
->nr_scan_inactive
= 0;
2430 zone
->nr_active
= 0;
2431 zone
->nr_inactive
= 0;
2432 zap_zone_vm_stats(zone
);
2433 atomic_set(&zone
->reclaim_in_progress
, 0);
2437 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2438 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2440 zone_start_pfn
+= size
;
2444 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2446 /* Skip empty nodes */
2447 if (!pgdat
->node_spanned_pages
)
2450 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2451 /* ia64 gets its own node_mem_map, before this, without bootmem */
2452 if (!pgdat
->node_mem_map
) {
2453 unsigned long size
, start
, end
;
2457 * The zone's endpoints aren't required to be MAX_ORDER
2458 * aligned but the node_mem_map endpoints must be in order
2459 * for the buddy allocator to function correctly.
2461 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
2462 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
2463 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
2464 size
= (end
- start
) * sizeof(struct page
);
2465 map
= alloc_remap(pgdat
->node_id
, size
);
2467 map
= alloc_bootmem_node(pgdat
, size
);
2468 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
2470 #ifdef CONFIG_FLATMEM
2472 * With no DISCONTIG, the global mem_map is just set as node 0's
2474 if (pgdat
== NODE_DATA(0)) {
2475 mem_map
= NODE_DATA(0)->node_mem_map
;
2476 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2477 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
2478 mem_map
-= pgdat
->node_start_pfn
;
2479 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2482 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2485 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2486 unsigned long *zones_size
, unsigned long node_start_pfn
,
2487 unsigned long *zholes_size
)
2489 pgdat
->node_id
= nid
;
2490 pgdat
->node_start_pfn
= node_start_pfn
;
2491 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
2493 alloc_node_mem_map(pgdat
);
2495 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2498 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2500 * add_active_range - Register a range of PFNs backed by physical memory
2501 * @nid: The node ID the range resides on
2502 * @start_pfn: The start PFN of the available physical memory
2503 * @end_pfn: The end PFN of the available physical memory
2505 * These ranges are stored in an early_node_map[] and later used by
2506 * free_area_init_nodes() to calculate zone sizes and holes. If the
2507 * range spans a memory hole, it is up to the architecture to ensure
2508 * the memory is not freed by the bootmem allocator. If possible
2509 * the range being registered will be merged with existing ranges.
2511 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
2512 unsigned long end_pfn
)
2516 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
2517 "%d entries of %d used\n",
2518 nid
, start_pfn
, end_pfn
,
2519 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
2521 /* Merge with existing active regions if possible */
2522 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2523 if (early_node_map
[i
].nid
!= nid
)
2526 /* Skip if an existing region covers this new one */
2527 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
2528 end_pfn
<= early_node_map
[i
].end_pfn
)
2531 /* Merge forward if suitable */
2532 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
2533 end_pfn
> early_node_map
[i
].end_pfn
) {
2534 early_node_map
[i
].end_pfn
= end_pfn
;
2538 /* Merge backward if suitable */
2539 if (start_pfn
< early_node_map
[i
].end_pfn
&&
2540 end_pfn
>= early_node_map
[i
].start_pfn
) {
2541 early_node_map
[i
].start_pfn
= start_pfn
;
2546 /* Check that early_node_map is large enough */
2547 if (i
>= MAX_ACTIVE_REGIONS
) {
2548 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
2549 MAX_ACTIVE_REGIONS
);
2553 early_node_map
[i
].nid
= nid
;
2554 early_node_map
[i
].start_pfn
= start_pfn
;
2555 early_node_map
[i
].end_pfn
= end_pfn
;
2556 nr_nodemap_entries
= i
+ 1;
2560 * shrink_active_range - Shrink an existing registered range of PFNs
2561 * @nid: The node id the range is on that should be shrunk
2562 * @old_end_pfn: The old end PFN of the range
2563 * @new_end_pfn: The new PFN of the range
2565 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2566 * The map is kept at the end physical page range that has already been
2567 * registered with add_active_range(). This function allows an arch to shrink
2568 * an existing registered range.
2570 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
2571 unsigned long new_end_pfn
)
2575 /* Find the old active region end and shrink */
2576 for_each_active_range_index_in_nid(i
, nid
)
2577 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
2578 early_node_map
[i
].end_pfn
= new_end_pfn
;
2584 * remove_all_active_ranges - Remove all currently registered regions
2585 * During discovery, it may be found that a table like SRAT is invalid
2586 * and an alternative discovery method must be used. This function removes
2587 * all currently registered regions.
2589 void __init
remove_all_active_ranges()
2591 memset(early_node_map
, 0, sizeof(early_node_map
));
2592 nr_nodemap_entries
= 0;
2593 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2594 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
2595 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
2596 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
2599 /* Compare two active node_active_regions */
2600 static int __init
cmp_node_active_region(const void *a
, const void *b
)
2602 struct node_active_region
*arange
= (struct node_active_region
*)a
;
2603 struct node_active_region
*brange
= (struct node_active_region
*)b
;
2605 /* Done this way to avoid overflows */
2606 if (arange
->start_pfn
> brange
->start_pfn
)
2608 if (arange
->start_pfn
< brange
->start_pfn
)
2614 /* sort the node_map by start_pfn */
2615 static void __init
sort_node_map(void)
2617 sort(early_node_map
, (size_t)nr_nodemap_entries
,
2618 sizeof(struct node_active_region
),
2619 cmp_node_active_region
, NULL
);
2622 /* Find the lowest pfn for a node. This depends on a sorted early_node_map */
2623 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
2627 /* Assuming a sorted map, the first range found has the starting pfn */
2628 for_each_active_range_index_in_nid(i
, nid
)
2629 return early_node_map
[i
].start_pfn
;
2631 printk(KERN_WARNING
"Could not find start_pfn for node %lu\n", nid
);
2636 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2638 * It returns the minimum PFN based on information provided via
2639 * add_active_range()
2641 unsigned long __init
find_min_pfn_with_active_regions(void)
2643 return find_min_pfn_for_node(MAX_NUMNODES
);
2647 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2649 * It returns the maximum PFN based on information provided via
2650 * add_active_range()
2652 unsigned long __init
find_max_pfn_with_active_regions(void)
2655 unsigned long max_pfn
= 0;
2657 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2658 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
2664 * free_area_init_nodes - Initialise all pg_data_t and zone data
2665 * @arch_max_dma_pfn: The maximum PFN usable for ZONE_DMA
2666 * @arch_max_dma32_pfn: The maximum PFN usable for ZONE_DMA32
2667 * @arch_max_low_pfn: The maximum PFN usable for ZONE_NORMAL
2668 * @arch_max_high_pfn: The maximum PFN usable for ZONE_HIGHMEM
2670 * This will call free_area_init_node() for each active node in the system.
2671 * Using the page ranges provided by add_active_range(), the size of each
2672 * zone in each node and their holes is calculated. If the maximum PFN
2673 * between two adjacent zones match, it is assumed that the zone is empty.
2674 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2675 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2676 * starts where the previous one ended. For example, ZONE_DMA32 starts
2677 * at arch_max_dma_pfn.
2679 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
2684 /* Record where the zone boundaries are */
2685 memset(arch_zone_lowest_possible_pfn
, 0,
2686 sizeof(arch_zone_lowest_possible_pfn
));
2687 memset(arch_zone_highest_possible_pfn
, 0,
2688 sizeof(arch_zone_highest_possible_pfn
));
2689 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
2690 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
2691 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
2692 arch_zone_lowest_possible_pfn
[i
] =
2693 arch_zone_highest_possible_pfn
[i
-1];
2694 arch_zone_highest_possible_pfn
[i
] =
2695 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
2698 /* Regions in the early_node_map can be in any order */
2701 /* Print out the zone ranges */
2702 printk("Zone PFN ranges:\n");
2703 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2704 printk(" %-8s %8lu -> %8lu\n",
2706 arch_zone_lowest_possible_pfn
[i
],
2707 arch_zone_highest_possible_pfn
[i
]);
2709 /* Print out the early_node_map[] */
2710 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
2711 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2712 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
2713 early_node_map
[i
].start_pfn
,
2714 early_node_map
[i
].end_pfn
);
2716 /* Initialise every node */
2717 for_each_online_node(nid
) {
2718 pg_data_t
*pgdat
= NODE_DATA(nid
);
2719 free_area_init_node(nid
, pgdat
, NULL
,
2720 find_min_pfn_for_node(nid
), NULL
);
2723 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2726 * set_dma_reserve - Account the specified number of pages reserved in ZONE_DMA
2727 * @new_dma_reserve - The number of pages to mark reserved
2729 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2730 * In the DMA zone, a significant percentage may be consumed by kernel image
2731 * and other unfreeable allocations which can skew the watermarks badly. This
2732 * function may optionally be used to account for unfreeable pages in
2733 * ZONE_DMA. The effect will be lower watermarks and smaller per-cpu batchsize
2735 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
2737 dma_reserve
= new_dma_reserve
;
2740 #ifndef CONFIG_NEED_MULTIPLE_NODES
2741 static bootmem_data_t contig_bootmem_data
;
2742 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2744 EXPORT_SYMBOL(contig_page_data
);
2747 void __init
free_area_init(unsigned long *zones_size
)
2749 free_area_init_node(0, NODE_DATA(0), zones_size
,
2750 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2753 #ifdef CONFIG_HOTPLUG_CPU
2754 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2755 unsigned long action
, void *hcpu
)
2757 int cpu
= (unsigned long)hcpu
;
2759 if (action
== CPU_DEAD
) {
2760 local_irq_disable();
2762 vm_events_fold_cpu(cpu
);
2764 refresh_cpu_vm_stats(cpu
);
2768 #endif /* CONFIG_HOTPLUG_CPU */
2770 void __init
page_alloc_init(void)
2772 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2776 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2777 * or min_free_kbytes changes.
2779 static void calculate_totalreserve_pages(void)
2781 struct pglist_data
*pgdat
;
2782 unsigned long reserve_pages
= 0;
2783 enum zone_type i
, j
;
2785 for_each_online_pgdat(pgdat
) {
2786 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2787 struct zone
*zone
= pgdat
->node_zones
+ i
;
2788 unsigned long max
= 0;
2790 /* Find valid and maximum lowmem_reserve in the zone */
2791 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
2792 if (zone
->lowmem_reserve
[j
] > max
)
2793 max
= zone
->lowmem_reserve
[j
];
2796 /* we treat pages_high as reserved pages. */
2797 max
+= zone
->pages_high
;
2799 if (max
> zone
->present_pages
)
2800 max
= zone
->present_pages
;
2801 reserve_pages
+= max
;
2804 totalreserve_pages
= reserve_pages
;
2808 * setup_per_zone_lowmem_reserve - called whenever
2809 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2810 * has a correct pages reserved value, so an adequate number of
2811 * pages are left in the zone after a successful __alloc_pages().
2813 static void setup_per_zone_lowmem_reserve(void)
2815 struct pglist_data
*pgdat
;
2816 enum zone_type j
, idx
;
2818 for_each_online_pgdat(pgdat
) {
2819 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2820 struct zone
*zone
= pgdat
->node_zones
+ j
;
2821 unsigned long present_pages
= zone
->present_pages
;
2823 zone
->lowmem_reserve
[j
] = 0;
2827 struct zone
*lower_zone
;
2831 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2832 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2834 lower_zone
= pgdat
->node_zones
+ idx
;
2835 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2836 sysctl_lowmem_reserve_ratio
[idx
];
2837 present_pages
+= lower_zone
->present_pages
;
2842 /* update totalreserve_pages */
2843 calculate_totalreserve_pages();
2847 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2848 * that the pages_{min,low,high} values for each zone are set correctly
2849 * with respect to min_free_kbytes.
2851 void setup_per_zone_pages_min(void)
2853 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2854 unsigned long lowmem_pages
= 0;
2856 unsigned long flags
;
2858 /* Calculate total number of !ZONE_HIGHMEM pages */
2859 for_each_zone(zone
) {
2860 if (!is_highmem(zone
))
2861 lowmem_pages
+= zone
->present_pages
;
2864 for_each_zone(zone
) {
2867 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2868 tmp
= (u64
)pages_min
* zone
->present_pages
;
2869 do_div(tmp
, lowmem_pages
);
2870 if (is_highmem(zone
)) {
2872 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2873 * need highmem pages, so cap pages_min to a small
2876 * The (pages_high-pages_low) and (pages_low-pages_min)
2877 * deltas controls asynch page reclaim, and so should
2878 * not be capped for highmem.
2882 min_pages
= zone
->present_pages
/ 1024;
2883 if (min_pages
< SWAP_CLUSTER_MAX
)
2884 min_pages
= SWAP_CLUSTER_MAX
;
2885 if (min_pages
> 128)
2887 zone
->pages_min
= min_pages
;
2890 * If it's a lowmem zone, reserve a number of pages
2891 * proportionate to the zone's size.
2893 zone
->pages_min
= tmp
;
2896 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
2897 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
2898 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2901 /* update totalreserve_pages */
2902 calculate_totalreserve_pages();
2906 * Initialise min_free_kbytes.
2908 * For small machines we want it small (128k min). For large machines
2909 * we want it large (64MB max). But it is not linear, because network
2910 * bandwidth does not increase linearly with machine size. We use
2912 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2913 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2929 static int __init
init_per_zone_pages_min(void)
2931 unsigned long lowmem_kbytes
;
2933 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2935 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2936 if (min_free_kbytes
< 128)
2937 min_free_kbytes
= 128;
2938 if (min_free_kbytes
> 65536)
2939 min_free_kbytes
= 65536;
2940 setup_per_zone_pages_min();
2941 setup_per_zone_lowmem_reserve();
2944 module_init(init_per_zone_pages_min
)
2947 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2948 * that we can call two helper functions whenever min_free_kbytes
2951 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2952 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2954 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2955 setup_per_zone_pages_min();
2960 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
2961 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2966 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2971 zone
->min_unmapped_pages
= (zone
->present_pages
*
2972 sysctl_min_unmapped_ratio
) / 100;
2976 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
2977 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2982 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2987 zone
->min_slab_pages
= (zone
->present_pages
*
2988 sysctl_min_slab_ratio
) / 100;
2994 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2995 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2996 * whenever sysctl_lowmem_reserve_ratio changes.
2998 * The reserve ratio obviously has absolutely no relation with the
2999 * pages_min watermarks. The lowmem reserve ratio can only make sense
3000 * if in function of the boot time zone sizes.
3002 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
3003 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3005 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3006 setup_per_zone_lowmem_reserve();
3011 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3012 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3013 * can have before it gets flushed back to buddy allocator.
3016 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
3017 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
3023 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
3024 if (!write
|| (ret
== -EINVAL
))
3026 for_each_zone(zone
) {
3027 for_each_online_cpu(cpu
) {
3029 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
3030 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
3036 int hashdist
= HASHDIST_DEFAULT
;
3039 static int __init
set_hashdist(char *str
)
3043 hashdist
= simple_strtoul(str
, &str
, 0);
3046 __setup("hashdist=", set_hashdist
);
3050 * allocate a large system hash table from bootmem
3051 * - it is assumed that the hash table must contain an exact power-of-2
3052 * quantity of entries
3053 * - limit is the number of hash buckets, not the total allocation size
3055 void *__init
alloc_large_system_hash(const char *tablename
,
3056 unsigned long bucketsize
,
3057 unsigned long numentries
,
3060 unsigned int *_hash_shift
,
3061 unsigned int *_hash_mask
,
3062 unsigned long limit
)
3064 unsigned long long max
= limit
;
3065 unsigned long log2qty
, size
;
3068 /* allow the kernel cmdline to have a say */
3070 /* round applicable memory size up to nearest megabyte */
3071 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
3072 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
3073 numentries
>>= 20 - PAGE_SHIFT
;
3074 numentries
<<= 20 - PAGE_SHIFT
;
3076 /* limit to 1 bucket per 2^scale bytes of low memory */
3077 if (scale
> PAGE_SHIFT
)
3078 numentries
>>= (scale
- PAGE_SHIFT
);
3080 numentries
<<= (PAGE_SHIFT
- scale
);
3082 numentries
= roundup_pow_of_two(numentries
);
3084 /* limit allocation size to 1/16 total memory by default */
3086 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
3087 do_div(max
, bucketsize
);
3090 if (numentries
> max
)
3093 log2qty
= long_log2(numentries
);
3096 size
= bucketsize
<< log2qty
;
3097 if (flags
& HASH_EARLY
)
3098 table
= alloc_bootmem(size
);
3100 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
3102 unsigned long order
;
3103 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
3105 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
3107 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
3110 panic("Failed to allocate %s hash table\n", tablename
);
3112 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3115 long_log2(size
) - PAGE_SHIFT
,
3119 *_hash_shift
= log2qty
;
3121 *_hash_mask
= (1 << log2qty
) - 1;
3126 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
3127 struct page
*pfn_to_page(unsigned long pfn
)
3129 return __pfn_to_page(pfn
);
3131 unsigned long page_to_pfn(struct page
*page
)
3133 return __page_to_pfn(page
);
3135 EXPORT_SYMBOL(pfn_to_page
);
3136 EXPORT_SYMBOL(page_to_pfn
);
3137 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */