2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/config.h>
18 #include <linux/stddef.h>
20 #include <linux/swap.h>
21 #include <linux/interrupt.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/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/nodemask.h>
36 #include <asm/tlbflush.h>
38 nodemask_t node_online_map
= NODE_MASK_NONE
;
39 nodemask_t node_possible_map
= NODE_MASK_ALL
;
40 struct pglist_data
*pgdat_list
;
41 unsigned long totalram_pages
;
42 unsigned long totalhigh_pages
;
45 int sysctl_lower_zone_protection
= 0;
47 EXPORT_SYMBOL(totalram_pages
);
48 EXPORT_SYMBOL(nr_swap_pages
);
51 * Used by page_zone() to look up the address of the struct zone whose
52 * id is encoded in the upper bits of page->flags
54 struct zone
*zone_table
[1 << (ZONES_SHIFT
+ NODES_SHIFT
)];
55 EXPORT_SYMBOL(zone_table
);
57 static char *zone_names
[MAX_NR_ZONES
] = { "DMA", "Normal", "HighMem" };
58 int min_free_kbytes
= 1024;
60 unsigned long __initdata nr_kernel_pages
;
61 unsigned long __initdata nr_all_pages
;
64 * Temporary debugging check for pages not lying within a given zone.
66 static int bad_range(struct zone
*zone
, struct page
*page
)
68 if (page_to_pfn(page
) >= zone
->zone_start_pfn
+ zone
->spanned_pages
)
70 if (page_to_pfn(page
) < zone
->zone_start_pfn
)
72 if (zone
!= page_zone(page
))
77 static void bad_page(const char *function
, struct page
*page
)
79 printk(KERN_EMERG
"Bad page state at %s (in process '%s', page %p)\n",
80 function
, current
->comm
, page
);
81 printk(KERN_EMERG
"flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
82 (int)(2*sizeof(page_flags_t
)), (unsigned long)page
->flags
,
83 page
->mapping
, page_mapcount(page
), page_count(page
));
84 printk(KERN_EMERG
"Backtrace:\n");
86 printk(KERN_EMERG
"Trying to fix it up, but a reboot is needed\n");
87 page
->flags
&= ~(1 << PG_private
|
94 set_page_count(page
, 0);
95 reset_page_mapcount(page
);
97 tainted
|= TAINT_BAD_PAGE
;
100 #ifndef CONFIG_HUGETLB_PAGE
101 #define prep_compound_page(page, order) do { } while (0)
102 #define destroy_compound_page(page, order) do { } while (0)
105 * Higher-order pages are called "compound pages". They are structured thusly:
107 * The first PAGE_SIZE page is called the "head page".
109 * The remaining PAGE_SIZE pages are called "tail pages".
111 * All pages have PG_compound set. All pages have their ->private pointing at
112 * the head page (even the head page has this).
114 * The first tail page's ->mapping, if non-zero, holds the address of the
115 * compound page's put_page() function.
117 * The order of the allocation is stored in the first tail page's ->index
118 * This is only for debug at present. This usage means that zero-order pages
119 * may not be compound.
121 static void prep_compound_page(struct page
*page
, unsigned long order
)
124 int nr_pages
= 1 << order
;
126 page
[1].mapping
= NULL
;
127 page
[1].index
= order
;
128 for (i
= 0; i
< nr_pages
; i
++) {
129 struct page
*p
= page
+ i
;
132 p
->private = (unsigned long)page
;
136 static void destroy_compound_page(struct page
*page
, unsigned long order
)
139 int nr_pages
= 1 << order
;
141 if (!PageCompound(page
))
144 if (page
[1].index
!= order
)
145 bad_page(__FUNCTION__
, page
);
147 for (i
= 0; i
< nr_pages
; i
++) {
148 struct page
*p
= page
+ i
;
150 if (!PageCompound(p
))
151 bad_page(__FUNCTION__
, page
);
152 if (p
->private != (unsigned long)page
)
153 bad_page(__FUNCTION__
, page
);
154 ClearPageCompound(p
);
157 #endif /* CONFIG_HUGETLB_PAGE */
160 * Freeing function for a buddy system allocator.
162 * The concept of a buddy system is to maintain direct-mapped table
163 * (containing bit values) for memory blocks of various "orders".
164 * The bottom level table contains the map for the smallest allocatable
165 * units of memory (here, pages), and each level above it describes
166 * pairs of units from the levels below, hence, "buddies".
167 * At a high level, all that happens here is marking the table entry
168 * at the bottom level available, and propagating the changes upward
169 * as necessary, plus some accounting needed to play nicely with other
170 * parts of the VM system.
171 * At each level, we keep one bit for each pair of blocks, which
172 * is set to 1 iff only one of the pair is allocated. So when we
173 * are allocating or freeing one, we can derive the state of the
174 * other. That is, if we allocate a small block, and both were
175 * free, the remainder of the region must be split into blocks.
176 * If a block is freed, and its buddy is also free, then this
177 * triggers coalescing into a block of larger size.
182 static inline void __free_pages_bulk (struct page
*page
, struct page
*base
,
183 struct zone
*zone
, struct free_area
*area
, unsigned int order
)
185 unsigned long page_idx
, index
, mask
;
188 destroy_compound_page(page
, order
);
189 mask
= (~0UL) << order
;
190 page_idx
= page
- base
;
191 if (page_idx
& ~mask
)
193 index
= page_idx
>> (1 + order
);
195 zone
->free_pages
+= 1 << order
;
196 while (order
< MAX_ORDER
-1) {
197 struct page
*buddy1
, *buddy2
;
199 BUG_ON(area
>= zone
->free_area
+ MAX_ORDER
);
200 if (!__test_and_change_bit(index
, area
->map
))
202 * the buddy page is still allocated.
206 /* Move the buddy up one level. */
207 buddy1
= base
+ (page_idx
^ (1 << order
));
208 buddy2
= base
+ page_idx
;
209 BUG_ON(bad_range(zone
, buddy1
));
210 BUG_ON(bad_range(zone
, buddy2
));
211 list_del(&buddy1
->lru
);
218 list_add(&(base
+ page_idx
)->lru
, &area
->free_list
);
221 static inline void free_pages_check(const char *function
, struct page
*page
)
223 if ( page_mapped(page
) ||
224 page
->mapping
!= NULL
||
225 page_count(page
) != 0 ||
234 1 << PG_writeback
)))
235 bad_page(function
, page
);
237 ClearPageDirty(page
);
241 * Frees a list of pages.
242 * Assumes all pages on list are in same zone, and of same order.
243 * count is the number of pages to free, or 0 for all on the list.
245 * If the zone was previously in an "all pages pinned" state then look to
246 * see if this freeing clears that state.
248 * And clear the zone's pages_scanned counter, to hold off the "all pages are
249 * pinned" detection logic.
252 free_pages_bulk(struct zone
*zone
, int count
,
253 struct list_head
*list
, unsigned int order
)
256 struct free_area
*area
;
257 struct page
*base
, *page
= NULL
;
260 base
= zone
->zone_mem_map
;
261 area
= zone
->free_area
+ order
;
262 spin_lock_irqsave(&zone
->lock
, flags
);
263 zone
->all_unreclaimable
= 0;
264 zone
->pages_scanned
= 0;
265 while (!list_empty(list
) && count
--) {
266 page
= list_entry(list
->prev
, struct page
, lru
);
267 /* have to delete it as __free_pages_bulk list manipulates */
268 list_del(&page
->lru
);
269 __free_pages_bulk(page
, base
, zone
, area
, order
);
272 spin_unlock_irqrestore(&zone
->lock
, flags
);
276 void __free_pages_ok(struct page
*page
, unsigned int order
)
281 arch_free_page(page
, order
);
283 mod_page_state(pgfree
, 1 << order
);
284 for (i
= 0 ; i
< (1 << order
) ; ++i
)
285 free_pages_check(__FUNCTION__
, page
+ i
);
286 list_add(&page
->lru
, &list
);
287 kernel_map_pages(page
, 1<<order
, 0);
288 free_pages_bulk(page_zone(page
), 1, &list
, order
);
291 #define MARK_USED(index, order, area) \
292 __change_bit((index) >> (1+(order)), (area)->map)
295 * The order of subdivision here is critical for the IO subsystem.
296 * Please do not alter this order without good reasons and regression
297 * testing. Specifically, as large blocks of memory are subdivided,
298 * the order in which smaller blocks are delivered depends on the order
299 * they're subdivided in this function. This is the primary factor
300 * influencing the order in which pages are delivered to the IO
301 * subsystem according to empirical testing, and this is also justified
302 * by considering the behavior of a buddy system containing a single
303 * large block of memory acted on by a series of small allocations.
304 * This behavior is a critical factor in sglist merging's success.
308 static inline struct page
*
309 expand(struct zone
*zone
, struct page
*page
,
310 unsigned long index
, int low
, int high
, struct free_area
*area
)
312 unsigned long size
= 1 << high
;
318 BUG_ON(bad_range(zone
, &page
[size
]));
319 list_add(&page
[size
].lru
, &area
->free_list
);
320 MARK_USED(index
+ size
, high
, area
);
325 static inline void set_page_refs(struct page
*page
, int order
)
328 set_page_count(page
, 1);
333 * We need to reference all the pages for this order, otherwise if
334 * anyone accesses one of the pages with (get/put) it will be freed.
336 for (i
= 0; i
< (1 << order
); i
++)
337 set_page_count(page
+i
, 1);
338 #endif /* CONFIG_MMU */
342 * This page is about to be returned from the page allocator
344 static void prep_new_page(struct page
*page
, int order
)
346 if (page
->mapping
|| page_mapped(page
) ||
355 1 << PG_writeback
)))
356 bad_page(__FUNCTION__
, page
);
358 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
359 1 << PG_referenced
| 1 << PG_arch_1
|
360 1 << PG_checked
| 1 << PG_mappedtodisk
);
362 set_page_refs(page
, order
);
366 * Do the hard work of removing an element from the buddy allocator.
367 * Call me with the zone->lock already held.
369 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
371 struct free_area
* area
;
372 unsigned int current_order
;
376 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
377 area
= zone
->free_area
+ current_order
;
378 if (list_empty(&area
->free_list
))
381 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
382 list_del(&page
->lru
);
383 index
= page
- zone
->zone_mem_map
;
384 if (current_order
!= MAX_ORDER
-1)
385 MARK_USED(index
, current_order
, area
);
386 zone
->free_pages
-= 1UL << order
;
387 return expand(zone
, page
, index
, order
, current_order
, area
);
394 * Obtain a specified number of elements from the buddy allocator, all under
395 * a single hold of the lock, for efficiency. Add them to the supplied list.
396 * Returns the number of new pages which were placed at *list.
398 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
399 unsigned long count
, struct list_head
*list
)
406 spin_lock_irqsave(&zone
->lock
, flags
);
407 for (i
= 0; i
< count
; ++i
) {
408 page
= __rmqueue(zone
, order
);
412 list_add_tail(&page
->lru
, list
);
414 spin_unlock_irqrestore(&zone
->lock
, flags
);
418 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
419 static void __drain_pages(unsigned int cpu
)
424 for_each_zone(zone
) {
425 struct per_cpu_pageset
*pset
;
427 pset
= &zone
->pageset
[cpu
];
428 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
429 struct per_cpu_pages
*pcp
;
432 pcp
->count
-= free_pages_bulk(zone
, pcp
->count
,
437 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
440 int is_head_of_free_region(struct page
*page
)
442 struct zone
*zone
= page_zone(page
);
445 struct list_head
*curr
;
448 * Should not matter as we need quiescent system for
449 * suspend anyway, but...
451 spin_lock_irqsave(&zone
->lock
, flags
);
452 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
453 list_for_each(curr
, &zone
->free_area
[order
].free_list
)
454 if (page
== list_entry(curr
, struct page
, lru
)) {
455 spin_unlock_irqrestore(&zone
->lock
, flags
);
458 spin_unlock_irqrestore(&zone
->lock
, flags
);
463 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
465 void drain_local_pages(void)
469 local_irq_save(flags
);
470 __drain_pages(smp_processor_id());
471 local_irq_restore(flags
);
473 #endif /* CONFIG_PM */
475 static void zone_statistics(struct zonelist
*zonelist
, struct zone
*z
)
480 pg_data_t
*pg
= z
->zone_pgdat
;
481 pg_data_t
*orig
= zonelist
->zones
[0]->zone_pgdat
;
482 struct per_cpu_pageset
*p
;
484 local_irq_save(flags
);
485 cpu
= smp_processor_id();
486 p
= &z
->pageset
[cpu
];
488 z
->pageset
[cpu
].numa_hit
++;
491 zonelist
->zones
[0]->pageset
[cpu
].numa_foreign
++;
493 if (pg
== NODE_DATA(numa_node_id()))
497 local_irq_restore(flags
);
502 * Free a 0-order page
504 static void FASTCALL(free_hot_cold_page(struct page
*page
, int cold
));
505 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
507 struct zone
*zone
= page_zone(page
);
508 struct per_cpu_pages
*pcp
;
511 arch_free_page(page
, 0);
513 kernel_map_pages(page
, 1, 0);
514 inc_page_state(pgfree
);
516 page
->mapping
= NULL
;
517 free_pages_check(__FUNCTION__
, page
);
518 pcp
= &zone
->pageset
[get_cpu()].pcp
[cold
];
519 local_irq_save(flags
);
520 if (pcp
->count
>= pcp
->high
)
521 pcp
->count
-= free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
522 list_add(&page
->lru
, &pcp
->list
);
524 local_irq_restore(flags
);
528 void fastcall
free_hot_page(struct page
*page
)
530 free_hot_cold_page(page
, 0);
533 void fastcall
free_cold_page(struct page
*page
)
535 free_hot_cold_page(page
, 1);
539 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
540 * we cheat by calling it from here, in the order > 0 path. Saves a branch
545 buffered_rmqueue(struct zone
*zone
, int order
, int gfp_flags
)
548 struct page
*page
= NULL
;
549 int cold
= !!(gfp_flags
& __GFP_COLD
);
552 struct per_cpu_pages
*pcp
;
554 pcp
= &zone
->pageset
[get_cpu()].pcp
[cold
];
555 local_irq_save(flags
);
556 if (pcp
->count
<= pcp
->low
)
557 pcp
->count
+= rmqueue_bulk(zone
, 0,
558 pcp
->batch
, &pcp
->list
);
560 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
561 list_del(&page
->lru
);
564 local_irq_restore(flags
);
569 spin_lock_irqsave(&zone
->lock
, flags
);
570 page
= __rmqueue(zone
, order
);
571 spin_unlock_irqrestore(&zone
->lock
, flags
);
575 BUG_ON(bad_range(zone
, page
));
576 mod_page_state_zone(zone
, pgalloc
, 1 << order
);
577 prep_new_page(page
, order
);
578 if (order
&& (gfp_flags
& __GFP_COMP
))
579 prep_compound_page(page
, order
);
585 * This is the 'heart' of the zoned buddy allocator.
587 * Herein lies the mysterious "incremental min". That's the
589 * local_low = z->pages_low;
592 * thing. The intent here is to provide additional protection to low zones for
593 * allocation requests which _could_ use higher zones. So a GFP_HIGHMEM
594 * request is not allowed to dip as deeply into the normal zone as a GFP_KERNEL
595 * request. This preserves additional space in those lower zones for requests
596 * which really do need memory from those zones. It means that on a decent
597 * sized machine, GFP_HIGHMEM and GFP_KERNEL requests basically leave the DMA
600 struct page
* fastcall
601 __alloc_pages(unsigned int gfp_mask
, unsigned int order
,
602 struct zonelist
*zonelist
)
604 const int wait
= gfp_mask
& __GFP_WAIT
;
606 struct zone
**zones
, *z
;
608 struct reclaim_state reclaim_state
;
609 struct task_struct
*p
= current
;
615 might_sleep_if(wait
);
618 * The caller may dip into page reserves a bit more if the caller
619 * cannot run direct reclaim, or is the caller has realtime scheduling
622 can_try_harder
= (unlikely(rt_task(p
)) && !in_interrupt()) || !wait
;
624 zones
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
626 if (unlikely(zones
[0] == NULL
)) {
627 /* Should this ever happen?? */
631 alloc_type
= zone_idx(zones
[0]);
633 /* Go through the zonelist once, looking for a zone with enough free */
634 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++) {
635 min
= z
->pages_low
+ (1<<order
) + z
->protection
[alloc_type
];
637 if (z
->free_pages
< min
)
640 page
= buffered_rmqueue(z
, order
, gfp_mask
);
645 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++)
649 * Go through the zonelist again. Let __GFP_HIGH and allocations
650 * coming from realtime tasks to go deeper into reserves
652 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++) {
654 if (gfp_mask
& __GFP_HIGH
)
658 min
+= (1<<order
) + z
->protection
[alloc_type
];
660 if (z
->free_pages
< min
)
663 page
= buffered_rmqueue(z
, order
, gfp_mask
);
668 /* This allocation should allow future memory freeing. */
669 if ((p
->flags
& (PF_MEMALLOC
| PF_MEMDIE
)) && !in_interrupt()) {
670 /* go through the zonelist yet again, ignoring mins */
671 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++) {
672 page
= buffered_rmqueue(z
, order
, gfp_mask
);
679 /* Atomic allocations - we can't balance anything */
684 /* We now go into synchronous reclaim */
685 p
->flags
|= PF_MEMALLOC
;
686 reclaim_state
.reclaimed_slab
= 0;
687 p
->reclaim_state
= &reclaim_state
;
689 try_to_free_pages(zones
, gfp_mask
, order
);
691 p
->reclaim_state
= NULL
;
692 p
->flags
&= ~PF_MEMALLOC
;
694 /* go through the zonelist yet one more time */
695 for (i
= 0; (z
= zones
[i
]) != NULL
; i
++) {
697 if (gfp_mask
& __GFP_HIGH
)
701 min
+= (1<<order
) + z
->protection
[alloc_type
];
703 if (z
->free_pages
< min
)
706 page
= buffered_rmqueue(z
, order
, gfp_mask
);
712 * Don't let big-order allocations loop unless the caller explicitly
713 * requests that. Wait for some write requests to complete then retry.
715 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
716 * <= 3, but that may not be true in other implementations.
719 if (!(gfp_mask
& __GFP_NORETRY
)) {
720 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
722 if (gfp_mask
& __GFP_NOFAIL
)
726 blk_congestion_wait(WRITE
, HZ
/50);
731 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
732 printk(KERN_WARNING
"%s: page allocation failure."
733 " order:%d, mode:0x%x\n",
734 p
->comm
, order
, gfp_mask
);
739 zone_statistics(zonelist
, z
);
740 kernel_map_pages(page
, 1 << order
, 1);
744 EXPORT_SYMBOL(__alloc_pages
);
747 * Common helper functions.
749 fastcall
unsigned long __get_free_pages(unsigned int gfp_mask
, unsigned int order
)
752 page
= alloc_pages(gfp_mask
, order
);
755 return (unsigned long) page_address(page
);
758 EXPORT_SYMBOL(__get_free_pages
);
760 fastcall
unsigned long get_zeroed_page(unsigned int gfp_mask
)
765 * get_zeroed_page() returns a 32-bit address, which cannot represent
768 BUG_ON(gfp_mask
& __GFP_HIGHMEM
);
770 page
= alloc_pages(gfp_mask
, 0);
772 void *address
= page_address(page
);
774 return (unsigned long) address
;
779 EXPORT_SYMBOL(get_zeroed_page
);
781 void __pagevec_free(struct pagevec
*pvec
)
783 int i
= pagevec_count(pvec
);
786 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
789 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
791 if (!PageReserved(page
) && put_page_testzero(page
)) {
795 __free_pages_ok(page
, order
);
799 EXPORT_SYMBOL(__free_pages
);
801 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
804 BUG_ON(!virt_addr_valid((void *)addr
));
805 __free_pages(virt_to_page((void *)addr
), order
);
809 EXPORT_SYMBOL(free_pages
);
812 * Total amount of free (allocatable) RAM:
814 unsigned int nr_free_pages(void)
816 unsigned int sum
= 0;
820 sum
+= zone
->free_pages
;
825 EXPORT_SYMBOL(nr_free_pages
);
828 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
830 unsigned int i
, sum
= 0;
832 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
833 sum
+= pgdat
->node_zones
[i
].free_pages
;
839 static unsigned int nr_free_zone_pages(int offset
)
842 unsigned int sum
= 0;
844 for_each_pgdat(pgdat
) {
845 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
846 struct zone
**zonep
= zonelist
->zones
;
849 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
850 unsigned long size
= zone
->present_pages
;
851 unsigned long high
= zone
->pages_high
;
861 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
863 unsigned int nr_free_buffer_pages(void)
865 return nr_free_zone_pages(GFP_USER
& GFP_ZONEMASK
);
869 * Amount of free RAM allocatable within all zones
871 unsigned int nr_free_pagecache_pages(void)
873 return nr_free_zone_pages(GFP_HIGHUSER
& GFP_ZONEMASK
);
876 #ifdef CONFIG_HIGHMEM
877 unsigned int nr_free_highpages (void)
880 unsigned int pages
= 0;
882 for_each_pgdat(pgdat
)
883 pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
890 static void show_node(struct zone
*zone
)
892 printk("Node %d ", zone
->zone_pgdat
->node_id
);
895 #define show_node(zone) do { } while (0)
899 * Accumulate the page_state information across all CPUs.
900 * The result is unavoidably approximate - it can change
901 * during and after execution of this function.
903 DEFINE_PER_CPU(struct page_state
, page_states
) = {0};
904 EXPORT_PER_CPU_SYMBOL(page_states
);
906 atomic_t nr_pagecache
= ATOMIC_INIT(0);
907 EXPORT_SYMBOL(nr_pagecache
);
909 DEFINE_PER_CPU(long, nr_pagecache_local
) = 0;
912 void __get_page_state(struct page_state
*ret
, int nr
)
916 memset(ret
, 0, sizeof(*ret
));
917 while (cpu
< NR_CPUS
) {
918 unsigned long *in
, *out
, off
;
920 if (!cpu_possible(cpu
)) {
925 in
= (unsigned long *)&per_cpu(page_states
, cpu
);
927 if (cpu
< NR_CPUS
&& cpu_possible(cpu
))
928 prefetch(&per_cpu(page_states
, cpu
));
929 out
= (unsigned long *)ret
;
930 for (off
= 0; off
< nr
; off
++)
935 void get_page_state(struct page_state
*ret
)
939 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
940 nr
/= sizeof(unsigned long);
942 __get_page_state(ret
, nr
+ 1);
945 void get_full_page_state(struct page_state
*ret
)
947 __get_page_state(ret
, sizeof(*ret
) / sizeof(unsigned long));
950 unsigned long __read_page_state(unsigned offset
)
952 unsigned long ret
= 0;
955 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
958 if (!cpu_possible(cpu
))
961 in
= (unsigned long)&per_cpu(page_states
, cpu
) + offset
;
962 ret
+= *((unsigned long *)in
);
967 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
968 unsigned long *free
, struct pglist_data
*pgdat
)
970 struct zone
*zones
= pgdat
->node_zones
;
976 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
977 *active
+= zones
[i
].nr_active
;
978 *inactive
+= zones
[i
].nr_inactive
;
979 *free
+= zones
[i
].free_pages
;
983 void get_zone_counts(unsigned long *active
,
984 unsigned long *inactive
, unsigned long *free
)
986 struct pglist_data
*pgdat
;
991 for_each_pgdat(pgdat
) {
992 unsigned long l
, m
, n
;
993 __get_zone_counts(&l
, &m
, &n
, pgdat
);
1000 void si_meminfo(struct sysinfo
*val
)
1002 val
->totalram
= totalram_pages
;
1004 val
->freeram
= nr_free_pages();
1005 val
->bufferram
= nr_blockdev_pages();
1006 #ifdef CONFIG_HIGHMEM
1007 val
->totalhigh
= totalhigh_pages
;
1008 val
->freehigh
= nr_free_highpages();
1013 val
->mem_unit
= PAGE_SIZE
;
1016 EXPORT_SYMBOL(si_meminfo
);
1019 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1021 pg_data_t
*pgdat
= NODE_DATA(nid
);
1023 val
->totalram
= pgdat
->node_present_pages
;
1024 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1025 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1026 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1027 val
->mem_unit
= PAGE_SIZE
;
1031 #define K(x) ((x) << (PAGE_SHIFT-10))
1034 * Show free area list (used inside shift_scroll-lock stuff)
1035 * We also calculate the percentage fragmentation. We do this by counting the
1036 * memory on each free list with the exception of the first item on the list.
1038 void show_free_areas(void)
1040 struct page_state ps
;
1041 int cpu
, temperature
;
1042 unsigned long active
;
1043 unsigned long inactive
;
1047 for_each_zone(zone
) {
1049 printk("%s per-cpu:", zone
->name
);
1051 if (!zone
->present_pages
) {
1057 for (cpu
= 0; cpu
< NR_CPUS
; ++cpu
) {
1058 struct per_cpu_pageset
*pageset
;
1060 if (!cpu_possible(cpu
))
1063 pageset
= zone
->pageset
+ cpu
;
1065 for (temperature
= 0; temperature
< 2; temperature
++)
1066 printk("cpu %d %s: low %d, high %d, batch %d\n",
1068 temperature
? "cold" : "hot",
1069 pageset
->pcp
[temperature
].low
,
1070 pageset
->pcp
[temperature
].high
,
1071 pageset
->pcp
[temperature
].batch
);
1075 get_page_state(&ps
);
1076 get_zone_counts(&active
, &inactive
, &free
);
1078 printk("\nFree pages: %11ukB (%ukB HighMem)\n",
1080 K(nr_free_highpages()));
1082 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1083 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1092 ps
.nr_page_table_pages
);
1094 for_each_zone(zone
) {
1108 K(zone
->free_pages
),
1111 K(zone
->pages_high
),
1113 K(zone
->nr_inactive
),
1114 K(zone
->present_pages
)
1116 printk("protections[]:");
1117 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1118 printk(" %lu", zone
->protection
[i
]);
1122 for_each_zone(zone
) {
1123 struct list_head
*elem
;
1124 unsigned long nr
, flags
, order
, total
= 0;
1127 printk("%s: ", zone
->name
);
1128 if (!zone
->present_pages
) {
1133 spin_lock_irqsave(&zone
->lock
, flags
);
1134 for (order
= 0; order
< MAX_ORDER
; order
++) {
1136 list_for_each(elem
, &zone
->free_area
[order
].free_list
)
1138 total
+= nr
<< order
;
1139 printk("%lu*%lukB ", nr
, K(1UL) << order
);
1141 spin_unlock_irqrestore(&zone
->lock
, flags
);
1142 printk("= %lukB\n", K(total
));
1145 show_swap_cache_info();
1149 * Builds allocation fallback zone lists.
1151 static int __init
build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
, int j
, int k
)
1158 zone
= pgdat
->node_zones
+ ZONE_HIGHMEM
;
1159 if (zone
->present_pages
) {
1160 #ifndef CONFIG_HIGHMEM
1163 zonelist
->zones
[j
++] = zone
;
1166 zone
= pgdat
->node_zones
+ ZONE_NORMAL
;
1167 if (zone
->present_pages
)
1168 zonelist
->zones
[j
++] = zone
;
1170 zone
= pgdat
->node_zones
+ ZONE_DMA
;
1171 if (zone
->present_pages
)
1172 zonelist
->zones
[j
++] = zone
;
1179 #define MAX_NODE_LOAD (numnodes)
1180 static int __initdata node_load
[MAX_NUMNODES
];
1182 * find_next_best_node - find the next node that should appear in a given
1183 * node's fallback list
1184 * @node: node whose fallback list we're appending
1185 * @used_node_mask: pointer to the bitmap of already used nodes
1187 * We use a number of factors to determine which is the next node that should
1188 * appear on a given node's fallback list. The node should not have appeared
1189 * already in @node's fallback list, and it should be the next closest node
1190 * according to the distance array (which contains arbitrary distance values
1191 * from each node to each node in the system), and should also prefer nodes
1192 * with no CPUs, since presumably they'll have very little allocation pressure
1193 * on them otherwise.
1194 * It returns -1 if no node is found.
1196 static int __init
find_next_best_node(int node
, void *used_node_mask
)
1199 int min_val
= INT_MAX
;
1202 for (i
= 0; i
< numnodes
; i
++) {
1205 /* Start from local node */
1206 n
= (node
+i
)%numnodes
;
1208 /* Don't want a node to appear more than once */
1209 if (test_bit(n
, used_node_mask
))
1212 /* Use the distance array to find the distance */
1213 val
= node_distance(node
, n
);
1215 /* Give preference to headless and unused nodes */
1216 tmp
= node_to_cpumask(n
);
1217 if (!cpus_empty(tmp
))
1218 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1220 /* Slight preference for less loaded node */
1221 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1222 val
+= node_load
[n
];
1224 if (val
< min_val
) {
1231 set_bit(best_node
, used_node_mask
);
1236 static void __init
build_zonelists(pg_data_t
*pgdat
)
1238 int i
, j
, k
, node
, local_node
;
1239 int prev_node
, load
;
1240 struct zonelist
*zonelist
;
1241 DECLARE_BITMAP(used_mask
, MAX_NUMNODES
);
1243 /* initialize zonelists */
1244 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1245 zonelist
= pgdat
->node_zonelists
+ i
;
1246 memset(zonelist
, 0, sizeof(*zonelist
));
1247 zonelist
->zones
[0] = NULL
;
1250 /* NUMA-aware ordering of nodes */
1251 local_node
= pgdat
->node_id
;
1253 prev_node
= local_node
;
1254 bitmap_zero(used_mask
, MAX_NUMNODES
);
1255 while ((node
= find_next_best_node(local_node
, used_mask
)) >= 0) {
1257 * We don't want to pressure a particular node.
1258 * So adding penalty to the first node in same
1259 * distance group to make it round-robin.
1261 if (node_distance(local_node
, node
) !=
1262 node_distance(local_node
, prev_node
))
1263 node_load
[node
] += load
;
1266 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1267 zonelist
= pgdat
->node_zonelists
+ i
;
1268 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1271 if (i
& __GFP_HIGHMEM
)
1276 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1277 zonelist
->zones
[j
] = NULL
;
1282 #else /* CONFIG_NUMA */
1284 static void __init
build_zonelists(pg_data_t
*pgdat
)
1286 int i
, j
, k
, node
, local_node
;
1288 local_node
= pgdat
->node_id
;
1289 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1290 struct zonelist
*zonelist
;
1292 zonelist
= pgdat
->node_zonelists
+ i
;
1293 memset(zonelist
, 0, sizeof(*zonelist
));
1297 if (i
& __GFP_HIGHMEM
)
1302 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1304 * Now we build the zonelist so that it contains the zones
1305 * of all the other nodes.
1306 * We don't want to pressure a particular node, so when
1307 * building the zones for node N, we make sure that the
1308 * zones coming right after the local ones are those from
1309 * node N+1 (modulo N)
1311 for (node
= local_node
+ 1; node
< numnodes
; node
++)
1312 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1313 for (node
= 0; node
< local_node
; node
++)
1314 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1316 zonelist
->zones
[j
] = NULL
;
1320 #endif /* CONFIG_NUMA */
1322 void __init
build_all_zonelists(void)
1326 for(i
= 0 ; i
< numnodes
; i
++)
1327 build_zonelists(NODE_DATA(i
));
1328 printk("Built %i zonelists\n", numnodes
);
1332 * Helper functions to size the waitqueue hash table.
1333 * Essentially these want to choose hash table sizes sufficiently
1334 * large so that collisions trying to wait on pages are rare.
1335 * But in fact, the number of active page waitqueues on typical
1336 * systems is ridiculously low, less than 200. So this is even
1337 * conservative, even though it seems large.
1339 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1340 * waitqueues, i.e. the size of the waitq table given the number of pages.
1342 #define PAGES_PER_WAITQUEUE 256
1344 static inline unsigned long wait_table_size(unsigned long pages
)
1346 unsigned long size
= 1;
1348 pages
/= PAGES_PER_WAITQUEUE
;
1350 while (size
< pages
)
1354 * Once we have dozens or even hundreds of threads sleeping
1355 * on IO we've got bigger problems than wait queue collision.
1356 * Limit the size of the wait table to a reasonable size.
1358 size
= min(size
, 4096UL);
1360 return max(size
, 4UL);
1364 * This is an integer logarithm so that shifts can be used later
1365 * to extract the more random high bits from the multiplicative
1366 * hash function before the remainder is taken.
1368 static inline unsigned long wait_table_bits(unsigned long size
)
1373 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1375 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1376 unsigned long *zones_size
, unsigned long *zholes_size
)
1378 unsigned long realtotalpages
, totalpages
= 0;
1381 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1382 totalpages
+= zones_size
[i
];
1383 pgdat
->node_spanned_pages
= totalpages
;
1385 realtotalpages
= totalpages
;
1387 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1388 realtotalpages
-= zholes_size
[i
];
1389 pgdat
->node_present_pages
= realtotalpages
;
1390 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1395 * Initially all pages are reserved - free ones are freed
1396 * up by free_all_bootmem() once the early boot process is
1397 * done. Non-atomic initialization, single-pass.
1399 void __init
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1400 unsigned long start_pfn
)
1402 struct page
*start
= pfn_to_page(start_pfn
);
1405 for (page
= start
; page
< (start
+ size
); page
++) {
1406 set_page_zone(page
, NODEZONE(nid
, zone
));
1407 set_page_count(page
, 0);
1408 reset_page_mapcount(page
);
1409 SetPageReserved(page
);
1410 INIT_LIST_HEAD(&page
->lru
);
1411 #ifdef WANT_PAGE_VIRTUAL
1412 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1413 if (!is_highmem_idx(zone
))
1414 set_page_address(page
, __va(start_pfn
<< PAGE_SHIFT
));
1421 * Page buddy system uses "index >> (i+1)", where "index" is
1424 * The extra "+3" is to round down to byte size (8 bits per byte
1425 * assumption). Thus we get "(size-1) >> (i+4)" as the last byte
1428 * The "+1" is because we want to round the byte allocation up
1429 * rather than down. So we should have had a "+7" before we shifted
1430 * down by three. Also, we have to add one as we actually _use_ the
1431 * last bit (it's [0,n] inclusive, not [0,n[).
1433 * So we actually had +7+1 before we shift down by 3. But
1434 * (n+8) >> 3 == (n >> 3) + 1 (modulo overflows, which we do not have).
1436 * Finally, we LONG_ALIGN because all bitmap operations are on longs.
1438 unsigned long pages_to_bitmap_size(unsigned long order
, unsigned long nr_pages
)
1440 unsigned long bitmap_size
;
1442 bitmap_size
= (nr_pages
-1) >> (order
+4);
1443 bitmap_size
= LONG_ALIGN(bitmap_size
+1);
1448 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
, unsigned long size
)
1451 for (order
= 0; ; order
++) {
1452 unsigned long bitmap_size
;
1454 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1455 if (order
== MAX_ORDER
-1) {
1456 zone
->free_area
[order
].map
= NULL
;
1460 bitmap_size
= pages_to_bitmap_size(order
, size
);
1461 zone
->free_area
[order
].map
=
1462 (unsigned long *) alloc_bootmem_node(pgdat
, bitmap_size
);
1466 #ifndef __HAVE_ARCH_MEMMAP_INIT
1467 #define memmap_init(size, nid, zone, start_pfn) \
1468 memmap_init_zone((size), (nid), (zone), (start_pfn))
1472 * Set up the zone data structures:
1473 * - mark all pages reserved
1474 * - mark all memory queues empty
1475 * - clear the memory bitmaps
1477 static void __init
free_area_init_core(struct pglist_data
*pgdat
,
1478 unsigned long *zones_size
, unsigned long *zholes_size
)
1481 const unsigned long zone_required_alignment
= 1UL << (MAX_ORDER
-1);
1482 int cpu
, nid
= pgdat
->node_id
;
1483 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
1485 pgdat
->nr_zones
= 0;
1486 init_waitqueue_head(&pgdat
->kswapd_wait
);
1488 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
1489 struct zone
*zone
= pgdat
->node_zones
+ j
;
1490 unsigned long size
, realsize
;
1491 unsigned long batch
;
1493 zone_table
[NODEZONE(nid
, j
)] = zone
;
1494 realsize
= size
= zones_size
[j
];
1496 realsize
-= zholes_size
[j
];
1498 if (j
== ZONE_DMA
|| j
== ZONE_NORMAL
)
1499 nr_kernel_pages
+= realsize
;
1500 nr_all_pages
+= realsize
;
1502 zone
->spanned_pages
= size
;
1503 zone
->present_pages
= realsize
;
1504 zone
->name
= zone_names
[j
];
1505 spin_lock_init(&zone
->lock
);
1506 spin_lock_init(&zone
->lru_lock
);
1507 zone
->zone_pgdat
= pgdat
;
1508 zone
->free_pages
= 0;
1510 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
1513 * The per-cpu-pages pools are set to around 1000th of the
1514 * size of the zone. But no more than 1/4 of a meg - there's
1515 * no point in going beyond the size of L2 cache.
1517 * OK, so we don't know how big the cache is. So guess.
1519 batch
= zone
->present_pages
/ 1024;
1520 if (batch
* PAGE_SIZE
> 256 * 1024)
1521 batch
= (256 * 1024) / PAGE_SIZE
;
1522 batch
/= 4; /* We effectively *= 4 below */
1526 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1527 struct per_cpu_pages
*pcp
;
1529 pcp
= &zone
->pageset
[cpu
].pcp
[0]; /* hot */
1531 pcp
->low
= 2 * batch
;
1532 pcp
->high
= 6 * batch
;
1533 pcp
->batch
= 1 * batch
;
1534 INIT_LIST_HEAD(&pcp
->list
);
1536 pcp
= &zone
->pageset
[cpu
].pcp
[1]; /* cold */
1539 pcp
->high
= 2 * batch
;
1540 pcp
->batch
= 1 * batch
;
1541 INIT_LIST_HEAD(&pcp
->list
);
1543 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1544 zone_names
[j
], realsize
, batch
);
1545 INIT_LIST_HEAD(&zone
->active_list
);
1546 INIT_LIST_HEAD(&zone
->inactive_list
);
1547 zone
->nr_scan_active
= 0;
1548 zone
->nr_scan_inactive
= 0;
1549 zone
->nr_active
= 0;
1550 zone
->nr_inactive
= 0;
1555 * The per-page waitqueue mechanism uses hashed waitqueues
1558 zone
->wait_table_size
= wait_table_size(size
);
1559 zone
->wait_table_bits
=
1560 wait_table_bits(zone
->wait_table_size
);
1561 zone
->wait_table
= (wait_queue_head_t
*)
1562 alloc_bootmem_node(pgdat
, zone
->wait_table_size
1563 * sizeof(wait_queue_head_t
));
1565 for(i
= 0; i
< zone
->wait_table_size
; ++i
)
1566 init_waitqueue_head(zone
->wait_table
+ i
);
1568 pgdat
->nr_zones
= j
+1;
1570 zone
->zone_mem_map
= pfn_to_page(zone_start_pfn
);
1571 zone
->zone_start_pfn
= zone_start_pfn
;
1573 if ((zone_start_pfn
) & (zone_required_alignment
-1))
1574 printk("BUG: wrong zone alignment, it will crash\n");
1576 memmap_init(size
, nid
, j
, zone_start_pfn
);
1578 zone_start_pfn
+= size
;
1580 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
1584 void __init
node_alloc_mem_map(struct pglist_data
*pgdat
)
1588 size
= (pgdat
->node_spanned_pages
+ 1) * sizeof(struct page
);
1589 pgdat
->node_mem_map
= alloc_bootmem_node(pgdat
, size
);
1590 #ifndef CONFIG_DISCONTIGMEM
1591 mem_map
= contig_page_data
.node_mem_map
;
1595 void __init
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
1596 unsigned long *zones_size
, unsigned long node_start_pfn
,
1597 unsigned long *zholes_size
)
1599 pgdat
->node_id
= nid
;
1600 pgdat
->node_start_pfn
= node_start_pfn
;
1601 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
1603 if (!pfn_to_page(node_start_pfn
))
1604 node_alloc_mem_map(pgdat
);
1606 free_area_init_core(pgdat
, zones_size
, zholes_size
);
1609 #ifndef CONFIG_DISCONTIGMEM
1610 static bootmem_data_t contig_bootmem_data
;
1611 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
1613 EXPORT_SYMBOL(contig_page_data
);
1615 void __init
free_area_init(unsigned long *zones_size
)
1617 free_area_init_node(0, &contig_page_data
, zones_size
,
1618 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
1622 #ifdef CONFIG_PROC_FS
1624 #include <linux/seq_file.h>
1626 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
1631 for (pgdat
= pgdat_list
; pgdat
&& node
; pgdat
= pgdat
->pgdat_next
)
1637 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
1639 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
1642 return pgdat
->pgdat_next
;
1645 static void frag_stop(struct seq_file
*m
, void *arg
)
1650 * This walks the freelist for each zone. Whilst this is slow, I'd rather
1651 * be slow here than slow down the fast path by keeping stats - mjbligh
1653 static int frag_show(struct seq_file
*m
, void *arg
)
1655 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
1657 struct zone
*node_zones
= pgdat
->node_zones
;
1658 unsigned long flags
;
1661 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
1662 if (!zone
->present_pages
)
1665 spin_lock_irqsave(&zone
->lock
, flags
);
1666 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
1667 for (order
= 0; order
< MAX_ORDER
; ++order
) {
1668 unsigned long nr_bufs
= 0;
1669 struct list_head
*elem
;
1671 list_for_each(elem
, &(zone
->free_area
[order
].free_list
))
1673 seq_printf(m
, "%6lu ", nr_bufs
);
1675 spin_unlock_irqrestore(&zone
->lock
, flags
);
1681 struct seq_operations fragmentation_op
= {
1682 .start
= frag_start
,
1688 static char *vmstat_text
[] = {
1692 "nr_page_table_pages",
1717 "pgscan_kswapd_high",
1718 "pgscan_kswapd_normal",
1720 "pgscan_kswapd_dma",
1721 "pgscan_direct_high",
1722 "pgscan_direct_normal",
1723 "pgscan_direct_dma",
1728 "kswapd_inodesteal",
1735 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
1737 struct page_state
*ps
;
1739 if (*pos
>= ARRAY_SIZE(vmstat_text
))
1742 ps
= kmalloc(sizeof(*ps
), GFP_KERNEL
);
1745 return ERR_PTR(-ENOMEM
);
1746 get_full_page_state(ps
);
1747 ps
->pgpgin
/= 2; /* sectors -> kbytes */
1749 return (unsigned long *)ps
+ *pos
;
1752 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
1755 if (*pos
>= ARRAY_SIZE(vmstat_text
))
1757 return (unsigned long *)m
->private + *pos
;
1760 static int vmstat_show(struct seq_file
*m
, void *arg
)
1762 unsigned long *l
= arg
;
1763 unsigned long off
= l
- (unsigned long *)m
->private;
1765 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
1769 static void vmstat_stop(struct seq_file
*m
, void *arg
)
1775 struct seq_operations vmstat_op
= {
1776 .start
= vmstat_start
,
1777 .next
= vmstat_next
,
1778 .stop
= vmstat_stop
,
1779 .show
= vmstat_show
,
1782 #endif /* CONFIG_PROC_FS */
1784 #ifdef CONFIG_HOTPLUG_CPU
1785 static int page_alloc_cpu_notify(struct notifier_block
*self
,
1786 unsigned long action
, void *hcpu
)
1788 int cpu
= (unsigned long)hcpu
;
1791 if (action
== CPU_DEAD
) {
1792 /* Drain local pagecache count. */
1793 count
= &per_cpu(nr_pagecache_local
, cpu
);
1794 atomic_add(*count
, &nr_pagecache
);
1796 local_irq_disable();
1802 #endif /* CONFIG_HOTPLUG_CPU */
1804 void __init
page_alloc_init(void)
1806 hotcpu_notifier(page_alloc_cpu_notify
, 0);
1809 static unsigned long higherzone_val(struct zone
*z
, int max_zone
,
1812 int z_idx
= zone_idx(z
);
1813 struct zone
*higherzone
;
1814 unsigned long pages
;
1816 /* there is no higher zone to get a contribution from */
1817 if (z_idx
== MAX_NR_ZONES
-1)
1820 higherzone
= &z
->zone_pgdat
->node_zones
[z_idx
+1];
1822 /* We always start with the higher zone's protection value */
1823 pages
= higherzone
->protection
[alloc_type
];
1826 * We get a lower-zone-protection contribution only if there are
1827 * pages in the higher zone and if we're not the highest zone
1828 * in the current zonelist. e.g., never happens for GFP_DMA. Happens
1829 * only for ZONE_DMA in a GFP_KERNEL allocation and happens for ZONE_DMA
1830 * and ZONE_NORMAL for a GFP_HIGHMEM allocation.
1832 if (higherzone
->present_pages
&& z_idx
< alloc_type
)
1833 pages
+= higherzone
->pages_low
* sysctl_lower_zone_protection
;
1839 * setup_per_zone_protection - called whenver min_free_kbytes or
1840 * sysctl_lower_zone_protection changes. Ensures that each zone
1841 * has a correct pages_protected value, so an adequate number of
1842 * pages are left in the zone after a successful __alloc_pages().
1844 * This algorithm is way confusing. I tries to keep the same behavior
1845 * as we had with the incremental min iterative algorithm.
1847 static void setup_per_zone_protection(void)
1849 struct pglist_data
*pgdat
;
1850 struct zone
*zones
, *zone
;
1854 for_each_pgdat(pgdat
) {
1855 zones
= pgdat
->node_zones
;
1857 for (i
= 0, max_zone
= 0; i
< MAX_NR_ZONES
; i
++)
1858 if (zones
[i
].present_pages
)
1862 * For each of the different allocation types:
1863 * GFP_DMA -> GFP_KERNEL -> GFP_HIGHMEM
1865 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1867 * For each of the zones:
1868 * ZONE_HIGHMEM -> ZONE_NORMAL -> ZONE_DMA
1870 for (j
= MAX_NR_ZONES
-1; j
>= 0; j
--) {
1874 * We never protect zones that don't have memory
1875 * in them (j>max_zone) or zones that aren't in
1876 * the zonelists for a certain type of
1877 * allocation (j>=i). We have to assign these
1878 * to zero because the lower zones take
1879 * contributions from the higher zones.
1881 if (j
> max_zone
|| j
>= i
) {
1882 zone
->protection
[i
] = 0;
1886 * The contribution of the next higher zone
1888 zone
->protection
[i
] = higherzone_val(zone
,
1896 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
1897 * that the pages_{min,low,high} values for each zone are set correctly
1898 * with respect to min_free_kbytes.
1900 static void setup_per_zone_pages_min(void)
1902 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
1903 unsigned long lowmem_pages
= 0;
1905 unsigned long flags
;
1907 /* Calculate total number of !ZONE_HIGHMEM pages */
1908 for_each_zone(zone
) {
1909 if (!is_highmem(zone
))
1910 lowmem_pages
+= zone
->present_pages
;
1913 for_each_zone(zone
) {
1914 spin_lock_irqsave(&zone
->lru_lock
, flags
);
1915 if (is_highmem(zone
)) {
1917 * Often, highmem doesn't need to reserve any pages.
1918 * But the pages_min/low/high values are also used for
1919 * batching up page reclaim activity so we need a
1920 * decent value here.
1924 min_pages
= zone
->present_pages
/ 1024;
1925 if (min_pages
< SWAP_CLUSTER_MAX
)
1926 min_pages
= SWAP_CLUSTER_MAX
;
1927 if (min_pages
> 128)
1929 zone
->pages_min
= min_pages
;
1931 /* if it's a lowmem zone, reserve a number of pages
1932 * proportionate to the zone's size.
1934 zone
->pages_min
= (pages_min
* zone
->present_pages
) /
1938 zone
->pages_low
= zone
->pages_min
* 2;
1939 zone
->pages_high
= zone
->pages_min
* 3;
1940 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
1945 * Initialise min_free_kbytes.
1947 * For small machines we want it small (128k min). For large machines
1948 * we want it large (16MB max). But it is not linear, because network
1949 * bandwidth does not increase linearly with machine size. We use
1951 * min_free_kbytes = sqrt(lowmem_kbytes)
1967 static int __init
init_per_zone_pages_min(void)
1969 unsigned long lowmem_kbytes
;
1971 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
1973 min_free_kbytes
= int_sqrt(lowmem_kbytes
);
1974 if (min_free_kbytes
< 128)
1975 min_free_kbytes
= 128;
1976 if (min_free_kbytes
> 16384)
1977 min_free_kbytes
= 16384;
1978 setup_per_zone_pages_min();
1979 setup_per_zone_protection();
1982 module_init(init_per_zone_pages_min
)
1985 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
1986 * that we can call two helper functions whenever min_free_kbytes
1989 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
1990 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
1992 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
1993 setup_per_zone_pages_min();
1994 setup_per_zone_protection();
1999 * lower_zone_protection_sysctl_handler - just a wrapper around
2000 * proc_dointvec() so that we can call setup_per_zone_protection()
2001 * whenever sysctl_lower_zone_protection changes.
2003 int lower_zone_protection_sysctl_handler(ctl_table
*table
, int write
,
2004 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2006 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2007 setup_per_zone_protection();
2012 * allocate a large system hash table from bootmem
2013 * - it is assumed that the hash table must contain an exact power-of-2
2014 * quantity of entries
2016 void *__init
alloc_large_system_hash(const char *tablename
,
2017 unsigned long bucketsize
,
2018 unsigned long numentries
,
2020 int consider_highmem
,
2021 unsigned int *_hash_shift
,
2022 unsigned int *_hash_mask
)
2024 unsigned long long max
;
2025 unsigned long log2qty
, size
;
2028 /* allow the kernel cmdline to have a say */
2030 /* round applicable memory size up to nearest megabyte */
2031 numentries
= consider_highmem
? nr_all_pages
: nr_kernel_pages
;
2032 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2033 numentries
>>= 20 - PAGE_SHIFT
;
2034 numentries
<<= 20 - PAGE_SHIFT
;
2036 /* limit to 1 bucket per 2^scale bytes of low memory */
2037 if (scale
> PAGE_SHIFT
)
2038 numentries
>>= (scale
- PAGE_SHIFT
);
2040 numentries
<<= (PAGE_SHIFT
- scale
);
2042 /* rounded up to nearest power of 2 in size */
2043 numentries
= 1UL << (long_log2(numentries
) + 1);
2045 /* limit allocation size to 1/16 total memory */
2046 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2047 do_div(max
, bucketsize
);
2049 if (numentries
> max
)
2052 log2qty
= long_log2(numentries
);
2055 size
= bucketsize
<< log2qty
;
2056 table
= alloc_bootmem(size
);
2057 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2060 panic("Failed to allocate %s hash table\n", tablename
);
2062 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2065 long_log2(size
) - PAGE_SHIFT
,
2069 *_hash_shift
= log2qty
;
2071 *_hash_mask
= (1 << log2qty
) - 1;