4 * Manages VM statistics
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Copyright (C) 2006 Silicon Graphics, Inc.,
9 * Christoph Lameter <christoph@lameter.com>
13 #include <linux/err.h>
14 #include <linux/module.h>
15 #include <linux/slab.h>
16 #include <linux/cpu.h>
17 #include <linux/vmstat.h>
18 #include <linux/sched.h>
19 #include <linux/math64.h>
21 #ifdef CONFIG_VM_EVENT_COUNTERS
22 DEFINE_PER_CPU(struct vm_event_state
, vm_event_states
) = {{0}};
23 EXPORT_PER_CPU_SYMBOL(vm_event_states
);
25 static void sum_vm_events(unsigned long *ret
)
30 memset(ret
, 0, NR_VM_EVENT_ITEMS
* sizeof(unsigned long));
32 for_each_online_cpu(cpu
) {
33 struct vm_event_state
*this = &per_cpu(vm_event_states
, cpu
);
35 for (i
= 0; i
< NR_VM_EVENT_ITEMS
; i
++)
36 ret
[i
] += this->event
[i
];
41 * Accumulate the vm event counters across all CPUs.
42 * The result is unavoidably approximate - it can change
43 * during and after execution of this function.
45 void all_vm_events(unsigned long *ret
)
51 EXPORT_SYMBOL_GPL(all_vm_events
);
55 * Fold the foreign cpu events into our own.
57 * This is adding to the events on one processor
58 * but keeps the global counts constant.
60 void vm_events_fold_cpu(int cpu
)
62 struct vm_event_state
*fold_state
= &per_cpu(vm_event_states
, cpu
);
65 for (i
= 0; i
< NR_VM_EVENT_ITEMS
; i
++) {
66 count_vm_events(i
, fold_state
->event
[i
]);
67 fold_state
->event
[i
] = 0;
70 #endif /* CONFIG_HOTPLUG */
72 #endif /* CONFIG_VM_EVENT_COUNTERS */
75 * Manage combined zone based / global counters
77 * vm_stat contains the global counters
79 atomic_long_t vm_stat
[NR_VM_ZONE_STAT_ITEMS
];
80 EXPORT_SYMBOL(vm_stat
);
84 static int calculate_threshold(struct zone
*zone
)
87 int mem
; /* memory in 128 MB units */
90 * The threshold scales with the number of processors and the amount
91 * of memory per zone. More memory means that we can defer updates for
92 * longer, more processors could lead to more contention.
93 * fls() is used to have a cheap way of logarithmic scaling.
95 * Some sample thresholds:
97 * Threshold Processors (fls) Zonesize fls(mem+1)
98 * ------------------------------------------------------------------
115 * 125 1024 10 8-16 GB 8
116 * 125 1024 10 16-32 GB 9
119 mem
= zone
->present_pages
>> (27 - PAGE_SHIFT
);
121 threshold
= 2 * fls(num_online_cpus()) * (1 + fls(mem
));
124 * Maximum threshold is 125
126 threshold
= min(125, threshold
);
132 * Refresh the thresholds for each zone.
134 static void refresh_zone_stat_thresholds(void)
140 for_each_populated_zone(zone
) {
141 unsigned long max_drift
, tolerate_drift
;
143 threshold
= calculate_threshold(zone
);
145 for_each_online_cpu(cpu
)
146 per_cpu_ptr(zone
->pageset
, cpu
)->stat_threshold
150 * Only set percpu_drift_mark if there is a danger that
151 * NR_FREE_PAGES reports the low watermark is ok when in fact
152 * the min watermark could be breached by an allocation
154 tolerate_drift
= low_wmark_pages(zone
) - min_wmark_pages(zone
);
155 max_drift
= num_online_cpus() * threshold
;
156 if (max_drift
> tolerate_drift
)
157 zone
->percpu_drift_mark
= high_wmark_pages(zone
) +
163 * For use when we know that interrupts are disabled.
165 void __mod_zone_page_state(struct zone
*zone
, enum zone_stat_item item
,
168 struct per_cpu_pageset
*pcp
= this_cpu_ptr(zone
->pageset
);
170 s8
*p
= pcp
->vm_stat_diff
+ item
;
175 if (unlikely(x
> pcp
->stat_threshold
|| x
< -pcp
->stat_threshold
)) {
176 zone_page_state_add(x
, zone
, item
);
181 EXPORT_SYMBOL(__mod_zone_page_state
);
184 * For an unknown interrupt state
186 void mod_zone_page_state(struct zone
*zone
, enum zone_stat_item item
,
191 local_irq_save(flags
);
192 __mod_zone_page_state(zone
, item
, delta
);
193 local_irq_restore(flags
);
195 EXPORT_SYMBOL(mod_zone_page_state
);
198 * Optimized increment and decrement functions.
200 * These are only for a single page and therefore can take a struct page *
201 * argument instead of struct zone *. This allows the inclusion of the code
202 * generated for page_zone(page) into the optimized functions.
204 * No overflow check is necessary and therefore the differential can be
205 * incremented or decremented in place which may allow the compilers to
206 * generate better code.
207 * The increment or decrement is known and therefore one boundary check can
210 * NOTE: These functions are very performance sensitive. Change only
213 * Some processors have inc/dec instructions that are atomic vs an interrupt.
214 * However, the code must first determine the differential location in a zone
215 * based on the processor number and then inc/dec the counter. There is no
216 * guarantee without disabling preemption that the processor will not change
217 * in between and therefore the atomicity vs. interrupt cannot be exploited
218 * in a useful way here.
220 void __inc_zone_state(struct zone
*zone
, enum zone_stat_item item
)
222 struct per_cpu_pageset
*pcp
= this_cpu_ptr(zone
->pageset
);
223 s8
*p
= pcp
->vm_stat_diff
+ item
;
227 if (unlikely(*p
> pcp
->stat_threshold
)) {
228 int overstep
= pcp
->stat_threshold
/ 2;
230 zone_page_state_add(*p
+ overstep
, zone
, item
);
235 void __inc_zone_page_state(struct page
*page
, enum zone_stat_item item
)
237 __inc_zone_state(page_zone(page
), item
);
239 EXPORT_SYMBOL(__inc_zone_page_state
);
241 void __dec_zone_state(struct zone
*zone
, enum zone_stat_item item
)
243 struct per_cpu_pageset
*pcp
= this_cpu_ptr(zone
->pageset
);
244 s8
*p
= pcp
->vm_stat_diff
+ item
;
248 if (unlikely(*p
< - pcp
->stat_threshold
)) {
249 int overstep
= pcp
->stat_threshold
/ 2;
251 zone_page_state_add(*p
- overstep
, zone
, item
);
256 void __dec_zone_page_state(struct page
*page
, enum zone_stat_item item
)
258 __dec_zone_state(page_zone(page
), item
);
260 EXPORT_SYMBOL(__dec_zone_page_state
);
262 void inc_zone_state(struct zone
*zone
, enum zone_stat_item item
)
266 local_irq_save(flags
);
267 __inc_zone_state(zone
, item
);
268 local_irq_restore(flags
);
271 void inc_zone_page_state(struct page
*page
, enum zone_stat_item item
)
276 zone
= page_zone(page
);
277 local_irq_save(flags
);
278 __inc_zone_state(zone
, item
);
279 local_irq_restore(flags
);
281 EXPORT_SYMBOL(inc_zone_page_state
);
283 void dec_zone_page_state(struct page
*page
, enum zone_stat_item item
)
287 local_irq_save(flags
);
288 __dec_zone_page_state(page
, item
);
289 local_irq_restore(flags
);
291 EXPORT_SYMBOL(dec_zone_page_state
);
294 * Update the zone counters for one cpu.
296 * The cpu specified must be either the current cpu or a processor that
297 * is not online. If it is the current cpu then the execution thread must
298 * be pinned to the current cpu.
300 * Note that refresh_cpu_vm_stats strives to only access
301 * node local memory. The per cpu pagesets on remote zones are placed
302 * in the memory local to the processor using that pageset. So the
303 * loop over all zones will access a series of cachelines local to
306 * The call to zone_page_state_add updates the cachelines with the
307 * statistics in the remote zone struct as well as the global cachelines
308 * with the global counters. These could cause remote node cache line
309 * bouncing and will have to be only done when necessary.
311 void refresh_cpu_vm_stats(int cpu
)
315 int global_diff
[NR_VM_ZONE_STAT_ITEMS
] = { 0, };
317 for_each_populated_zone(zone
) {
318 struct per_cpu_pageset
*p
;
320 p
= per_cpu_ptr(zone
->pageset
, cpu
);
322 for (i
= 0; i
< NR_VM_ZONE_STAT_ITEMS
; i
++)
323 if (p
->vm_stat_diff
[i
]) {
327 local_irq_save(flags
);
328 v
= p
->vm_stat_diff
[i
];
329 p
->vm_stat_diff
[i
] = 0;
330 local_irq_restore(flags
);
331 atomic_long_add(v
, &zone
->vm_stat
[i
]);
334 /* 3 seconds idle till flush */
341 * Deal with draining the remote pageset of this
344 * Check if there are pages remaining in this pageset
345 * if not then there is nothing to expire.
347 if (!p
->expire
|| !p
->pcp
.count
)
351 * We never drain zones local to this processor.
353 if (zone_to_nid(zone
) == numa_node_id()) {
363 drain_zone_pages(zone
, &p
->pcp
);
367 for (i
= 0; i
< NR_VM_ZONE_STAT_ITEMS
; i
++)
369 atomic_long_add(global_diff
[i
], &vm_stat
[i
]);
376 * zonelist = the list of zones passed to the allocator
377 * z = the zone from which the allocation occurred.
379 * Must be called with interrupts disabled.
381 void zone_statistics(struct zone
*preferred_zone
, struct zone
*z
)
383 if (z
->zone_pgdat
== preferred_zone
->zone_pgdat
) {
384 __inc_zone_state(z
, NUMA_HIT
);
386 __inc_zone_state(z
, NUMA_MISS
);
387 __inc_zone_state(preferred_zone
, NUMA_FOREIGN
);
389 if (z
->node
== numa_node_id())
390 __inc_zone_state(z
, NUMA_LOCAL
);
392 __inc_zone_state(z
, NUMA_OTHER
);
396 #ifdef CONFIG_COMPACTION
397 struct contig_page_info
{
398 unsigned long free_pages
;
399 unsigned long free_blocks_total
;
400 unsigned long free_blocks_suitable
;
404 * Calculate the number of free pages in a zone, how many contiguous
405 * pages are free and how many are large enough to satisfy an allocation of
406 * the target size. Note that this function makes no attempt to estimate
407 * how many suitable free blocks there *might* be if MOVABLE pages were
408 * migrated. Calculating that is possible, but expensive and can be
409 * figured out from userspace
411 static void fill_contig_page_info(struct zone
*zone
,
412 unsigned int suitable_order
,
413 struct contig_page_info
*info
)
417 info
->free_pages
= 0;
418 info
->free_blocks_total
= 0;
419 info
->free_blocks_suitable
= 0;
421 for (order
= 0; order
< MAX_ORDER
; order
++) {
422 unsigned long blocks
;
424 /* Count number of free blocks */
425 blocks
= zone
->free_area
[order
].nr_free
;
426 info
->free_blocks_total
+= blocks
;
428 /* Count free base pages */
429 info
->free_pages
+= blocks
<< order
;
431 /* Count the suitable free blocks */
432 if (order
>= suitable_order
)
433 info
->free_blocks_suitable
+= blocks
<<
434 (order
- suitable_order
);
439 * A fragmentation index only makes sense if an allocation of a requested
440 * size would fail. If that is true, the fragmentation index indicates
441 * whether external fragmentation or a lack of memory was the problem.
442 * The value can be used to determine if page reclaim or compaction
445 static int __fragmentation_index(unsigned int order
, struct contig_page_info
*info
)
447 unsigned long requested
= 1UL << order
;
449 if (!info
->free_blocks_total
)
452 /* Fragmentation index only makes sense when a request would fail */
453 if (info
->free_blocks_suitable
)
457 * Index is between 0 and 1 so return within 3 decimal places
459 * 0 => allocation would fail due to lack of memory
460 * 1 => allocation would fail due to fragmentation
462 return 1000 - div_u64( (1000+(div_u64(info
->free_pages
* 1000ULL, requested
))), info
->free_blocks_total
);
465 /* Same as __fragmentation index but allocs contig_page_info on stack */
466 int fragmentation_index(struct zone
*zone
, unsigned int order
)
468 struct contig_page_info info
;
470 fill_contig_page_info(zone
, order
, &info
);
471 return __fragmentation_index(order
, &info
);
475 #if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
476 #include <linux/proc_fs.h>
477 #include <linux/seq_file.h>
479 static char * const migratetype_names
[MIGRATE_TYPES
] = {
487 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
491 for (pgdat
= first_online_pgdat();
493 pgdat
= next_online_pgdat(pgdat
))
499 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
501 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
504 return next_online_pgdat(pgdat
);
507 static void frag_stop(struct seq_file
*m
, void *arg
)
511 /* Walk all the zones in a node and print using a callback */
512 static void walk_zones_in_node(struct seq_file
*m
, pg_data_t
*pgdat
,
513 void (*print
)(struct seq_file
*m
, pg_data_t
*, struct zone
*))
516 struct zone
*node_zones
= pgdat
->node_zones
;
519 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
520 if (!populated_zone(zone
))
523 spin_lock_irqsave(&zone
->lock
, flags
);
524 print(m
, pgdat
, zone
);
525 spin_unlock_irqrestore(&zone
->lock
, flags
);
530 #ifdef CONFIG_PROC_FS
531 static void frag_show_print(struct seq_file
*m
, pg_data_t
*pgdat
,
536 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
537 for (order
= 0; order
< MAX_ORDER
; ++order
)
538 seq_printf(m
, "%6lu ", zone
->free_area
[order
].nr_free
);
543 * This walks the free areas for each zone.
545 static int frag_show(struct seq_file
*m
, void *arg
)
547 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
548 walk_zones_in_node(m
, pgdat
, frag_show_print
);
552 static void pagetypeinfo_showfree_print(struct seq_file
*m
,
553 pg_data_t
*pgdat
, struct zone
*zone
)
557 for (mtype
= 0; mtype
< MIGRATE_TYPES
; mtype
++) {
558 seq_printf(m
, "Node %4d, zone %8s, type %12s ",
561 migratetype_names
[mtype
]);
562 for (order
= 0; order
< MAX_ORDER
; ++order
) {
563 unsigned long freecount
= 0;
564 struct free_area
*area
;
565 struct list_head
*curr
;
567 area
= &(zone
->free_area
[order
]);
569 list_for_each(curr
, &area
->free_list
[mtype
])
571 seq_printf(m
, "%6lu ", freecount
);
577 /* Print out the free pages at each order for each migatetype */
578 static int pagetypeinfo_showfree(struct seq_file
*m
, void *arg
)
581 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
584 seq_printf(m
, "%-43s ", "Free pages count per migrate type at order");
585 for (order
= 0; order
< MAX_ORDER
; ++order
)
586 seq_printf(m
, "%6d ", order
);
589 walk_zones_in_node(m
, pgdat
, pagetypeinfo_showfree_print
);
594 static void pagetypeinfo_showblockcount_print(struct seq_file
*m
,
595 pg_data_t
*pgdat
, struct zone
*zone
)
599 unsigned long start_pfn
= zone
->zone_start_pfn
;
600 unsigned long end_pfn
= start_pfn
+ zone
->spanned_pages
;
601 unsigned long count
[MIGRATE_TYPES
] = { 0, };
603 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
609 page
= pfn_to_page(pfn
);
611 /* Watch for unexpected holes punched in the memmap */
612 if (!memmap_valid_within(pfn
, page
, zone
))
615 mtype
= get_pageblock_migratetype(page
);
617 if (mtype
< MIGRATE_TYPES
)
622 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
623 for (mtype
= 0; mtype
< MIGRATE_TYPES
; mtype
++)
624 seq_printf(m
, "%12lu ", count
[mtype
]);
628 /* Print out the free pages at each order for each migratetype */
629 static int pagetypeinfo_showblockcount(struct seq_file
*m
, void *arg
)
632 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
634 seq_printf(m
, "\n%-23s", "Number of blocks type ");
635 for (mtype
= 0; mtype
< MIGRATE_TYPES
; mtype
++)
636 seq_printf(m
, "%12s ", migratetype_names
[mtype
]);
638 walk_zones_in_node(m
, pgdat
, pagetypeinfo_showblockcount_print
);
644 * This prints out statistics in relation to grouping pages by mobility.
645 * It is expensive to collect so do not constantly read the file.
647 static int pagetypeinfo_show(struct seq_file
*m
, void *arg
)
649 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
651 /* check memoryless node */
652 if (!node_state(pgdat
->node_id
, N_HIGH_MEMORY
))
655 seq_printf(m
, "Page block order: %d\n", pageblock_order
);
656 seq_printf(m
, "Pages per block: %lu\n", pageblock_nr_pages
);
658 pagetypeinfo_showfree(m
, pgdat
);
659 pagetypeinfo_showblockcount(m
, pgdat
);
664 static const struct seq_operations fragmentation_op
= {
671 static int fragmentation_open(struct inode
*inode
, struct file
*file
)
673 return seq_open(file
, &fragmentation_op
);
676 static const struct file_operations fragmentation_file_operations
= {
677 .open
= fragmentation_open
,
680 .release
= seq_release
,
683 static const struct seq_operations pagetypeinfo_op
= {
687 .show
= pagetypeinfo_show
,
690 static int pagetypeinfo_open(struct inode
*inode
, struct file
*file
)
692 return seq_open(file
, &pagetypeinfo_op
);
695 static const struct file_operations pagetypeinfo_file_ops
= {
696 .open
= pagetypeinfo_open
,
699 .release
= seq_release
,
702 #ifdef CONFIG_ZONE_DMA
703 #define TEXT_FOR_DMA(xx) xx "_dma",
705 #define TEXT_FOR_DMA(xx)
708 #ifdef CONFIG_ZONE_DMA32
709 #define TEXT_FOR_DMA32(xx) xx "_dma32",
711 #define TEXT_FOR_DMA32(xx)
714 #ifdef CONFIG_HIGHMEM
715 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
717 #define TEXT_FOR_HIGHMEM(xx)
720 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
721 TEXT_FOR_HIGHMEM(xx) xx "_movable",
723 static const char * const vmstat_text
[] = {
724 /* Zoned VM counters */
737 "nr_slab_reclaimable",
738 "nr_slab_unreclaimable",
739 "nr_page_table_pages",
760 #ifdef CONFIG_VM_EVENT_COUNTERS
766 TEXTS_FOR_ZONES("pgalloc")
775 TEXTS_FOR_ZONES("pgrefill")
776 TEXTS_FOR_ZONES("pgsteal")
777 TEXTS_FOR_ZONES("pgscan_kswapd")
778 TEXTS_FOR_ZONES("pgscan_direct")
781 "zone_reclaim_failed",
787 "kswapd_low_wmark_hit_quickly",
788 "kswapd_high_wmark_hit_quickly",
789 "kswapd_skip_congestion_wait",
795 #ifdef CONFIG_COMPACTION
796 "compact_blocks_moved",
797 "compact_pages_moved",
798 "compact_pagemigrate_failed",
804 #ifdef CONFIG_HUGETLB_PAGE
805 "htlb_buddy_alloc_success",
806 "htlb_buddy_alloc_fail",
808 "unevictable_pgs_culled",
809 "unevictable_pgs_scanned",
810 "unevictable_pgs_rescued",
811 "unevictable_pgs_mlocked",
812 "unevictable_pgs_munlocked",
813 "unevictable_pgs_cleared",
814 "unevictable_pgs_stranded",
815 "unevictable_pgs_mlockfreed",
819 static void zoneinfo_show_print(struct seq_file
*m
, pg_data_t
*pgdat
,
823 seq_printf(m
, "Node %d, zone %8s", pgdat
->node_id
, zone
->name
);
832 zone_nr_free_pages(zone
),
833 min_wmark_pages(zone
),
834 low_wmark_pages(zone
),
835 high_wmark_pages(zone
),
838 zone
->present_pages
);
840 for (i
= 0; i
< NR_VM_ZONE_STAT_ITEMS
; i
++)
841 seq_printf(m
, "\n %-12s %lu", vmstat_text
[i
],
842 zone_page_state(zone
, i
));
845 "\n protection: (%lu",
846 zone
->lowmem_reserve
[0]);
847 for (i
= 1; i
< ARRAY_SIZE(zone
->lowmem_reserve
); i
++)
848 seq_printf(m
, ", %lu", zone
->lowmem_reserve
[i
]);
852 for_each_online_cpu(i
) {
853 struct per_cpu_pageset
*pageset
;
855 pageset
= per_cpu_ptr(zone
->pageset
, i
);
866 seq_printf(m
, "\n vm stats threshold: %d",
867 pageset
->stat_threshold
);
871 "\n all_unreclaimable: %u"
873 "\n inactive_ratio: %u",
874 zone
->all_unreclaimable
,
875 zone
->zone_start_pfn
,
876 zone
->inactive_ratio
);
881 * Output information about zones in @pgdat.
883 static int zoneinfo_show(struct seq_file
*m
, void *arg
)
885 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
886 walk_zones_in_node(m
, pgdat
, zoneinfo_show_print
);
890 static const struct seq_operations zoneinfo_op
= {
891 .start
= frag_start
, /* iterate over all zones. The same as in
895 .show
= zoneinfo_show
,
898 static int zoneinfo_open(struct inode
*inode
, struct file
*file
)
900 return seq_open(file
, &zoneinfo_op
);
903 static const struct file_operations proc_zoneinfo_file_operations
= {
904 .open
= zoneinfo_open
,
907 .release
= seq_release
,
910 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
913 #ifdef CONFIG_VM_EVENT_COUNTERS
918 if (*pos
>= ARRAY_SIZE(vmstat_text
))
921 #ifdef CONFIG_VM_EVENT_COUNTERS
922 v
= kmalloc(NR_VM_ZONE_STAT_ITEMS
* sizeof(unsigned long)
923 + sizeof(struct vm_event_state
), GFP_KERNEL
);
925 v
= kmalloc(NR_VM_ZONE_STAT_ITEMS
* sizeof(unsigned long),
930 return ERR_PTR(-ENOMEM
);
931 for (i
= 0; i
< NR_VM_ZONE_STAT_ITEMS
; i
++)
932 v
[i
] = global_page_state(i
);
933 #ifdef CONFIG_VM_EVENT_COUNTERS
934 e
= v
+ NR_VM_ZONE_STAT_ITEMS
;
936 e
[PGPGIN
] /= 2; /* sectors -> kbytes */
942 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
945 if (*pos
>= ARRAY_SIZE(vmstat_text
))
947 return (unsigned long *)m
->private + *pos
;
950 static int vmstat_show(struct seq_file
*m
, void *arg
)
952 unsigned long *l
= arg
;
953 unsigned long off
= l
- (unsigned long *)m
->private;
955 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
959 static void vmstat_stop(struct seq_file
*m
, void *arg
)
965 static const struct seq_operations vmstat_op
= {
966 .start
= vmstat_start
,
972 static int vmstat_open(struct inode
*inode
, struct file
*file
)
974 return seq_open(file
, &vmstat_op
);
977 static const struct file_operations proc_vmstat_file_operations
= {
981 .release
= seq_release
,
983 #endif /* CONFIG_PROC_FS */
986 static DEFINE_PER_CPU(struct delayed_work
, vmstat_work
);
987 int sysctl_stat_interval __read_mostly
= HZ
;
989 static void vmstat_update(struct work_struct
*w
)
991 refresh_cpu_vm_stats(smp_processor_id());
992 schedule_delayed_work(&__get_cpu_var(vmstat_work
),
993 round_jiffies_relative(sysctl_stat_interval
));
996 static void __cpuinit
start_cpu_timer(int cpu
)
998 struct delayed_work
*work
= &per_cpu(vmstat_work
, cpu
);
1000 INIT_DELAYED_WORK_DEFERRABLE(work
, vmstat_update
);
1001 schedule_delayed_work_on(cpu
, work
, __round_jiffies_relative(HZ
, cpu
));
1005 * Use the cpu notifier to insure that the thresholds are recalculated
1008 static int __cpuinit
vmstat_cpuup_callback(struct notifier_block
*nfb
,
1009 unsigned long action
,
1012 long cpu
= (long)hcpu
;
1016 case CPU_ONLINE_FROZEN
:
1017 refresh_zone_stat_thresholds();
1018 start_cpu_timer(cpu
);
1019 node_set_state(cpu_to_node(cpu
), N_CPU
);
1021 case CPU_DOWN_PREPARE
:
1022 case CPU_DOWN_PREPARE_FROZEN
:
1023 cancel_rearming_delayed_work(&per_cpu(vmstat_work
, cpu
));
1024 per_cpu(vmstat_work
, cpu
).work
.func
= NULL
;
1026 case CPU_DOWN_FAILED
:
1027 case CPU_DOWN_FAILED_FROZEN
:
1028 start_cpu_timer(cpu
);
1031 case CPU_DEAD_FROZEN
:
1032 refresh_zone_stat_thresholds();
1040 static struct notifier_block __cpuinitdata vmstat_notifier
=
1041 { &vmstat_cpuup_callback
, NULL
, 0 };
1044 static int __init
setup_vmstat(void)
1049 refresh_zone_stat_thresholds();
1050 register_cpu_notifier(&vmstat_notifier
);
1052 for_each_online_cpu(cpu
)
1053 start_cpu_timer(cpu
);
1055 #ifdef CONFIG_PROC_FS
1056 proc_create("buddyinfo", S_IRUGO
, NULL
, &fragmentation_file_operations
);
1057 proc_create("pagetypeinfo", S_IRUGO
, NULL
, &pagetypeinfo_file_ops
);
1058 proc_create("vmstat", S_IRUGO
, NULL
, &proc_vmstat_file_operations
);
1059 proc_create("zoneinfo", S_IRUGO
, NULL
, &proc_zoneinfo_file_operations
);
1063 module_init(setup_vmstat
)
1065 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1066 #include <linux/debugfs.h>
1068 static struct dentry
*extfrag_debug_root
;
1071 * Return an index indicating how much of the available free memory is
1072 * unusable for an allocation of the requested size.
1074 static int unusable_free_index(unsigned int order
,
1075 struct contig_page_info
*info
)
1077 /* No free memory is interpreted as all free memory is unusable */
1078 if (info
->free_pages
== 0)
1082 * Index should be a value between 0 and 1. Return a value to 3
1085 * 0 => no fragmentation
1086 * 1 => high fragmentation
1088 return div_u64((info
->free_pages
- (info
->free_blocks_suitable
<< order
)) * 1000ULL, info
->free_pages
);
1092 static void unusable_show_print(struct seq_file
*m
,
1093 pg_data_t
*pgdat
, struct zone
*zone
)
1097 struct contig_page_info info
;
1099 seq_printf(m
, "Node %d, zone %8s ",
1102 for (order
= 0; order
< MAX_ORDER
; ++order
) {
1103 fill_contig_page_info(zone
, order
, &info
);
1104 index
= unusable_free_index(order
, &info
);
1105 seq_printf(m
, "%d.%03d ", index
/ 1000, index
% 1000);
1112 * Display unusable free space index
1114 * The unusable free space index measures how much of the available free
1115 * memory cannot be used to satisfy an allocation of a given size and is a
1116 * value between 0 and 1. The higher the value, the more of free memory is
1117 * unusable and by implication, the worse the external fragmentation is. This
1118 * can be expressed as a percentage by multiplying by 100.
1120 static int unusable_show(struct seq_file
*m
, void *arg
)
1122 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
1124 /* check memoryless node */
1125 if (!node_state(pgdat
->node_id
, N_HIGH_MEMORY
))
1128 walk_zones_in_node(m
, pgdat
, unusable_show_print
);
1133 static const struct seq_operations unusable_op
= {
1134 .start
= frag_start
,
1137 .show
= unusable_show
,
1140 static int unusable_open(struct inode
*inode
, struct file
*file
)
1142 return seq_open(file
, &unusable_op
);
1145 static const struct file_operations unusable_file_ops
= {
1146 .open
= unusable_open
,
1148 .llseek
= seq_lseek
,
1149 .release
= seq_release
,
1152 static void extfrag_show_print(struct seq_file
*m
,
1153 pg_data_t
*pgdat
, struct zone
*zone
)
1158 /* Alloc on stack as interrupts are disabled for zone walk */
1159 struct contig_page_info info
;
1161 seq_printf(m
, "Node %d, zone %8s ",
1164 for (order
= 0; order
< MAX_ORDER
; ++order
) {
1165 fill_contig_page_info(zone
, order
, &info
);
1166 index
= __fragmentation_index(order
, &info
);
1167 seq_printf(m
, "%d.%03d ", index
/ 1000, index
% 1000);
1174 * Display fragmentation index for orders that allocations would fail for
1176 static int extfrag_show(struct seq_file
*m
, void *arg
)
1178 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
1180 walk_zones_in_node(m
, pgdat
, extfrag_show_print
);
1185 static const struct seq_operations extfrag_op
= {
1186 .start
= frag_start
,
1189 .show
= extfrag_show
,
1192 static int extfrag_open(struct inode
*inode
, struct file
*file
)
1194 return seq_open(file
, &extfrag_op
);
1197 static const struct file_operations extfrag_file_ops
= {
1198 .open
= extfrag_open
,
1200 .llseek
= seq_lseek
,
1201 .release
= seq_release
,
1204 static int __init
extfrag_debug_init(void)
1206 extfrag_debug_root
= debugfs_create_dir("extfrag", NULL
);
1207 if (!extfrag_debug_root
)
1210 if (!debugfs_create_file("unusable_index", 0444,
1211 extfrag_debug_root
, NULL
, &unusable_file_ops
))
1214 if (!debugfs_create_file("extfrag_index", 0444,
1215 extfrag_debug_root
, NULL
, &extfrag_file_ops
))
1221 module_init(extfrag_debug_init
);