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>
20 #include <linux/writeback.h>
21 #include <linux/compaction.h>
23 #ifdef CONFIG_VM_EVENT_COUNTERS
24 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
25 EXPORT_PER_CPU_SYMBOL(vm_event_states);
27 static void sum_vm_events(unsigned long *ret)
32 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
34 for_each_online_cpu(cpu) {
35 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
37 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
38 ret[i] += this->event[i];
43 * Accumulate the vm event counters across all CPUs.
44 * The result is unavoidably approximate - it can change
45 * during and after execution of this function.
47 void all_vm_events(unsigned long *ret)
53 EXPORT_SYMBOL_GPL(all_vm_events);
57 * Fold the foreign cpu events into our own.
59 * This is adding to the events on one processor
60 * but keeps the global counts constant.
62 void vm_events_fold_cpu(int cpu)
64 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
67 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
68 count_vm_events(i, fold_state->event[i]);
69 fold_state->event[i] = 0;
72 #endif /* CONFIG_HOTPLUG */
74 #endif /* CONFIG_VM_EVENT_COUNTERS */
77 * Manage combined zone based / global counters
79 * vm_stat contains the global counters
81 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
82 EXPORT_SYMBOL(vm_stat);
86 static int calculate_threshold(struct zone *zone)
89 int mem; /* memory in 128 MB units */
92 * The threshold scales with the number of processors and the amount
93 * of memory per zone. More memory means that we can defer updates for
94 * longer, more processors could lead to more contention.
95 * fls() is used to have a cheap way of logarithmic scaling.
97 * Some sample thresholds:
99 * Threshold Processors (fls) Zonesize fls(mem+1)
100 * ------------------------------------------------------------------
117 * 125 1024 10 8-16 GB 8
118 * 125 1024 10 16-32 GB 9
121 mem = zone->present_pages >> (27 - PAGE_SHIFT);
123 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
126 * Maximum threshold is 125
128 threshold = min(125, threshold);
134 * Refresh the thresholds for each zone.
136 static void refresh_zone_stat_thresholds(void)
142 for_each_populated_zone(zone) {
143 unsigned long max_drift, tolerate_drift;
145 threshold = calculate_threshold(zone);
147 for_each_online_cpu(cpu)
148 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
152 * Only set percpu_drift_mark if there is a danger that
153 * NR_FREE_PAGES reports the low watermark is ok when in fact
154 * the min watermark could be breached by an allocation
156 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
157 max_drift = num_online_cpus() * threshold;
158 if (max_drift > tolerate_drift)
159 zone->percpu_drift_mark = high_wmark_pages(zone) +
165 * For use when we know that interrupts are disabled.
167 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
170 struct per_cpu_pageset __percpu *pcp = zone->pageset;
171 s8 __percpu *p = pcp->vm_stat_diff + item;
175 x = delta + __this_cpu_read(*p);
177 t = __this_cpu_read(pcp->stat_threshold);
179 if (unlikely(x > t || x < -t)) {
180 zone_page_state_add(x, zone, item);
183 __this_cpu_write(*p, x);
185 EXPORT_SYMBOL(__mod_zone_page_state);
188 * For an unknown interrupt state
190 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
195 local_irq_save(flags);
196 __mod_zone_page_state(zone, item, delta);
197 local_irq_restore(flags);
199 EXPORT_SYMBOL(mod_zone_page_state);
202 * Optimized increment and decrement functions.
204 * These are only for a single page and therefore can take a struct page *
205 * argument instead of struct zone *. This allows the inclusion of the code
206 * generated for page_zone(page) into the optimized functions.
208 * No overflow check is necessary and therefore the differential can be
209 * incremented or decremented in place which may allow the compilers to
210 * generate better code.
211 * The increment or decrement is known and therefore one boundary check can
214 * NOTE: These functions are very performance sensitive. Change only
217 * Some processors have inc/dec instructions that are atomic vs an interrupt.
218 * However, the code must first determine the differential location in a zone
219 * based on the processor number and then inc/dec the counter. There is no
220 * guarantee without disabling preemption that the processor will not change
221 * in between and therefore the atomicity vs. interrupt cannot be exploited
222 * in a useful way here.
224 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
226 struct per_cpu_pageset __percpu *pcp = zone->pageset;
227 s8 __percpu *p = pcp->vm_stat_diff + item;
232 v = __this_cpu_read(*p);
233 t = __this_cpu_read(pcp->stat_threshold);
234 if (unlikely(v > t)) {
235 s8 overstep = t >> 1;
237 zone_page_state_add(v + overstep, zone, item);
238 __this_cpu_write(*p, -overstep);
242 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
244 __inc_zone_state(page_zone(page), item);
246 EXPORT_SYMBOL(__inc_zone_page_state);
248 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
250 struct per_cpu_pageset __percpu *pcp = zone->pageset;
251 s8 __percpu *p = pcp->vm_stat_diff + item;
256 v = __this_cpu_read(*p);
257 t = __this_cpu_read(pcp->stat_threshold);
258 if (unlikely(v < - t)) {
259 s8 overstep = t >> 1;
261 zone_page_state_add(v - overstep, zone, item);
262 __this_cpu_write(*p, overstep);
266 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
268 __dec_zone_state(page_zone(page), item);
270 EXPORT_SYMBOL(__dec_zone_page_state);
272 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
276 local_irq_save(flags);
277 __inc_zone_state(zone, item);
278 local_irq_restore(flags);
281 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
286 zone = page_zone(page);
287 local_irq_save(flags);
288 __inc_zone_state(zone, item);
289 local_irq_restore(flags);
291 EXPORT_SYMBOL(inc_zone_page_state);
293 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
297 local_irq_save(flags);
298 __dec_zone_page_state(page, item);
299 local_irq_restore(flags);
301 EXPORT_SYMBOL(dec_zone_page_state);
304 * Update the zone counters for one cpu.
306 * The cpu specified must be either the current cpu or a processor that
307 * is not online. If it is the current cpu then the execution thread must
308 * be pinned to the current cpu.
310 * Note that refresh_cpu_vm_stats strives to only access
311 * node local memory. The per cpu pagesets on remote zones are placed
312 * in the memory local to the processor using that pageset. So the
313 * loop over all zones will access a series of cachelines local to
316 * The call to zone_page_state_add updates the cachelines with the
317 * statistics in the remote zone struct as well as the global cachelines
318 * with the global counters. These could cause remote node cache line
319 * bouncing and will have to be only done when necessary.
321 void refresh_cpu_vm_stats(int cpu)
325 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
327 for_each_populated_zone(zone) {
328 struct per_cpu_pageset *p;
330 p = per_cpu_ptr(zone->pageset, cpu);
332 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
333 if (p->vm_stat_diff[i]) {
337 local_irq_save(flags);
338 v = p->vm_stat_diff[i];
339 p->vm_stat_diff[i] = 0;
340 local_irq_restore(flags);
341 atomic_long_add(v, &zone->vm_stat[i]);
344 /* 3 seconds idle till flush */
351 * Deal with draining the remote pageset of this
354 * Check if there are pages remaining in this pageset
355 * if not then there is nothing to expire.
357 if (!p->expire || !p->pcp.count)
361 * We never drain zones local to this processor.
363 if (zone_to_nid(zone) == numa_node_id()) {
373 drain_zone_pages(zone, &p->pcp);
377 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
379 atomic_long_add(global_diff[i], &vm_stat[i]);
386 * zonelist = the list of zones passed to the allocator
387 * z = the zone from which the allocation occurred.
389 * Must be called with interrupts disabled.
391 void zone_statistics(struct zone *preferred_zone, struct zone *z)
393 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
394 __inc_zone_state(z, NUMA_HIT);
396 __inc_zone_state(z, NUMA_MISS);
397 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
399 if (z->node == numa_node_id())
400 __inc_zone_state(z, NUMA_LOCAL);
402 __inc_zone_state(z, NUMA_OTHER);
406 #ifdef CONFIG_COMPACTION
408 struct contig_page_info {
409 unsigned long free_pages;
410 unsigned long free_blocks_total;
411 unsigned long free_blocks_suitable;
415 * Calculate the number of free pages in a zone, how many contiguous
416 * pages are free and how many are large enough to satisfy an allocation of
417 * the target size. Note that this function makes no attempt to estimate
418 * how many suitable free blocks there *might* be if MOVABLE pages were
419 * migrated. Calculating that is possible, but expensive and can be
420 * figured out from userspace
422 static void fill_contig_page_info(struct zone *zone,
423 unsigned int suitable_order,
424 struct contig_page_info *info)
428 info->free_pages = 0;
429 info->free_blocks_total = 0;
430 info->free_blocks_suitable = 0;
432 for (order = 0; order < MAX_ORDER; order++) {
433 unsigned long blocks;
435 /* Count number of free blocks */
436 blocks = zone->free_area[order].nr_free;
437 info->free_blocks_total += blocks;
439 /* Count free base pages */
440 info->free_pages += blocks << order;
442 /* Count the suitable free blocks */
443 if (order >= suitable_order)
444 info->free_blocks_suitable += blocks <<
445 (order - suitable_order);
450 * A fragmentation index only makes sense if an allocation of a requested
451 * size would fail. If that is true, the fragmentation index indicates
452 * whether external fragmentation or a lack of memory was the problem.
453 * The value can be used to determine if page reclaim or compaction
456 static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
458 unsigned long requested = 1UL << order;
460 if (!info->free_blocks_total)
463 /* Fragmentation index only makes sense when a request would fail */
464 if (info->free_blocks_suitable)
468 * Index is between 0 and 1 so return within 3 decimal places
470 * 0 => allocation would fail due to lack of memory
471 * 1 => allocation would fail due to fragmentation
473 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
476 /* Same as __fragmentation index but allocs contig_page_info on stack */
477 int fragmentation_index(struct zone *zone, unsigned int order)
479 struct contig_page_info info;
481 fill_contig_page_info(zone, order, &info);
482 return __fragmentation_index(order, &info);
486 #if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
487 #include <linux/proc_fs.h>
488 #include <linux/seq_file.h>
490 static char * const migratetype_names[MIGRATE_TYPES] = {
498 static void *frag_start(struct seq_file *m, loff_t *pos)
502 for (pgdat = first_online_pgdat();
504 pgdat = next_online_pgdat(pgdat))
510 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
512 pg_data_t *pgdat = (pg_data_t *)arg;
515 return next_online_pgdat(pgdat);
518 static void frag_stop(struct seq_file *m, void *arg)
522 /* Walk all the zones in a node and print using a callback */
523 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
524 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
527 struct zone *node_zones = pgdat->node_zones;
530 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
531 if (!populated_zone(zone))
534 spin_lock_irqsave(&zone->lock, flags);
535 print(m, pgdat, zone);
536 spin_unlock_irqrestore(&zone->lock, flags);
541 #ifdef CONFIG_PROC_FS
542 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
547 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
548 for (order = 0; order < MAX_ORDER; ++order)
549 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
554 * This walks the free areas for each zone.
556 static int frag_show(struct seq_file *m, void *arg)
558 pg_data_t *pgdat = (pg_data_t *)arg;
559 walk_zones_in_node(m, pgdat, frag_show_print);
563 static void pagetypeinfo_showfree_print(struct seq_file *m,
564 pg_data_t *pgdat, struct zone *zone)
568 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
569 seq_printf(m, "Node %4d, zone %8s, type %12s ",
572 migratetype_names[mtype]);
573 for (order = 0; order < MAX_ORDER; ++order) {
574 unsigned long freecount = 0;
575 struct free_area *area;
576 struct list_head *curr;
578 area = &(zone->free_area[order]);
580 list_for_each(curr, &area->free_list[mtype])
582 seq_printf(m, "%6lu ", freecount);
588 /* Print out the free pages at each order for each migatetype */
589 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
592 pg_data_t *pgdat = (pg_data_t *)arg;
595 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
596 for (order = 0; order < MAX_ORDER; ++order)
597 seq_printf(m, "%6d ", order);
600 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
605 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
606 pg_data_t *pgdat, struct zone *zone)
610 unsigned long start_pfn = zone->zone_start_pfn;
611 unsigned long end_pfn = start_pfn + zone->spanned_pages;
612 unsigned long count[MIGRATE_TYPES] = { 0, };
614 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
620 page = pfn_to_page(pfn);
622 /* Watch for unexpected holes punched in the memmap */
623 if (!memmap_valid_within(pfn, page, zone))
626 mtype = get_pageblock_migratetype(page);
628 if (mtype < MIGRATE_TYPES)
633 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
634 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
635 seq_printf(m, "%12lu ", count[mtype]);
639 /* Print out the free pages at each order for each migratetype */
640 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
643 pg_data_t *pgdat = (pg_data_t *)arg;
645 seq_printf(m, "\n%-23s", "Number of blocks type ");
646 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
647 seq_printf(m, "%12s ", migratetype_names[mtype]);
649 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
655 * This prints out statistics in relation to grouping pages by mobility.
656 * It is expensive to collect so do not constantly read the file.
658 static int pagetypeinfo_show(struct seq_file *m, void *arg)
660 pg_data_t *pgdat = (pg_data_t *)arg;
662 /* check memoryless node */
663 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
666 seq_printf(m, "Page block order: %d\n", pageblock_order);
667 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
669 pagetypeinfo_showfree(m, pgdat);
670 pagetypeinfo_showblockcount(m, pgdat);
675 static const struct seq_operations fragmentation_op = {
682 static int fragmentation_open(struct inode *inode, struct file *file)
684 return seq_open(file, &fragmentation_op);
687 static const struct file_operations fragmentation_file_operations = {
688 .open = fragmentation_open,
691 .release = seq_release,
694 static const struct seq_operations pagetypeinfo_op = {
698 .show = pagetypeinfo_show,
701 static int pagetypeinfo_open(struct inode *inode, struct file *file)
703 return seq_open(file, &pagetypeinfo_op);
706 static const struct file_operations pagetypeinfo_file_ops = {
707 .open = pagetypeinfo_open,
710 .release = seq_release,
713 #ifdef CONFIG_ZONE_DMA
714 #define TEXT_FOR_DMA(xx) xx "_dma",
716 #define TEXT_FOR_DMA(xx)
719 #ifdef CONFIG_ZONE_DMA32
720 #define TEXT_FOR_DMA32(xx) xx "_dma32",
722 #define TEXT_FOR_DMA32(xx)
725 #ifdef CONFIG_HIGHMEM
726 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
728 #define TEXT_FOR_HIGHMEM(xx)
731 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
732 TEXT_FOR_HIGHMEM(xx) xx "_movable",
734 static const char * const vmstat_text[] = {
735 /* Zoned VM counters */
748 "nr_slab_reclaimable",
749 "nr_slab_unreclaimable",
750 "nr_page_table_pages",
770 "nr_dirty_threshold",
771 "nr_dirty_background_threshold",
773 #ifdef CONFIG_VM_EVENT_COUNTERS
779 TEXTS_FOR_ZONES("pgalloc")
788 TEXTS_FOR_ZONES("pgrefill")
789 TEXTS_FOR_ZONES("pgsteal")
790 TEXTS_FOR_ZONES("pgscan_kswapd")
791 TEXTS_FOR_ZONES("pgscan_direct")
794 "zone_reclaim_failed",
800 "kswapd_low_wmark_hit_quickly",
801 "kswapd_high_wmark_hit_quickly",
802 "kswapd_skip_congestion_wait",
808 #ifdef CONFIG_COMPACTION
809 "compact_blocks_moved",
810 "compact_pages_moved",
811 "compact_pagemigrate_failed",
817 #ifdef CONFIG_HUGETLB_PAGE
818 "htlb_buddy_alloc_success",
819 "htlb_buddy_alloc_fail",
821 "unevictable_pgs_culled",
822 "unevictable_pgs_scanned",
823 "unevictable_pgs_rescued",
824 "unevictable_pgs_mlocked",
825 "unevictable_pgs_munlocked",
826 "unevictable_pgs_cleared",
827 "unevictable_pgs_stranded",
828 "unevictable_pgs_mlockfreed",
832 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
836 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
845 zone_nr_free_pages(zone),
846 min_wmark_pages(zone),
847 low_wmark_pages(zone),
848 high_wmark_pages(zone),
851 zone->present_pages);
853 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
854 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
855 zone_page_state(zone, i));
858 "\n protection: (%lu",
859 zone->lowmem_reserve[0]);
860 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
861 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
865 for_each_online_cpu(i) {
866 struct per_cpu_pageset *pageset;
868 pageset = per_cpu_ptr(zone->pageset, i);
879 seq_printf(m, "\n vm stats threshold: %d",
880 pageset->stat_threshold);
884 "\n all_unreclaimable: %u"
886 "\n inactive_ratio: %u",
887 zone->all_unreclaimable,
888 zone->zone_start_pfn,
889 zone->inactive_ratio);
894 * Output information about zones in @pgdat.
896 static int zoneinfo_show(struct seq_file *m, void *arg)
898 pg_data_t *pgdat = (pg_data_t *)arg;
899 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
903 static const struct seq_operations zoneinfo_op = {
904 .start = frag_start, /* iterate over all zones. The same as in
908 .show = zoneinfo_show,
911 static int zoneinfo_open(struct inode *inode, struct file *file)
913 return seq_open(file, &zoneinfo_op);
916 static const struct file_operations proc_zoneinfo_file_operations = {
917 .open = zoneinfo_open,
920 .release = seq_release,
923 enum writeback_stat_item {
925 NR_DIRTY_BG_THRESHOLD,
926 NR_VM_WRITEBACK_STAT_ITEMS,
929 static void *vmstat_start(struct seq_file *m, loff_t *pos)
932 int i, stat_items_size;
934 if (*pos >= ARRAY_SIZE(vmstat_text))
936 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
937 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
939 #ifdef CONFIG_VM_EVENT_COUNTERS
940 stat_items_size += sizeof(struct vm_event_state);
943 v = kmalloc(stat_items_size, GFP_KERNEL);
946 return ERR_PTR(-ENOMEM);
947 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
948 v[i] = global_page_state(i);
949 v += NR_VM_ZONE_STAT_ITEMS;
951 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
952 v + NR_DIRTY_THRESHOLD);
953 v += NR_VM_WRITEBACK_STAT_ITEMS;
955 #ifdef CONFIG_VM_EVENT_COUNTERS
957 v[PGPGIN] /= 2; /* sectors -> kbytes */
960 return (unsigned long *)m->private + *pos;
963 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
966 if (*pos >= ARRAY_SIZE(vmstat_text))
968 return (unsigned long *)m->private + *pos;
971 static int vmstat_show(struct seq_file *m, void *arg)
973 unsigned long *l = arg;
974 unsigned long off = l - (unsigned long *)m->private;
976 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
980 static void vmstat_stop(struct seq_file *m, void *arg)
986 static const struct seq_operations vmstat_op = {
987 .start = vmstat_start,
993 static int vmstat_open(struct inode *inode, struct file *file)
995 return seq_open(file, &vmstat_op);
998 static const struct file_operations proc_vmstat_file_operations = {
1001 .llseek = seq_lseek,
1002 .release = seq_release,
1004 #endif /* CONFIG_PROC_FS */
1007 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1008 int sysctl_stat_interval __read_mostly = HZ;
1010 static void vmstat_update(struct work_struct *w)
1012 refresh_cpu_vm_stats(smp_processor_id());
1013 schedule_delayed_work(&__get_cpu_var(vmstat_work),
1014 round_jiffies_relative(sysctl_stat_interval));
1017 static void __cpuinit start_cpu_timer(int cpu)
1019 struct delayed_work *work = &per_cpu(vmstat_work, cpu);
1021 INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
1022 schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
1026 * Use the cpu notifier to insure that the thresholds are recalculated
1029 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
1030 unsigned long action,
1033 long cpu = (long)hcpu;
1037 case CPU_ONLINE_FROZEN:
1038 refresh_zone_stat_thresholds();
1039 start_cpu_timer(cpu);
1040 node_set_state(cpu_to_node(cpu), N_CPU);
1042 case CPU_DOWN_PREPARE:
1043 case CPU_DOWN_PREPARE_FROZEN:
1044 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
1045 per_cpu(vmstat_work, cpu).work.func = NULL;
1047 case CPU_DOWN_FAILED:
1048 case CPU_DOWN_FAILED_FROZEN:
1049 start_cpu_timer(cpu);
1052 case CPU_DEAD_FROZEN:
1053 refresh_zone_stat_thresholds();
1061 static struct notifier_block __cpuinitdata vmstat_notifier =
1062 { &vmstat_cpuup_callback, NULL, 0 };
1065 static int __init setup_vmstat(void)
1070 refresh_zone_stat_thresholds();
1071 register_cpu_notifier(&vmstat_notifier);
1073 for_each_online_cpu(cpu)
1074 start_cpu_timer(cpu);
1076 #ifdef CONFIG_PROC_FS
1077 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1078 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1079 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1080 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1084 module_init(setup_vmstat)
1086 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1087 #include <linux/debugfs.h>
1089 static struct dentry *extfrag_debug_root;
1092 * Return an index indicating how much of the available free memory is
1093 * unusable for an allocation of the requested size.
1095 static int unusable_free_index(unsigned int order,
1096 struct contig_page_info *info)
1098 /* No free memory is interpreted as all free memory is unusable */
1099 if (info->free_pages == 0)
1103 * Index should be a value between 0 and 1. Return a value to 3
1106 * 0 => no fragmentation
1107 * 1 => high fragmentation
1109 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1113 static void unusable_show_print(struct seq_file *m,
1114 pg_data_t *pgdat, struct zone *zone)
1118 struct contig_page_info info;
1120 seq_printf(m, "Node %d, zone %8s ",
1123 for (order = 0; order < MAX_ORDER; ++order) {
1124 fill_contig_page_info(zone, order, &info);
1125 index = unusable_free_index(order, &info);
1126 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1133 * Display unusable free space index
1135 * The unusable free space index measures how much of the available free
1136 * memory cannot be used to satisfy an allocation of a given size and is a
1137 * value between 0 and 1. The higher the value, the more of free memory is
1138 * unusable and by implication, the worse the external fragmentation is. This
1139 * can be expressed as a percentage by multiplying by 100.
1141 static int unusable_show(struct seq_file *m, void *arg)
1143 pg_data_t *pgdat = (pg_data_t *)arg;
1145 /* check memoryless node */
1146 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
1149 walk_zones_in_node(m, pgdat, unusable_show_print);
1154 static const struct seq_operations unusable_op = {
1155 .start = frag_start,
1158 .show = unusable_show,
1161 static int unusable_open(struct inode *inode, struct file *file)
1163 return seq_open(file, &unusable_op);
1166 static const struct file_operations unusable_file_ops = {
1167 .open = unusable_open,
1169 .llseek = seq_lseek,
1170 .release = seq_release,
1173 static void extfrag_show_print(struct seq_file *m,
1174 pg_data_t *pgdat, struct zone *zone)
1179 /* Alloc on stack as interrupts are disabled for zone walk */
1180 struct contig_page_info info;
1182 seq_printf(m, "Node %d, zone %8s ",
1185 for (order = 0; order < MAX_ORDER; ++order) {
1186 fill_contig_page_info(zone, order, &info);
1187 index = __fragmentation_index(order, &info);
1188 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1195 * Display fragmentation index for orders that allocations would fail for
1197 static int extfrag_show(struct seq_file *m, void *arg)
1199 pg_data_t *pgdat = (pg_data_t *)arg;
1201 walk_zones_in_node(m, pgdat, extfrag_show_print);
1206 static const struct seq_operations extfrag_op = {
1207 .start = frag_start,
1210 .show = extfrag_show,
1213 static int extfrag_open(struct inode *inode, struct file *file)
1215 return seq_open(file, &extfrag_op);
1218 static const struct file_operations extfrag_file_ops = {
1219 .open = extfrag_open,
1221 .llseek = seq_lseek,
1222 .release = seq_release,
1225 static int __init extfrag_debug_init(void)
1227 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1228 if (!extfrag_debug_root)
1231 if (!debugfs_create_file("unusable_index", 0444,
1232 extfrag_debug_root, NULL, &unusable_file_ops))
1235 if (!debugfs_create_file("extfrag_index", 0444,
1236 extfrag_debug_root, NULL, &extfrag_file_ops))
1242 module_init(extfrag_debug_init);