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>
22 #include <linux/mm_inline.h>
26 #ifdef CONFIG_VM_EVENT_COUNTERS
27 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
28 EXPORT_PER_CPU_SYMBOL(vm_event_states);
30 static void sum_vm_events(unsigned long *ret)
35 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
37 for_each_online_cpu(cpu) {
38 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
40 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
41 ret[i] += this->event[i];
46 * Accumulate the vm event counters across all CPUs.
47 * The result is unavoidably approximate - it can change
48 * during and after execution of this function.
50 void all_vm_events(unsigned long *ret)
56 EXPORT_SYMBOL_GPL(all_vm_events);
59 * Fold the foreign cpu events into our own.
61 * This is adding to the events on one processor
62 * but keeps the global counts constant.
64 void vm_events_fold_cpu(int cpu)
66 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
69 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
70 count_vm_events(i, fold_state->event[i]);
71 fold_state->event[i] = 0;
75 #endif /* CONFIG_VM_EVENT_COUNTERS */
78 * Manage combined zone based / global counters
80 * vm_stat contains the global counters
82 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
83 EXPORT_SYMBOL(vm_stat);
87 int calculate_pressure_threshold(struct zone *zone)
90 int watermark_distance;
93 * As vmstats are not up to date, there is drift between the estimated
94 * and real values. For high thresholds and a high number of CPUs, it
95 * is possible for the min watermark to be breached while the estimated
96 * value looks fine. The pressure threshold is a reduced value such
97 * that even the maximum amount of drift will not accidentally breach
100 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
101 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
104 * Maximum threshold is 125
106 threshold = min(125, threshold);
111 int calculate_normal_threshold(struct zone *zone)
114 int mem; /* memory in 128 MB units */
117 * The threshold scales with the number of processors and the amount
118 * of memory per zone. More memory means that we can defer updates for
119 * longer, more processors could lead to more contention.
120 * fls() is used to have a cheap way of logarithmic scaling.
122 * Some sample thresholds:
124 * Threshold Processors (fls) Zonesize fls(mem+1)
125 * ------------------------------------------------------------------
142 * 125 1024 10 8-16 GB 8
143 * 125 1024 10 16-32 GB 9
146 mem = zone->managed_pages >> (27 - PAGE_SHIFT);
148 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
151 * Maximum threshold is 125
153 threshold = min(125, threshold);
159 * Refresh the thresholds for each zone.
161 void refresh_zone_stat_thresholds(void)
167 for_each_populated_zone(zone) {
168 unsigned long max_drift, tolerate_drift;
170 threshold = calculate_normal_threshold(zone);
172 for_each_online_cpu(cpu)
173 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
177 * Only set percpu_drift_mark if there is a danger that
178 * NR_FREE_PAGES reports the low watermark is ok when in fact
179 * the min watermark could be breached by an allocation
181 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
182 max_drift = num_online_cpus() * threshold;
183 if (max_drift > tolerate_drift)
184 zone->percpu_drift_mark = high_wmark_pages(zone) +
189 void set_pgdat_percpu_threshold(pg_data_t *pgdat,
190 int (*calculate_pressure)(struct zone *))
197 for (i = 0; i < pgdat->nr_zones; i++) {
198 zone = &pgdat->node_zones[i];
199 if (!zone->percpu_drift_mark)
202 threshold = (*calculate_pressure)(zone);
203 for_each_possible_cpu(cpu)
204 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
210 * For use when we know that interrupts are disabled.
212 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
215 struct per_cpu_pageset __percpu *pcp = zone->pageset;
216 s8 __percpu *p = pcp->vm_stat_diff + item;
220 x = delta + __this_cpu_read(*p);
222 t = __this_cpu_read(pcp->stat_threshold);
224 if (unlikely(x > t || x < -t)) {
225 zone_page_state_add(x, zone, item);
228 __this_cpu_write(*p, x);
230 EXPORT_SYMBOL(__mod_zone_page_state);
233 * Optimized increment and decrement functions.
235 * These are only for a single page and therefore can take a struct page *
236 * argument instead of struct zone *. This allows the inclusion of the code
237 * generated for page_zone(page) into the optimized functions.
239 * No overflow check is necessary and therefore the differential can be
240 * incremented or decremented in place which may allow the compilers to
241 * generate better code.
242 * The increment or decrement is known and therefore one boundary check can
245 * NOTE: These functions are very performance sensitive. Change only
248 * Some processors have inc/dec instructions that are atomic vs an interrupt.
249 * However, the code must first determine the differential location in a zone
250 * based on the processor number and then inc/dec the counter. There is no
251 * guarantee without disabling preemption that the processor will not change
252 * in between and therefore the atomicity vs. interrupt cannot be exploited
253 * in a useful way here.
255 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
257 struct per_cpu_pageset __percpu *pcp = zone->pageset;
258 s8 __percpu *p = pcp->vm_stat_diff + item;
261 v = __this_cpu_inc_return(*p);
262 t = __this_cpu_read(pcp->stat_threshold);
263 if (unlikely(v > t)) {
264 s8 overstep = t >> 1;
266 zone_page_state_add(v + overstep, zone, item);
267 __this_cpu_write(*p, -overstep);
271 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
273 __inc_zone_state(page_zone(page), item);
275 EXPORT_SYMBOL(__inc_zone_page_state);
277 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
279 struct per_cpu_pageset __percpu *pcp = zone->pageset;
280 s8 __percpu *p = pcp->vm_stat_diff + item;
283 v = __this_cpu_dec_return(*p);
284 t = __this_cpu_read(pcp->stat_threshold);
285 if (unlikely(v < - t)) {
286 s8 overstep = t >> 1;
288 zone_page_state_add(v - overstep, zone, item);
289 __this_cpu_write(*p, overstep);
293 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
295 __dec_zone_state(page_zone(page), item);
297 EXPORT_SYMBOL(__dec_zone_page_state);
299 #ifdef CONFIG_HAVE_CMPXCHG_LOCAL
301 * If we have cmpxchg_local support then we do not need to incur the overhead
302 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
304 * mod_state() modifies the zone counter state through atomic per cpu
307 * Overstep mode specifies how overstep should handled:
309 * 1 Overstepping half of threshold
310 * -1 Overstepping minus half of threshold
312 static inline void mod_state(struct zone *zone,
313 enum zone_stat_item item, int delta, int overstep_mode)
315 struct per_cpu_pageset __percpu *pcp = zone->pageset;
316 s8 __percpu *p = pcp->vm_stat_diff + item;
320 z = 0; /* overflow to zone counters */
323 * The fetching of the stat_threshold is racy. We may apply
324 * a counter threshold to the wrong the cpu if we get
325 * rescheduled while executing here. However, the next
326 * counter update will apply the threshold again and
327 * therefore bring the counter under the threshold again.
329 * Most of the time the thresholds are the same anyways
330 * for all cpus in a zone.
332 t = this_cpu_read(pcp->stat_threshold);
334 o = this_cpu_read(*p);
337 if (n > t || n < -t) {
338 int os = overstep_mode * (t >> 1) ;
340 /* Overflow must be added to zone counters */
344 } while (this_cpu_cmpxchg(*p, o, n) != o);
347 zone_page_state_add(z, zone, item);
350 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
353 mod_state(zone, item, delta, 0);
355 EXPORT_SYMBOL(mod_zone_page_state);
357 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
359 mod_state(zone, item, 1, 1);
362 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
364 mod_state(page_zone(page), item, 1, 1);
366 EXPORT_SYMBOL(inc_zone_page_state);
368 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
370 mod_state(page_zone(page), item, -1, -1);
372 EXPORT_SYMBOL(dec_zone_page_state);
375 * Use interrupt disable to serialize counter updates
377 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
382 local_irq_save(flags);
383 __mod_zone_page_state(zone, item, delta);
384 local_irq_restore(flags);
386 EXPORT_SYMBOL(mod_zone_page_state);
388 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
392 local_irq_save(flags);
393 __inc_zone_state(zone, item);
394 local_irq_restore(flags);
397 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
402 zone = page_zone(page);
403 local_irq_save(flags);
404 __inc_zone_state(zone, item);
405 local_irq_restore(flags);
407 EXPORT_SYMBOL(inc_zone_page_state);
409 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
413 local_irq_save(flags);
414 __dec_zone_page_state(page, item);
415 local_irq_restore(flags);
417 EXPORT_SYMBOL(dec_zone_page_state);
421 * Update the zone counters for one cpu.
423 * The cpu specified must be either the current cpu or a processor that
424 * is not online. If it is the current cpu then the execution thread must
425 * be pinned to the current cpu.
427 * Note that refresh_cpu_vm_stats strives to only access
428 * node local memory. The per cpu pagesets on remote zones are placed
429 * in the memory local to the processor using that pageset. So the
430 * loop over all zones will access a series of cachelines local to
433 * The call to zone_page_state_add updates the cachelines with the
434 * statistics in the remote zone struct as well as the global cachelines
435 * with the global counters. These could cause remote node cache line
436 * bouncing and will have to be only done when necessary.
438 void refresh_cpu_vm_stats(int cpu)
442 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
444 for_each_populated_zone(zone) {
445 struct per_cpu_pageset *p;
447 p = per_cpu_ptr(zone->pageset, cpu);
449 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
450 if (p->vm_stat_diff[i]) {
454 local_irq_save(flags);
455 v = p->vm_stat_diff[i];
456 p->vm_stat_diff[i] = 0;
457 local_irq_restore(flags);
458 atomic_long_add(v, &zone->vm_stat[i]);
461 /* 3 seconds idle till flush */
468 * Deal with draining the remote pageset of this
471 * Check if there are pages remaining in this pageset
472 * if not then there is nothing to expire.
474 if (!p->expire || !p->pcp.count)
478 * We never drain zones local to this processor.
480 if (zone_to_nid(zone) == numa_node_id()) {
490 drain_zone_pages(zone, &p->pcp);
494 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
496 atomic_long_add(global_diff[i], &vm_stat[i]);
500 * this is only called if !populated_zone(zone), which implies no other users of
501 * pset->vm_stat_diff[] exsist.
503 void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
507 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
508 if (pset->vm_stat_diff[i]) {
509 int v = pset->vm_stat_diff[i];
510 pset->vm_stat_diff[i] = 0;
511 atomic_long_add(v, &zone->vm_stat[i]);
512 atomic_long_add(v, &vm_stat[i]);
519 * zonelist = the list of zones passed to the allocator
520 * z = the zone from which the allocation occurred.
522 * Must be called with interrupts disabled.
524 * When __GFP_OTHER_NODE is set assume the node of the preferred
525 * zone is the local node. This is useful for daemons who allocate
526 * memory on behalf of other processes.
528 void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags)
530 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
531 __inc_zone_state(z, NUMA_HIT);
533 __inc_zone_state(z, NUMA_MISS);
534 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
536 if (z->node == ((flags & __GFP_OTHER_NODE) ?
537 preferred_zone->node : numa_node_id()))
538 __inc_zone_state(z, NUMA_LOCAL);
540 __inc_zone_state(z, NUMA_OTHER);
544 #ifdef CONFIG_COMPACTION
546 struct contig_page_info {
547 unsigned long free_pages;
548 unsigned long free_blocks_total;
549 unsigned long free_blocks_suitable;
553 * Calculate the number of free pages in a zone, how many contiguous
554 * pages are free and how many are large enough to satisfy an allocation of
555 * the target size. Note that this function makes no attempt to estimate
556 * how many suitable free blocks there *might* be if MOVABLE pages were
557 * migrated. Calculating that is possible, but expensive and can be
558 * figured out from userspace
560 static void fill_contig_page_info(struct zone *zone,
561 unsigned int suitable_order,
562 struct contig_page_info *info)
566 info->free_pages = 0;
567 info->free_blocks_total = 0;
568 info->free_blocks_suitable = 0;
570 for (order = 0; order < MAX_ORDER; order++) {
571 unsigned long blocks;
573 /* Count number of free blocks */
574 blocks = zone->free_area[order].nr_free;
575 info->free_blocks_total += blocks;
577 /* Count free base pages */
578 info->free_pages += blocks << order;
580 /* Count the suitable free blocks */
581 if (order >= suitable_order)
582 info->free_blocks_suitable += blocks <<
583 (order - suitable_order);
588 * A fragmentation index only makes sense if an allocation of a requested
589 * size would fail. If that is true, the fragmentation index indicates
590 * whether external fragmentation or a lack of memory was the problem.
591 * The value can be used to determine if page reclaim or compaction
594 static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
596 unsigned long requested = 1UL << order;
598 if (!info->free_blocks_total)
601 /* Fragmentation index only makes sense when a request would fail */
602 if (info->free_blocks_suitable)
606 * Index is between 0 and 1 so return within 3 decimal places
608 * 0 => allocation would fail due to lack of memory
609 * 1 => allocation would fail due to fragmentation
611 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
614 /* Same as __fragmentation index but allocs contig_page_info on stack */
615 int fragmentation_index(struct zone *zone, unsigned int order)
617 struct contig_page_info info;
619 fill_contig_page_info(zone, order, &info);
620 return __fragmentation_index(order, &info);
624 #if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
625 #include <linux/proc_fs.h>
626 #include <linux/seq_file.h>
628 static char * const migratetype_names[MIGRATE_TYPES] = {
636 #ifdef CONFIG_MEMORY_ISOLATION
641 static void *frag_start(struct seq_file *m, loff_t *pos)
645 for (pgdat = first_online_pgdat();
647 pgdat = next_online_pgdat(pgdat))
653 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
655 pg_data_t *pgdat = (pg_data_t *)arg;
658 return next_online_pgdat(pgdat);
661 static void frag_stop(struct seq_file *m, void *arg)
665 /* Walk all the zones in a node and print using a callback */
666 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
667 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
670 struct zone *node_zones = pgdat->node_zones;
673 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
674 if (!populated_zone(zone))
677 spin_lock_irqsave(&zone->lock, flags);
678 print(m, pgdat, zone);
679 spin_unlock_irqrestore(&zone->lock, flags);
684 #if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
685 #ifdef CONFIG_ZONE_DMA
686 #define TEXT_FOR_DMA(xx) xx "_dma",
688 #define TEXT_FOR_DMA(xx)
691 #ifdef CONFIG_ZONE_DMA32
692 #define TEXT_FOR_DMA32(xx) xx "_dma32",
694 #define TEXT_FOR_DMA32(xx)
697 #ifdef CONFIG_HIGHMEM
698 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
700 #define TEXT_FOR_HIGHMEM(xx)
703 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
704 TEXT_FOR_HIGHMEM(xx) xx "_movable",
706 const char * const vmstat_text[] = {
707 /* Zoned VM counters */
720 "nr_slab_reclaimable",
721 "nr_slab_unreclaimable",
722 "nr_page_table_pages",
727 "nr_vmscan_immediate_reclaim",
743 "nr_anon_transparent_hugepages",
745 "nr_dirty_threshold",
746 "nr_dirty_background_threshold",
748 #ifdef CONFIG_VM_EVENT_COUNTERS
754 TEXTS_FOR_ZONES("pgalloc")
763 TEXTS_FOR_ZONES("pgrefill")
764 TEXTS_FOR_ZONES("pgsteal_kswapd")
765 TEXTS_FOR_ZONES("pgsteal_direct")
766 TEXTS_FOR_ZONES("pgscan_kswapd")
767 TEXTS_FOR_ZONES("pgscan_direct")
768 "pgscan_direct_throttle",
771 "zone_reclaim_failed",
776 "kswapd_low_wmark_hit_quickly",
777 "kswapd_high_wmark_hit_quickly",
783 #ifdef CONFIG_NUMA_BALANCING
786 "numa_hint_faults_local",
787 "numa_pages_migrated",
789 #ifdef CONFIG_MIGRATION
793 #ifdef CONFIG_COMPACTION
794 "compact_migrate_scanned",
795 "compact_free_scanned",
802 #ifdef CONFIG_HUGETLB_PAGE
803 "htlb_buddy_alloc_success",
804 "htlb_buddy_alloc_fail",
806 "unevictable_pgs_culled",
807 "unevictable_pgs_scanned",
808 "unevictable_pgs_rescued",
809 "unevictable_pgs_mlocked",
810 "unevictable_pgs_munlocked",
811 "unevictable_pgs_cleared",
812 "unevictable_pgs_stranded",
814 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
816 "thp_fault_fallback",
817 "thp_collapse_alloc",
818 "thp_collapse_alloc_failed",
820 "thp_zero_page_alloc",
821 "thp_zero_page_alloc_failed",
824 #endif /* CONFIG_VM_EVENTS_COUNTERS */
826 #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
829 #ifdef CONFIG_PROC_FS
830 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
835 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
836 for (order = 0; order < MAX_ORDER; ++order)
837 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
842 * This walks the free areas for each zone.
844 static int frag_show(struct seq_file *m, void *arg)
846 pg_data_t *pgdat = (pg_data_t *)arg;
847 walk_zones_in_node(m, pgdat, frag_show_print);
851 static void pagetypeinfo_showfree_print(struct seq_file *m,
852 pg_data_t *pgdat, struct zone *zone)
856 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
857 seq_printf(m, "Node %4d, zone %8s, type %12s ",
860 migratetype_names[mtype]);
861 for (order = 0; order < MAX_ORDER; ++order) {
862 unsigned long freecount = 0;
863 struct free_area *area;
864 struct list_head *curr;
866 area = &(zone->free_area[order]);
868 list_for_each(curr, &area->free_list[mtype])
870 seq_printf(m, "%6lu ", freecount);
876 /* Print out the free pages at each order for each migatetype */
877 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
880 pg_data_t *pgdat = (pg_data_t *)arg;
883 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
884 for (order = 0; order < MAX_ORDER; ++order)
885 seq_printf(m, "%6d ", order);
888 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
893 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
894 pg_data_t *pgdat, struct zone *zone)
898 unsigned long start_pfn = zone->zone_start_pfn;
899 unsigned long end_pfn = zone_end_pfn(zone);
900 unsigned long count[MIGRATE_TYPES] = { 0, };
902 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
908 page = pfn_to_page(pfn);
910 /* Watch for unexpected holes punched in the memmap */
911 if (!memmap_valid_within(pfn, page, zone))
914 mtype = get_pageblock_migratetype(page);
916 if (mtype < MIGRATE_TYPES)
921 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
922 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
923 seq_printf(m, "%12lu ", count[mtype]);
927 /* Print out the free pages at each order for each migratetype */
928 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
931 pg_data_t *pgdat = (pg_data_t *)arg;
933 seq_printf(m, "\n%-23s", "Number of blocks type ");
934 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
935 seq_printf(m, "%12s ", migratetype_names[mtype]);
937 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
943 * This prints out statistics in relation to grouping pages by mobility.
944 * It is expensive to collect so do not constantly read the file.
946 static int pagetypeinfo_show(struct seq_file *m, void *arg)
948 pg_data_t *pgdat = (pg_data_t *)arg;
950 /* check memoryless node */
951 if (!node_state(pgdat->node_id, N_MEMORY))
954 seq_printf(m, "Page block order: %d\n", pageblock_order);
955 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
957 pagetypeinfo_showfree(m, pgdat);
958 pagetypeinfo_showblockcount(m, pgdat);
963 static const struct seq_operations fragmentation_op = {
970 static int fragmentation_open(struct inode *inode, struct file *file)
972 return seq_open(file, &fragmentation_op);
975 static const struct file_operations fragmentation_file_operations = {
976 .open = fragmentation_open,
979 .release = seq_release,
982 static const struct seq_operations pagetypeinfo_op = {
986 .show = pagetypeinfo_show,
989 static int pagetypeinfo_open(struct inode *inode, struct file *file)
991 return seq_open(file, &pagetypeinfo_op);
994 static const struct file_operations pagetypeinfo_file_ops = {
995 .open = pagetypeinfo_open,
998 .release = seq_release,
1001 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1005 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1015 zone_page_state(zone, NR_FREE_PAGES),
1016 min_wmark_pages(zone),
1017 low_wmark_pages(zone),
1018 high_wmark_pages(zone),
1019 zone->pages_scanned,
1020 zone->spanned_pages,
1021 zone->present_pages,
1022 zone->managed_pages);
1024 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1025 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1026 zone_page_state(zone, i));
1029 "\n protection: (%lu",
1030 zone->lowmem_reserve[0]);
1031 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1032 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
1036 for_each_online_cpu(i) {
1037 struct per_cpu_pageset *pageset;
1039 pageset = per_cpu_ptr(zone->pageset, i);
1048 pageset->pcp.batch);
1050 seq_printf(m, "\n vm stats threshold: %d",
1051 pageset->stat_threshold);
1055 "\n all_unreclaimable: %u"
1057 "\n inactive_ratio: %u",
1058 !zone_reclaimable(zone),
1059 zone->zone_start_pfn,
1060 zone->inactive_ratio);
1065 * Output information about zones in @pgdat.
1067 static int zoneinfo_show(struct seq_file *m, void *arg)
1069 pg_data_t *pgdat = (pg_data_t *)arg;
1070 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
1074 static const struct seq_operations zoneinfo_op = {
1075 .start = frag_start, /* iterate over all zones. The same as in
1079 .show = zoneinfo_show,
1082 static int zoneinfo_open(struct inode *inode, struct file *file)
1084 return seq_open(file, &zoneinfo_op);
1087 static const struct file_operations proc_zoneinfo_file_operations = {
1088 .open = zoneinfo_open,
1090 .llseek = seq_lseek,
1091 .release = seq_release,
1094 enum writeback_stat_item {
1096 NR_DIRTY_BG_THRESHOLD,
1097 NR_VM_WRITEBACK_STAT_ITEMS,
1100 static void *vmstat_start(struct seq_file *m, loff_t *pos)
1103 int i, stat_items_size;
1105 if (*pos >= ARRAY_SIZE(vmstat_text))
1107 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1108 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
1110 #ifdef CONFIG_VM_EVENT_COUNTERS
1111 stat_items_size += sizeof(struct vm_event_state);
1114 v = kmalloc(stat_items_size, GFP_KERNEL);
1117 return ERR_PTR(-ENOMEM);
1118 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1119 v[i] = global_page_state(i);
1120 v += NR_VM_ZONE_STAT_ITEMS;
1122 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1123 v + NR_DIRTY_THRESHOLD);
1124 v += NR_VM_WRITEBACK_STAT_ITEMS;
1126 #ifdef CONFIG_VM_EVENT_COUNTERS
1128 v[PGPGIN] /= 2; /* sectors -> kbytes */
1131 return (unsigned long *)m->private + *pos;
1134 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1137 if (*pos >= ARRAY_SIZE(vmstat_text))
1139 return (unsigned long *)m->private + *pos;
1142 static int vmstat_show(struct seq_file *m, void *arg)
1144 unsigned long *l = arg;
1145 unsigned long off = l - (unsigned long *)m->private;
1147 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1151 static void vmstat_stop(struct seq_file *m, void *arg)
1157 static const struct seq_operations vmstat_op = {
1158 .start = vmstat_start,
1159 .next = vmstat_next,
1160 .stop = vmstat_stop,
1161 .show = vmstat_show,
1164 static int vmstat_open(struct inode *inode, struct file *file)
1166 return seq_open(file, &vmstat_op);
1169 static const struct file_operations proc_vmstat_file_operations = {
1170 .open = vmstat_open,
1172 .llseek = seq_lseek,
1173 .release = seq_release,
1175 #endif /* CONFIG_PROC_FS */
1178 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1179 int sysctl_stat_interval __read_mostly = HZ;
1181 static void vmstat_update(struct work_struct *w)
1183 refresh_cpu_vm_stats(smp_processor_id());
1184 schedule_delayed_work(&__get_cpu_var(vmstat_work),
1185 round_jiffies_relative(sysctl_stat_interval));
1188 static void __cpuinit start_cpu_timer(int cpu)
1190 struct delayed_work *work = &per_cpu(vmstat_work, cpu);
1192 INIT_DEFERRABLE_WORK(work, vmstat_update);
1193 schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
1197 * Use the cpu notifier to insure that the thresholds are recalculated
1200 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
1201 unsigned long action,
1204 long cpu = (long)hcpu;
1208 case CPU_ONLINE_FROZEN:
1209 refresh_zone_stat_thresholds();
1210 start_cpu_timer(cpu);
1211 node_set_state(cpu_to_node(cpu), N_CPU);
1213 case CPU_DOWN_PREPARE:
1214 case CPU_DOWN_PREPARE_FROZEN:
1215 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1216 per_cpu(vmstat_work, cpu).work.func = NULL;
1218 case CPU_DOWN_FAILED:
1219 case CPU_DOWN_FAILED_FROZEN:
1220 start_cpu_timer(cpu);
1223 case CPU_DEAD_FROZEN:
1224 refresh_zone_stat_thresholds();
1232 static struct notifier_block __cpuinitdata vmstat_notifier =
1233 { &vmstat_cpuup_callback, NULL, 0 };
1236 static int __init setup_vmstat(void)
1241 register_cpu_notifier(&vmstat_notifier);
1243 for_each_online_cpu(cpu)
1244 start_cpu_timer(cpu);
1246 #ifdef CONFIG_PROC_FS
1247 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1248 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1249 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1250 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1254 module_init(setup_vmstat)
1256 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1257 #include <linux/debugfs.h>
1261 * Return an index indicating how much of the available free memory is
1262 * unusable for an allocation of the requested size.
1264 static int unusable_free_index(unsigned int order,
1265 struct contig_page_info *info)
1267 /* No free memory is interpreted as all free memory is unusable */
1268 if (info->free_pages == 0)
1272 * Index should be a value between 0 and 1. Return a value to 3
1275 * 0 => no fragmentation
1276 * 1 => high fragmentation
1278 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1282 static void unusable_show_print(struct seq_file *m,
1283 pg_data_t *pgdat, struct zone *zone)
1287 struct contig_page_info info;
1289 seq_printf(m, "Node %d, zone %8s ",
1292 for (order = 0; order < MAX_ORDER; ++order) {
1293 fill_contig_page_info(zone, order, &info);
1294 index = unusable_free_index(order, &info);
1295 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1302 * Display unusable free space index
1304 * The unusable free space index measures how much of the available free
1305 * memory cannot be used to satisfy an allocation of a given size and is a
1306 * value between 0 and 1. The higher the value, the more of free memory is
1307 * unusable and by implication, the worse the external fragmentation is. This
1308 * can be expressed as a percentage by multiplying by 100.
1310 static int unusable_show(struct seq_file *m, void *arg)
1312 pg_data_t *pgdat = (pg_data_t *)arg;
1314 /* check memoryless node */
1315 if (!node_state(pgdat->node_id, N_MEMORY))
1318 walk_zones_in_node(m, pgdat, unusable_show_print);
1323 static const struct seq_operations unusable_op = {
1324 .start = frag_start,
1327 .show = unusable_show,
1330 static int unusable_open(struct inode *inode, struct file *file)
1332 return seq_open(file, &unusable_op);
1335 static const struct file_operations unusable_file_ops = {
1336 .open = unusable_open,
1338 .llseek = seq_lseek,
1339 .release = seq_release,
1342 static void extfrag_show_print(struct seq_file *m,
1343 pg_data_t *pgdat, struct zone *zone)
1348 /* Alloc on stack as interrupts are disabled for zone walk */
1349 struct contig_page_info info;
1351 seq_printf(m, "Node %d, zone %8s ",
1354 for (order = 0; order < MAX_ORDER; ++order) {
1355 fill_contig_page_info(zone, order, &info);
1356 index = __fragmentation_index(order, &info);
1357 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1364 * Display fragmentation index for orders that allocations would fail for
1366 static int extfrag_show(struct seq_file *m, void *arg)
1368 pg_data_t *pgdat = (pg_data_t *)arg;
1370 walk_zones_in_node(m, pgdat, extfrag_show_print);
1375 static const struct seq_operations extfrag_op = {
1376 .start = frag_start,
1379 .show = extfrag_show,
1382 static int extfrag_open(struct inode *inode, struct file *file)
1384 return seq_open(file, &extfrag_op);
1387 static const struct file_operations extfrag_file_ops = {
1388 .open = extfrag_open,
1390 .llseek = seq_lseek,
1391 .release = seq_release,
1394 static int __init extfrag_debug_init(void)
1396 struct dentry *extfrag_debug_root;
1398 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1399 if (!extfrag_debug_root)
1402 if (!debugfs_create_file("unusable_index", 0444,
1403 extfrag_debug_root, NULL, &unusable_file_ops))
1406 if (!debugfs_create_file("extfrag_index", 0444,
1407 extfrag_debug_root, NULL, &extfrag_file_ops))
1412 debugfs_remove_recursive(extfrag_debug_root);
1416 module_init(extfrag_debug_init);