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);
56 * Fold the foreign cpu events into our own.
58 * This is adding to the events on one processor
59 * but keeps the global counts constant.
61 void vm_events_fold_cpu(int cpu)
63 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
66 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
67 count_vm_events(i, fold_state->event[i]);
68 fold_state->event[i] = 0;
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] __cacheline_aligned_in_smp;
80 EXPORT_SYMBOL(vm_stat);
84 int calculate_pressure_threshold(struct zone *zone)
87 int watermark_distance;
90 * As vmstats are not up to date, there is drift between the estimated
91 * and real values. For high thresholds and a high number of CPUs, it
92 * is possible for the min watermark to be breached while the estimated
93 * value looks fine. The pressure threshold is a reduced value such
94 * that even the maximum amount of drift will not accidentally breach
97 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
98 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
101 * Maximum threshold is 125
103 threshold = min(125, threshold);
108 int calculate_normal_threshold(struct zone *zone)
111 int mem; /* memory in 128 MB units */
114 * The threshold scales with the number of processors and the amount
115 * of memory per zone. More memory means that we can defer updates for
116 * longer, more processors could lead to more contention.
117 * fls() is used to have a cheap way of logarithmic scaling.
119 * Some sample thresholds:
121 * Threshold Processors (fls) Zonesize fls(mem+1)
122 * ------------------------------------------------------------------
139 * 125 1024 10 8-16 GB 8
140 * 125 1024 10 16-32 GB 9
143 mem = zone->managed_pages >> (27 - PAGE_SHIFT);
145 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
148 * Maximum threshold is 125
150 threshold = min(125, threshold);
156 * Refresh the thresholds for each zone.
158 void refresh_zone_stat_thresholds(void)
164 for_each_populated_zone(zone) {
165 unsigned long max_drift, tolerate_drift;
167 threshold = calculate_normal_threshold(zone);
169 for_each_online_cpu(cpu)
170 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
174 * Only set percpu_drift_mark if there is a danger that
175 * NR_FREE_PAGES reports the low watermark is ok when in fact
176 * the min watermark could be breached by an allocation
178 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
179 max_drift = num_online_cpus() * threshold;
180 if (max_drift > tolerate_drift)
181 zone->percpu_drift_mark = high_wmark_pages(zone) +
186 void set_pgdat_percpu_threshold(pg_data_t *pgdat,
187 int (*calculate_pressure)(struct zone *))
194 for (i = 0; i < pgdat->nr_zones; i++) {
195 zone = &pgdat->node_zones[i];
196 if (!zone->percpu_drift_mark)
199 threshold = (*calculate_pressure)(zone);
200 for_each_possible_cpu(cpu)
201 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
207 * For use when we know that interrupts are disabled.
209 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
212 struct per_cpu_pageset __percpu *pcp = zone->pageset;
213 s8 __percpu *p = pcp->vm_stat_diff + item;
217 x = delta + __this_cpu_read(*p);
219 t = __this_cpu_read(pcp->stat_threshold);
221 if (unlikely(x > t || x < -t)) {
222 zone_page_state_add(x, zone, item);
225 __this_cpu_write(*p, x);
227 EXPORT_SYMBOL(__mod_zone_page_state);
230 * Optimized increment and decrement functions.
232 * These are only for a single page and therefore can take a struct page *
233 * argument instead of struct zone *. This allows the inclusion of the code
234 * generated for page_zone(page) into the optimized functions.
236 * No overflow check is necessary and therefore the differential can be
237 * incremented or decremented in place which may allow the compilers to
238 * generate better code.
239 * The increment or decrement is known and therefore one boundary check can
242 * NOTE: These functions are very performance sensitive. Change only
245 * Some processors have inc/dec instructions that are atomic vs an interrupt.
246 * However, the code must first determine the differential location in a zone
247 * based on the processor number and then inc/dec the counter. There is no
248 * guarantee without disabling preemption that the processor will not change
249 * in between and therefore the atomicity vs. interrupt cannot be exploited
250 * in a useful way here.
252 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
254 struct per_cpu_pageset __percpu *pcp = zone->pageset;
255 s8 __percpu *p = pcp->vm_stat_diff + item;
258 v = __this_cpu_inc_return(*p);
259 t = __this_cpu_read(pcp->stat_threshold);
260 if (unlikely(v > t)) {
261 s8 overstep = t >> 1;
263 zone_page_state_add(v + overstep, zone, item);
264 __this_cpu_write(*p, -overstep);
268 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
270 __inc_zone_state(page_zone(page), item);
272 EXPORT_SYMBOL(__inc_zone_page_state);
274 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
276 struct per_cpu_pageset __percpu *pcp = zone->pageset;
277 s8 __percpu *p = pcp->vm_stat_diff + item;
280 v = __this_cpu_dec_return(*p);
281 t = __this_cpu_read(pcp->stat_threshold);
282 if (unlikely(v < - t)) {
283 s8 overstep = t >> 1;
285 zone_page_state_add(v - overstep, zone, item);
286 __this_cpu_write(*p, overstep);
290 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
292 __dec_zone_state(page_zone(page), item);
294 EXPORT_SYMBOL(__dec_zone_page_state);
296 #ifdef CONFIG_HAVE_CMPXCHG_LOCAL
298 * If we have cmpxchg_local support then we do not need to incur the overhead
299 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
301 * mod_state() modifies the zone counter state through atomic per cpu
304 * Overstep mode specifies how overstep should handled:
306 * 1 Overstepping half of threshold
307 * -1 Overstepping minus half of threshold
309 static inline void mod_state(struct zone *zone,
310 enum zone_stat_item item, int delta, int overstep_mode)
312 struct per_cpu_pageset __percpu *pcp = zone->pageset;
313 s8 __percpu *p = pcp->vm_stat_diff + item;
317 z = 0; /* overflow to zone counters */
320 * The fetching of the stat_threshold is racy. We may apply
321 * a counter threshold to the wrong the cpu if we get
322 * rescheduled while executing here. However, the next
323 * counter update will apply the threshold again and
324 * therefore bring the counter under the threshold again.
326 * Most of the time the thresholds are the same anyways
327 * for all cpus in a zone.
329 t = this_cpu_read(pcp->stat_threshold);
331 o = this_cpu_read(*p);
334 if (n > t || n < -t) {
335 int os = overstep_mode * (t >> 1) ;
337 /* Overflow must be added to zone counters */
341 } while (this_cpu_cmpxchg(*p, o, n) != o);
344 zone_page_state_add(z, zone, item);
347 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
350 mod_state(zone, item, delta, 0);
352 EXPORT_SYMBOL(mod_zone_page_state);
354 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
356 mod_state(zone, item, 1, 1);
359 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
361 mod_state(page_zone(page), item, 1, 1);
363 EXPORT_SYMBOL(inc_zone_page_state);
365 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
367 mod_state(page_zone(page), item, -1, -1);
369 EXPORT_SYMBOL(dec_zone_page_state);
372 * Use interrupt disable to serialize counter updates
374 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
379 local_irq_save(flags);
380 __mod_zone_page_state(zone, item, delta);
381 local_irq_restore(flags);
383 EXPORT_SYMBOL(mod_zone_page_state);
385 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
389 local_irq_save(flags);
390 __inc_zone_state(zone, item);
391 local_irq_restore(flags);
394 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
399 zone = page_zone(page);
400 local_irq_save(flags);
401 __inc_zone_state(zone, item);
402 local_irq_restore(flags);
404 EXPORT_SYMBOL(inc_zone_page_state);
406 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
410 local_irq_save(flags);
411 __dec_zone_page_state(page, item);
412 local_irq_restore(flags);
414 EXPORT_SYMBOL(dec_zone_page_state);
418 * Update the zone counters for the current cpu.
420 * Note that refresh_cpu_vm_stats strives to only access
421 * node local memory. The per cpu pagesets on remote zones are placed
422 * in the memory local to the processor using that pageset. So the
423 * loop over all zones will access a series of cachelines local to
426 * The call to zone_page_state_add updates the cachelines with the
427 * statistics in the remote zone struct as well as the global cachelines
428 * with the global counters. These could cause remote node cache line
429 * bouncing and will have to be only done when necessary.
431 static void refresh_cpu_vm_stats(int cpu)
435 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
437 for_each_populated_zone(zone) {
438 struct per_cpu_pageset *p;
440 p = per_cpu_ptr(zone->pageset, cpu);
442 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
443 if (p->vm_stat_diff[i]) {
447 local_irq_save(flags);
448 v = p->vm_stat_diff[i];
449 p->vm_stat_diff[i] = 0;
450 local_irq_restore(flags);
451 atomic_long_add(v, &zone->vm_stat[i]);
454 /* 3 seconds idle till flush */
461 * Deal with draining the remote pageset of this
464 * Check if there are pages remaining in this pageset
465 * if not then there is nothing to expire.
467 if (!p->expire || !p->pcp.count)
471 * We never drain zones local to this processor.
473 if (zone_to_nid(zone) == numa_node_id()) {
483 drain_zone_pages(zone, &p->pcp);
487 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
489 atomic_long_add(global_diff[i], &vm_stat[i]);
493 * Fold the data for an offline cpu into the global array.
494 * There cannot be any access by the offline cpu and therefore
495 * synchronization is simplified.
497 void cpu_vm_stats_fold(int cpu)
501 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
503 for_each_populated_zone(zone) {
504 struct per_cpu_pageset *p;
506 p = per_cpu_ptr(zone->pageset, cpu);
508 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
509 if (p->vm_stat_diff[i]) {
512 v = p->vm_stat_diff[i];
513 p->vm_stat_diff[i] = 0;
514 atomic_long_add(v, &zone->vm_stat[i]);
519 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
521 atomic_long_add(global_diff[i], &vm_stat[i]);
525 * this is only called if !populated_zone(zone), which implies no other users of
526 * pset->vm_stat_diff[] exsist.
528 void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
532 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
533 if (pset->vm_stat_diff[i]) {
534 int v = pset->vm_stat_diff[i];
535 pset->vm_stat_diff[i] = 0;
536 atomic_long_add(v, &zone->vm_stat[i]);
537 atomic_long_add(v, &vm_stat[i]);
544 * zonelist = the list of zones passed to the allocator
545 * z = the zone from which the allocation occurred.
547 * Must be called with interrupts disabled.
549 * When __GFP_OTHER_NODE is set assume the node of the preferred
550 * zone is the local node. This is useful for daemons who allocate
551 * memory on behalf of other processes.
553 void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags)
555 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
556 __inc_zone_state(z, NUMA_HIT);
558 __inc_zone_state(z, NUMA_MISS);
559 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
561 if (z->node == ((flags & __GFP_OTHER_NODE) ?
562 preferred_zone->node : numa_node_id()))
563 __inc_zone_state(z, NUMA_LOCAL);
565 __inc_zone_state(z, NUMA_OTHER);
569 #ifdef CONFIG_COMPACTION
571 struct contig_page_info {
572 unsigned long free_pages;
573 unsigned long free_blocks_total;
574 unsigned long free_blocks_suitable;
578 * Calculate the number of free pages in a zone, how many contiguous
579 * pages are free and how many are large enough to satisfy an allocation of
580 * the target size. Note that this function makes no attempt to estimate
581 * how many suitable free blocks there *might* be if MOVABLE pages were
582 * migrated. Calculating that is possible, but expensive and can be
583 * figured out from userspace
585 static void fill_contig_page_info(struct zone *zone,
586 unsigned int suitable_order,
587 struct contig_page_info *info)
591 info->free_pages = 0;
592 info->free_blocks_total = 0;
593 info->free_blocks_suitable = 0;
595 for (order = 0; order < MAX_ORDER; order++) {
596 unsigned long blocks;
598 /* Count number of free blocks */
599 blocks = zone->free_area[order].nr_free;
600 info->free_blocks_total += blocks;
602 /* Count free base pages */
603 info->free_pages += blocks << order;
605 /* Count the suitable free blocks */
606 if (order >= suitable_order)
607 info->free_blocks_suitable += blocks <<
608 (order - suitable_order);
613 * A fragmentation index only makes sense if an allocation of a requested
614 * size would fail. If that is true, the fragmentation index indicates
615 * whether external fragmentation or a lack of memory was the problem.
616 * The value can be used to determine if page reclaim or compaction
619 static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
621 unsigned long requested = 1UL << order;
623 if (!info->free_blocks_total)
626 /* Fragmentation index only makes sense when a request would fail */
627 if (info->free_blocks_suitable)
631 * Index is between 0 and 1 so return within 3 decimal places
633 * 0 => allocation would fail due to lack of memory
634 * 1 => allocation would fail due to fragmentation
636 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
639 /* Same as __fragmentation index but allocs contig_page_info on stack */
640 int fragmentation_index(struct zone *zone, unsigned int order)
642 struct contig_page_info info;
644 fill_contig_page_info(zone, order, &info);
645 return __fragmentation_index(order, &info);
649 #if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
650 #include <linux/proc_fs.h>
651 #include <linux/seq_file.h>
653 static char * const migratetype_names[MIGRATE_TYPES] = {
661 #ifdef CONFIG_MEMORY_ISOLATION
666 static void *frag_start(struct seq_file *m, loff_t *pos)
670 for (pgdat = first_online_pgdat();
672 pgdat = next_online_pgdat(pgdat))
678 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
680 pg_data_t *pgdat = (pg_data_t *)arg;
683 return next_online_pgdat(pgdat);
686 static void frag_stop(struct seq_file *m, void *arg)
690 /* Walk all the zones in a node and print using a callback */
691 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
692 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
695 struct zone *node_zones = pgdat->node_zones;
698 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
699 if (!populated_zone(zone))
702 spin_lock_irqsave(&zone->lock, flags);
703 print(m, pgdat, zone);
704 spin_unlock_irqrestore(&zone->lock, flags);
709 #if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
710 #ifdef CONFIG_ZONE_DMA
711 #define TEXT_FOR_DMA(xx) xx "_dma",
713 #define TEXT_FOR_DMA(xx)
716 #ifdef CONFIG_ZONE_DMA32
717 #define TEXT_FOR_DMA32(xx) xx "_dma32",
719 #define TEXT_FOR_DMA32(xx)
722 #ifdef CONFIG_HIGHMEM
723 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
725 #define TEXT_FOR_HIGHMEM(xx)
728 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
729 TEXT_FOR_HIGHMEM(xx) xx "_movable",
731 const char * const vmstat_text[] = {
732 /* Zoned VM counters */
746 "nr_slab_reclaimable",
747 "nr_slab_unreclaimable",
748 "nr_page_table_pages",
753 "nr_vmscan_immediate_reclaim",
769 "nr_anon_transparent_hugepages",
771 "nr_dirty_threshold",
772 "nr_dirty_background_threshold",
774 #ifdef CONFIG_VM_EVENT_COUNTERS
780 TEXTS_FOR_ZONES("pgalloc")
789 TEXTS_FOR_ZONES("pgrefill")
790 TEXTS_FOR_ZONES("pgsteal_kswapd")
791 TEXTS_FOR_ZONES("pgsteal_direct")
792 TEXTS_FOR_ZONES("pgscan_kswapd")
793 TEXTS_FOR_ZONES("pgscan_direct")
794 "pgscan_direct_throttle",
797 "zone_reclaim_failed",
802 "kswapd_low_wmark_hit_quickly",
803 "kswapd_high_wmark_hit_quickly",
809 #ifdef CONFIG_NUMA_BALANCING
812 "numa_hint_faults_local",
813 "numa_pages_migrated",
815 #ifdef CONFIG_MIGRATION
819 #ifdef CONFIG_COMPACTION
820 "compact_migrate_scanned",
821 "compact_free_scanned",
828 #ifdef CONFIG_HUGETLB_PAGE
829 "htlb_buddy_alloc_success",
830 "htlb_buddy_alloc_fail",
832 "unevictable_pgs_culled",
833 "unevictable_pgs_scanned",
834 "unevictable_pgs_rescued",
835 "unevictable_pgs_mlocked",
836 "unevictable_pgs_munlocked",
837 "unevictable_pgs_cleared",
838 "unevictable_pgs_stranded",
840 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
842 "thp_fault_fallback",
843 "thp_collapse_alloc",
844 "thp_collapse_alloc_failed",
846 "thp_zero_page_alloc",
847 "thp_zero_page_alloc_failed",
850 "nr_tlb_remote_flush",
851 "nr_tlb_remote_flush_received",
853 "nr_tlb_local_flush_all",
854 "nr_tlb_local_flush_one",
856 #endif /* CONFIG_VM_EVENTS_COUNTERS */
858 #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
861 #ifdef CONFIG_PROC_FS
862 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
867 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
868 for (order = 0; order < MAX_ORDER; ++order)
869 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
874 * This walks the free areas for each zone.
876 static int frag_show(struct seq_file *m, void *arg)
878 pg_data_t *pgdat = (pg_data_t *)arg;
879 walk_zones_in_node(m, pgdat, frag_show_print);
883 static void pagetypeinfo_showfree_print(struct seq_file *m,
884 pg_data_t *pgdat, struct zone *zone)
888 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
889 seq_printf(m, "Node %4d, zone %8s, type %12s ",
892 migratetype_names[mtype]);
893 for (order = 0; order < MAX_ORDER; ++order) {
894 unsigned long freecount = 0;
895 struct free_area *area;
896 struct list_head *curr;
898 area = &(zone->free_area[order]);
900 list_for_each(curr, &area->free_list[mtype])
902 seq_printf(m, "%6lu ", freecount);
908 /* Print out the free pages at each order for each migatetype */
909 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
912 pg_data_t *pgdat = (pg_data_t *)arg;
915 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
916 for (order = 0; order < MAX_ORDER; ++order)
917 seq_printf(m, "%6d ", order);
920 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
925 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
926 pg_data_t *pgdat, struct zone *zone)
930 unsigned long start_pfn = zone->zone_start_pfn;
931 unsigned long end_pfn = zone_end_pfn(zone);
932 unsigned long count[MIGRATE_TYPES] = { 0, };
934 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
940 page = pfn_to_page(pfn);
942 /* Watch for unexpected holes punched in the memmap */
943 if (!memmap_valid_within(pfn, page, zone))
946 mtype = get_pageblock_migratetype(page);
948 if (mtype < MIGRATE_TYPES)
953 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
954 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
955 seq_printf(m, "%12lu ", count[mtype]);
959 /* Print out the free pages at each order for each migratetype */
960 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
963 pg_data_t *pgdat = (pg_data_t *)arg;
965 seq_printf(m, "\n%-23s", "Number of blocks type ");
966 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
967 seq_printf(m, "%12s ", migratetype_names[mtype]);
969 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
975 * This prints out statistics in relation to grouping pages by mobility.
976 * It is expensive to collect so do not constantly read the file.
978 static int pagetypeinfo_show(struct seq_file *m, void *arg)
980 pg_data_t *pgdat = (pg_data_t *)arg;
982 /* check memoryless node */
983 if (!node_state(pgdat->node_id, N_MEMORY))
986 seq_printf(m, "Page block order: %d\n", pageblock_order);
987 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
989 pagetypeinfo_showfree(m, pgdat);
990 pagetypeinfo_showblockcount(m, pgdat);
995 static const struct seq_operations fragmentation_op = {
1002 static int fragmentation_open(struct inode *inode, struct file *file)
1004 return seq_open(file, &fragmentation_op);
1007 static const struct file_operations fragmentation_file_operations = {
1008 .open = fragmentation_open,
1010 .llseek = seq_lseek,
1011 .release = seq_release,
1014 static const struct seq_operations pagetypeinfo_op = {
1015 .start = frag_start,
1018 .show = pagetypeinfo_show,
1021 static int pagetypeinfo_open(struct inode *inode, struct file *file)
1023 return seq_open(file, &pagetypeinfo_op);
1026 static const struct file_operations pagetypeinfo_file_ops = {
1027 .open = pagetypeinfo_open,
1029 .llseek = seq_lseek,
1030 .release = seq_release,
1033 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1037 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1047 zone_page_state(zone, NR_FREE_PAGES),
1048 min_wmark_pages(zone),
1049 low_wmark_pages(zone),
1050 high_wmark_pages(zone),
1051 zone->pages_scanned,
1052 zone->spanned_pages,
1053 zone->present_pages,
1054 zone->managed_pages);
1056 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1057 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1058 zone_page_state(zone, i));
1061 "\n protection: (%lu",
1062 zone->lowmem_reserve[0]);
1063 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1064 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
1068 for_each_online_cpu(i) {
1069 struct per_cpu_pageset *pageset;
1071 pageset = per_cpu_ptr(zone->pageset, i);
1080 pageset->pcp.batch);
1082 seq_printf(m, "\n vm stats threshold: %d",
1083 pageset->stat_threshold);
1087 "\n all_unreclaimable: %u"
1089 "\n inactive_ratio: %u",
1090 zone->all_unreclaimable,
1091 zone->zone_start_pfn,
1092 zone->inactive_ratio);
1097 * Output information about zones in @pgdat.
1099 static int zoneinfo_show(struct seq_file *m, void *arg)
1101 pg_data_t *pgdat = (pg_data_t *)arg;
1102 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
1106 static const struct seq_operations zoneinfo_op = {
1107 .start = frag_start, /* iterate over all zones. The same as in
1111 .show = zoneinfo_show,
1114 static int zoneinfo_open(struct inode *inode, struct file *file)
1116 return seq_open(file, &zoneinfo_op);
1119 static const struct file_operations proc_zoneinfo_file_operations = {
1120 .open = zoneinfo_open,
1122 .llseek = seq_lseek,
1123 .release = seq_release,
1126 enum writeback_stat_item {
1128 NR_DIRTY_BG_THRESHOLD,
1129 NR_VM_WRITEBACK_STAT_ITEMS,
1132 static void *vmstat_start(struct seq_file *m, loff_t *pos)
1135 int i, stat_items_size;
1137 if (*pos >= ARRAY_SIZE(vmstat_text))
1139 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1140 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
1142 #ifdef CONFIG_VM_EVENT_COUNTERS
1143 stat_items_size += sizeof(struct vm_event_state);
1146 v = kmalloc(stat_items_size, GFP_KERNEL);
1149 return ERR_PTR(-ENOMEM);
1150 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1151 v[i] = global_page_state(i);
1152 v += NR_VM_ZONE_STAT_ITEMS;
1154 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1155 v + NR_DIRTY_THRESHOLD);
1156 v += NR_VM_WRITEBACK_STAT_ITEMS;
1158 #ifdef CONFIG_VM_EVENT_COUNTERS
1160 v[PGPGIN] /= 2; /* sectors -> kbytes */
1163 return (unsigned long *)m->private + *pos;
1166 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1169 if (*pos >= ARRAY_SIZE(vmstat_text))
1171 return (unsigned long *)m->private + *pos;
1174 static int vmstat_show(struct seq_file *m, void *arg)
1176 unsigned long *l = arg;
1177 unsigned long off = l - (unsigned long *)m->private;
1179 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1183 static void vmstat_stop(struct seq_file *m, void *arg)
1189 static const struct seq_operations vmstat_op = {
1190 .start = vmstat_start,
1191 .next = vmstat_next,
1192 .stop = vmstat_stop,
1193 .show = vmstat_show,
1196 static int vmstat_open(struct inode *inode, struct file *file)
1198 return seq_open(file, &vmstat_op);
1201 static const struct file_operations proc_vmstat_file_operations = {
1202 .open = vmstat_open,
1204 .llseek = seq_lseek,
1205 .release = seq_release,
1207 #endif /* CONFIG_PROC_FS */
1210 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1211 int sysctl_stat_interval __read_mostly = HZ;
1213 static void vmstat_update(struct work_struct *w)
1215 refresh_cpu_vm_stats(smp_processor_id());
1216 schedule_delayed_work(&__get_cpu_var(vmstat_work),
1217 round_jiffies_relative(sysctl_stat_interval));
1220 static void start_cpu_timer(int cpu)
1222 struct delayed_work *work = &per_cpu(vmstat_work, cpu);
1224 INIT_DEFERRABLE_WORK(work, vmstat_update);
1225 schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
1229 * Use the cpu notifier to insure that the thresholds are recalculated
1232 static int vmstat_cpuup_callback(struct notifier_block *nfb,
1233 unsigned long action,
1236 long cpu = (long)hcpu;
1240 case CPU_ONLINE_FROZEN:
1241 refresh_zone_stat_thresholds();
1242 start_cpu_timer(cpu);
1243 node_set_state(cpu_to_node(cpu), N_CPU);
1245 case CPU_DOWN_PREPARE:
1246 case CPU_DOWN_PREPARE_FROZEN:
1247 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1248 per_cpu(vmstat_work, cpu).work.func = NULL;
1250 case CPU_DOWN_FAILED:
1251 case CPU_DOWN_FAILED_FROZEN:
1252 start_cpu_timer(cpu);
1255 case CPU_DEAD_FROZEN:
1256 refresh_zone_stat_thresholds();
1264 static struct notifier_block vmstat_notifier =
1265 { &vmstat_cpuup_callback, NULL, 0 };
1268 static int __init setup_vmstat(void)
1273 register_cpu_notifier(&vmstat_notifier);
1275 for_each_online_cpu(cpu)
1276 start_cpu_timer(cpu);
1278 #ifdef CONFIG_PROC_FS
1279 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1280 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1281 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1282 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1286 module_init(setup_vmstat)
1288 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1289 #include <linux/debugfs.h>
1293 * Return an index indicating how much of the available free memory is
1294 * unusable for an allocation of the requested size.
1296 static int unusable_free_index(unsigned int order,
1297 struct contig_page_info *info)
1299 /* No free memory is interpreted as all free memory is unusable */
1300 if (info->free_pages == 0)
1304 * Index should be a value between 0 and 1. Return a value to 3
1307 * 0 => no fragmentation
1308 * 1 => high fragmentation
1310 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1314 static void unusable_show_print(struct seq_file *m,
1315 pg_data_t *pgdat, struct zone *zone)
1319 struct contig_page_info info;
1321 seq_printf(m, "Node %d, zone %8s ",
1324 for (order = 0; order < MAX_ORDER; ++order) {
1325 fill_contig_page_info(zone, order, &info);
1326 index = unusable_free_index(order, &info);
1327 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1334 * Display unusable free space index
1336 * The unusable free space index measures how much of the available free
1337 * memory cannot be used to satisfy an allocation of a given size and is a
1338 * value between 0 and 1. The higher the value, the more of free memory is
1339 * unusable and by implication, the worse the external fragmentation is. This
1340 * can be expressed as a percentage by multiplying by 100.
1342 static int unusable_show(struct seq_file *m, void *arg)
1344 pg_data_t *pgdat = (pg_data_t *)arg;
1346 /* check memoryless node */
1347 if (!node_state(pgdat->node_id, N_MEMORY))
1350 walk_zones_in_node(m, pgdat, unusable_show_print);
1355 static const struct seq_operations unusable_op = {
1356 .start = frag_start,
1359 .show = unusable_show,
1362 static int unusable_open(struct inode *inode, struct file *file)
1364 return seq_open(file, &unusable_op);
1367 static const struct file_operations unusable_file_ops = {
1368 .open = unusable_open,
1370 .llseek = seq_lseek,
1371 .release = seq_release,
1374 static void extfrag_show_print(struct seq_file *m,
1375 pg_data_t *pgdat, struct zone *zone)
1380 /* Alloc on stack as interrupts are disabled for zone walk */
1381 struct contig_page_info info;
1383 seq_printf(m, "Node %d, zone %8s ",
1386 for (order = 0; order < MAX_ORDER; ++order) {
1387 fill_contig_page_info(zone, order, &info);
1388 index = __fragmentation_index(order, &info);
1389 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1396 * Display fragmentation index for orders that allocations would fail for
1398 static int extfrag_show(struct seq_file *m, void *arg)
1400 pg_data_t *pgdat = (pg_data_t *)arg;
1402 walk_zones_in_node(m, pgdat, extfrag_show_print);
1407 static const struct seq_operations extfrag_op = {
1408 .start = frag_start,
1411 .show = extfrag_show,
1414 static int extfrag_open(struct inode *inode, struct file *file)
1416 return seq_open(file, &extfrag_op);
1419 static const struct file_operations extfrag_file_ops = {
1420 .open = extfrag_open,
1422 .llseek = seq_lseek,
1423 .release = seq_release,
1426 static int __init extfrag_debug_init(void)
1428 struct dentry *extfrag_debug_root;
1430 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1431 if (!extfrag_debug_root)
1434 if (!debugfs_create_file("unusable_index", 0444,
1435 extfrag_debug_root, NULL, &unusable_file_ops))
1438 if (!debugfs_create_file("extfrag_index", 0444,
1439 extfrag_debug_root, NULL, &extfrag_file_ops))
1444 debugfs_remove_recursive(extfrag_debug_root);
1448 module_init(extfrag_debug_init);