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);
417 static inline void fold_diff(int *diff)
421 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
423 atomic_long_add(diff[i], &vm_stat[i]);
427 * Update the zone counters for the current cpu.
429 * Note that refresh_cpu_vm_stats strives to only access
430 * node local memory. The per cpu pagesets on remote zones are placed
431 * in the memory local to the processor using that pageset. So the
432 * loop over all zones will access a series of cachelines local to
435 * The call to zone_page_state_add updates the cachelines with the
436 * statistics in the remote zone struct as well as the global cachelines
437 * with the global counters. These could cause remote node cache line
438 * bouncing and will have to be only done when necessary.
440 static void refresh_cpu_vm_stats(void)
444 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
446 for_each_populated_zone(zone) {
447 struct per_cpu_pageset __percpu *p = zone->pageset;
449 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
452 v = this_cpu_xchg(p->vm_stat_diff[i], 0);
455 atomic_long_add(v, &zone->vm_stat[i]);
458 /* 3 seconds idle till flush */
459 __this_cpu_write(p->expire, 3);
466 * Deal with draining the remote pageset of this
469 * Check if there are pages remaining in this pageset
470 * if not then there is nothing to expire.
472 if (!__this_cpu_read(p->expire) ||
473 !__this_cpu_read(p->pcp.count))
477 * We never drain zones local to this processor.
479 if (zone_to_nid(zone) == numa_node_id()) {
480 __this_cpu_write(p->expire, 0);
485 if (__this_cpu_dec_return(p->expire))
488 if (__this_cpu_read(p->pcp.count))
489 drain_zone_pages(zone, __this_cpu_ptr(&p->pcp));
492 fold_diff(global_diff);
496 * Fold the data for an offline cpu into the global array.
497 * There cannot be any access by the offline cpu and therefore
498 * synchronization is simplified.
500 void cpu_vm_stats_fold(int cpu)
504 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
506 for_each_populated_zone(zone) {
507 struct per_cpu_pageset *p;
509 p = per_cpu_ptr(zone->pageset, cpu);
511 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
512 if (p->vm_stat_diff[i]) {
515 v = p->vm_stat_diff[i];
516 p->vm_stat_diff[i] = 0;
517 atomic_long_add(v, &zone->vm_stat[i]);
522 fold_diff(global_diff);
526 * this is only called if !populated_zone(zone), which implies no other users of
527 * pset->vm_stat_diff[] exsist.
529 void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
533 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
534 if (pset->vm_stat_diff[i]) {
535 int v = pset->vm_stat_diff[i];
536 pset->vm_stat_diff[i] = 0;
537 atomic_long_add(v, &zone->vm_stat[i]);
538 atomic_long_add(v, &vm_stat[i]);
545 * zonelist = the list of zones passed to the allocator
546 * z = the zone from which the allocation occurred.
548 * Must be called with interrupts disabled.
550 * When __GFP_OTHER_NODE is set assume the node of the preferred
551 * zone is the local node. This is useful for daemons who allocate
552 * memory on behalf of other processes.
554 void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags)
556 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
557 __inc_zone_state(z, NUMA_HIT);
559 __inc_zone_state(z, NUMA_MISS);
560 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
562 if (z->node == ((flags & __GFP_OTHER_NODE) ?
563 preferred_zone->node : numa_node_id()))
564 __inc_zone_state(z, NUMA_LOCAL);
566 __inc_zone_state(z, NUMA_OTHER);
570 #ifdef CONFIG_COMPACTION
572 struct contig_page_info {
573 unsigned long free_pages;
574 unsigned long free_blocks_total;
575 unsigned long free_blocks_suitable;
579 * Calculate the number of free pages in a zone, how many contiguous
580 * pages are free and how many are large enough to satisfy an allocation of
581 * the target size. Note that this function makes no attempt to estimate
582 * how many suitable free blocks there *might* be if MOVABLE pages were
583 * migrated. Calculating that is possible, but expensive and can be
584 * figured out from userspace
586 static void fill_contig_page_info(struct zone *zone,
587 unsigned int suitable_order,
588 struct contig_page_info *info)
592 info->free_pages = 0;
593 info->free_blocks_total = 0;
594 info->free_blocks_suitable = 0;
596 for (order = 0; order < MAX_ORDER; order++) {
597 unsigned long blocks;
599 /* Count number of free blocks */
600 blocks = zone->free_area[order].nr_free;
601 info->free_blocks_total += blocks;
603 /* Count free base pages */
604 info->free_pages += blocks << order;
606 /* Count the suitable free blocks */
607 if (order >= suitable_order)
608 info->free_blocks_suitable += blocks <<
609 (order - suitable_order);
614 * A fragmentation index only makes sense if an allocation of a requested
615 * size would fail. If that is true, the fragmentation index indicates
616 * whether external fragmentation or a lack of memory was the problem.
617 * The value can be used to determine if page reclaim or compaction
620 static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
622 unsigned long requested = 1UL << order;
624 if (!info->free_blocks_total)
627 /* Fragmentation index only makes sense when a request would fail */
628 if (info->free_blocks_suitable)
632 * Index is between 0 and 1 so return within 3 decimal places
634 * 0 => allocation would fail due to lack of memory
635 * 1 => allocation would fail due to fragmentation
637 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
640 /* Same as __fragmentation index but allocs contig_page_info on stack */
641 int fragmentation_index(struct zone *zone, unsigned int order)
643 struct contig_page_info info;
645 fill_contig_page_info(zone, order, &info);
646 return __fragmentation_index(order, &info);
650 #if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
651 #include <linux/proc_fs.h>
652 #include <linux/seq_file.h>
654 static char * const migratetype_names[MIGRATE_TYPES] = {
662 #ifdef CONFIG_MEMORY_ISOLATION
667 static void *frag_start(struct seq_file *m, loff_t *pos)
671 for (pgdat = first_online_pgdat();
673 pgdat = next_online_pgdat(pgdat))
679 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
681 pg_data_t *pgdat = (pg_data_t *)arg;
684 return next_online_pgdat(pgdat);
687 static void frag_stop(struct seq_file *m, void *arg)
691 /* Walk all the zones in a node and print using a callback */
692 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
693 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
696 struct zone *node_zones = pgdat->node_zones;
699 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
700 if (!populated_zone(zone))
703 spin_lock_irqsave(&zone->lock, flags);
704 print(m, pgdat, zone);
705 spin_unlock_irqrestore(&zone->lock, flags);
710 #if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
711 #ifdef CONFIG_ZONE_DMA
712 #define TEXT_FOR_DMA(xx) xx "_dma",
714 #define TEXT_FOR_DMA(xx)
717 #ifdef CONFIG_ZONE_DMA32
718 #define TEXT_FOR_DMA32(xx) xx "_dma32",
720 #define TEXT_FOR_DMA32(xx)
723 #ifdef CONFIG_HIGHMEM
724 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
726 #define TEXT_FOR_HIGHMEM(xx)
729 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
730 TEXT_FOR_HIGHMEM(xx) xx "_movable",
732 const char * const vmstat_text[] = {
733 /* Zoned VM counters */
747 "nr_slab_reclaimable",
748 "nr_slab_unreclaimable",
749 "nr_page_table_pages",
754 "nr_vmscan_immediate_reclaim",
770 "nr_anon_transparent_hugepages",
772 "nr_dirty_threshold",
773 "nr_dirty_background_threshold",
775 #ifdef CONFIG_VM_EVENT_COUNTERS
781 TEXTS_FOR_ZONES("pgalloc")
790 TEXTS_FOR_ZONES("pgrefill")
791 TEXTS_FOR_ZONES("pgsteal_kswapd")
792 TEXTS_FOR_ZONES("pgsteal_direct")
793 TEXTS_FOR_ZONES("pgscan_kswapd")
794 TEXTS_FOR_ZONES("pgscan_direct")
795 "pgscan_direct_throttle",
798 "zone_reclaim_failed",
803 "kswapd_low_wmark_hit_quickly",
804 "kswapd_high_wmark_hit_quickly",
810 #ifdef CONFIG_NUMA_BALANCING
813 "numa_hint_faults_local",
814 "numa_pages_migrated",
816 #ifdef CONFIG_MIGRATION
820 #ifdef CONFIG_COMPACTION
821 "compact_migrate_scanned",
822 "compact_free_scanned",
829 #ifdef CONFIG_HUGETLB_PAGE
830 "htlb_buddy_alloc_success",
831 "htlb_buddy_alloc_fail",
833 "unevictable_pgs_culled",
834 "unevictable_pgs_scanned",
835 "unevictable_pgs_rescued",
836 "unevictable_pgs_mlocked",
837 "unevictable_pgs_munlocked",
838 "unevictable_pgs_cleared",
839 "unevictable_pgs_stranded",
841 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
843 "thp_fault_fallback",
844 "thp_collapse_alloc",
845 "thp_collapse_alloc_failed",
847 "thp_zero_page_alloc",
848 "thp_zero_page_alloc_failed",
851 "nr_tlb_remote_flush",
852 "nr_tlb_remote_flush_received",
854 "nr_tlb_local_flush_all",
855 "nr_tlb_local_flush_one",
857 #endif /* CONFIG_VM_EVENTS_COUNTERS */
859 #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
862 #ifdef CONFIG_PROC_FS
863 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
868 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
869 for (order = 0; order < MAX_ORDER; ++order)
870 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
875 * This walks the free areas for each zone.
877 static int frag_show(struct seq_file *m, void *arg)
879 pg_data_t *pgdat = (pg_data_t *)arg;
880 walk_zones_in_node(m, pgdat, frag_show_print);
884 static void pagetypeinfo_showfree_print(struct seq_file *m,
885 pg_data_t *pgdat, struct zone *zone)
889 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
890 seq_printf(m, "Node %4d, zone %8s, type %12s ",
893 migratetype_names[mtype]);
894 for (order = 0; order < MAX_ORDER; ++order) {
895 unsigned long freecount = 0;
896 struct free_area *area;
897 struct list_head *curr;
899 area = &(zone->free_area[order]);
901 list_for_each(curr, &area->free_list[mtype])
903 seq_printf(m, "%6lu ", freecount);
909 /* Print out the free pages at each order for each migatetype */
910 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
913 pg_data_t *pgdat = (pg_data_t *)arg;
916 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
917 for (order = 0; order < MAX_ORDER; ++order)
918 seq_printf(m, "%6d ", order);
921 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
926 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
927 pg_data_t *pgdat, struct zone *zone)
931 unsigned long start_pfn = zone->zone_start_pfn;
932 unsigned long end_pfn = zone_end_pfn(zone);
933 unsigned long count[MIGRATE_TYPES] = { 0, };
935 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
941 page = pfn_to_page(pfn);
943 /* Watch for unexpected holes punched in the memmap */
944 if (!memmap_valid_within(pfn, page, zone))
947 mtype = get_pageblock_migratetype(page);
949 if (mtype < MIGRATE_TYPES)
954 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
955 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
956 seq_printf(m, "%12lu ", count[mtype]);
960 /* Print out the free pages at each order for each migratetype */
961 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
964 pg_data_t *pgdat = (pg_data_t *)arg;
966 seq_printf(m, "\n%-23s", "Number of blocks type ");
967 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
968 seq_printf(m, "%12s ", migratetype_names[mtype]);
970 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
976 * This prints out statistics in relation to grouping pages by mobility.
977 * It is expensive to collect so do not constantly read the file.
979 static int pagetypeinfo_show(struct seq_file *m, void *arg)
981 pg_data_t *pgdat = (pg_data_t *)arg;
983 /* check memoryless node */
984 if (!node_state(pgdat->node_id, N_MEMORY))
987 seq_printf(m, "Page block order: %d\n", pageblock_order);
988 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
990 pagetypeinfo_showfree(m, pgdat);
991 pagetypeinfo_showblockcount(m, pgdat);
996 static const struct seq_operations fragmentation_op = {
1003 static int fragmentation_open(struct inode *inode, struct file *file)
1005 return seq_open(file, &fragmentation_op);
1008 static const struct file_operations fragmentation_file_operations = {
1009 .open = fragmentation_open,
1011 .llseek = seq_lseek,
1012 .release = seq_release,
1015 static const struct seq_operations pagetypeinfo_op = {
1016 .start = frag_start,
1019 .show = pagetypeinfo_show,
1022 static int pagetypeinfo_open(struct inode *inode, struct file *file)
1024 return seq_open(file, &pagetypeinfo_op);
1027 static const struct file_operations pagetypeinfo_file_ops = {
1028 .open = pagetypeinfo_open,
1030 .llseek = seq_lseek,
1031 .release = seq_release,
1034 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1038 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1048 zone_page_state(zone, NR_FREE_PAGES),
1049 min_wmark_pages(zone),
1050 low_wmark_pages(zone),
1051 high_wmark_pages(zone),
1052 zone->pages_scanned,
1053 zone->spanned_pages,
1054 zone->present_pages,
1055 zone->managed_pages);
1057 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1058 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1059 zone_page_state(zone, i));
1062 "\n protection: (%lu",
1063 zone->lowmem_reserve[0]);
1064 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1065 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
1069 for_each_online_cpu(i) {
1070 struct per_cpu_pageset *pageset;
1072 pageset = per_cpu_ptr(zone->pageset, i);
1081 pageset->pcp.batch);
1083 seq_printf(m, "\n vm stats threshold: %d",
1084 pageset->stat_threshold);
1088 "\n all_unreclaimable: %u"
1090 "\n inactive_ratio: %u",
1091 zone->all_unreclaimable,
1092 zone->zone_start_pfn,
1093 zone->inactive_ratio);
1098 * Output information about zones in @pgdat.
1100 static int zoneinfo_show(struct seq_file *m, void *arg)
1102 pg_data_t *pgdat = (pg_data_t *)arg;
1103 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
1107 static const struct seq_operations zoneinfo_op = {
1108 .start = frag_start, /* iterate over all zones. The same as in
1112 .show = zoneinfo_show,
1115 static int zoneinfo_open(struct inode *inode, struct file *file)
1117 return seq_open(file, &zoneinfo_op);
1120 static const struct file_operations proc_zoneinfo_file_operations = {
1121 .open = zoneinfo_open,
1123 .llseek = seq_lseek,
1124 .release = seq_release,
1127 enum writeback_stat_item {
1129 NR_DIRTY_BG_THRESHOLD,
1130 NR_VM_WRITEBACK_STAT_ITEMS,
1133 static void *vmstat_start(struct seq_file *m, loff_t *pos)
1136 int i, stat_items_size;
1138 if (*pos >= ARRAY_SIZE(vmstat_text))
1140 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1141 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
1143 #ifdef CONFIG_VM_EVENT_COUNTERS
1144 stat_items_size += sizeof(struct vm_event_state);
1147 v = kmalloc(stat_items_size, GFP_KERNEL);
1150 return ERR_PTR(-ENOMEM);
1151 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1152 v[i] = global_page_state(i);
1153 v += NR_VM_ZONE_STAT_ITEMS;
1155 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1156 v + NR_DIRTY_THRESHOLD);
1157 v += NR_VM_WRITEBACK_STAT_ITEMS;
1159 #ifdef CONFIG_VM_EVENT_COUNTERS
1161 v[PGPGIN] /= 2; /* sectors -> kbytes */
1164 return (unsigned long *)m->private + *pos;
1167 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1170 if (*pos >= ARRAY_SIZE(vmstat_text))
1172 return (unsigned long *)m->private + *pos;
1175 static int vmstat_show(struct seq_file *m, void *arg)
1177 unsigned long *l = arg;
1178 unsigned long off = l - (unsigned long *)m->private;
1180 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1184 static void vmstat_stop(struct seq_file *m, void *arg)
1190 static const struct seq_operations vmstat_op = {
1191 .start = vmstat_start,
1192 .next = vmstat_next,
1193 .stop = vmstat_stop,
1194 .show = vmstat_show,
1197 static int vmstat_open(struct inode *inode, struct file *file)
1199 return seq_open(file, &vmstat_op);
1202 static const struct file_operations proc_vmstat_file_operations = {
1203 .open = vmstat_open,
1205 .llseek = seq_lseek,
1206 .release = seq_release,
1208 #endif /* CONFIG_PROC_FS */
1211 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1212 int sysctl_stat_interval __read_mostly = HZ;
1214 static void vmstat_update(struct work_struct *w)
1216 refresh_cpu_vm_stats();
1217 schedule_delayed_work(&__get_cpu_var(vmstat_work),
1218 round_jiffies_relative(sysctl_stat_interval));
1221 static void start_cpu_timer(int cpu)
1223 struct delayed_work *work = &per_cpu(vmstat_work, cpu);
1225 INIT_DEFERRABLE_WORK(work, vmstat_update);
1226 schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
1230 * Use the cpu notifier to insure that the thresholds are recalculated
1233 static int vmstat_cpuup_callback(struct notifier_block *nfb,
1234 unsigned long action,
1237 long cpu = (long)hcpu;
1241 case CPU_ONLINE_FROZEN:
1242 refresh_zone_stat_thresholds();
1243 start_cpu_timer(cpu);
1244 node_set_state(cpu_to_node(cpu), N_CPU);
1246 case CPU_DOWN_PREPARE:
1247 case CPU_DOWN_PREPARE_FROZEN:
1248 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1249 per_cpu(vmstat_work, cpu).work.func = NULL;
1251 case CPU_DOWN_FAILED:
1252 case CPU_DOWN_FAILED_FROZEN:
1253 start_cpu_timer(cpu);
1256 case CPU_DEAD_FROZEN:
1257 refresh_zone_stat_thresholds();
1265 static struct notifier_block vmstat_notifier =
1266 { &vmstat_cpuup_callback, NULL, 0 };
1269 static int __init setup_vmstat(void)
1274 register_cpu_notifier(&vmstat_notifier);
1276 for_each_online_cpu(cpu)
1277 start_cpu_timer(cpu);
1279 #ifdef CONFIG_PROC_FS
1280 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1281 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1282 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1283 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1287 module_init(setup_vmstat)
1289 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1290 #include <linux/debugfs.h>
1294 * Return an index indicating how much of the available free memory is
1295 * unusable for an allocation of the requested size.
1297 static int unusable_free_index(unsigned int order,
1298 struct contig_page_info *info)
1300 /* No free memory is interpreted as all free memory is unusable */
1301 if (info->free_pages == 0)
1305 * Index should be a value between 0 and 1. Return a value to 3
1308 * 0 => no fragmentation
1309 * 1 => high fragmentation
1311 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1315 static void unusable_show_print(struct seq_file *m,
1316 pg_data_t *pgdat, struct zone *zone)
1320 struct contig_page_info info;
1322 seq_printf(m, "Node %d, zone %8s ",
1325 for (order = 0; order < MAX_ORDER; ++order) {
1326 fill_contig_page_info(zone, order, &info);
1327 index = unusable_free_index(order, &info);
1328 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1335 * Display unusable free space index
1337 * The unusable free space index measures how much of the available free
1338 * memory cannot be used to satisfy an allocation of a given size and is a
1339 * value between 0 and 1. The higher the value, the more of free memory is
1340 * unusable and by implication, the worse the external fragmentation is. This
1341 * can be expressed as a percentage by multiplying by 100.
1343 static int unusable_show(struct seq_file *m, void *arg)
1345 pg_data_t *pgdat = (pg_data_t *)arg;
1347 /* check memoryless node */
1348 if (!node_state(pgdat->node_id, N_MEMORY))
1351 walk_zones_in_node(m, pgdat, unusable_show_print);
1356 static const struct seq_operations unusable_op = {
1357 .start = frag_start,
1360 .show = unusable_show,
1363 static int unusable_open(struct inode *inode, struct file *file)
1365 return seq_open(file, &unusable_op);
1368 static const struct file_operations unusable_file_ops = {
1369 .open = unusable_open,
1371 .llseek = seq_lseek,
1372 .release = seq_release,
1375 static void extfrag_show_print(struct seq_file *m,
1376 pg_data_t *pgdat, struct zone *zone)
1381 /* Alloc on stack as interrupts are disabled for zone walk */
1382 struct contig_page_info info;
1384 seq_printf(m, "Node %d, zone %8s ",
1387 for (order = 0; order < MAX_ORDER; ++order) {
1388 fill_contig_page_info(zone, order, &info);
1389 index = __fragmentation_index(order, &info);
1390 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1397 * Display fragmentation index for orders that allocations would fail for
1399 static int extfrag_show(struct seq_file *m, void *arg)
1401 pg_data_t *pgdat = (pg_data_t *)arg;
1403 walk_zones_in_node(m, pgdat, extfrag_show_print);
1408 static const struct seq_operations extfrag_op = {
1409 .start = frag_start,
1412 .show = extfrag_show,
1415 static int extfrag_open(struct inode *inode, struct file *file)
1417 return seq_open(file, &extfrag_op);
1420 static const struct file_operations extfrag_file_ops = {
1421 .open = extfrag_open,
1423 .llseek = seq_lseek,
1424 .release = seq_release,
1427 static int __init extfrag_debug_init(void)
1429 struct dentry *extfrag_debug_root;
1431 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1432 if (!extfrag_debug_root)
1435 if (!debugfs_create_file("unusable_index", 0444,
1436 extfrag_debug_root, NULL, &unusable_file_ops))
1439 if (!debugfs_create_file("extfrag_index", 0444,
1440 extfrag_debug_root, NULL, &extfrag_file_ops))
1445 debugfs_remove_recursive(extfrag_debug_root);
1449 module_init(extfrag_debug_init);