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>
10 * Copyright (C) 2008-2014 Christoph Lameter
14 #include <linux/err.h>
15 #include <linux/module.h>
16 #include <linux/slab.h>
17 #include <linux/cpu.h>
18 #include <linux/cpumask.h>
19 #include <linux/vmstat.h>
20 #include <linux/proc_fs.h>
21 #include <linux/seq_file.h>
22 #include <linux/debugfs.h>
23 #include <linux/sched.h>
24 #include <linux/math64.h>
25 #include <linux/writeback.h>
26 #include <linux/compaction.h>
27 #include <linux/mm_inline.h>
28 #include <linux/page_ext.h>
29 #include <linux/page_owner.h>
33 #ifdef CONFIG_VM_EVENT_COUNTERS
34 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
35 EXPORT_PER_CPU_SYMBOL(vm_event_states);
37 static void sum_vm_events(unsigned long *ret)
42 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
44 for_each_online_cpu(cpu) {
45 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
47 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
48 ret[i] += this->event[i];
53 * Accumulate the vm event counters across all CPUs.
54 * The result is unavoidably approximate - it can change
55 * during and after execution of this function.
57 void all_vm_events(unsigned long *ret)
63 EXPORT_SYMBOL_GPL(all_vm_events);
66 * Fold the foreign cpu events into our own.
68 * This is adding to the events on one processor
69 * but keeps the global counts constant.
71 void vm_events_fold_cpu(int cpu)
73 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
76 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
77 count_vm_events(i, fold_state->event[i]);
78 fold_state->event[i] = 0;
82 #endif /* CONFIG_VM_EVENT_COUNTERS */
85 * Manage combined zone based / global counters
87 * vm_stat contains the global counters
89 atomic_long_t vm_zone_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
90 atomic_long_t vm_node_stat[NR_VM_NODE_STAT_ITEMS] __cacheline_aligned_in_smp;
91 EXPORT_SYMBOL(vm_zone_stat);
92 EXPORT_SYMBOL(vm_node_stat);
96 int calculate_pressure_threshold(struct zone *zone)
99 int watermark_distance;
102 * As vmstats are not up to date, there is drift between the estimated
103 * and real values. For high thresholds and a high number of CPUs, it
104 * is possible for the min watermark to be breached while the estimated
105 * value looks fine. The pressure threshold is a reduced value such
106 * that even the maximum amount of drift will not accidentally breach
109 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
110 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
113 * Maximum threshold is 125
115 threshold = min(125, threshold);
120 int calculate_normal_threshold(struct zone *zone)
123 int mem; /* memory in 128 MB units */
126 * The threshold scales with the number of processors and the amount
127 * of memory per zone. More memory means that we can defer updates for
128 * longer, more processors could lead to more contention.
129 * fls() is used to have a cheap way of logarithmic scaling.
131 * Some sample thresholds:
133 * Threshold Processors (fls) Zonesize fls(mem+1)
134 * ------------------------------------------------------------------
151 * 125 1024 10 8-16 GB 8
152 * 125 1024 10 16-32 GB 9
155 mem = zone->managed_pages >> (27 - PAGE_SHIFT);
157 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
160 * Maximum threshold is 125
162 threshold = min(125, threshold);
168 * Refresh the thresholds for each zone.
170 void refresh_zone_stat_thresholds(void)
172 struct pglist_data *pgdat;
177 /* Zero current pgdat thresholds */
178 for_each_online_pgdat(pgdat) {
179 for_each_online_cpu(cpu) {
180 per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold = 0;
184 for_each_populated_zone(zone) {
185 struct pglist_data *pgdat = zone->zone_pgdat;
186 unsigned long max_drift, tolerate_drift;
188 threshold = calculate_normal_threshold(zone);
190 for_each_online_cpu(cpu) {
193 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
196 /* Base nodestat threshold on the largest populated zone. */
197 pgdat_threshold = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold;
198 per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold
199 = max(threshold, pgdat_threshold);
203 * Only set percpu_drift_mark if there is a danger that
204 * NR_FREE_PAGES reports the low watermark is ok when in fact
205 * the min watermark could be breached by an allocation
207 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
208 max_drift = num_online_cpus() * threshold;
209 if (max_drift > tolerate_drift)
210 zone->percpu_drift_mark = high_wmark_pages(zone) +
215 void set_pgdat_percpu_threshold(pg_data_t *pgdat,
216 int (*calculate_pressure)(struct zone *))
223 for (i = 0; i < pgdat->nr_zones; i++) {
224 zone = &pgdat->node_zones[i];
225 if (!zone->percpu_drift_mark)
228 threshold = (*calculate_pressure)(zone);
229 for_each_online_cpu(cpu)
230 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
236 * For use when we know that interrupts are disabled,
237 * or when we know that preemption is disabled and that
238 * particular counter cannot be updated from interrupt context.
240 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
243 struct per_cpu_pageset __percpu *pcp = zone->pageset;
244 s8 __percpu *p = pcp->vm_stat_diff + item;
248 x = delta + __this_cpu_read(*p);
250 t = __this_cpu_read(pcp->stat_threshold);
252 if (unlikely(x > t || x < -t)) {
253 zone_page_state_add(x, zone, item);
256 __this_cpu_write(*p, x);
258 EXPORT_SYMBOL(__mod_zone_page_state);
260 void __mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
263 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
264 s8 __percpu *p = pcp->vm_node_stat_diff + item;
268 x = delta + __this_cpu_read(*p);
270 t = __this_cpu_read(pcp->stat_threshold);
272 if (unlikely(x > t || x < -t)) {
273 node_page_state_add(x, pgdat, item);
276 __this_cpu_write(*p, x);
278 EXPORT_SYMBOL(__mod_node_page_state);
281 * Optimized increment and decrement functions.
283 * These are only for a single page and therefore can take a struct page *
284 * argument instead of struct zone *. This allows the inclusion of the code
285 * generated for page_zone(page) into the optimized functions.
287 * No overflow check is necessary and therefore the differential can be
288 * incremented or decremented in place which may allow the compilers to
289 * generate better code.
290 * The increment or decrement is known and therefore one boundary check can
293 * NOTE: These functions are very performance sensitive. Change only
296 * Some processors have inc/dec instructions that are atomic vs an interrupt.
297 * However, the code must first determine the differential location in a zone
298 * based on the processor number and then inc/dec the counter. There is no
299 * guarantee without disabling preemption that the processor will not change
300 * in between and therefore the atomicity vs. interrupt cannot be exploited
301 * in a useful way here.
303 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
305 struct per_cpu_pageset __percpu *pcp = zone->pageset;
306 s8 __percpu *p = pcp->vm_stat_diff + item;
309 v = __this_cpu_inc_return(*p);
310 t = __this_cpu_read(pcp->stat_threshold);
311 if (unlikely(v > t)) {
312 s8 overstep = t >> 1;
314 zone_page_state_add(v + overstep, zone, item);
315 __this_cpu_write(*p, -overstep);
319 void __inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
321 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
322 s8 __percpu *p = pcp->vm_node_stat_diff + item;
325 v = __this_cpu_inc_return(*p);
326 t = __this_cpu_read(pcp->stat_threshold);
327 if (unlikely(v > t)) {
328 s8 overstep = t >> 1;
330 node_page_state_add(v + overstep, pgdat, item);
331 __this_cpu_write(*p, -overstep);
335 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
337 __inc_zone_state(page_zone(page), item);
339 EXPORT_SYMBOL(__inc_zone_page_state);
341 void __inc_node_page_state(struct page *page, enum node_stat_item item)
343 __inc_node_state(page_pgdat(page), item);
345 EXPORT_SYMBOL(__inc_node_page_state);
347 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
349 struct per_cpu_pageset __percpu *pcp = zone->pageset;
350 s8 __percpu *p = pcp->vm_stat_diff + item;
353 v = __this_cpu_dec_return(*p);
354 t = __this_cpu_read(pcp->stat_threshold);
355 if (unlikely(v < - t)) {
356 s8 overstep = t >> 1;
358 zone_page_state_add(v - overstep, zone, item);
359 __this_cpu_write(*p, overstep);
363 void __dec_node_state(struct pglist_data *pgdat, enum node_stat_item item)
365 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
366 s8 __percpu *p = pcp->vm_node_stat_diff + item;
369 v = __this_cpu_dec_return(*p);
370 t = __this_cpu_read(pcp->stat_threshold);
371 if (unlikely(v < - t)) {
372 s8 overstep = t >> 1;
374 node_page_state_add(v - overstep, pgdat, item);
375 __this_cpu_write(*p, overstep);
379 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
381 __dec_zone_state(page_zone(page), item);
383 EXPORT_SYMBOL(__dec_zone_page_state);
385 void __dec_node_page_state(struct page *page, enum node_stat_item item)
387 __dec_node_state(page_pgdat(page), item);
389 EXPORT_SYMBOL(__dec_node_page_state);
391 #ifdef CONFIG_HAVE_CMPXCHG_LOCAL
393 * If we have cmpxchg_local support then we do not need to incur the overhead
394 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
396 * mod_state() modifies the zone counter state through atomic per cpu
399 * Overstep mode specifies how overstep should handled:
401 * 1 Overstepping half of threshold
402 * -1 Overstepping minus half of threshold
404 static inline void mod_zone_state(struct zone *zone,
405 enum zone_stat_item item, long delta, int overstep_mode)
407 struct per_cpu_pageset __percpu *pcp = zone->pageset;
408 s8 __percpu *p = pcp->vm_stat_diff + item;
412 z = 0; /* overflow to zone counters */
415 * The fetching of the stat_threshold is racy. We may apply
416 * a counter threshold to the wrong the cpu if we get
417 * rescheduled while executing here. However, the next
418 * counter update will apply the threshold again and
419 * therefore bring the counter under the threshold again.
421 * Most of the time the thresholds are the same anyways
422 * for all cpus in a zone.
424 t = this_cpu_read(pcp->stat_threshold);
426 o = this_cpu_read(*p);
429 if (n > t || n < -t) {
430 int os = overstep_mode * (t >> 1) ;
432 /* Overflow must be added to zone counters */
436 } while (this_cpu_cmpxchg(*p, o, n) != o);
439 zone_page_state_add(z, zone, item);
442 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
445 mod_zone_state(zone, item, delta, 0);
447 EXPORT_SYMBOL(mod_zone_page_state);
449 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
451 mod_zone_state(zone, item, 1, 1);
454 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
456 mod_zone_state(page_zone(page), item, 1, 1);
458 EXPORT_SYMBOL(inc_zone_page_state);
460 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
462 mod_zone_state(page_zone(page), item, -1, -1);
464 EXPORT_SYMBOL(dec_zone_page_state);
466 static inline void mod_node_state(struct pglist_data *pgdat,
467 enum node_stat_item item, int delta, int overstep_mode)
469 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
470 s8 __percpu *p = pcp->vm_node_stat_diff + item;
474 z = 0; /* overflow to node counters */
477 * The fetching of the stat_threshold is racy. We may apply
478 * a counter threshold to the wrong the cpu if we get
479 * rescheduled while executing here. However, the next
480 * counter update will apply the threshold again and
481 * therefore bring the counter under the threshold again.
483 * Most of the time the thresholds are the same anyways
484 * for all cpus in a node.
486 t = this_cpu_read(pcp->stat_threshold);
488 o = this_cpu_read(*p);
491 if (n > t || n < -t) {
492 int os = overstep_mode * (t >> 1) ;
494 /* Overflow must be added to node counters */
498 } while (this_cpu_cmpxchg(*p, o, n) != o);
501 node_page_state_add(z, pgdat, item);
504 void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
507 mod_node_state(pgdat, item, delta, 0);
509 EXPORT_SYMBOL(mod_node_page_state);
511 void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
513 mod_node_state(pgdat, item, 1, 1);
516 void inc_node_page_state(struct page *page, enum node_stat_item item)
518 mod_node_state(page_pgdat(page), item, 1, 1);
520 EXPORT_SYMBOL(inc_node_page_state);
522 void dec_node_page_state(struct page *page, enum node_stat_item item)
524 mod_node_state(page_pgdat(page), item, -1, -1);
526 EXPORT_SYMBOL(dec_node_page_state);
529 * Use interrupt disable to serialize counter updates
531 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
536 local_irq_save(flags);
537 __mod_zone_page_state(zone, item, delta);
538 local_irq_restore(flags);
540 EXPORT_SYMBOL(mod_zone_page_state);
542 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
546 local_irq_save(flags);
547 __inc_zone_state(zone, item);
548 local_irq_restore(flags);
551 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
556 zone = page_zone(page);
557 local_irq_save(flags);
558 __inc_zone_state(zone, item);
559 local_irq_restore(flags);
561 EXPORT_SYMBOL(inc_zone_page_state);
563 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
567 local_irq_save(flags);
568 __dec_zone_page_state(page, item);
569 local_irq_restore(flags);
571 EXPORT_SYMBOL(dec_zone_page_state);
573 void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
577 local_irq_save(flags);
578 __inc_node_state(pgdat, item);
579 local_irq_restore(flags);
581 EXPORT_SYMBOL(inc_node_state);
583 void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
588 local_irq_save(flags);
589 __mod_node_page_state(pgdat, item, delta);
590 local_irq_restore(flags);
592 EXPORT_SYMBOL(mod_node_page_state);
594 void inc_node_page_state(struct page *page, enum node_stat_item item)
597 struct pglist_data *pgdat;
599 pgdat = page_pgdat(page);
600 local_irq_save(flags);
601 __inc_node_state(pgdat, item);
602 local_irq_restore(flags);
604 EXPORT_SYMBOL(inc_node_page_state);
606 void dec_node_page_state(struct page *page, enum node_stat_item item)
610 local_irq_save(flags);
611 __dec_node_page_state(page, item);
612 local_irq_restore(flags);
614 EXPORT_SYMBOL(dec_node_page_state);
618 * Fold a differential into the global counters.
619 * Returns the number of counters updated.
621 static int fold_diff(int *zone_diff, int *node_diff)
626 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
628 atomic_long_add(zone_diff[i], &vm_zone_stat[i]);
632 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
634 atomic_long_add(node_diff[i], &vm_node_stat[i]);
641 * Update the zone counters for the current cpu.
643 * Note that refresh_cpu_vm_stats strives to only access
644 * node local memory. The per cpu pagesets on remote zones are placed
645 * in the memory local to the processor using that pageset. So the
646 * loop over all zones will access a series of cachelines local to
649 * The call to zone_page_state_add updates the cachelines with the
650 * statistics in the remote zone struct as well as the global cachelines
651 * with the global counters. These could cause remote node cache line
652 * bouncing and will have to be only done when necessary.
654 * The function returns the number of global counters updated.
656 static int refresh_cpu_vm_stats(bool do_pagesets)
658 struct pglist_data *pgdat;
661 int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
662 int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
665 for_each_populated_zone(zone) {
666 struct per_cpu_pageset __percpu *p = zone->pageset;
668 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
671 v = this_cpu_xchg(p->vm_stat_diff[i], 0);
674 atomic_long_add(v, &zone->vm_stat[i]);
675 global_zone_diff[i] += v;
677 /* 3 seconds idle till flush */
678 __this_cpu_write(p->expire, 3);
686 * Deal with draining the remote pageset of this
689 * Check if there are pages remaining in this pageset
690 * if not then there is nothing to expire.
692 if (!__this_cpu_read(p->expire) ||
693 !__this_cpu_read(p->pcp.count))
697 * We never drain zones local to this processor.
699 if (zone_to_nid(zone) == numa_node_id()) {
700 __this_cpu_write(p->expire, 0);
704 if (__this_cpu_dec_return(p->expire))
707 if (__this_cpu_read(p->pcp.count)) {
708 drain_zone_pages(zone, this_cpu_ptr(&p->pcp));
715 for_each_online_pgdat(pgdat) {
716 struct per_cpu_nodestat __percpu *p = pgdat->per_cpu_nodestats;
718 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
721 v = this_cpu_xchg(p->vm_node_stat_diff[i], 0);
723 atomic_long_add(v, &pgdat->vm_stat[i]);
724 global_node_diff[i] += v;
729 changes += fold_diff(global_zone_diff, global_node_diff);
734 * Fold the data for an offline cpu into the global array.
735 * There cannot be any access by the offline cpu and therefore
736 * synchronization is simplified.
738 void cpu_vm_stats_fold(int cpu)
740 struct pglist_data *pgdat;
743 int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
744 int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
746 for_each_populated_zone(zone) {
747 struct per_cpu_pageset *p;
749 p = per_cpu_ptr(zone->pageset, cpu);
751 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
752 if (p->vm_stat_diff[i]) {
755 v = p->vm_stat_diff[i];
756 p->vm_stat_diff[i] = 0;
757 atomic_long_add(v, &zone->vm_stat[i]);
758 global_zone_diff[i] += v;
762 for_each_online_pgdat(pgdat) {
763 struct per_cpu_nodestat *p;
765 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
767 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
768 if (p->vm_node_stat_diff[i]) {
771 v = p->vm_node_stat_diff[i];
772 p->vm_node_stat_diff[i] = 0;
773 atomic_long_add(v, &pgdat->vm_stat[i]);
774 global_node_diff[i] += v;
778 fold_diff(global_zone_diff, global_node_diff);
782 * this is only called if !populated_zone(zone), which implies no other users of
783 * pset->vm_stat_diff[] exsist.
785 void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
789 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
790 if (pset->vm_stat_diff[i]) {
791 int v = pset->vm_stat_diff[i];
792 pset->vm_stat_diff[i] = 0;
793 atomic_long_add(v, &zone->vm_stat[i]);
794 atomic_long_add(v, &vm_zone_stat[i]);
801 * Determine the per node value of a stat item. This function
802 * is called frequently in a NUMA machine, so try to be as
803 * frugal as possible.
805 unsigned long sum_zone_node_page_state(int node,
806 enum zone_stat_item item)
808 struct zone *zones = NODE_DATA(node)->node_zones;
810 unsigned long count = 0;
812 for (i = 0; i < MAX_NR_ZONES; i++)
813 count += zone_page_state(zones + i, item);
819 * Determine the per node value of a stat item.
821 unsigned long node_page_state(struct pglist_data *pgdat,
822 enum node_stat_item item)
824 long x = atomic_long_read(&pgdat->vm_stat[item]);
833 #ifdef CONFIG_COMPACTION
835 struct contig_page_info {
836 unsigned long free_pages;
837 unsigned long free_blocks_total;
838 unsigned long free_blocks_suitable;
842 * Calculate the number of free pages in a zone, how many contiguous
843 * pages are free and how many are large enough to satisfy an allocation of
844 * the target size. Note that this function makes no attempt to estimate
845 * how many suitable free blocks there *might* be if MOVABLE pages were
846 * migrated. Calculating that is possible, but expensive and can be
847 * figured out from userspace
849 static void fill_contig_page_info(struct zone *zone,
850 unsigned int suitable_order,
851 struct contig_page_info *info)
855 info->free_pages = 0;
856 info->free_blocks_total = 0;
857 info->free_blocks_suitable = 0;
859 for (order = 0; order < MAX_ORDER; order++) {
860 unsigned long blocks;
862 /* Count number of free blocks */
863 blocks = zone->free_area[order].nr_free;
864 info->free_blocks_total += blocks;
866 /* Count free base pages */
867 info->free_pages += blocks << order;
869 /* Count the suitable free blocks */
870 if (order >= suitable_order)
871 info->free_blocks_suitable += blocks <<
872 (order - suitable_order);
877 * A fragmentation index only makes sense if an allocation of a requested
878 * size would fail. If that is true, the fragmentation index indicates
879 * whether external fragmentation or a lack of memory was the problem.
880 * The value can be used to determine if page reclaim or compaction
883 static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
885 unsigned long requested = 1UL << order;
887 if (!info->free_blocks_total)
890 /* Fragmentation index only makes sense when a request would fail */
891 if (info->free_blocks_suitable)
895 * Index is between 0 and 1 so return within 3 decimal places
897 * 0 => allocation would fail due to lack of memory
898 * 1 => allocation would fail due to fragmentation
900 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
903 /* Same as __fragmentation index but allocs contig_page_info on stack */
904 int fragmentation_index(struct zone *zone, unsigned int order)
906 struct contig_page_info info;
908 fill_contig_page_info(zone, order, &info);
909 return __fragmentation_index(order, &info);
913 #if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
914 #ifdef CONFIG_ZONE_DMA
915 #define TEXT_FOR_DMA(xx) xx "_dma",
917 #define TEXT_FOR_DMA(xx)
920 #ifdef CONFIG_ZONE_DMA32
921 #define TEXT_FOR_DMA32(xx) xx "_dma32",
923 #define TEXT_FOR_DMA32(xx)
926 #ifdef CONFIG_HIGHMEM
927 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
929 #define TEXT_FOR_HIGHMEM(xx)
932 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
933 TEXT_FOR_HIGHMEM(xx) xx "_movable",
935 const char * const vmstat_text[] = {
936 /* enum zone_stat_item countes */
950 "nr_slab_reclaimable",
951 "nr_slab_unreclaimable",
952 "nr_page_table_pages",
957 "nr_vmscan_immediate_reclaim",
965 #if IS_ENABLED(CONFIG_ZSMALLOC)
976 "workingset_refault",
977 "workingset_activate",
978 "workingset_nodereclaim",
979 "nr_anon_transparent_hugepages",
980 "nr_shmem_hugepages",
981 "nr_shmem_pmdmapped",
984 /* enum writeback_stat_item counters */
985 "nr_dirty_threshold",
986 "nr_dirty_background_threshold",
988 #ifdef CONFIG_VM_EVENT_COUNTERS
989 /* enum vm_event_item counters */
995 TEXTS_FOR_ZONES("pgalloc")
1005 TEXTS_FOR_ZONES("pgrefill")
1006 TEXTS_FOR_ZONES("pgsteal_kswapd")
1007 TEXTS_FOR_ZONES("pgsteal_direct")
1008 TEXTS_FOR_ZONES("pgscan_kswapd")
1009 TEXTS_FOR_ZONES("pgscan_direct")
1010 "pgscan_direct_throttle",
1013 "zone_reclaim_failed",
1017 "kswapd_inodesteal",
1018 "kswapd_low_wmark_hit_quickly",
1019 "kswapd_high_wmark_hit_quickly",
1028 #ifdef CONFIG_NUMA_BALANCING
1030 "numa_huge_pte_updates",
1032 "numa_hint_faults_local",
1033 "numa_pages_migrated",
1035 #ifdef CONFIG_MIGRATION
1036 "pgmigrate_success",
1039 #ifdef CONFIG_COMPACTION
1040 "compact_migrate_scanned",
1041 "compact_free_scanned",
1046 "compact_daemon_wake",
1049 #ifdef CONFIG_HUGETLB_PAGE
1050 "htlb_buddy_alloc_success",
1051 "htlb_buddy_alloc_fail",
1053 "unevictable_pgs_culled",
1054 "unevictable_pgs_scanned",
1055 "unevictable_pgs_rescued",
1056 "unevictable_pgs_mlocked",
1057 "unevictable_pgs_munlocked",
1058 "unevictable_pgs_cleared",
1059 "unevictable_pgs_stranded",
1061 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1063 "thp_fault_fallback",
1064 "thp_collapse_alloc",
1065 "thp_collapse_alloc_failed",
1069 "thp_split_page_failed",
1070 "thp_deferred_split_page",
1072 "thp_zero_page_alloc",
1073 "thp_zero_page_alloc_failed",
1075 #ifdef CONFIG_MEMORY_BALLOON
1078 #ifdef CONFIG_BALLOON_COMPACTION
1081 #endif /* CONFIG_MEMORY_BALLOON */
1082 #ifdef CONFIG_DEBUG_TLBFLUSH
1084 "nr_tlb_remote_flush",
1085 "nr_tlb_remote_flush_received",
1086 #endif /* CONFIG_SMP */
1087 "nr_tlb_local_flush_all",
1088 "nr_tlb_local_flush_one",
1089 #endif /* CONFIG_DEBUG_TLBFLUSH */
1091 #ifdef CONFIG_DEBUG_VM_VMACACHE
1092 "vmacache_find_calls",
1093 "vmacache_find_hits",
1094 "vmacache_full_flushes",
1096 #endif /* CONFIG_VM_EVENTS_COUNTERS */
1098 #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
1101 #if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
1102 defined(CONFIG_PROC_FS)
1103 static void *frag_start(struct seq_file *m, loff_t *pos)
1108 for (pgdat = first_online_pgdat();
1110 pgdat = next_online_pgdat(pgdat))
1116 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
1118 pg_data_t *pgdat = (pg_data_t *)arg;
1121 return next_online_pgdat(pgdat);
1124 static void frag_stop(struct seq_file *m, void *arg)
1128 /* Walk all the zones in a node and print using a callback */
1129 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
1130 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
1133 struct zone *node_zones = pgdat->node_zones;
1134 unsigned long flags;
1136 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
1137 if (!populated_zone(zone))
1140 spin_lock_irqsave(&zone->lock, flags);
1141 print(m, pgdat, zone);
1142 spin_unlock_irqrestore(&zone->lock, flags);
1147 #ifdef CONFIG_PROC_FS
1148 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
1153 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1154 for (order = 0; order < MAX_ORDER; ++order)
1155 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
1160 * This walks the free areas for each zone.
1162 static int frag_show(struct seq_file *m, void *arg)
1164 pg_data_t *pgdat = (pg_data_t *)arg;
1165 walk_zones_in_node(m, pgdat, frag_show_print);
1169 static void pagetypeinfo_showfree_print(struct seq_file *m,
1170 pg_data_t *pgdat, struct zone *zone)
1174 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
1175 seq_printf(m, "Node %4d, zone %8s, type %12s ",
1178 migratetype_names[mtype]);
1179 for (order = 0; order < MAX_ORDER; ++order) {
1180 unsigned long freecount = 0;
1181 struct free_area *area;
1182 struct list_head *curr;
1184 area = &(zone->free_area[order]);
1186 list_for_each(curr, &area->free_list[mtype])
1188 seq_printf(m, "%6lu ", freecount);
1194 /* Print out the free pages at each order for each migatetype */
1195 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
1198 pg_data_t *pgdat = (pg_data_t *)arg;
1201 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
1202 for (order = 0; order < MAX_ORDER; ++order)
1203 seq_printf(m, "%6d ", order);
1206 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
1211 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
1212 pg_data_t *pgdat, struct zone *zone)
1216 unsigned long start_pfn = zone->zone_start_pfn;
1217 unsigned long end_pfn = zone_end_pfn(zone);
1218 unsigned long count[MIGRATE_TYPES] = { 0, };
1220 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
1223 if (!pfn_valid(pfn))
1226 page = pfn_to_page(pfn);
1228 /* Watch for unexpected holes punched in the memmap */
1229 if (!memmap_valid_within(pfn, page, zone))
1232 if (page_zone(page) != zone)
1235 mtype = get_pageblock_migratetype(page);
1237 if (mtype < MIGRATE_TYPES)
1242 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1243 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1244 seq_printf(m, "%12lu ", count[mtype]);
1248 /* Print out the free pages at each order for each migratetype */
1249 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1252 pg_data_t *pgdat = (pg_data_t *)arg;
1254 seq_printf(m, "\n%-23s", "Number of blocks type ");
1255 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1256 seq_printf(m, "%12s ", migratetype_names[mtype]);
1258 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
1263 #ifdef CONFIG_PAGE_OWNER
1264 static void pagetypeinfo_showmixedcount_print(struct seq_file *m,
1269 struct page_ext *page_ext;
1270 unsigned long pfn = zone->zone_start_pfn, block_end_pfn;
1271 unsigned long end_pfn = pfn + zone->spanned_pages;
1272 unsigned long count[MIGRATE_TYPES] = { 0, };
1273 int pageblock_mt, page_mt;
1276 /* Scan block by block. First and last block may be incomplete */
1277 pfn = zone->zone_start_pfn;
1280 * Walk the zone in pageblock_nr_pages steps. If a page block spans
1281 * a zone boundary, it will be double counted between zones. This does
1282 * not matter as the mixed block count will still be correct
1284 for (; pfn < end_pfn; ) {
1285 if (!pfn_valid(pfn)) {
1286 pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
1290 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
1291 block_end_pfn = min(block_end_pfn, end_pfn);
1293 page = pfn_to_page(pfn);
1294 pageblock_mt = get_pageblock_migratetype(page);
1296 for (; pfn < block_end_pfn; pfn++) {
1297 if (!pfn_valid_within(pfn))
1300 page = pfn_to_page(pfn);
1302 if (page_zone(page) != zone)
1305 if (PageBuddy(page)) {
1306 pfn += (1UL << page_order(page)) - 1;
1310 if (PageReserved(page))
1313 page_ext = lookup_page_ext(page);
1314 if (unlikely(!page_ext))
1317 if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
1320 page_mt = gfpflags_to_migratetype(page_ext->gfp_mask);
1321 if (pageblock_mt != page_mt) {
1322 if (is_migrate_cma(pageblock_mt))
1323 count[MIGRATE_MOVABLE]++;
1325 count[pageblock_mt]++;
1327 pfn = block_end_pfn;
1330 pfn += (1UL << page_ext->order) - 1;
1335 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1336 for (i = 0; i < MIGRATE_TYPES; i++)
1337 seq_printf(m, "%12lu ", count[i]);
1340 #endif /* CONFIG_PAGE_OWNER */
1343 * Print out the number of pageblocks for each migratetype that contain pages
1344 * of other types. This gives an indication of how well fallbacks are being
1345 * contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1346 * to determine what is going on
1348 static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1350 #ifdef CONFIG_PAGE_OWNER
1353 if (!static_branch_unlikely(&page_owner_inited))
1356 drain_all_pages(NULL);
1358 seq_printf(m, "\n%-23s", "Number of mixed blocks ");
1359 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1360 seq_printf(m, "%12s ", migratetype_names[mtype]);
1363 walk_zones_in_node(m, pgdat, pagetypeinfo_showmixedcount_print);
1364 #endif /* CONFIG_PAGE_OWNER */
1368 * This prints out statistics in relation to grouping pages by mobility.
1369 * It is expensive to collect so do not constantly read the file.
1371 static int pagetypeinfo_show(struct seq_file *m, void *arg)
1373 pg_data_t *pgdat = (pg_data_t *)arg;
1375 /* check memoryless node */
1376 if (!node_state(pgdat->node_id, N_MEMORY))
1379 seq_printf(m, "Page block order: %d\n", pageblock_order);
1380 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
1382 pagetypeinfo_showfree(m, pgdat);
1383 pagetypeinfo_showblockcount(m, pgdat);
1384 pagetypeinfo_showmixedcount(m, pgdat);
1389 static const struct seq_operations fragmentation_op = {
1390 .start = frag_start,
1396 static int fragmentation_open(struct inode *inode, struct file *file)
1398 return seq_open(file, &fragmentation_op);
1401 static const struct file_operations fragmentation_file_operations = {
1402 .open = fragmentation_open,
1404 .llseek = seq_lseek,
1405 .release = seq_release,
1408 static const struct seq_operations pagetypeinfo_op = {
1409 .start = frag_start,
1412 .show = pagetypeinfo_show,
1415 static int pagetypeinfo_open(struct inode *inode, struct file *file)
1417 return seq_open(file, &pagetypeinfo_op);
1420 static const struct file_operations pagetypeinfo_file_ops = {
1421 .open = pagetypeinfo_open,
1423 .llseek = seq_lseek,
1424 .release = seq_release,
1427 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1431 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1441 zone_page_state(zone, NR_FREE_PAGES),
1442 min_wmark_pages(zone),
1443 low_wmark_pages(zone),
1444 high_wmark_pages(zone),
1445 zone_page_state(zone, NR_PAGES_SCANNED),
1446 zone->spanned_pages,
1447 zone->present_pages,
1448 zone->managed_pages);
1450 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1451 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1452 zone_page_state(zone, i));
1455 "\n protection: (%ld",
1456 zone->lowmem_reserve[0]);
1457 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1458 seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
1462 for_each_online_cpu(i) {
1463 struct per_cpu_pageset *pageset;
1465 pageset = per_cpu_ptr(zone->pageset, i);
1474 pageset->pcp.batch);
1476 seq_printf(m, "\n vm stats threshold: %d",
1477 pageset->stat_threshold);
1481 "\n all_unreclaimable: %u"
1483 "\n inactive_ratio: %u",
1484 !zone_reclaimable(zone),
1485 zone->zone_start_pfn,
1486 zone->inactive_ratio);
1491 * Output information about zones in @pgdat.
1493 static int zoneinfo_show(struct seq_file *m, void *arg)
1495 pg_data_t *pgdat = (pg_data_t *)arg;
1496 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
1500 static const struct seq_operations zoneinfo_op = {
1501 .start = frag_start, /* iterate over all zones. The same as in
1505 .show = zoneinfo_show,
1508 static int zoneinfo_open(struct inode *inode, struct file *file)
1510 return seq_open(file, &zoneinfo_op);
1513 static const struct file_operations proc_zoneinfo_file_operations = {
1514 .open = zoneinfo_open,
1516 .llseek = seq_lseek,
1517 .release = seq_release,
1520 enum writeback_stat_item {
1522 NR_DIRTY_BG_THRESHOLD,
1523 NR_VM_WRITEBACK_STAT_ITEMS,
1526 static void *vmstat_start(struct seq_file *m, loff_t *pos)
1529 int i, stat_items_size;
1531 if (*pos >= ARRAY_SIZE(vmstat_text))
1533 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1534 NR_VM_NODE_STAT_ITEMS * sizeof(unsigned long) +
1535 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
1537 #ifdef CONFIG_VM_EVENT_COUNTERS
1538 stat_items_size += sizeof(struct vm_event_state);
1541 v = kmalloc(stat_items_size, GFP_KERNEL);
1544 return ERR_PTR(-ENOMEM);
1545 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1546 v[i] = global_page_state(i);
1547 v += NR_VM_ZONE_STAT_ITEMS;
1549 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
1550 v[i] = global_node_page_state(i);
1551 v += NR_VM_NODE_STAT_ITEMS;
1553 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1554 v + NR_DIRTY_THRESHOLD);
1555 v += NR_VM_WRITEBACK_STAT_ITEMS;
1557 #ifdef CONFIG_VM_EVENT_COUNTERS
1559 v[PGPGIN] /= 2; /* sectors -> kbytes */
1562 return (unsigned long *)m->private + *pos;
1565 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1568 if (*pos >= ARRAY_SIZE(vmstat_text))
1570 return (unsigned long *)m->private + *pos;
1573 static int vmstat_show(struct seq_file *m, void *arg)
1575 unsigned long *l = arg;
1576 unsigned long off = l - (unsigned long *)m->private;
1578 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1582 static void vmstat_stop(struct seq_file *m, void *arg)
1588 static const struct seq_operations vmstat_op = {
1589 .start = vmstat_start,
1590 .next = vmstat_next,
1591 .stop = vmstat_stop,
1592 .show = vmstat_show,
1595 static int vmstat_open(struct inode *inode, struct file *file)
1597 return seq_open(file, &vmstat_op);
1600 static const struct file_operations proc_vmstat_file_operations = {
1601 .open = vmstat_open,
1603 .llseek = seq_lseek,
1604 .release = seq_release,
1606 #endif /* CONFIG_PROC_FS */
1609 static struct workqueue_struct *vmstat_wq;
1610 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1611 int sysctl_stat_interval __read_mostly = HZ;
1613 #ifdef CONFIG_PROC_FS
1614 static void refresh_vm_stats(struct work_struct *work)
1616 refresh_cpu_vm_stats(true);
1619 int vmstat_refresh(struct ctl_table *table, int write,
1620 void __user *buffer, size_t *lenp, loff_t *ppos)
1627 * The regular update, every sysctl_stat_interval, may come later
1628 * than expected: leaving a significant amount in per_cpu buckets.
1629 * This is particularly misleading when checking a quantity of HUGE
1630 * pages, immediately after running a test. /proc/sys/vm/stat_refresh,
1631 * which can equally be echo'ed to or cat'ted from (by root),
1632 * can be used to update the stats just before reading them.
1634 * Oh, and since global_page_state() etc. are so careful to hide
1635 * transiently negative values, report an error here if any of
1636 * the stats is negative, so we know to go looking for imbalance.
1638 err = schedule_on_each_cpu(refresh_vm_stats);
1641 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
1642 val = atomic_long_read(&vm_zone_stat[i]);
1645 case NR_ALLOC_BATCH:
1646 case NR_PAGES_SCANNED:
1648 * These are often seen to go negative in
1649 * recent kernels, but not to go permanently
1650 * negative. Whilst it would be nicer not to
1651 * have exceptions, rooting them out would be
1652 * another task, of rather low priority.
1656 pr_warn("%s: %s %ld\n",
1657 __func__, vmstat_text[i], val);
1671 #endif /* CONFIG_PROC_FS */
1673 static void vmstat_update(struct work_struct *w)
1675 if (refresh_cpu_vm_stats(true)) {
1677 * Counters were updated so we expect more updates
1678 * to occur in the future. Keep on running the
1679 * update worker thread.
1681 queue_delayed_work_on(smp_processor_id(), vmstat_wq,
1682 this_cpu_ptr(&vmstat_work),
1683 round_jiffies_relative(sysctl_stat_interval));
1688 * Switch off vmstat processing and then fold all the remaining differentials
1689 * until the diffs stay at zero. The function is used by NOHZ and can only be
1690 * invoked when tick processing is not active.
1693 * Check if the diffs for a certain cpu indicate that
1694 * an update is needed.
1696 static bool need_update(int cpu)
1700 for_each_populated_zone(zone) {
1701 struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu);
1703 BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1);
1705 * The fast way of checking if there are any vmstat diffs.
1706 * This works because the diffs are byte sized items.
1708 if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS))
1716 * Switch off vmstat processing and then fold all the remaining differentials
1717 * until the diffs stay at zero. The function is used by NOHZ and can only be
1718 * invoked when tick processing is not active.
1720 void quiet_vmstat(void)
1722 if (system_state != SYSTEM_RUNNING)
1725 if (!delayed_work_pending(this_cpu_ptr(&vmstat_work)))
1728 if (!need_update(smp_processor_id()))
1732 * Just refresh counters and do not care about the pending delayed
1733 * vmstat_update. It doesn't fire that often to matter and canceling
1734 * it would be too expensive from this path.
1735 * vmstat_shepherd will take care about that for us.
1737 refresh_cpu_vm_stats(false);
1741 * Shepherd worker thread that checks the
1742 * differentials of processors that have their worker
1743 * threads for vm statistics updates disabled because of
1746 static void vmstat_shepherd(struct work_struct *w);
1748 static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
1750 static void vmstat_shepherd(struct work_struct *w)
1755 /* Check processors whose vmstat worker threads have been disabled */
1756 for_each_online_cpu(cpu) {
1757 struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
1759 if (!delayed_work_pending(dw) && need_update(cpu))
1760 queue_delayed_work_on(cpu, vmstat_wq, dw, 0);
1764 schedule_delayed_work(&shepherd,
1765 round_jiffies_relative(sysctl_stat_interval));
1768 static void __init start_shepherd_timer(void)
1772 for_each_possible_cpu(cpu)
1773 INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
1776 vmstat_wq = alloc_workqueue("vmstat", WQ_FREEZABLE|WQ_MEM_RECLAIM, 0);
1777 schedule_delayed_work(&shepherd,
1778 round_jiffies_relative(sysctl_stat_interval));
1781 static void vmstat_cpu_dead(int node)
1786 for_each_online_cpu(cpu)
1787 if (cpu_to_node(cpu) == node)
1790 node_clear_state(node, N_CPU);
1796 * Use the cpu notifier to insure that the thresholds are recalculated
1799 static int vmstat_cpuup_callback(struct notifier_block *nfb,
1800 unsigned long action,
1803 long cpu = (long)hcpu;
1807 case CPU_ONLINE_FROZEN:
1808 refresh_zone_stat_thresholds();
1809 node_set_state(cpu_to_node(cpu), N_CPU);
1811 case CPU_DOWN_PREPARE:
1812 case CPU_DOWN_PREPARE_FROZEN:
1813 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1815 case CPU_DOWN_FAILED:
1816 case CPU_DOWN_FAILED_FROZEN:
1819 case CPU_DEAD_FROZEN:
1820 refresh_zone_stat_thresholds();
1821 vmstat_cpu_dead(cpu_to_node(cpu));
1829 static struct notifier_block vmstat_notifier =
1830 { &vmstat_cpuup_callback, NULL, 0 };
1833 static int __init setup_vmstat(void)
1836 cpu_notifier_register_begin();
1837 __register_cpu_notifier(&vmstat_notifier);
1839 start_shepherd_timer();
1840 cpu_notifier_register_done();
1842 #ifdef CONFIG_PROC_FS
1843 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1844 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1845 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1846 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1850 module_init(setup_vmstat)
1852 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1855 * Return an index indicating how much of the available free memory is
1856 * unusable for an allocation of the requested size.
1858 static int unusable_free_index(unsigned int order,
1859 struct contig_page_info *info)
1861 /* No free memory is interpreted as all free memory is unusable */
1862 if (info->free_pages == 0)
1866 * Index should be a value between 0 and 1. Return a value to 3
1869 * 0 => no fragmentation
1870 * 1 => high fragmentation
1872 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1876 static void unusable_show_print(struct seq_file *m,
1877 pg_data_t *pgdat, struct zone *zone)
1881 struct contig_page_info info;
1883 seq_printf(m, "Node %d, zone %8s ",
1886 for (order = 0; order < MAX_ORDER; ++order) {
1887 fill_contig_page_info(zone, order, &info);
1888 index = unusable_free_index(order, &info);
1889 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1896 * Display unusable free space index
1898 * The unusable free space index measures how much of the available free
1899 * memory cannot be used to satisfy an allocation of a given size and is a
1900 * value between 0 and 1. The higher the value, the more of free memory is
1901 * unusable and by implication, the worse the external fragmentation is. This
1902 * can be expressed as a percentage by multiplying by 100.
1904 static int unusable_show(struct seq_file *m, void *arg)
1906 pg_data_t *pgdat = (pg_data_t *)arg;
1908 /* check memoryless node */
1909 if (!node_state(pgdat->node_id, N_MEMORY))
1912 walk_zones_in_node(m, pgdat, unusable_show_print);
1917 static const struct seq_operations unusable_op = {
1918 .start = frag_start,
1921 .show = unusable_show,
1924 static int unusable_open(struct inode *inode, struct file *file)
1926 return seq_open(file, &unusable_op);
1929 static const struct file_operations unusable_file_ops = {
1930 .open = unusable_open,
1932 .llseek = seq_lseek,
1933 .release = seq_release,
1936 static void extfrag_show_print(struct seq_file *m,
1937 pg_data_t *pgdat, struct zone *zone)
1942 /* Alloc on stack as interrupts are disabled for zone walk */
1943 struct contig_page_info info;
1945 seq_printf(m, "Node %d, zone %8s ",
1948 for (order = 0; order < MAX_ORDER; ++order) {
1949 fill_contig_page_info(zone, order, &info);
1950 index = __fragmentation_index(order, &info);
1951 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1958 * Display fragmentation index for orders that allocations would fail for
1960 static int extfrag_show(struct seq_file *m, void *arg)
1962 pg_data_t *pgdat = (pg_data_t *)arg;
1964 walk_zones_in_node(m, pgdat, extfrag_show_print);
1969 static const struct seq_operations extfrag_op = {
1970 .start = frag_start,
1973 .show = extfrag_show,
1976 static int extfrag_open(struct inode *inode, struct file *file)
1978 return seq_open(file, &extfrag_op);
1981 static const struct file_operations extfrag_file_ops = {
1982 .open = extfrag_open,
1984 .llseek = seq_lseek,
1985 .release = seq_release,
1988 static int __init extfrag_debug_init(void)
1990 struct dentry *extfrag_debug_root;
1992 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1993 if (!extfrag_debug_root)
1996 if (!debugfs_create_file("unusable_index", 0444,
1997 extfrag_debug_root, NULL, &unusable_file_ops))
2000 if (!debugfs_create_file("extfrag_index", 0444,
2001 extfrag_debug_root, NULL, &extfrag_file_ops))
2006 debugfs_remove_recursive(extfrag_debug_root);
2010 module_init(extfrag_debug_init);