thermal: fix Mediatek thermal controller build
[linux/fpc-iii.git] / mm / vmstat.c
blob5e43004828971e778a185ec69f2ef9710f085fb4
1 /*
2 * linux/mm/vmstat.c
4 * Manages VM statistics
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
7 * zoned VM statistics
8 * Copyright (C) 2006 Silicon Graphics, Inc.,
9 * Christoph Lameter <christoph@lameter.com>
10 * Copyright (C) 2008-2014 Christoph Lameter
12 #include <linux/fs.h>
13 #include <linux/mm.h>
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>
31 #include "internal.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)
39 int cpu;
40 int i;
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)
59 get_online_cpus();
60 sum_vm_events(ret);
61 put_online_cpus();
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);
74 int i;
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_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
90 EXPORT_SYMBOL(vm_stat);
92 #ifdef CONFIG_SMP
94 int calculate_pressure_threshold(struct zone *zone)
96 int threshold;
97 int watermark_distance;
100 * As vmstats are not up to date, there is drift between the estimated
101 * and real values. For high thresholds and a high number of CPUs, it
102 * is possible for the min watermark to be breached while the estimated
103 * value looks fine. The pressure threshold is a reduced value such
104 * that even the maximum amount of drift will not accidentally breach
105 * the min watermark
107 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
108 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
111 * Maximum threshold is 125
113 threshold = min(125, threshold);
115 return threshold;
118 int calculate_normal_threshold(struct zone *zone)
120 int threshold;
121 int mem; /* memory in 128 MB units */
124 * The threshold scales with the number of processors and the amount
125 * of memory per zone. More memory means that we can defer updates for
126 * longer, more processors could lead to more contention.
127 * fls() is used to have a cheap way of logarithmic scaling.
129 * Some sample thresholds:
131 * Threshold Processors (fls) Zonesize fls(mem+1)
132 * ------------------------------------------------------------------
133 * 8 1 1 0.9-1 GB 4
134 * 16 2 2 0.9-1 GB 4
135 * 20 2 2 1-2 GB 5
136 * 24 2 2 2-4 GB 6
137 * 28 2 2 4-8 GB 7
138 * 32 2 2 8-16 GB 8
139 * 4 2 2 <128M 1
140 * 30 4 3 2-4 GB 5
141 * 48 4 3 8-16 GB 8
142 * 32 8 4 1-2 GB 4
143 * 32 8 4 0.9-1GB 4
144 * 10 16 5 <128M 1
145 * 40 16 5 900M 4
146 * 70 64 7 2-4 GB 5
147 * 84 64 7 4-8 GB 6
148 * 108 512 9 4-8 GB 6
149 * 125 1024 10 8-16 GB 8
150 * 125 1024 10 16-32 GB 9
153 mem = zone->managed_pages >> (27 - PAGE_SHIFT);
155 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
158 * Maximum threshold is 125
160 threshold = min(125, threshold);
162 return threshold;
166 * Refresh the thresholds for each zone.
168 void refresh_zone_stat_thresholds(void)
170 struct zone *zone;
171 int cpu;
172 int threshold;
174 for_each_populated_zone(zone) {
175 unsigned long max_drift, tolerate_drift;
177 threshold = calculate_normal_threshold(zone);
179 for_each_online_cpu(cpu)
180 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
181 = threshold;
184 * Only set percpu_drift_mark if there is a danger that
185 * NR_FREE_PAGES reports the low watermark is ok when in fact
186 * the min watermark could be breached by an allocation
188 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
189 max_drift = num_online_cpus() * threshold;
190 if (max_drift > tolerate_drift)
191 zone->percpu_drift_mark = high_wmark_pages(zone) +
192 max_drift;
196 void set_pgdat_percpu_threshold(pg_data_t *pgdat,
197 int (*calculate_pressure)(struct zone *))
199 struct zone *zone;
200 int cpu;
201 int threshold;
202 int i;
204 for (i = 0; i < pgdat->nr_zones; i++) {
205 zone = &pgdat->node_zones[i];
206 if (!zone->percpu_drift_mark)
207 continue;
209 threshold = (*calculate_pressure)(zone);
210 for_each_online_cpu(cpu)
211 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
212 = threshold;
217 * For use when we know that interrupts are disabled,
218 * or when we know that preemption is disabled and that
219 * particular counter cannot be updated from interrupt context.
221 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
222 long delta)
224 struct per_cpu_pageset __percpu *pcp = zone->pageset;
225 s8 __percpu *p = pcp->vm_stat_diff + item;
226 long x;
227 long t;
229 x = delta + __this_cpu_read(*p);
231 t = __this_cpu_read(pcp->stat_threshold);
233 if (unlikely(x > t || x < -t)) {
234 zone_page_state_add(x, zone, item);
235 x = 0;
237 __this_cpu_write(*p, x);
239 EXPORT_SYMBOL(__mod_zone_page_state);
242 * Optimized increment and decrement functions.
244 * These are only for a single page and therefore can take a struct page *
245 * argument instead of struct zone *. This allows the inclusion of the code
246 * generated for page_zone(page) into the optimized functions.
248 * No overflow check is necessary and therefore the differential can be
249 * incremented or decremented in place which may allow the compilers to
250 * generate better code.
251 * The increment or decrement is known and therefore one boundary check can
252 * be omitted.
254 * NOTE: These functions are very performance sensitive. Change only
255 * with care.
257 * Some processors have inc/dec instructions that are atomic vs an interrupt.
258 * However, the code must first determine the differential location in a zone
259 * based on the processor number and then inc/dec the counter. There is no
260 * guarantee without disabling preemption that the processor will not change
261 * in between and therefore the atomicity vs. interrupt cannot be exploited
262 * in a useful way here.
264 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
266 struct per_cpu_pageset __percpu *pcp = zone->pageset;
267 s8 __percpu *p = pcp->vm_stat_diff + item;
268 s8 v, t;
270 v = __this_cpu_inc_return(*p);
271 t = __this_cpu_read(pcp->stat_threshold);
272 if (unlikely(v > t)) {
273 s8 overstep = t >> 1;
275 zone_page_state_add(v + overstep, zone, item);
276 __this_cpu_write(*p, -overstep);
280 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
282 __inc_zone_state(page_zone(page), item);
284 EXPORT_SYMBOL(__inc_zone_page_state);
286 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
288 struct per_cpu_pageset __percpu *pcp = zone->pageset;
289 s8 __percpu *p = pcp->vm_stat_diff + item;
290 s8 v, t;
292 v = __this_cpu_dec_return(*p);
293 t = __this_cpu_read(pcp->stat_threshold);
294 if (unlikely(v < - t)) {
295 s8 overstep = t >> 1;
297 zone_page_state_add(v - overstep, zone, item);
298 __this_cpu_write(*p, overstep);
302 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
304 __dec_zone_state(page_zone(page), item);
306 EXPORT_SYMBOL(__dec_zone_page_state);
308 #ifdef CONFIG_HAVE_CMPXCHG_LOCAL
310 * If we have cmpxchg_local support then we do not need to incur the overhead
311 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
313 * mod_state() modifies the zone counter state through atomic per cpu
314 * operations.
316 * Overstep mode specifies how overstep should handled:
317 * 0 No overstepping
318 * 1 Overstepping half of threshold
319 * -1 Overstepping minus half of threshold
321 static inline void mod_state(struct zone *zone, enum zone_stat_item item,
322 long delta, int overstep_mode)
324 struct per_cpu_pageset __percpu *pcp = zone->pageset;
325 s8 __percpu *p = pcp->vm_stat_diff + item;
326 long o, n, t, z;
328 do {
329 z = 0; /* overflow to zone counters */
332 * The fetching of the stat_threshold is racy. We may apply
333 * a counter threshold to the wrong the cpu if we get
334 * rescheduled while executing here. However, the next
335 * counter update will apply the threshold again and
336 * therefore bring the counter under the threshold again.
338 * Most of the time the thresholds are the same anyways
339 * for all cpus in a zone.
341 t = this_cpu_read(pcp->stat_threshold);
343 o = this_cpu_read(*p);
344 n = delta + o;
346 if (n > t || n < -t) {
347 int os = overstep_mode * (t >> 1) ;
349 /* Overflow must be added to zone counters */
350 z = n + os;
351 n = -os;
353 } while (this_cpu_cmpxchg(*p, o, n) != o);
355 if (z)
356 zone_page_state_add(z, zone, item);
359 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
360 long delta)
362 mod_state(zone, item, delta, 0);
364 EXPORT_SYMBOL(mod_zone_page_state);
366 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
368 mod_state(zone, item, 1, 1);
371 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
373 mod_state(page_zone(page), item, 1, 1);
375 EXPORT_SYMBOL(inc_zone_page_state);
377 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
379 mod_state(page_zone(page), item, -1, -1);
381 EXPORT_SYMBOL(dec_zone_page_state);
382 #else
384 * Use interrupt disable to serialize counter updates
386 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
387 long delta)
389 unsigned long flags;
391 local_irq_save(flags);
392 __mod_zone_page_state(zone, item, delta);
393 local_irq_restore(flags);
395 EXPORT_SYMBOL(mod_zone_page_state);
397 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
399 unsigned long flags;
401 local_irq_save(flags);
402 __inc_zone_state(zone, item);
403 local_irq_restore(flags);
406 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
408 unsigned long flags;
409 struct zone *zone;
411 zone = page_zone(page);
412 local_irq_save(flags);
413 __inc_zone_state(zone, item);
414 local_irq_restore(flags);
416 EXPORT_SYMBOL(inc_zone_page_state);
418 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
420 unsigned long flags;
422 local_irq_save(flags);
423 __dec_zone_page_state(page, item);
424 local_irq_restore(flags);
426 EXPORT_SYMBOL(dec_zone_page_state);
427 #endif
431 * Fold a differential into the global counters.
432 * Returns the number of counters updated.
434 static int fold_diff(int *diff)
436 int i;
437 int changes = 0;
439 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
440 if (diff[i]) {
441 atomic_long_add(diff[i], &vm_stat[i]);
442 changes++;
444 return changes;
448 * Update the zone counters for the current cpu.
450 * Note that refresh_cpu_vm_stats strives to only access
451 * node local memory. The per cpu pagesets on remote zones are placed
452 * in the memory local to the processor using that pageset. So the
453 * loop over all zones will access a series of cachelines local to
454 * the processor.
456 * The call to zone_page_state_add updates the cachelines with the
457 * statistics in the remote zone struct as well as the global cachelines
458 * with the global counters. These could cause remote node cache line
459 * bouncing and will have to be only done when necessary.
461 * The function returns the number of global counters updated.
463 static int refresh_cpu_vm_stats(bool do_pagesets)
465 struct zone *zone;
466 int i;
467 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
468 int changes = 0;
470 for_each_populated_zone(zone) {
471 struct per_cpu_pageset __percpu *p = zone->pageset;
473 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
474 int v;
476 v = this_cpu_xchg(p->vm_stat_diff[i], 0);
477 if (v) {
479 atomic_long_add(v, &zone->vm_stat[i]);
480 global_diff[i] += v;
481 #ifdef CONFIG_NUMA
482 /* 3 seconds idle till flush */
483 __this_cpu_write(p->expire, 3);
484 #endif
487 #ifdef CONFIG_NUMA
488 if (do_pagesets) {
489 cond_resched();
491 * Deal with draining the remote pageset of this
492 * processor
494 * Check if there are pages remaining in this pageset
495 * if not then there is nothing to expire.
497 if (!__this_cpu_read(p->expire) ||
498 !__this_cpu_read(p->pcp.count))
499 continue;
502 * We never drain zones local to this processor.
504 if (zone_to_nid(zone) == numa_node_id()) {
505 __this_cpu_write(p->expire, 0);
506 continue;
509 if (__this_cpu_dec_return(p->expire))
510 continue;
512 if (__this_cpu_read(p->pcp.count)) {
513 drain_zone_pages(zone, this_cpu_ptr(&p->pcp));
514 changes++;
517 #endif
519 changes += fold_diff(global_diff);
520 return changes;
524 * Fold the data for an offline cpu into the global array.
525 * There cannot be any access by the offline cpu and therefore
526 * synchronization is simplified.
528 void cpu_vm_stats_fold(int cpu)
530 struct zone *zone;
531 int i;
532 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
534 for_each_populated_zone(zone) {
535 struct per_cpu_pageset *p;
537 p = per_cpu_ptr(zone->pageset, cpu);
539 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
540 if (p->vm_stat_diff[i]) {
541 int v;
543 v = p->vm_stat_diff[i];
544 p->vm_stat_diff[i] = 0;
545 atomic_long_add(v, &zone->vm_stat[i]);
546 global_diff[i] += v;
550 fold_diff(global_diff);
554 * this is only called if !populated_zone(zone), which implies no other users of
555 * pset->vm_stat_diff[] exsist.
557 void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
559 int i;
561 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
562 if (pset->vm_stat_diff[i]) {
563 int v = pset->vm_stat_diff[i];
564 pset->vm_stat_diff[i] = 0;
565 atomic_long_add(v, &zone->vm_stat[i]);
566 atomic_long_add(v, &vm_stat[i]);
569 #endif
571 #ifdef CONFIG_NUMA
573 * zonelist = the list of zones passed to the allocator
574 * z = the zone from which the allocation occurred.
576 * Must be called with interrupts disabled.
578 * When __GFP_OTHER_NODE is set assume the node of the preferred
579 * zone is the local node. This is useful for daemons who allocate
580 * memory on behalf of other processes.
582 void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags)
584 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
585 __inc_zone_state(z, NUMA_HIT);
586 } else {
587 __inc_zone_state(z, NUMA_MISS);
588 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
590 if (z->node == ((flags & __GFP_OTHER_NODE) ?
591 preferred_zone->node : numa_node_id()))
592 __inc_zone_state(z, NUMA_LOCAL);
593 else
594 __inc_zone_state(z, NUMA_OTHER);
598 * Determine the per node value of a stat item.
600 unsigned long node_page_state(int node, enum zone_stat_item item)
602 struct zone *zones = NODE_DATA(node)->node_zones;
604 return
605 #ifdef CONFIG_ZONE_DMA
606 zone_page_state(&zones[ZONE_DMA], item) +
607 #endif
608 #ifdef CONFIG_ZONE_DMA32
609 zone_page_state(&zones[ZONE_DMA32], item) +
610 #endif
611 #ifdef CONFIG_HIGHMEM
612 zone_page_state(&zones[ZONE_HIGHMEM], item) +
613 #endif
614 zone_page_state(&zones[ZONE_NORMAL], item) +
615 zone_page_state(&zones[ZONE_MOVABLE], item);
618 #endif
620 #ifdef CONFIG_COMPACTION
622 struct contig_page_info {
623 unsigned long free_pages;
624 unsigned long free_blocks_total;
625 unsigned long free_blocks_suitable;
629 * Calculate the number of free pages in a zone, how many contiguous
630 * pages are free and how many are large enough to satisfy an allocation of
631 * the target size. Note that this function makes no attempt to estimate
632 * how many suitable free blocks there *might* be if MOVABLE pages were
633 * migrated. Calculating that is possible, but expensive and can be
634 * figured out from userspace
636 static void fill_contig_page_info(struct zone *zone,
637 unsigned int suitable_order,
638 struct contig_page_info *info)
640 unsigned int order;
642 info->free_pages = 0;
643 info->free_blocks_total = 0;
644 info->free_blocks_suitable = 0;
646 for (order = 0; order < MAX_ORDER; order++) {
647 unsigned long blocks;
649 /* Count number of free blocks */
650 blocks = zone->free_area[order].nr_free;
651 info->free_blocks_total += blocks;
653 /* Count free base pages */
654 info->free_pages += blocks << order;
656 /* Count the suitable free blocks */
657 if (order >= suitable_order)
658 info->free_blocks_suitable += blocks <<
659 (order - suitable_order);
664 * A fragmentation index only makes sense if an allocation of a requested
665 * size would fail. If that is true, the fragmentation index indicates
666 * whether external fragmentation or a lack of memory was the problem.
667 * The value can be used to determine if page reclaim or compaction
668 * should be used
670 static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
672 unsigned long requested = 1UL << order;
674 if (!info->free_blocks_total)
675 return 0;
677 /* Fragmentation index only makes sense when a request would fail */
678 if (info->free_blocks_suitable)
679 return -1000;
682 * Index is between 0 and 1 so return within 3 decimal places
684 * 0 => allocation would fail due to lack of memory
685 * 1 => allocation would fail due to fragmentation
687 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
690 /* Same as __fragmentation index but allocs contig_page_info on stack */
691 int fragmentation_index(struct zone *zone, unsigned int order)
693 struct contig_page_info info;
695 fill_contig_page_info(zone, order, &info);
696 return __fragmentation_index(order, &info);
698 #endif
700 #if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
701 #ifdef CONFIG_ZONE_DMA
702 #define TEXT_FOR_DMA(xx) xx "_dma",
703 #else
704 #define TEXT_FOR_DMA(xx)
705 #endif
707 #ifdef CONFIG_ZONE_DMA32
708 #define TEXT_FOR_DMA32(xx) xx "_dma32",
709 #else
710 #define TEXT_FOR_DMA32(xx)
711 #endif
713 #ifdef CONFIG_HIGHMEM
714 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
715 #else
716 #define TEXT_FOR_HIGHMEM(xx)
717 #endif
719 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
720 TEXT_FOR_HIGHMEM(xx) xx "_movable",
722 const char * const vmstat_text[] = {
723 /* enum zone_stat_item countes */
724 "nr_free_pages",
725 "nr_alloc_batch",
726 "nr_inactive_anon",
727 "nr_active_anon",
728 "nr_inactive_file",
729 "nr_active_file",
730 "nr_unevictable",
731 "nr_mlock",
732 "nr_anon_pages",
733 "nr_mapped",
734 "nr_file_pages",
735 "nr_dirty",
736 "nr_writeback",
737 "nr_slab_reclaimable",
738 "nr_slab_unreclaimable",
739 "nr_page_table_pages",
740 "nr_kernel_stack",
741 "nr_unstable",
742 "nr_bounce",
743 "nr_vmscan_write",
744 "nr_vmscan_immediate_reclaim",
745 "nr_writeback_temp",
746 "nr_isolated_anon",
747 "nr_isolated_file",
748 "nr_shmem",
749 "nr_dirtied",
750 "nr_written",
751 "nr_pages_scanned",
753 #ifdef CONFIG_NUMA
754 "numa_hit",
755 "numa_miss",
756 "numa_foreign",
757 "numa_interleave",
758 "numa_local",
759 "numa_other",
760 #endif
761 "workingset_refault",
762 "workingset_activate",
763 "workingset_nodereclaim",
764 "nr_anon_transparent_hugepages",
765 "nr_free_cma",
767 /* enum writeback_stat_item counters */
768 "nr_dirty_threshold",
769 "nr_dirty_background_threshold",
771 #ifdef CONFIG_VM_EVENT_COUNTERS
772 /* enum vm_event_item counters */
773 "pgpgin",
774 "pgpgout",
775 "pswpin",
776 "pswpout",
778 TEXTS_FOR_ZONES("pgalloc")
780 "pgfree",
781 "pgactivate",
782 "pgdeactivate",
784 "pgfault",
785 "pgmajfault",
786 "pglazyfreed",
788 TEXTS_FOR_ZONES("pgrefill")
789 TEXTS_FOR_ZONES("pgsteal_kswapd")
790 TEXTS_FOR_ZONES("pgsteal_direct")
791 TEXTS_FOR_ZONES("pgscan_kswapd")
792 TEXTS_FOR_ZONES("pgscan_direct")
793 "pgscan_direct_throttle",
795 #ifdef CONFIG_NUMA
796 "zone_reclaim_failed",
797 #endif
798 "pginodesteal",
799 "slabs_scanned",
800 "kswapd_inodesteal",
801 "kswapd_low_wmark_hit_quickly",
802 "kswapd_high_wmark_hit_quickly",
803 "pageoutrun",
804 "allocstall",
806 "pgrotated",
808 "drop_pagecache",
809 "drop_slab",
811 #ifdef CONFIG_NUMA_BALANCING
812 "numa_pte_updates",
813 "numa_huge_pte_updates",
814 "numa_hint_faults",
815 "numa_hint_faults_local",
816 "numa_pages_migrated",
817 #endif
818 #ifdef CONFIG_MIGRATION
819 "pgmigrate_success",
820 "pgmigrate_fail",
821 #endif
822 #ifdef CONFIG_COMPACTION
823 "compact_migrate_scanned",
824 "compact_free_scanned",
825 "compact_isolated",
826 "compact_stall",
827 "compact_fail",
828 "compact_success",
829 "compact_daemon_wake",
830 #endif
832 #ifdef CONFIG_HUGETLB_PAGE
833 "htlb_buddy_alloc_success",
834 "htlb_buddy_alloc_fail",
835 #endif
836 "unevictable_pgs_culled",
837 "unevictable_pgs_scanned",
838 "unevictable_pgs_rescued",
839 "unevictable_pgs_mlocked",
840 "unevictable_pgs_munlocked",
841 "unevictable_pgs_cleared",
842 "unevictable_pgs_stranded",
844 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
845 "thp_fault_alloc",
846 "thp_fault_fallback",
847 "thp_collapse_alloc",
848 "thp_collapse_alloc_failed",
849 "thp_split_page",
850 "thp_split_page_failed",
851 "thp_deferred_split_page",
852 "thp_split_pmd",
853 "thp_zero_page_alloc",
854 "thp_zero_page_alloc_failed",
855 #endif
856 #ifdef CONFIG_MEMORY_BALLOON
857 "balloon_inflate",
858 "balloon_deflate",
859 #ifdef CONFIG_BALLOON_COMPACTION
860 "balloon_migrate",
861 #endif
862 #endif /* CONFIG_MEMORY_BALLOON */
863 #ifdef CONFIG_DEBUG_TLBFLUSH
864 #ifdef CONFIG_SMP
865 "nr_tlb_remote_flush",
866 "nr_tlb_remote_flush_received",
867 #endif /* CONFIG_SMP */
868 "nr_tlb_local_flush_all",
869 "nr_tlb_local_flush_one",
870 #endif /* CONFIG_DEBUG_TLBFLUSH */
872 #ifdef CONFIG_DEBUG_VM_VMACACHE
873 "vmacache_find_calls",
874 "vmacache_find_hits",
875 "vmacache_full_flushes",
876 #endif
877 #endif /* CONFIG_VM_EVENTS_COUNTERS */
879 #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
882 #if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
883 defined(CONFIG_PROC_FS)
884 static void *frag_start(struct seq_file *m, loff_t *pos)
886 pg_data_t *pgdat;
887 loff_t node = *pos;
889 for (pgdat = first_online_pgdat();
890 pgdat && node;
891 pgdat = next_online_pgdat(pgdat))
892 --node;
894 return pgdat;
897 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
899 pg_data_t *pgdat = (pg_data_t *)arg;
901 (*pos)++;
902 return next_online_pgdat(pgdat);
905 static void frag_stop(struct seq_file *m, void *arg)
909 /* Walk all the zones in a node and print using a callback */
910 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
911 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
913 struct zone *zone;
914 struct zone *node_zones = pgdat->node_zones;
915 unsigned long flags;
917 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
918 if (!populated_zone(zone))
919 continue;
921 spin_lock_irqsave(&zone->lock, flags);
922 print(m, pgdat, zone);
923 spin_unlock_irqrestore(&zone->lock, flags);
926 #endif
928 #ifdef CONFIG_PROC_FS
929 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
930 struct zone *zone)
932 int order;
934 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
935 for (order = 0; order < MAX_ORDER; ++order)
936 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
937 seq_putc(m, '\n');
941 * This walks the free areas for each zone.
943 static int frag_show(struct seq_file *m, void *arg)
945 pg_data_t *pgdat = (pg_data_t *)arg;
946 walk_zones_in_node(m, pgdat, frag_show_print);
947 return 0;
950 static void pagetypeinfo_showfree_print(struct seq_file *m,
951 pg_data_t *pgdat, struct zone *zone)
953 int order, mtype;
955 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
956 seq_printf(m, "Node %4d, zone %8s, type %12s ",
957 pgdat->node_id,
958 zone->name,
959 migratetype_names[mtype]);
960 for (order = 0; order < MAX_ORDER; ++order) {
961 unsigned long freecount = 0;
962 struct free_area *area;
963 struct list_head *curr;
965 area = &(zone->free_area[order]);
967 list_for_each(curr, &area->free_list[mtype])
968 freecount++;
969 seq_printf(m, "%6lu ", freecount);
971 seq_putc(m, '\n');
975 /* Print out the free pages at each order for each migatetype */
976 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
978 int order;
979 pg_data_t *pgdat = (pg_data_t *)arg;
981 /* Print header */
982 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
983 for (order = 0; order < MAX_ORDER; ++order)
984 seq_printf(m, "%6d ", order);
985 seq_putc(m, '\n');
987 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
989 return 0;
992 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
993 pg_data_t *pgdat, struct zone *zone)
995 int mtype;
996 unsigned long pfn;
997 unsigned long start_pfn = zone->zone_start_pfn;
998 unsigned long end_pfn = zone_end_pfn(zone);
999 unsigned long count[MIGRATE_TYPES] = { 0, };
1001 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
1002 struct page *page;
1004 if (!pfn_valid(pfn))
1005 continue;
1007 page = pfn_to_page(pfn);
1009 /* Watch for unexpected holes punched in the memmap */
1010 if (!memmap_valid_within(pfn, page, zone))
1011 continue;
1013 mtype = get_pageblock_migratetype(page);
1015 if (mtype < MIGRATE_TYPES)
1016 count[mtype]++;
1019 /* Print counts */
1020 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1021 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1022 seq_printf(m, "%12lu ", count[mtype]);
1023 seq_putc(m, '\n');
1026 /* Print out the free pages at each order for each migratetype */
1027 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1029 int mtype;
1030 pg_data_t *pgdat = (pg_data_t *)arg;
1032 seq_printf(m, "\n%-23s", "Number of blocks type ");
1033 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1034 seq_printf(m, "%12s ", migratetype_names[mtype]);
1035 seq_putc(m, '\n');
1036 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
1038 return 0;
1041 #ifdef CONFIG_PAGE_OWNER
1042 static void pagetypeinfo_showmixedcount_print(struct seq_file *m,
1043 pg_data_t *pgdat,
1044 struct zone *zone)
1046 struct page *page;
1047 struct page_ext *page_ext;
1048 unsigned long pfn = zone->zone_start_pfn, block_end_pfn;
1049 unsigned long end_pfn = pfn + zone->spanned_pages;
1050 unsigned long count[MIGRATE_TYPES] = { 0, };
1051 int pageblock_mt, page_mt;
1052 int i;
1054 /* Scan block by block. First and last block may be incomplete */
1055 pfn = zone->zone_start_pfn;
1058 * Walk the zone in pageblock_nr_pages steps. If a page block spans
1059 * a zone boundary, it will be double counted between zones. This does
1060 * not matter as the mixed block count will still be correct
1062 for (; pfn < end_pfn; ) {
1063 if (!pfn_valid(pfn)) {
1064 pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
1065 continue;
1068 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
1069 block_end_pfn = min(block_end_pfn, end_pfn);
1071 page = pfn_to_page(pfn);
1072 pageblock_mt = get_pfnblock_migratetype(page, pfn);
1074 for (; pfn < block_end_pfn; pfn++) {
1075 if (!pfn_valid_within(pfn))
1076 continue;
1078 page = pfn_to_page(pfn);
1079 if (PageBuddy(page)) {
1080 pfn += (1UL << page_order(page)) - 1;
1081 continue;
1084 if (PageReserved(page))
1085 continue;
1087 page_ext = lookup_page_ext(page);
1089 if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
1090 continue;
1092 page_mt = gfpflags_to_migratetype(page_ext->gfp_mask);
1093 if (pageblock_mt != page_mt) {
1094 if (is_migrate_cma(pageblock_mt))
1095 count[MIGRATE_MOVABLE]++;
1096 else
1097 count[pageblock_mt]++;
1099 pfn = block_end_pfn;
1100 break;
1102 pfn += (1UL << page_ext->order) - 1;
1106 /* Print counts */
1107 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1108 for (i = 0; i < MIGRATE_TYPES; i++)
1109 seq_printf(m, "%12lu ", count[i]);
1110 seq_putc(m, '\n');
1112 #endif /* CONFIG_PAGE_OWNER */
1115 * Print out the number of pageblocks for each migratetype that contain pages
1116 * of other types. This gives an indication of how well fallbacks are being
1117 * contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1118 * to determine what is going on
1120 static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1122 #ifdef CONFIG_PAGE_OWNER
1123 int mtype;
1125 if (!static_branch_unlikely(&page_owner_inited))
1126 return;
1128 drain_all_pages(NULL);
1130 seq_printf(m, "\n%-23s", "Number of mixed blocks ");
1131 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1132 seq_printf(m, "%12s ", migratetype_names[mtype]);
1133 seq_putc(m, '\n');
1135 walk_zones_in_node(m, pgdat, pagetypeinfo_showmixedcount_print);
1136 #endif /* CONFIG_PAGE_OWNER */
1140 * This prints out statistics in relation to grouping pages by mobility.
1141 * It is expensive to collect so do not constantly read the file.
1143 static int pagetypeinfo_show(struct seq_file *m, void *arg)
1145 pg_data_t *pgdat = (pg_data_t *)arg;
1147 /* check memoryless node */
1148 if (!node_state(pgdat->node_id, N_MEMORY))
1149 return 0;
1151 seq_printf(m, "Page block order: %d\n", pageblock_order);
1152 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
1153 seq_putc(m, '\n');
1154 pagetypeinfo_showfree(m, pgdat);
1155 pagetypeinfo_showblockcount(m, pgdat);
1156 pagetypeinfo_showmixedcount(m, pgdat);
1158 return 0;
1161 static const struct seq_operations fragmentation_op = {
1162 .start = frag_start,
1163 .next = frag_next,
1164 .stop = frag_stop,
1165 .show = frag_show,
1168 static int fragmentation_open(struct inode *inode, struct file *file)
1170 return seq_open(file, &fragmentation_op);
1173 static const struct file_operations fragmentation_file_operations = {
1174 .open = fragmentation_open,
1175 .read = seq_read,
1176 .llseek = seq_lseek,
1177 .release = seq_release,
1180 static const struct seq_operations pagetypeinfo_op = {
1181 .start = frag_start,
1182 .next = frag_next,
1183 .stop = frag_stop,
1184 .show = pagetypeinfo_show,
1187 static int pagetypeinfo_open(struct inode *inode, struct file *file)
1189 return seq_open(file, &pagetypeinfo_op);
1192 static const struct file_operations pagetypeinfo_file_ops = {
1193 .open = pagetypeinfo_open,
1194 .read = seq_read,
1195 .llseek = seq_lseek,
1196 .release = seq_release,
1199 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1200 struct zone *zone)
1202 int i;
1203 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1204 seq_printf(m,
1205 "\n pages free %lu"
1206 "\n min %lu"
1207 "\n low %lu"
1208 "\n high %lu"
1209 "\n scanned %lu"
1210 "\n spanned %lu"
1211 "\n present %lu"
1212 "\n managed %lu",
1213 zone_page_state(zone, NR_FREE_PAGES),
1214 min_wmark_pages(zone),
1215 low_wmark_pages(zone),
1216 high_wmark_pages(zone),
1217 zone_page_state(zone, NR_PAGES_SCANNED),
1218 zone->spanned_pages,
1219 zone->present_pages,
1220 zone->managed_pages);
1222 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1223 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1224 zone_page_state(zone, i));
1226 seq_printf(m,
1227 "\n protection: (%ld",
1228 zone->lowmem_reserve[0]);
1229 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1230 seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
1231 seq_printf(m,
1233 "\n pagesets");
1234 for_each_online_cpu(i) {
1235 struct per_cpu_pageset *pageset;
1237 pageset = per_cpu_ptr(zone->pageset, i);
1238 seq_printf(m,
1239 "\n cpu: %i"
1240 "\n count: %i"
1241 "\n high: %i"
1242 "\n batch: %i",
1244 pageset->pcp.count,
1245 pageset->pcp.high,
1246 pageset->pcp.batch);
1247 #ifdef CONFIG_SMP
1248 seq_printf(m, "\n vm stats threshold: %d",
1249 pageset->stat_threshold);
1250 #endif
1252 seq_printf(m,
1253 "\n all_unreclaimable: %u"
1254 "\n start_pfn: %lu"
1255 "\n inactive_ratio: %u",
1256 !zone_reclaimable(zone),
1257 zone->zone_start_pfn,
1258 zone->inactive_ratio);
1259 seq_putc(m, '\n');
1263 * Output information about zones in @pgdat.
1265 static int zoneinfo_show(struct seq_file *m, void *arg)
1267 pg_data_t *pgdat = (pg_data_t *)arg;
1268 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
1269 return 0;
1272 static const struct seq_operations zoneinfo_op = {
1273 .start = frag_start, /* iterate over all zones. The same as in
1274 * fragmentation. */
1275 .next = frag_next,
1276 .stop = frag_stop,
1277 .show = zoneinfo_show,
1280 static int zoneinfo_open(struct inode *inode, struct file *file)
1282 return seq_open(file, &zoneinfo_op);
1285 static const struct file_operations proc_zoneinfo_file_operations = {
1286 .open = zoneinfo_open,
1287 .read = seq_read,
1288 .llseek = seq_lseek,
1289 .release = seq_release,
1292 enum writeback_stat_item {
1293 NR_DIRTY_THRESHOLD,
1294 NR_DIRTY_BG_THRESHOLD,
1295 NR_VM_WRITEBACK_STAT_ITEMS,
1298 static void *vmstat_start(struct seq_file *m, loff_t *pos)
1300 unsigned long *v;
1301 int i, stat_items_size;
1303 if (*pos >= ARRAY_SIZE(vmstat_text))
1304 return NULL;
1305 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1306 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
1308 #ifdef CONFIG_VM_EVENT_COUNTERS
1309 stat_items_size += sizeof(struct vm_event_state);
1310 #endif
1312 v = kmalloc(stat_items_size, GFP_KERNEL);
1313 m->private = v;
1314 if (!v)
1315 return ERR_PTR(-ENOMEM);
1316 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1317 v[i] = global_page_state(i);
1318 v += NR_VM_ZONE_STAT_ITEMS;
1320 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1321 v + NR_DIRTY_THRESHOLD);
1322 v += NR_VM_WRITEBACK_STAT_ITEMS;
1324 #ifdef CONFIG_VM_EVENT_COUNTERS
1325 all_vm_events(v);
1326 v[PGPGIN] /= 2; /* sectors -> kbytes */
1327 v[PGPGOUT] /= 2;
1328 #endif
1329 return (unsigned long *)m->private + *pos;
1332 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1334 (*pos)++;
1335 if (*pos >= ARRAY_SIZE(vmstat_text))
1336 return NULL;
1337 return (unsigned long *)m->private + *pos;
1340 static int vmstat_show(struct seq_file *m, void *arg)
1342 unsigned long *l = arg;
1343 unsigned long off = l - (unsigned long *)m->private;
1345 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1346 return 0;
1349 static void vmstat_stop(struct seq_file *m, void *arg)
1351 kfree(m->private);
1352 m->private = NULL;
1355 static const struct seq_operations vmstat_op = {
1356 .start = vmstat_start,
1357 .next = vmstat_next,
1358 .stop = vmstat_stop,
1359 .show = vmstat_show,
1362 static int vmstat_open(struct inode *inode, struct file *file)
1364 return seq_open(file, &vmstat_op);
1367 static const struct file_operations proc_vmstat_file_operations = {
1368 .open = vmstat_open,
1369 .read = seq_read,
1370 .llseek = seq_lseek,
1371 .release = seq_release,
1373 #endif /* CONFIG_PROC_FS */
1375 #ifdef CONFIG_SMP
1376 static struct workqueue_struct *vmstat_wq;
1377 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1378 int sysctl_stat_interval __read_mostly = HZ;
1379 static cpumask_var_t cpu_stat_off;
1381 static void vmstat_update(struct work_struct *w)
1383 if (refresh_cpu_vm_stats(true)) {
1385 * Counters were updated so we expect more updates
1386 * to occur in the future. Keep on running the
1387 * update worker thread.
1388 * If we were marked on cpu_stat_off clear the flag
1389 * so that vmstat_shepherd doesn't schedule us again.
1391 if (!cpumask_test_and_clear_cpu(smp_processor_id(),
1392 cpu_stat_off)) {
1393 queue_delayed_work_on(smp_processor_id(), vmstat_wq,
1394 this_cpu_ptr(&vmstat_work),
1395 round_jiffies_relative(sysctl_stat_interval));
1397 } else {
1399 * We did not update any counters so the app may be in
1400 * a mode where it does not cause counter updates.
1401 * We may be uselessly running vmstat_update.
1402 * Defer the checking for differentials to the
1403 * shepherd thread on a different processor.
1405 cpumask_set_cpu(smp_processor_id(), cpu_stat_off);
1410 * Switch off vmstat processing and then fold all the remaining differentials
1411 * until the diffs stay at zero. The function is used by NOHZ and can only be
1412 * invoked when tick processing is not active.
1415 * Check if the diffs for a certain cpu indicate that
1416 * an update is needed.
1418 static bool need_update(int cpu)
1420 struct zone *zone;
1422 for_each_populated_zone(zone) {
1423 struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu);
1425 BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1);
1427 * The fast way of checking if there are any vmstat diffs.
1428 * This works because the diffs are byte sized items.
1430 if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS))
1431 return true;
1434 return false;
1437 void quiet_vmstat(void)
1439 if (system_state != SYSTEM_RUNNING)
1440 return;
1443 * If we are already in hands of the shepherd then there
1444 * is nothing for us to do here.
1446 if (cpumask_test_and_set_cpu(smp_processor_id(), cpu_stat_off))
1447 return;
1449 if (!need_update(smp_processor_id()))
1450 return;
1453 * Just refresh counters and do not care about the pending delayed
1454 * vmstat_update. It doesn't fire that often to matter and canceling
1455 * it would be too expensive from this path.
1456 * vmstat_shepherd will take care about that for us.
1458 refresh_cpu_vm_stats(false);
1463 * Shepherd worker thread that checks the
1464 * differentials of processors that have their worker
1465 * threads for vm statistics updates disabled because of
1466 * inactivity.
1468 static void vmstat_shepherd(struct work_struct *w);
1470 static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
1472 static void vmstat_shepherd(struct work_struct *w)
1474 int cpu;
1476 get_online_cpus();
1477 /* Check processors whose vmstat worker threads have been disabled */
1478 for_each_cpu(cpu, cpu_stat_off) {
1479 struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
1481 if (need_update(cpu)) {
1482 if (cpumask_test_and_clear_cpu(cpu, cpu_stat_off))
1483 queue_delayed_work_on(cpu, vmstat_wq, dw, 0);
1484 } else {
1486 * Cancel the work if quiet_vmstat has put this
1487 * cpu on cpu_stat_off because the work item might
1488 * be still scheduled
1490 cancel_delayed_work(dw);
1493 put_online_cpus();
1495 schedule_delayed_work(&shepherd,
1496 round_jiffies_relative(sysctl_stat_interval));
1499 static void __init start_shepherd_timer(void)
1501 int cpu;
1503 for_each_possible_cpu(cpu)
1504 INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
1505 vmstat_update);
1507 if (!alloc_cpumask_var(&cpu_stat_off, GFP_KERNEL))
1508 BUG();
1509 cpumask_copy(cpu_stat_off, cpu_online_mask);
1511 vmstat_wq = alloc_workqueue("vmstat", WQ_FREEZABLE|WQ_MEM_RECLAIM, 0);
1512 schedule_delayed_work(&shepherd,
1513 round_jiffies_relative(sysctl_stat_interval));
1516 static void vmstat_cpu_dead(int node)
1518 int cpu;
1520 get_online_cpus();
1521 for_each_online_cpu(cpu)
1522 if (cpu_to_node(cpu) == node)
1523 goto end;
1525 node_clear_state(node, N_CPU);
1526 end:
1527 put_online_cpus();
1531 * Use the cpu notifier to insure that the thresholds are recalculated
1532 * when necessary.
1534 static int vmstat_cpuup_callback(struct notifier_block *nfb,
1535 unsigned long action,
1536 void *hcpu)
1538 long cpu = (long)hcpu;
1540 switch (action) {
1541 case CPU_ONLINE:
1542 case CPU_ONLINE_FROZEN:
1543 refresh_zone_stat_thresholds();
1544 node_set_state(cpu_to_node(cpu), N_CPU);
1545 cpumask_set_cpu(cpu, cpu_stat_off);
1546 break;
1547 case CPU_DOWN_PREPARE:
1548 case CPU_DOWN_PREPARE_FROZEN:
1549 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1550 cpumask_clear_cpu(cpu, cpu_stat_off);
1551 break;
1552 case CPU_DOWN_FAILED:
1553 case CPU_DOWN_FAILED_FROZEN:
1554 cpumask_set_cpu(cpu, cpu_stat_off);
1555 break;
1556 case CPU_DEAD:
1557 case CPU_DEAD_FROZEN:
1558 refresh_zone_stat_thresholds();
1559 vmstat_cpu_dead(cpu_to_node(cpu));
1560 break;
1561 default:
1562 break;
1564 return NOTIFY_OK;
1567 static struct notifier_block vmstat_notifier =
1568 { &vmstat_cpuup_callback, NULL, 0 };
1569 #endif
1571 static int __init setup_vmstat(void)
1573 #ifdef CONFIG_SMP
1574 cpu_notifier_register_begin();
1575 __register_cpu_notifier(&vmstat_notifier);
1577 start_shepherd_timer();
1578 cpu_notifier_register_done();
1579 #endif
1580 #ifdef CONFIG_PROC_FS
1581 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1582 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1583 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1584 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1585 #endif
1586 return 0;
1588 module_init(setup_vmstat)
1590 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1593 * Return an index indicating how much of the available free memory is
1594 * unusable for an allocation of the requested size.
1596 static int unusable_free_index(unsigned int order,
1597 struct contig_page_info *info)
1599 /* No free memory is interpreted as all free memory is unusable */
1600 if (info->free_pages == 0)
1601 return 1000;
1604 * Index should be a value between 0 and 1. Return a value to 3
1605 * decimal places.
1607 * 0 => no fragmentation
1608 * 1 => high fragmentation
1610 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1614 static void unusable_show_print(struct seq_file *m,
1615 pg_data_t *pgdat, struct zone *zone)
1617 unsigned int order;
1618 int index;
1619 struct contig_page_info info;
1621 seq_printf(m, "Node %d, zone %8s ",
1622 pgdat->node_id,
1623 zone->name);
1624 for (order = 0; order < MAX_ORDER; ++order) {
1625 fill_contig_page_info(zone, order, &info);
1626 index = unusable_free_index(order, &info);
1627 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1630 seq_putc(m, '\n');
1634 * Display unusable free space index
1636 * The unusable free space index measures how much of the available free
1637 * memory cannot be used to satisfy an allocation of a given size and is a
1638 * value between 0 and 1. The higher the value, the more of free memory is
1639 * unusable and by implication, the worse the external fragmentation is. This
1640 * can be expressed as a percentage by multiplying by 100.
1642 static int unusable_show(struct seq_file *m, void *arg)
1644 pg_data_t *pgdat = (pg_data_t *)arg;
1646 /* check memoryless node */
1647 if (!node_state(pgdat->node_id, N_MEMORY))
1648 return 0;
1650 walk_zones_in_node(m, pgdat, unusable_show_print);
1652 return 0;
1655 static const struct seq_operations unusable_op = {
1656 .start = frag_start,
1657 .next = frag_next,
1658 .stop = frag_stop,
1659 .show = unusable_show,
1662 static int unusable_open(struct inode *inode, struct file *file)
1664 return seq_open(file, &unusable_op);
1667 static const struct file_operations unusable_file_ops = {
1668 .open = unusable_open,
1669 .read = seq_read,
1670 .llseek = seq_lseek,
1671 .release = seq_release,
1674 static void extfrag_show_print(struct seq_file *m,
1675 pg_data_t *pgdat, struct zone *zone)
1677 unsigned int order;
1678 int index;
1680 /* Alloc on stack as interrupts are disabled for zone walk */
1681 struct contig_page_info info;
1683 seq_printf(m, "Node %d, zone %8s ",
1684 pgdat->node_id,
1685 zone->name);
1686 for (order = 0; order < MAX_ORDER; ++order) {
1687 fill_contig_page_info(zone, order, &info);
1688 index = __fragmentation_index(order, &info);
1689 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1692 seq_putc(m, '\n');
1696 * Display fragmentation index for orders that allocations would fail for
1698 static int extfrag_show(struct seq_file *m, void *arg)
1700 pg_data_t *pgdat = (pg_data_t *)arg;
1702 walk_zones_in_node(m, pgdat, extfrag_show_print);
1704 return 0;
1707 static const struct seq_operations extfrag_op = {
1708 .start = frag_start,
1709 .next = frag_next,
1710 .stop = frag_stop,
1711 .show = extfrag_show,
1714 static int extfrag_open(struct inode *inode, struct file *file)
1716 return seq_open(file, &extfrag_op);
1719 static const struct file_operations extfrag_file_ops = {
1720 .open = extfrag_open,
1721 .read = seq_read,
1722 .llseek = seq_lseek,
1723 .release = seq_release,
1726 static int __init extfrag_debug_init(void)
1728 struct dentry *extfrag_debug_root;
1730 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1731 if (!extfrag_debug_root)
1732 return -ENOMEM;
1734 if (!debugfs_create_file("unusable_index", 0444,
1735 extfrag_debug_root, NULL, &unusable_file_ops))
1736 goto fail;
1738 if (!debugfs_create_file("extfrag_index", 0444,
1739 extfrag_debug_root, NULL, &extfrag_file_ops))
1740 goto fail;
1742 return 0;
1743 fail:
1744 debugfs_remove_recursive(extfrag_debug_root);
1745 return -ENOMEM;
1748 module_init(extfrag_debug_init);
1749 #endif