1/*
2 *  linux/mm/vmstat.c
3 *
4 *  Manages VM statistics
5 *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
6 *
7 *  zoned VM statistics
8 *  Copyright (C) 2006 Silicon Graphics, Inc.,
9 *		Christoph Lameter <christoph@lameter.com>
10 *  Copyright (C) 2008-2014 Christoph Lameter
11 */
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>
30
31#include "internal.h"
32
33#ifdef CONFIG_VM_EVENT_COUNTERS
34DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
35EXPORT_PER_CPU_SYMBOL(vm_event_states);
36
37static void sum_vm_events(unsigned long *ret)
38{
39	int cpu;
40	int i;
41
42	memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
43
44	for_each_online_cpu(cpu) {
45		struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
46
47		for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
48			ret[i] += this->event[i];
49	}
50}
51
52/*
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.
56*/
57void all_vm_events(unsigned long *ret)
58{
59	get_online_cpus();
60	sum_vm_events(ret);
61	put_online_cpus();
62}
63EXPORT_SYMBOL_GPL(all_vm_events);
64
65/*
66 * Fold the foreign cpu events into our own.
67 *
68 * This is adding to the events on one processor
69 * but keeps the global counts constant.
70 */
71void vm_events_fold_cpu(int cpu)
72{
73	struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
74	int i;
75
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;
79	}
80}
81
82#endif /* CONFIG_VM_EVENT_COUNTERS */
83
84/*
85 * Manage combined zone based / global counters
86 *
87 * vm_stat contains the global counters
88 */
89atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
90EXPORT_SYMBOL(vm_stat);
91
92#ifdef CONFIG_SMP
93
94int calculate_pressure_threshold(struct zone *zone)
95{
96	int threshold;
97	int watermark_distance;
98
99	/*
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
106	 */
107	watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
108	threshold = max(1, (int)(watermark_distance / num_online_cpus()));
109
110	/*
111	 * Maximum threshold is 125
112	 */
113	threshold = min(125, threshold);
114
115	return threshold;
116}
117
118int calculate_normal_threshold(struct zone *zone)
119{
120	int threshold;
121	int mem;	/* memory in 128 MB units */
122
123	/*
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.
128	 *
129	 * Some sample thresholds:
130	 *
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
151	 */
152
153	mem = zone->managed_pages >> (27 - PAGE_SHIFT);
154
155	threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
156
157	/*
158	 * Maximum threshold is 125
159	 */
160	threshold = min(125, threshold);
161
162	return threshold;
163}
164
165/*
166 * Refresh the thresholds for each zone.
167 */
168void refresh_zone_stat_thresholds(void)
169{
170	struct zone *zone;
171	int cpu;
172	int threshold;
173
174	for_each_populated_zone(zone) {
175		unsigned long max_drift, tolerate_drift;
176
177		threshold = calculate_normal_threshold(zone);
178
179		for_each_online_cpu(cpu)
180			per_cpu_ptr(zone->pageset, cpu)->stat_threshold
181							= threshold;
182
183		/*
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
187		 */
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;
193	}
194}
195
196void set_pgdat_percpu_threshold(pg_data_t *pgdat,
197				int (*calculate_pressure)(struct zone *))
198{
199	struct zone *zone;
200	int cpu;
201	int threshold;
202	int i;
203
204	for (i = 0; i < pgdat->nr_zones; i++) {
205		zone = &pgdat->node_zones[i];
206		if (!zone->percpu_drift_mark)
207			continue;
208
209		threshold = (*calculate_pressure)(zone);
210		for_each_online_cpu(cpu)
211			per_cpu_ptr(zone->pageset, cpu)->stat_threshold
212							= threshold;
213	}
214}
215
216/*
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.
220 */
221void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
222				int delta)
223{
224	struct per_cpu_pageset __percpu *pcp = zone->pageset;
225	s8 __percpu *p = pcp->vm_stat_diff + item;
226	long x;
227	long t;
228
229	x = delta + __this_cpu_read(*p);
230
231	t = __this_cpu_read(pcp->stat_threshold);
232
233	if (unlikely(x > t || x < -t)) {
234		zone_page_state_add(x, zone, item);
235		x = 0;
236	}
237	__this_cpu_write(*p, x);
238}
239EXPORT_SYMBOL(__mod_zone_page_state);
240
241/*
242 * Optimized increment and decrement functions.
243 *
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.
247 *
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.
253 *
254 * NOTE: These functions are very performance sensitive. Change only
255 * with care.
256 *
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.
263 */
264void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
265{
266	struct per_cpu_pageset __percpu *pcp = zone->pageset;
267	s8 __percpu *p = pcp->vm_stat_diff + item;
268	s8 v, t;
269
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;
274
275		zone_page_state_add(v + overstep, zone, item);
276		__this_cpu_write(*p, -overstep);
277	}
278}
279
280void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
281{
282	__inc_zone_state(page_zone(page), item);
283}
284EXPORT_SYMBOL(__inc_zone_page_state);
285
286void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
287{
288	struct per_cpu_pageset __percpu *pcp = zone->pageset;
289	s8 __percpu *p = pcp->vm_stat_diff + item;
290	s8 v, t;
291
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;
296
297		zone_page_state_add(v - overstep, zone, item);
298		__this_cpu_write(*p, overstep);
299	}
300}
301
302void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
303{
304	__dec_zone_state(page_zone(page), item);
305}
306EXPORT_SYMBOL(__dec_zone_page_state);
307
308#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
309/*
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.
312 *
313 * mod_state() modifies the zone counter state through atomic per cpu
314 * operations.
315 *
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
320*/
321static inline void mod_state(struct zone *zone,
322       enum zone_stat_item item, int delta, int overstep_mode)
323{
324	struct per_cpu_pageset __percpu *pcp = zone->pageset;
325	s8 __percpu *p = pcp->vm_stat_diff + item;
326	long o, n, t, z;
327
328	do {
329		z = 0;  /* overflow to zone counters */
330
331		/*
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.
337		 *
338		 * Most of the time the thresholds are the same anyways
339		 * for all cpus in a zone.
340		 */
341		t = this_cpu_read(pcp->stat_threshold);
342
343		o = this_cpu_read(*p);
344		n = delta + o;
345
346		if (n > t || n < -t) {
347			int os = overstep_mode * (t >> 1) ;
348
349			/* Overflow must be added to zone counters */
350			z = n + os;
351			n = -os;
352		}
353	} while (this_cpu_cmpxchg(*p, o, n) != o);
354
355	if (z)
356		zone_page_state_add(z, zone, item);
357}
358
359void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
360					int delta)
361{
362	mod_state(zone, item, delta, 0);
363}
364EXPORT_SYMBOL(mod_zone_page_state);
365
366void inc_zone_state(struct zone *zone, enum zone_stat_item item)
367{
368	mod_state(zone, item, 1, 1);
369}
370
371void inc_zone_page_state(struct page *page, enum zone_stat_item item)
372{
373	mod_state(page_zone(page), item, 1, 1);
374}
375EXPORT_SYMBOL(inc_zone_page_state);
376
377void dec_zone_page_state(struct page *page, enum zone_stat_item item)
378{
379	mod_state(page_zone(page), item, -1, -1);
380}
381EXPORT_SYMBOL(dec_zone_page_state);
382#else
383/*
384 * Use interrupt disable to serialize counter updates
385 */
386void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
387					int delta)
388{
389	unsigned long flags;
390
391	local_irq_save(flags);
392	__mod_zone_page_state(zone, item, delta);
393	local_irq_restore(flags);
394}
395EXPORT_SYMBOL(mod_zone_page_state);
396
397void inc_zone_state(struct zone *zone, enum zone_stat_item item)
398{
399	unsigned long flags;
400
401	local_irq_save(flags);
402	__inc_zone_state(zone, item);
403	local_irq_restore(flags);
404}
405
406void inc_zone_page_state(struct page *page, enum zone_stat_item item)
407{
408	unsigned long flags;
409	struct zone *zone;
410
411	zone = page_zone(page);
412	local_irq_save(flags);
413	__inc_zone_state(zone, item);
414	local_irq_restore(flags);
415}
416EXPORT_SYMBOL(inc_zone_page_state);
417
418void dec_zone_page_state(struct page *page, enum zone_stat_item item)
419{
420	unsigned long flags;
421
422	local_irq_save(flags);
423	__dec_zone_page_state(page, item);
424	local_irq_restore(flags);
425}
426EXPORT_SYMBOL(dec_zone_page_state);
427#endif
428
429
430/*
431 * Fold a differential into the global counters.
432 * Returns the number of counters updated.
433 */
434static int fold_diff(int *diff)
435{
436	int i;
437	int changes = 0;
438
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++;
443	}
444	return changes;
445}
446
447/*
448 * Update the zone counters for the current cpu.
449 *
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.
455 *
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.
460 *
461 * The function returns the number of global counters updated.
462 */
463static int refresh_cpu_vm_stats(void)
464{
465	struct zone *zone;
466	int i;
467	int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
468	int changes = 0;
469
470	for_each_populated_zone(zone) {
471		struct per_cpu_pageset __percpu *p = zone->pageset;
472
473		for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
474			int v;
475
476			v = this_cpu_xchg(p->vm_stat_diff[i], 0);
477			if (v) {
478
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
485			}
486		}
487		cond_resched();
488#ifdef CONFIG_NUMA
489		/*
490		 * Deal with draining the remote pageset of this
491		 * processor
492		 *
493		 * Check if there are pages remaining in this pageset
494		 * if not then there is nothing to expire.
495		 */
496		if (!__this_cpu_read(p->expire) ||
497			       !__this_cpu_read(p->pcp.count))
498			continue;
499
500		/*
501		 * We never drain zones local to this processor.
502		 */
503		if (zone_to_nid(zone) == numa_node_id()) {
504			__this_cpu_write(p->expire, 0);
505			continue;
506		}
507
508		if (__this_cpu_dec_return(p->expire))
509			continue;
510
511		if (__this_cpu_read(p->pcp.count)) {
512			drain_zone_pages(zone, this_cpu_ptr(&p->pcp));
513			changes++;
514		}
515#endif
516	}
517	changes += fold_diff(global_diff);
518	return changes;
519}
520
521/*
522 * Fold the data for an offline cpu into the global array.
523 * There cannot be any access by the offline cpu and therefore
524 * synchronization is simplified.
525 */
526void cpu_vm_stats_fold(int cpu)
527{
528	struct zone *zone;
529	int i;
530	int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
531
532	for_each_populated_zone(zone) {
533		struct per_cpu_pageset *p;
534
535		p = per_cpu_ptr(zone->pageset, cpu);
536
537		for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
538			if (p->vm_stat_diff[i]) {
539				int v;
540
541				v = p->vm_stat_diff[i];
542				p->vm_stat_diff[i] = 0;
543				atomic_long_add(v, &zone->vm_stat[i]);
544				global_diff[i] += v;
545			}
546	}
547
548	fold_diff(global_diff);
549}
550
551/*
552 * this is only called if !populated_zone(zone), which implies no other users of
553 * pset->vm_stat_diff[] exsist.
554 */
555void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
556{
557	int i;
558
559	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
560		if (pset->vm_stat_diff[i]) {
561			int v = pset->vm_stat_diff[i];
562			pset->vm_stat_diff[i] = 0;
563			atomic_long_add(v, &zone->vm_stat[i]);
564			atomic_long_add(v, &vm_stat[i]);
565		}
566}
567#endif
568
569#ifdef CONFIG_NUMA
570/*
571 * zonelist = the list of zones passed to the allocator
572 * z 	    = the zone from which the allocation occurred.
573 *
574 * Must be called with interrupts disabled.
575 *
576 * When __GFP_OTHER_NODE is set assume the node of the preferred
577 * zone is the local node. This is useful for daemons who allocate
578 * memory on behalf of other processes.
579 */
580void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags)
581{
582	if (z->zone_pgdat == preferred_zone->zone_pgdat) {
583		__inc_zone_state(z, NUMA_HIT);
584	} else {
585		__inc_zone_state(z, NUMA_MISS);
586		__inc_zone_state(preferred_zone, NUMA_FOREIGN);
587	}
588	if (z->node == ((flags & __GFP_OTHER_NODE) ?
589			preferred_zone->node : numa_node_id()))
590		__inc_zone_state(z, NUMA_LOCAL);
591	else
592		__inc_zone_state(z, NUMA_OTHER);
593}
594#endif
595
596#ifdef CONFIG_COMPACTION
597
598struct contig_page_info {
599	unsigned long free_pages;
600	unsigned long free_blocks_total;
601	unsigned long free_blocks_suitable;
602};
603
604/*
605 * Calculate the number of free pages in a zone, how many contiguous
606 * pages are free and how many are large enough to satisfy an allocation of
607 * the target size. Note that this function makes no attempt to estimate
608 * how many suitable free blocks there *might* be if MOVABLE pages were
609 * migrated. Calculating that is possible, but expensive and can be
610 * figured out from userspace
611 */
612static void fill_contig_page_info(struct zone *zone,
613				unsigned int suitable_order,
614				struct contig_page_info *info)
615{
616	unsigned int order;
617
618	info->free_pages = 0;
619	info->free_blocks_total = 0;
620	info->free_blocks_suitable = 0;
621
622	for (order = 0; order < MAX_ORDER; order++) {
623		unsigned long blocks;
624
625		/* Count number of free blocks */
626		blocks = zone->free_area[order].nr_free;
627		info->free_blocks_total += blocks;
628
629		/* Count free base pages */
630		info->free_pages += blocks << order;
631
632		/* Count the suitable free blocks */
633		if (order >= suitable_order)
634			info->free_blocks_suitable += blocks <<
635						(order - suitable_order);
636	}
637}
638
639/*
640 * A fragmentation index only makes sense if an allocation of a requested
641 * size would fail. If that is true, the fragmentation index indicates
642 * whether external fragmentation or a lack of memory was the problem.
643 * The value can be used to determine if page reclaim or compaction
644 * should be used
645 */
646static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
647{
648	unsigned long requested = 1UL << order;
649
650	if (!info->free_blocks_total)
651		return 0;
652
653	/* Fragmentation index only makes sense when a request would fail */
654	if (info->free_blocks_suitable)
655		return -1000;
656
657	/*
658	 * Index is between 0 and 1 so return within 3 decimal places
659	 *
660	 * 0 => allocation would fail due to lack of memory
661	 * 1 => allocation would fail due to fragmentation
662	 */
663	return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
664}
665
666/* Same as __fragmentation index but allocs contig_page_info on stack */
667int fragmentation_index(struct zone *zone, unsigned int order)
668{
669	struct contig_page_info info;
670
671	fill_contig_page_info(zone, order, &info);
672	return __fragmentation_index(order, &info);
673}
674#endif
675
676#if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
677#ifdef CONFIG_ZONE_DMA
678#define TEXT_FOR_DMA(xx) xx "_dma",
679#else
680#define TEXT_FOR_DMA(xx)
681#endif
682
683#ifdef CONFIG_ZONE_DMA32
684#define TEXT_FOR_DMA32(xx) xx "_dma32",
685#else
686#define TEXT_FOR_DMA32(xx)
687#endif
688
689#ifdef CONFIG_HIGHMEM
690#define TEXT_FOR_HIGHMEM(xx) xx "_high",
691#else
692#define TEXT_FOR_HIGHMEM(xx)
693#endif
694
695#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
696					TEXT_FOR_HIGHMEM(xx) xx "_movable",
697
698const char * const vmstat_text[] = {
699	/* enum zone_stat_item countes */
700	"nr_free_pages",
701	"nr_alloc_batch",
702	"nr_inactive_anon",
703	"nr_active_anon",
704	"nr_inactive_file",
705	"nr_active_file",
706	"nr_unevictable",
707	"nr_mlock",
708	"nr_anon_pages",
709	"nr_mapped",
710	"nr_file_pages",
711	"nr_dirty",
712	"nr_writeback",
713	"nr_slab_reclaimable",
714	"nr_slab_unreclaimable",
715	"nr_page_table_pages",
716	"nr_kernel_stack",
717	"nr_unstable",
718	"nr_bounce",
719	"nr_vmscan_write",
720	"nr_vmscan_immediate_reclaim",
721	"nr_writeback_temp",
722	"nr_isolated_anon",
723	"nr_isolated_file",
724	"nr_shmem",
725	"nr_dirtied",
726	"nr_written",
727	"nr_pages_scanned",
728
729#ifdef CONFIG_NUMA
730	"numa_hit",
731	"numa_miss",
732	"numa_foreign",
733	"numa_interleave",
734	"numa_local",
735	"numa_other",
736#endif
737	"workingset_refault",
738	"workingset_activate",
739	"workingset_nodereclaim",
740	"nr_anon_transparent_hugepages",
741	"nr_free_cma",
742
743	/* enum writeback_stat_item counters */
744	"nr_dirty_threshold",
745	"nr_dirty_background_threshold",
746
747#ifdef CONFIG_VM_EVENT_COUNTERS
748	/* enum vm_event_item counters */
749	"pgpgin",
750	"pgpgout",
751	"pswpin",
752	"pswpout",
753
754	TEXTS_FOR_ZONES("pgalloc")
755
756	"pgfree",
757	"pgactivate",
758	"pgdeactivate",
759
760	"pgfault",
761	"pgmajfault",
762
763	TEXTS_FOR_ZONES("pgrefill")
764	TEXTS_FOR_ZONES("pgsteal_kswapd")
765	TEXTS_FOR_ZONES("pgsteal_direct")
766	TEXTS_FOR_ZONES("pgscan_kswapd")
767	TEXTS_FOR_ZONES("pgscan_direct")
768	"pgscan_direct_throttle",
769
770#ifdef CONFIG_NUMA
771	"zone_reclaim_failed",
772#endif
773	"pginodesteal",
774	"slabs_scanned",
775	"kswapd_inodesteal",
776	"kswapd_low_wmark_hit_quickly",
777	"kswapd_high_wmark_hit_quickly",
778	"pageoutrun",
779	"allocstall",
780
781	"pgrotated",
782
783	"drop_pagecache",
784	"drop_slab",
785
786#ifdef CONFIG_NUMA_BALANCING
787	"numa_pte_updates",
788	"numa_huge_pte_updates",
789	"numa_hint_faults",
790	"numa_hint_faults_local",
791	"numa_pages_migrated",
792#endif
793#ifdef CONFIG_MIGRATION
794	"pgmigrate_success",
795	"pgmigrate_fail",
796#endif
797#ifdef CONFIG_COMPACTION
798	"compact_migrate_scanned",
799	"compact_free_scanned",
800	"compact_isolated",
801	"compact_stall",
802	"compact_fail",
803	"compact_success",
804#endif
805
806#ifdef CONFIG_HUGETLB_PAGE
807	"htlb_buddy_alloc_success",
808	"htlb_buddy_alloc_fail",
809#endif
810	"unevictable_pgs_culled",
811	"unevictable_pgs_scanned",
812	"unevictable_pgs_rescued",
813	"unevictable_pgs_mlocked",
814	"unevictable_pgs_munlocked",
815	"unevictable_pgs_cleared",
816	"unevictable_pgs_stranded",
817
818#ifdef CONFIG_TRANSPARENT_HUGEPAGE
819	"thp_fault_alloc",
820	"thp_fault_fallback",
821	"thp_collapse_alloc",
822	"thp_collapse_alloc_failed",
823	"thp_split",
824	"thp_zero_page_alloc",
825	"thp_zero_page_alloc_failed",
826#endif
827#ifdef CONFIG_MEMORY_BALLOON
828	"balloon_inflate",
829	"balloon_deflate",
830#ifdef CONFIG_BALLOON_COMPACTION
831	"balloon_migrate",
832#endif
833#endif /* CONFIG_MEMORY_BALLOON */
834#ifdef CONFIG_DEBUG_TLBFLUSH
835#ifdef CONFIG_SMP
836	"nr_tlb_remote_flush",
837	"nr_tlb_remote_flush_received",
838#endif /* CONFIG_SMP */
839	"nr_tlb_local_flush_all",
840	"nr_tlb_local_flush_one",
841#endif /* CONFIG_DEBUG_TLBFLUSH */
842
843#ifdef CONFIG_DEBUG_VM_VMACACHE
844	"vmacache_find_calls",
845	"vmacache_find_hits",
846	"vmacache_full_flushes",
847#endif
848#endif /* CONFIG_VM_EVENTS_COUNTERS */
849};
850#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
851
852
853#if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
854     defined(CONFIG_PROC_FS)
855static void *frag_start(struct seq_file *m, loff_t *pos)
856{
857	pg_data_t *pgdat;
858	loff_t node = *pos;
859
860	for (pgdat = first_online_pgdat();
861	     pgdat && node;
862	     pgdat = next_online_pgdat(pgdat))
863		--node;
864
865	return pgdat;
866}
867
868static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
869{
870	pg_data_t *pgdat = (pg_data_t *)arg;
871
872	(*pos)++;
873	return next_online_pgdat(pgdat);
874}
875
876static void frag_stop(struct seq_file *m, void *arg)
877{
878}
879
880/* Walk all the zones in a node and print using a callback */
881static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
882		void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
883{
884	struct zone *zone;
885	struct zone *node_zones = pgdat->node_zones;
886	unsigned long flags;
887
888	for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
889		if (!populated_zone(zone))
890			continue;
891
892		spin_lock_irqsave(&zone->lock, flags);
893		print(m, pgdat, zone);
894		spin_unlock_irqrestore(&zone->lock, flags);
895	}
896}
897#endif
898
899#ifdef CONFIG_PROC_FS
900static char * const migratetype_names[MIGRATE_TYPES] = {
901	"Unmovable",
902	"Reclaimable",
903	"Movable",
904	"Reserve",
905#ifdef CONFIG_CMA
906	"CMA",
907#endif
908#ifdef CONFIG_MEMORY_ISOLATION
909	"Isolate",
910#endif
911};
912
913static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
914						struct zone *zone)
915{
916	int order;
917
918	seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
919	for (order = 0; order < MAX_ORDER; ++order)
920		seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
921	seq_putc(m, '\n');
922}
923
924/*
925 * This walks the free areas for each zone.
926 */
927static int frag_show(struct seq_file *m, void *arg)
928{
929	pg_data_t *pgdat = (pg_data_t *)arg;
930	walk_zones_in_node(m, pgdat, frag_show_print);
931	return 0;
932}
933
934static void pagetypeinfo_showfree_print(struct seq_file *m,
935					pg_data_t *pgdat, struct zone *zone)
936{
937	int order, mtype;
938
939	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
940		seq_printf(m, "Node %4d, zone %8s, type %12s ",
941					pgdat->node_id,
942					zone->name,
943					migratetype_names[mtype]);
944		for (order = 0; order < MAX_ORDER; ++order) {
945			unsigned long freecount = 0;
946			struct free_area *area;
947			struct list_head *curr;
948
949			area = &(zone->free_area[order]);
950
951			list_for_each(curr, &area->free_list[mtype])
952				freecount++;
953			seq_printf(m, "%6lu ", freecount);
954		}
955		seq_putc(m, '\n');
956	}
957}
958
959/* Print out the free pages at each order for each migatetype */
960static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
961{
962	int order;
963	pg_data_t *pgdat = (pg_data_t *)arg;
964
965	/* Print header */
966	seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
967	for (order = 0; order < MAX_ORDER; ++order)
968		seq_printf(m, "%6d ", order);
969	seq_putc(m, '\n');
970
971	walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
972
973	return 0;
974}
975
976static void pagetypeinfo_showblockcount_print(struct seq_file *m,
977					pg_data_t *pgdat, struct zone *zone)
978{
979	int mtype;
980	unsigned long pfn;
981	unsigned long start_pfn = zone->zone_start_pfn;
982	unsigned long end_pfn = zone_end_pfn(zone);
983	unsigned long count[MIGRATE_TYPES] = { 0, };
984
985	for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
986		struct page *page;
987
988		if (!pfn_valid(pfn))
989			continue;
990
991		page = pfn_to_page(pfn);
992
993		/* Watch for unexpected holes punched in the memmap */
994		if (!memmap_valid_within(pfn, page, zone))
995			continue;
996
997		mtype = get_pageblock_migratetype(page);
998
999		if (mtype < MIGRATE_TYPES)
1000			count[mtype]++;
1001	}
1002
1003	/* Print counts */
1004	seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1005	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1006		seq_printf(m, "%12lu ", count[mtype]);
1007	seq_putc(m, '\n');
1008}
1009
1010/* Print out the free pages at each order for each migratetype */
1011static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1012{
1013	int mtype;
1014	pg_data_t *pgdat = (pg_data_t *)arg;
1015
1016	seq_printf(m, "\n%-23s", "Number of blocks type ");
1017	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1018		seq_printf(m, "%12s ", migratetype_names[mtype]);
1019	seq_putc(m, '\n');
1020	walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
1021
1022	return 0;
1023}
1024
1025#ifdef CONFIG_PAGE_OWNER
1026static void pagetypeinfo_showmixedcount_print(struct seq_file *m,
1027							pg_data_t *pgdat,
1028							struct zone *zone)
1029{
1030	struct page *page;
1031	struct page_ext *page_ext;
1032	unsigned long pfn = zone->zone_start_pfn, block_end_pfn;
1033	unsigned long end_pfn = pfn + zone->spanned_pages;
1034	unsigned long count[MIGRATE_TYPES] = { 0, };
1035	int pageblock_mt, page_mt;
1036	int i;
1037
1038	/* Scan block by block. First and last block may be incomplete */
1039	pfn = zone->zone_start_pfn;
1040
1041	/*
1042	 * Walk the zone in pageblock_nr_pages steps. If a page block spans
1043	 * a zone boundary, it will be double counted between zones. This does
1044	 * not matter as the mixed block count will still be correct
1045	 */
1046	for (; pfn < end_pfn; ) {
1047		if (!pfn_valid(pfn)) {
1048			pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
1049			continue;
1050		}
1051
1052		block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
1053		block_end_pfn = min(block_end_pfn, end_pfn);
1054
1055		page = pfn_to_page(pfn);
1056		pageblock_mt = get_pfnblock_migratetype(page, pfn);
1057
1058		for (; pfn < block_end_pfn; pfn++) {
1059			if (!pfn_valid_within(pfn))
1060				continue;
1061
1062			page = pfn_to_page(pfn);
1063			if (PageBuddy(page)) {
1064				pfn += (1UL << page_order(page)) - 1;
1065				continue;
1066			}
1067
1068			if (PageReserved(page))
1069				continue;
1070
1071			page_ext = lookup_page_ext(page);
1072
1073			if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
1074				continue;
1075
1076			page_mt = gfpflags_to_migratetype(page_ext->gfp_mask);
1077			if (pageblock_mt != page_mt) {
1078				if (is_migrate_cma(pageblock_mt))
1079					count[MIGRATE_MOVABLE]++;
1080				else
1081					count[pageblock_mt]++;
1082
1083				pfn = block_end_pfn;
1084				break;
1085			}
1086			pfn += (1UL << page_ext->order) - 1;
1087		}
1088	}
1089
1090	/* Print counts */
1091	seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1092	for (i = 0; i < MIGRATE_TYPES; i++)
1093		seq_printf(m, "%12lu ", count[i]);
1094	seq_putc(m, '\n');
1095}
1096#endif /* CONFIG_PAGE_OWNER */
1097
1098/*
1099 * Print out the number of pageblocks for each migratetype that contain pages
1100 * of other types. This gives an indication of how well fallbacks are being
1101 * contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1102 * to determine what is going on
1103 */
1104static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1105{
1106#ifdef CONFIG_PAGE_OWNER
1107	int mtype;
1108
1109	if (!page_owner_inited)
1110		return;
1111
1112	drain_all_pages(NULL);
1113
1114	seq_printf(m, "\n%-23s", "Number of mixed blocks ");
1115	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1116		seq_printf(m, "%12s ", migratetype_names[mtype]);
1117	seq_putc(m, '\n');
1118
1119	walk_zones_in_node(m, pgdat, pagetypeinfo_showmixedcount_print);
1120#endif /* CONFIG_PAGE_OWNER */
1121}
1122
1123/*
1124 * This prints out statistics in relation to grouping pages by mobility.
1125 * It is expensive to collect so do not constantly read the file.
1126 */
1127static int pagetypeinfo_show(struct seq_file *m, void *arg)
1128{
1129	pg_data_t *pgdat = (pg_data_t *)arg;
1130
1131	/* check memoryless node */
1132	if (!node_state(pgdat->node_id, N_MEMORY))
1133		return 0;
1134
1135	seq_printf(m, "Page block order: %d\n", pageblock_order);
1136	seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
1137	seq_putc(m, '\n');
1138	pagetypeinfo_showfree(m, pgdat);
1139	pagetypeinfo_showblockcount(m, pgdat);
1140	pagetypeinfo_showmixedcount(m, pgdat);
1141
1142	return 0;
1143}
1144
1145static const struct seq_operations fragmentation_op = {
1146	.start	= frag_start,
1147	.next	= frag_next,
1148	.stop	= frag_stop,
1149	.show	= frag_show,
1150};
1151
1152static int fragmentation_open(struct inode *inode, struct file *file)
1153{
1154	return seq_open(file, &fragmentation_op);
1155}
1156
1157static const struct file_operations fragmentation_file_operations = {
1158	.open		= fragmentation_open,
1159	.read		= seq_read,
1160	.llseek		= seq_lseek,
1161	.release	= seq_release,
1162};
1163
1164static const struct seq_operations pagetypeinfo_op = {
1165	.start	= frag_start,
1166	.next	= frag_next,
1167	.stop	= frag_stop,
1168	.show	= pagetypeinfo_show,
1169};
1170
1171static int pagetypeinfo_open(struct inode *inode, struct file *file)
1172{
1173	return seq_open(file, &pagetypeinfo_op);
1174}
1175
1176static const struct file_operations pagetypeinfo_file_ops = {
1177	.open		= pagetypeinfo_open,
1178	.read		= seq_read,
1179	.llseek		= seq_lseek,
1180	.release	= seq_release,
1181};
1182
1183static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1184							struct zone *zone)
1185{
1186	int i;
1187	seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1188	seq_printf(m,
1189		   "\n  pages free     %lu"
1190		   "\n        min      %lu"
1191		   "\n        low      %lu"
1192		   "\n        high     %lu"
1193		   "\n        scanned  %lu"
1194		   "\n        spanned  %lu"
1195		   "\n        present  %lu"
1196		   "\n        managed  %lu",
1197		   zone_page_state(zone, NR_FREE_PAGES),
1198		   min_wmark_pages(zone),
1199		   low_wmark_pages(zone),
1200		   high_wmark_pages(zone),
1201		   zone_page_state(zone, NR_PAGES_SCANNED),
1202		   zone->spanned_pages,
1203		   zone->present_pages,
1204		   zone->managed_pages);
1205
1206	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1207		seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
1208				zone_page_state(zone, i));
1209
1210	seq_printf(m,
1211		   "\n        protection: (%ld",
1212		   zone->lowmem_reserve[0]);
1213	for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1214		seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
1215	seq_printf(m,
1216		   ")"
1217		   "\n  pagesets");
1218	for_each_online_cpu(i) {
1219		struct per_cpu_pageset *pageset;
1220
1221		pageset = per_cpu_ptr(zone->pageset, i);
1222		seq_printf(m,
1223			   "\n    cpu: %i"
1224			   "\n              count: %i"
1225			   "\n              high:  %i"
1226			   "\n              batch: %i",
1227			   i,
1228			   pageset->pcp.count,
1229			   pageset->pcp.high,
1230			   pageset->pcp.batch);
1231#ifdef CONFIG_SMP
1232		seq_printf(m, "\n  vm stats threshold: %d",
1233				pageset->stat_threshold);
1234#endif
1235	}
1236	seq_printf(m,
1237		   "\n  all_unreclaimable: %u"
1238		   "\n  start_pfn:         %lu"
1239		   "\n  inactive_ratio:    %u",
1240		   !zone_reclaimable(zone),
1241		   zone->zone_start_pfn,
1242		   zone->inactive_ratio);
1243	seq_putc(m, '\n');
1244}
1245
1246/*
1247 * Output information about zones in @pgdat.
1248 */
1249static int zoneinfo_show(struct seq_file *m, void *arg)
1250{
1251	pg_data_t *pgdat = (pg_data_t *)arg;
1252	walk_zones_in_node(m, pgdat, zoneinfo_show_print);
1253	return 0;
1254}
1255
1256static const struct seq_operations zoneinfo_op = {
1257	.start	= frag_start, /* iterate over all zones. The same as in
1258			       * fragmentation. */
1259	.next	= frag_next,
1260	.stop	= frag_stop,
1261	.show	= zoneinfo_show,
1262};
1263
1264static int zoneinfo_open(struct inode *inode, struct file *file)
1265{
1266	return seq_open(file, &zoneinfo_op);
1267}
1268
1269static const struct file_operations proc_zoneinfo_file_operations = {
1270	.open		= zoneinfo_open,
1271	.read		= seq_read,
1272	.llseek		= seq_lseek,
1273	.release	= seq_release,
1274};
1275
1276enum writeback_stat_item {
1277	NR_DIRTY_THRESHOLD,
1278	NR_DIRTY_BG_THRESHOLD,
1279	NR_VM_WRITEBACK_STAT_ITEMS,
1280};
1281
1282static void *vmstat_start(struct seq_file *m, loff_t *pos)
1283{
1284	unsigned long *v;
1285	int i, stat_items_size;
1286
1287	if (*pos >= ARRAY_SIZE(vmstat_text))
1288		return NULL;
1289	stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1290			  NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
1291
1292#ifdef CONFIG_VM_EVENT_COUNTERS
1293	stat_items_size += sizeof(struct vm_event_state);
1294#endif
1295
1296	v = kmalloc(stat_items_size, GFP_KERNEL);
1297	m->private = v;
1298	if (!v)
1299		return ERR_PTR(-ENOMEM);
1300	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1301		v[i] = global_page_state(i);
1302	v += NR_VM_ZONE_STAT_ITEMS;
1303
1304	global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1305			    v + NR_DIRTY_THRESHOLD);
1306	v += NR_VM_WRITEBACK_STAT_ITEMS;
1307
1308#ifdef CONFIG_VM_EVENT_COUNTERS
1309	all_vm_events(v);
1310	v[PGPGIN] /= 2;		/* sectors -> kbytes */
1311	v[PGPGOUT] /= 2;
1312#endif
1313	return (unsigned long *)m->private + *pos;
1314}
1315
1316static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1317{
1318	(*pos)++;
1319	if (*pos >= ARRAY_SIZE(vmstat_text))
1320		return NULL;
1321	return (unsigned long *)m->private + *pos;
1322}
1323
1324static int vmstat_show(struct seq_file *m, void *arg)
1325{
1326	unsigned long *l = arg;
1327	unsigned long off = l - (unsigned long *)m->private;
1328
1329	seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1330	return 0;
1331}
1332
1333static void vmstat_stop(struct seq_file *m, void *arg)
1334{
1335	kfree(m->private);
1336	m->private = NULL;
1337}
1338
1339static const struct seq_operations vmstat_op = {
1340	.start	= vmstat_start,
1341	.next	= vmstat_next,
1342	.stop	= vmstat_stop,
1343	.show	= vmstat_show,
1344};
1345
1346static int vmstat_open(struct inode *inode, struct file *file)
1347{
1348	return seq_open(file, &vmstat_op);
1349}
1350
1351static const struct file_operations proc_vmstat_file_operations = {
1352	.open		= vmstat_open,
1353	.read		= seq_read,
1354	.llseek		= seq_lseek,
1355	.release	= seq_release,
1356};
1357#endif /* CONFIG_PROC_FS */
1358
1359#ifdef CONFIG_SMP
1360static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1361int sysctl_stat_interval __read_mostly = HZ;
1362static cpumask_var_t cpu_stat_off;
1363
1364static void vmstat_update(struct work_struct *w)
1365{
1366	if (refresh_cpu_vm_stats())
1367		/*
1368		 * Counters were updated so we expect more updates
1369		 * to occur in the future. Keep on running the
1370		 * update worker thread.
1371		 */
1372		schedule_delayed_work(this_cpu_ptr(&vmstat_work),
1373			round_jiffies_relative(sysctl_stat_interval));
1374	else {
1375		/*
1376		 * We did not update any counters so the app may be in
1377		 * a mode where it does not cause counter updates.
1378		 * We may be uselessly running vmstat_update.
1379		 * Defer the checking for differentials to the
1380		 * shepherd thread on a different processor.
1381		 */
1382		int r;
1383		/*
1384		 * Shepherd work thread does not race since it never
1385		 * changes the bit if its zero but the cpu
1386		 * online / off line code may race if
1387		 * worker threads are still allowed during
1388		 * shutdown / startup.
1389		 */
1390		r = cpumask_test_and_set_cpu(smp_processor_id(),
1391			cpu_stat_off);
1392		VM_BUG_ON(r);
1393	}
1394}
1395
1396/*
1397 * Check if the diffs for a certain cpu indicate that
1398 * an update is needed.
1399 */
1400static bool need_update(int cpu)
1401{
1402	struct zone *zone;
1403
1404	for_each_populated_zone(zone) {
1405		struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu);
1406
1407		BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1);
1408		/*
1409		 * The fast way of checking if there are any vmstat diffs.
1410		 * This works because the diffs are byte sized items.
1411		 */
1412		if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS))
1413			return true;
1414
1415	}
1416	return false;
1417}
1418
1419
1420/*
1421 * Shepherd worker thread that checks the
1422 * differentials of processors that have their worker
1423 * threads for vm statistics updates disabled because of
1424 * inactivity.
1425 */
1426static void vmstat_shepherd(struct work_struct *w);
1427
1428static DECLARE_DELAYED_WORK(shepherd, vmstat_shepherd);
1429
1430static void vmstat_shepherd(struct work_struct *w)
1431{
1432	int cpu;
1433
1434	get_online_cpus();
1435	/* Check processors whose vmstat worker threads have been disabled */
1436	for_each_cpu(cpu, cpu_stat_off)
1437		if (need_update(cpu) &&
1438			cpumask_test_and_clear_cpu(cpu, cpu_stat_off))
1439
1440			schedule_delayed_work_on(cpu,
1441				&per_cpu(vmstat_work, cpu), 0);
1442
1443	put_online_cpus();
1444
1445	schedule_delayed_work(&shepherd,
1446		round_jiffies_relative(sysctl_stat_interval));
1447
1448}
1449
1450static void __init start_shepherd_timer(void)
1451{
1452	int cpu;
1453
1454	for_each_possible_cpu(cpu)
1455		INIT_DELAYED_WORK(per_cpu_ptr(&vmstat_work, cpu),
1456			vmstat_update);
1457
1458	if (!alloc_cpumask_var(&cpu_stat_off, GFP_KERNEL))
1459		BUG();
1460	cpumask_copy(cpu_stat_off, cpu_online_mask);
1461
1462	schedule_delayed_work(&shepherd,
1463		round_jiffies_relative(sysctl_stat_interval));
1464}
1465
1466static void vmstat_cpu_dead(int node)
1467{
1468	int cpu;
1469
1470	get_online_cpus();
1471	for_each_online_cpu(cpu)
1472		if (cpu_to_node(cpu) == node)
1473			goto end;
1474
1475	node_clear_state(node, N_CPU);
1476end:
1477	put_online_cpus();
1478}
1479
1480/*
1481 * Use the cpu notifier to insure that the thresholds are recalculated
1482 * when necessary.
1483 */
1484static int vmstat_cpuup_callback(struct notifier_block *nfb,
1485		unsigned long action,
1486		void *hcpu)
1487{
1488	long cpu = (long)hcpu;
1489
1490	switch (action) {
1491	case CPU_ONLINE:
1492	case CPU_ONLINE_FROZEN:
1493		refresh_zone_stat_thresholds();
1494		node_set_state(cpu_to_node(cpu), N_CPU);
1495		cpumask_set_cpu(cpu, cpu_stat_off);
1496		break;
1497	case CPU_DOWN_PREPARE:
1498	case CPU_DOWN_PREPARE_FROZEN:
1499		cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1500		cpumask_clear_cpu(cpu, cpu_stat_off);
1501		break;
1502	case CPU_DOWN_FAILED:
1503	case CPU_DOWN_FAILED_FROZEN:
1504		cpumask_set_cpu(cpu, cpu_stat_off);
1505		break;
1506	case CPU_DEAD:
1507	case CPU_DEAD_FROZEN:
1508		refresh_zone_stat_thresholds();
1509		vmstat_cpu_dead(cpu_to_node(cpu));
1510		break;
1511	default:
1512		break;
1513	}
1514	return NOTIFY_OK;
1515}
1516
1517static struct notifier_block vmstat_notifier =
1518	{ &vmstat_cpuup_callback, NULL, 0 };
1519#endif
1520
1521static int __init setup_vmstat(void)
1522{
1523#ifdef CONFIG_SMP
1524	cpu_notifier_register_begin();
1525	__register_cpu_notifier(&vmstat_notifier);
1526
1527	start_shepherd_timer();
1528	cpu_notifier_register_done();
1529#endif
1530#ifdef CONFIG_PROC_FS
1531	proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1532	proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1533	proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1534	proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1535#endif
1536	return 0;
1537}
1538module_init(setup_vmstat)
1539
1540#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1541
1542/*
1543 * Return an index indicating how much of the available free memory is
1544 * unusable for an allocation of the requested size.
1545 */
1546static int unusable_free_index(unsigned int order,
1547				struct contig_page_info *info)
1548{
1549	/* No free memory is interpreted as all free memory is unusable */
1550	if (info->free_pages == 0)
1551		return 1000;
1552
1553	/*
1554	 * Index should be a value between 0 and 1. Return a value to 3
1555	 * decimal places.
1556	 *
1557	 * 0 => no fragmentation
1558	 * 1 => high fragmentation
1559	 */
1560	return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1561
1562}
1563
1564static void unusable_show_print(struct seq_file *m,
1565					pg_data_t *pgdat, struct zone *zone)
1566{
1567	unsigned int order;
1568	int index;
1569	struct contig_page_info info;
1570
1571	seq_printf(m, "Node %d, zone %8s ",
1572				pgdat->node_id,
1573				zone->name);
1574	for (order = 0; order < MAX_ORDER; ++order) {
1575		fill_contig_page_info(zone, order, &info);
1576		index = unusable_free_index(order, &info);
1577		seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1578	}
1579
1580	seq_putc(m, '\n');
1581}
1582
1583/*
1584 * Display unusable free space index
1585 *
1586 * The unusable free space index measures how much of the available free
1587 * memory cannot be used to satisfy an allocation of a given size and is a
1588 * value between 0 and 1. The higher the value, the more of free memory is
1589 * unusable and by implication, the worse the external fragmentation is. This
1590 * can be expressed as a percentage by multiplying by 100.
1591 */
1592static int unusable_show(struct seq_file *m, void *arg)
1593{
1594	pg_data_t *pgdat = (pg_data_t *)arg;
1595
1596	/* check memoryless node */
1597	if (!node_state(pgdat->node_id, N_MEMORY))
1598		return 0;
1599
1600	walk_zones_in_node(m, pgdat, unusable_show_print);
1601
1602	return 0;
1603}
1604
1605static const struct seq_operations unusable_op = {
1606	.start	= frag_start,
1607	.next	= frag_next,
1608	.stop	= frag_stop,
1609	.show	= unusable_show,
1610};
1611
1612static int unusable_open(struct inode *inode, struct file *file)
1613{
1614	return seq_open(file, &unusable_op);
1615}
1616
1617static const struct file_operations unusable_file_ops = {
1618	.open		= unusable_open,
1619	.read		= seq_read,
1620	.llseek		= seq_lseek,
1621	.release	= seq_release,
1622};
1623
1624static void extfrag_show_print(struct seq_file *m,
1625					pg_data_t *pgdat, struct zone *zone)
1626{
1627	unsigned int order;
1628	int index;
1629
1630	/* Alloc on stack as interrupts are disabled for zone walk */
1631	struct contig_page_info info;
1632
1633	seq_printf(m, "Node %d, zone %8s ",
1634				pgdat->node_id,
1635				zone->name);
1636	for (order = 0; order < MAX_ORDER; ++order) {
1637		fill_contig_page_info(zone, order, &info);
1638		index = __fragmentation_index(order, &info);
1639		seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1640	}
1641
1642	seq_putc(m, '\n');
1643}
1644
1645/*
1646 * Display fragmentation index for orders that allocations would fail for
1647 */
1648static int extfrag_show(struct seq_file *m, void *arg)
1649{
1650	pg_data_t *pgdat = (pg_data_t *)arg;
1651
1652	walk_zones_in_node(m, pgdat, extfrag_show_print);
1653
1654	return 0;
1655}
1656
1657static const struct seq_operations extfrag_op = {
1658	.start	= frag_start,
1659	.next	= frag_next,
1660	.stop	= frag_stop,
1661	.show	= extfrag_show,
1662};
1663
1664static int extfrag_open(struct inode *inode, struct file *file)
1665{
1666	return seq_open(file, &extfrag_op);
1667}
1668
1669static const struct file_operations extfrag_file_ops = {
1670	.open		= extfrag_open,
1671	.read		= seq_read,
1672	.llseek		= seq_lseek,
1673	.release	= seq_release,
1674};
1675
1676static int __init extfrag_debug_init(void)
1677{
1678	struct dentry *extfrag_debug_root;
1679
1680	extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1681	if (!extfrag_debug_root)
1682		return -ENOMEM;
1683
1684	if (!debugfs_create_file("unusable_index", 0444,
1685			extfrag_debug_root, NULL, &unusable_file_ops))
1686		goto fail;
1687
1688	if (!debugfs_create_file("extfrag_index", 0444,
1689			extfrag_debug_root, NULL, &extfrag_file_ops))
1690		goto fail;
1691
1692	return 0;
1693fail:
1694	debugfs_remove_recursive(extfrag_debug_root);
1695	return -ENOMEM;
1696}
1697
1698module_init(extfrag_debug_init);
1699#endif
1700