1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* linux/mm/vmstat.c
4	*
5	* Manages VM statistics
6	* Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
7	*
8	* zoned VM statistics
9	* Copyright (C) 2006 Silicon Graphics, Inc.,
10	* Christoph Lameter <christoph@lameter.com>
11	* Copyright (C) 2008-2014 Christoph Lameter
12	*/
13	#include <linux/fs.h>
14	#include <linux/mm.h>
15	#include <linux/err.h>
16	#include <linux/module.h>
17	#include <linux/slab.h>
18	#include <linux/cpu.h>
19	#include <linux/cpumask.h>
20	#include <linux/vmstat.h>
21	#include <linux/proc_fs.h>
22	#include <linux/seq_file.h>
23	#include <linux/debugfs.h>
24	#include <linux/sched.h>
25	#include <linux/math64.h>
26	#include <linux/writeback.h>
27	#include <linux/compaction.h>
28	#include <linux/mm_inline.h>
29	#include <linux/page_owner.h>
30	#include <linux/sched/isolation.h>
31
32	#include "internal.h"
33
34	#ifdef CONFIG_NUMA
35	int sysctl_vm_numa_stat = ENABLE_NUMA_STAT;
36
37	/* zero numa counters within a zone */
38	static void zero_zone_numa_counters(struct zone *zone)
39	{
40	int item, cpu;
41
42	for (item = 0; item < NR_VM_NUMA_EVENT_ITEMS; item++) {
43	atomic_long_set(v: &zone->vm_numa_event[item], i: 0);
44	for_each_online_cpu(cpu) {
45	per_cpu_ptr(zone->per_cpu_zonestats, cpu)->vm_numa_event[item]
46	= 0;
47	}
48	}
49	}
50
51	/* zero numa counters of all the populated zones */
52	static void zero_zones_numa_counters(void)
53	{
54	struct zone *zone;
55
56	for_each_populated_zone(zone)
57	zero_zone_numa_counters(zone);
58	}
59
60	/* zero global numa counters */
61	static void zero_global_numa_counters(void)
62	{
63	int item;
64
65	for (item = 0; item < NR_VM_NUMA_EVENT_ITEMS; item++)
66	atomic_long_set(v: &vm_numa_event[item], i: 0);
67	}
68
69	static void invalid_numa_statistics(void)
70	{
71	zero_zones_numa_counters();
72	zero_global_numa_counters();
73	}
74
75	static DEFINE_MUTEX(vm_numa_stat_lock);
76
77	int sysctl_vm_numa_stat_handler(struct ctl_table *table, int write,
78	void *buffer, size_t *length, loff_t *ppos)
79	{
80	int ret, oldval;
81
82	mutex_lock(&vm_numa_stat_lock);
83	if (write)
84	oldval = sysctl_vm_numa_stat;
85	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
86	if (ret || !write)
87	goto out;
88
89	if (oldval == sysctl_vm_numa_stat)
90	goto out;
91	else if (sysctl_vm_numa_stat == ENABLE_NUMA_STAT) {
92	static_branch_enable(&vm_numa_stat_key);
93	pr_info("enable numa statistics\n");
94	} else {
95	static_branch_disable(&vm_numa_stat_key);
96	invalid_numa_statistics();
97	pr_info("disable numa statistics, and clear numa counters\n");
98	}
99
100	out:
101	mutex_unlock(lock: &vm_numa_stat_lock);
102	return ret;
103	}
104	#endif
105
106	#ifdef CONFIG_VM_EVENT_COUNTERS
107	DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
108	EXPORT_PER_CPU_SYMBOL(vm_event_states);
109
110	static void sum_vm_events(unsigned long *ret)
111	{
112	int cpu;
113	int i;
114
115	memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
116
117	for_each_online_cpu(cpu) {
118	struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
119
120	for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
121	ret[i] += this->event[i];
122	}
123	}
124
125	/*
126	* Accumulate the vm event counters across all CPUs.
127	* The result is unavoidably approximate - it can change
128	* during and after execution of this function.
129	*/
130	void all_vm_events(unsigned long *ret)
131	{
132	cpus_read_lock();
133	sum_vm_events(ret);
134	cpus_read_unlock();
135	}
136	EXPORT_SYMBOL_GPL(all_vm_events);
137
138	/*
139	* Fold the foreign cpu events into our own.
140	*
141	* This is adding to the events on one processor
142	* but keeps the global counts constant.
143	*/
144	void vm_events_fold_cpu(int cpu)
145	{
146	struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
147	int i;
148
149	for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
150	count_vm_events(item: i, delta: fold_state->event[i]);
151	fold_state->event[i] = 0;
152	}
153	}
154
155	#endif /* CONFIG_VM_EVENT_COUNTERS */
156
157	/*
158	* Manage combined zone based / global counters
159	*
160	* vm_stat contains the global counters
161	*/
162	atomic_long_t vm_zone_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
163	atomic_long_t vm_node_stat[NR_VM_NODE_STAT_ITEMS] __cacheline_aligned_in_smp;
164	atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS] __cacheline_aligned_in_smp;
165	EXPORT_SYMBOL(vm_zone_stat);
166	EXPORT_SYMBOL(vm_node_stat);
167
168	#ifdef CONFIG_NUMA
169	static void fold_vm_zone_numa_events(struct zone *zone)
170	{
171	unsigned long zone_numa_events[NR_VM_NUMA_EVENT_ITEMS] = { 0, };
172	int cpu;
173	enum numa_stat_item item;
174
175	for_each_online_cpu(cpu) {
176	struct per_cpu_zonestat *pzstats;
177
178	pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
179	for (item = 0; item < NR_VM_NUMA_EVENT_ITEMS; item++)
180	zone_numa_events[item] += xchg(&pzstats->vm_numa_event[item], 0);
181	}
182
183	for (item = 0; item < NR_VM_NUMA_EVENT_ITEMS; item++)
184	zone_numa_event_add(x: zone_numa_events[item], zone, item);
185	}
186
187	void fold_vm_numa_events(void)
188	{
189	struct zone *zone;
190
191	for_each_populated_zone(zone)
192	fold_vm_zone_numa_events(zone);
193	}
194	#endif
195
196	#ifdef CONFIG_SMP
197
198	int calculate_pressure_threshold(struct zone *zone)
199	{
200	int threshold;
201	int watermark_distance;
202
203	/*
204	* As vmstats are not up to date, there is drift between the estimated
205	* and real values. For high thresholds and a high number of CPUs, it
206	* is possible for the min watermark to be breached while the estimated
207	* value looks fine. The pressure threshold is a reduced value such
208	* that even the maximum amount of drift will not accidentally breach
209	* the min watermark
210	*/
211	watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
212	threshold = max(1, (int)(watermark_distance / num_online_cpus()));
213
214	/*
215	* Maximum threshold is 125
216	*/
217	threshold = min(125, threshold);
218
219	return threshold;
220	}
221
222	int calculate_normal_threshold(struct zone *zone)
223	{
224	int threshold;
225	int mem; /* memory in 128 MB units */
226
227	/*
228	* The threshold scales with the number of processors and the amount
229	* of memory per zone. More memory means that we can defer updates for
230	* longer, more processors could lead to more contention.
231	* fls() is used to have a cheap way of logarithmic scaling.
232	*
233	* Some sample thresholds:
234	*
235	* Threshold Processors (fls) Zonesize fls(mem)+1
236	* ------------------------------------------------------------------
237	* 8 1 1 0.9-1 GB 4
238	* 16 2 2 0.9-1 GB 4
239	* 20 2 2 1-2 GB 5
240	* 24 2 2 2-4 GB 6
241	* 28 2 2 4-8 GB 7
242	* 32 2 2 8-16 GB 8
243	* 4 2 2 <128M 1
244	* 30 4 3 2-4 GB 5
245	* 48 4 3 8-16 GB 8
246	* 32 8 4 1-2 GB 4
247	* 32 8 4 0.9-1GB 4
248	* 10 16 5 <128M 1
249	* 40 16 5 900M 4
250	* 70 64 7 2-4 GB 5
251	* 84 64 7 4-8 GB 6
252	* 108 512 9 4-8 GB 6
253	* 125 1024 10 8-16 GB 8
254	* 125 1024 10 16-32 GB 9
255	*/
256
257	mem = zone_managed_pages(zone) >> (27 - PAGE_SHIFT);
258
259	threshold = 2 * fls(x: num_online_cpus()) * (1 + fls(x: mem));
260
261	/*
262	* Maximum threshold is 125
263	*/
264	threshold = min(125, threshold);
265
266	return threshold;
267	}
268
269	/*
270	* Refresh the thresholds for each zone.
271	*/
272	void refresh_zone_stat_thresholds(void)
273	{
274	struct pglist_data *pgdat;
275	struct zone *zone;
276	int cpu;
277	int threshold;
278
279	/* Zero current pgdat thresholds */
280	for_each_online_pgdat(pgdat) {
281	for_each_online_cpu(cpu) {
282	per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold = 0;
283	}
284	}
285
286	for_each_populated_zone(zone) {
287	struct pglist_data *pgdat = zone->zone_pgdat;
288	unsigned long max_drift, tolerate_drift;
289
290	threshold = calculate_normal_threshold(zone);
291
292	for_each_online_cpu(cpu) {
293	int pgdat_threshold;
294
295	per_cpu_ptr(zone->per_cpu_zonestats, cpu)->stat_threshold
296	= threshold;
297
298	/* Base nodestat threshold on the largest populated zone. */
299	pgdat_threshold = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold;
300	per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold
301	= max(threshold, pgdat_threshold);
302	}
303
304	/*
305	* Only set percpu_drift_mark if there is a danger that
306	* NR_FREE_PAGES reports the low watermark is ok when in fact
307	* the min watermark could be breached by an allocation
308	*/
309	tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
310	max_drift = num_online_cpus() * threshold;
311	if (max_drift > tolerate_drift)
312	zone->percpu_drift_mark = high_wmark_pages(zone) +
313	max_drift;
314	}
315	}
316
317	void set_pgdat_percpu_threshold(pg_data_t *pgdat,
318	int (*calculate_pressure)(struct zone *))
319	{
320	struct zone *zone;
321	int cpu;
322	int threshold;
323	int i;
324
325	for (i = 0; i < pgdat->nr_zones; i++) {
326	zone = &pgdat->node_zones[i];
327	if (!zone->percpu_drift_mark)
328	continue;
329
330	threshold = (*calculate_pressure)(zone);
331	for_each_online_cpu(cpu)
332	per_cpu_ptr(zone->per_cpu_zonestats, cpu)->stat_threshold
333	= threshold;
334	}
335	}
336
337	/*
338	* For use when we know that interrupts are disabled,
339	* or when we know that preemption is disabled and that
340	* particular counter cannot be updated from interrupt context.
341	*/
342	void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
343	long delta)
344	{
345	struct per_cpu_zonestat __percpu *pcp = zone->per_cpu_zonestats;
346	s8 __percpu *p = pcp->vm_stat_diff + item;
347	long x;
348	long t;
349
350	/*
351	* Accurate vmstat updates require a RMW. On !PREEMPT_RT kernels,
352	* atomicity is provided by IRQs being disabled -- either explicitly
353	* or via local_lock_irq. On PREEMPT_RT, local_lock_irq only disables
354	* CPU migrations and preemption potentially corrupts a counter so
355	* disable preemption.
356	*/
357	preempt_disable_nested();
358
359	x = delta + __this_cpu_read(*p);
360
361	t = __this_cpu_read(pcp->stat_threshold);
362
363	if (unlikely(abs(x) > t)) {
364	zone_page_state_add(x, zone, item);
365	x = 0;
366	}
367	__this_cpu_write(*p, x);
368
369	preempt_enable_nested();
370	}
371	EXPORT_SYMBOL(__mod_zone_page_state);
372
373	void __mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
374	long delta)
375	{
376	struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
377	s8 __percpu *p = pcp->vm_node_stat_diff + item;
378	long x;
379	long t;
380
381	if (vmstat_item_in_bytes(idx: item)) {
382	/*
383	* Only cgroups use subpage accounting right now; at
384	* the global level, these items still change in
385	* multiples of whole pages. Store them as pages
386	* internally to keep the per-cpu counters compact.
387	*/
388	VM_WARN_ON_ONCE(delta & (PAGE_SIZE - 1));
389	delta >>= PAGE_SHIFT;
390	}
391
392	/* See __mod_node_page_state */
393	preempt_disable_nested();
394
395	x = delta + __this_cpu_read(*p);
396
397	t = __this_cpu_read(pcp->stat_threshold);
398
399	if (unlikely(abs(x) > t)) {
400	node_page_state_add(x, pgdat, item);
401	x = 0;
402	}
403	__this_cpu_write(*p, x);
404
405	preempt_enable_nested();
406	}
407	EXPORT_SYMBOL(__mod_node_page_state);
408
409	/*
410	* Optimized increment and decrement functions.
411	*
412	* These are only for a single page and therefore can take a struct page *
413	* argument instead of struct zone *. This allows the inclusion of the code
414	* generated for page_zone(page) into the optimized functions.
415	*
416	* No overflow check is necessary and therefore the differential can be
417	* incremented or decremented in place which may allow the compilers to
418	* generate better code.
419	* The increment or decrement is known and therefore one boundary check can
420	* be omitted.
421	*
422	* NOTE: These functions are very performance sensitive. Change only
423	* with care.
424	*
425	* Some processors have inc/dec instructions that are atomic vs an interrupt.
426	* However, the code must first determine the differential location in a zone
427	* based on the processor number and then inc/dec the counter. There is no
428	* guarantee without disabling preemption that the processor will not change
429	* in between and therefore the atomicity vs. interrupt cannot be exploited
430	* in a useful way here.
431	*/
432	void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
433	{
434	struct per_cpu_zonestat __percpu *pcp = zone->per_cpu_zonestats;
435	s8 __percpu *p = pcp->vm_stat_diff + item;
436	s8 v, t;
437
438	/* See __mod_node_page_state */
439	preempt_disable_nested();
440
441	v = __this_cpu_inc_return(*p);
442	t = __this_cpu_read(pcp->stat_threshold);
443	if (unlikely(v > t)) {
444	s8 overstep = t >> 1;
445
446	zone_page_state_add(x: v + overstep, zone, item);
447	__this_cpu_write(*p, -overstep);
448	}
449
450	preempt_enable_nested();
451	}
452
453	void __inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
454	{
455	struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
456	s8 __percpu *p = pcp->vm_node_stat_diff + item;
457	s8 v, t;
458
459	VM_WARN_ON_ONCE(vmstat_item_in_bytes(item));
460
461	/* See __mod_node_page_state */
462	preempt_disable_nested();
463
464	v = __this_cpu_inc_return(*p);
465	t = __this_cpu_read(pcp->stat_threshold);
466	if (unlikely(v > t)) {
467	s8 overstep = t >> 1;
468
469	node_page_state_add(x: v + overstep, pgdat, item);
470	__this_cpu_write(*p, -overstep);
471	}
472
473	preempt_enable_nested();
474	}
475
476	void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
477	{
478	__inc_zone_state(zone: page_zone(page), item);
479	}
480	EXPORT_SYMBOL(__inc_zone_page_state);
481
482	void __inc_node_page_state(struct page *page, enum node_stat_item item)
483	{
484	__inc_node_state(pgdat: page_pgdat(page), item);
485	}
486	EXPORT_SYMBOL(__inc_node_page_state);
487
488	void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
489	{
490	struct per_cpu_zonestat __percpu *pcp = zone->per_cpu_zonestats;
491	s8 __percpu *p = pcp->vm_stat_diff + item;
492	s8 v, t;
493
494	/* See __mod_node_page_state */
495	preempt_disable_nested();
496
497	v = __this_cpu_dec_return(*p);
498	t = __this_cpu_read(pcp->stat_threshold);
499	if (unlikely(v < - t)) {
500	s8 overstep = t >> 1;
501
502	zone_page_state_add(x: v - overstep, zone, item);
503	__this_cpu_write(*p, overstep);
504	}
505
506	preempt_enable_nested();
507	}
508
509	void __dec_node_state(struct pglist_data *pgdat, enum node_stat_item item)
510	{
511	struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
512	s8 __percpu *p = pcp->vm_node_stat_diff + item;
513	s8 v, t;
514
515	VM_WARN_ON_ONCE(vmstat_item_in_bytes(item));
516
517	/* See __mod_node_page_state */
518	preempt_disable_nested();
519
520	v = __this_cpu_dec_return(*p);
521	t = __this_cpu_read(pcp->stat_threshold);
522	if (unlikely(v < - t)) {
523	s8 overstep = t >> 1;
524
525	node_page_state_add(x: v - overstep, pgdat, item);
526	__this_cpu_write(*p, overstep);
527	}
528
529	preempt_enable_nested();
530	}
531
532	void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
533	{
534	__dec_zone_state(zone: page_zone(page), item);
535	}
536	EXPORT_SYMBOL(__dec_zone_page_state);
537
538	void __dec_node_page_state(struct page *page, enum node_stat_item item)
539	{
540	__dec_node_state(pgdat: page_pgdat(page), item);
541	}
542	EXPORT_SYMBOL(__dec_node_page_state);
543
544	#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
545	/*
546	* If we have cmpxchg_local support then we do not need to incur the overhead
547	* that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
548	*
549	* mod_state() modifies the zone counter state through atomic per cpu
550	* operations.
551	*
552	* Overstep mode specifies how overstep should handled:
553	* 0 No overstepping
554	* 1 Overstepping half of threshold
555	* -1 Overstepping minus half of threshold
556	*/
557	static inline void mod_zone_state(struct zone *zone,
558	enum zone_stat_item item, long delta, int overstep_mode)
559	{
560	struct per_cpu_zonestat __percpu *pcp = zone->per_cpu_zonestats;
561	s8 __percpu *p = pcp->vm_stat_diff + item;
562	long n, t, z;
563	s8 o;
564
565	o = this_cpu_read(*p);
566	do {
567	z = 0; /* overflow to zone counters */
568
569	/*
570	* The fetching of the stat_threshold is racy. We may apply
571	* a counter threshold to the wrong the cpu if we get
572	* rescheduled while executing here. However, the next
573	* counter update will apply the threshold again and
574	* therefore bring the counter under the threshold again.
575	*
576	* Most of the time the thresholds are the same anyways
577	* for all cpus in a zone.
578	*/
579	t = this_cpu_read(pcp->stat_threshold);
580
581	n = delta + (long)o;
582
583	if (abs(n) > t) {
584	int os = overstep_mode * (t >> 1) ;
585
586	/* Overflow must be added to zone counters */
587	z = n + os;
588	n = -os;
589	}
590	} while (!this_cpu_try_cmpxchg(*p, &o, n));
591
592	if (z)
593	zone_page_state_add(x: z, zone, item);
594	}
595
596	void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
597	long delta)
598	{
599	mod_zone_state(zone, item, delta, overstep_mode: 0);
600	}
601	EXPORT_SYMBOL(mod_zone_page_state);
602
603	void inc_zone_page_state(struct page *page, enum zone_stat_item item)
604	{
605	mod_zone_state(zone: page_zone(page), item, delta: 1, overstep_mode: 1);
606	}
607	EXPORT_SYMBOL(inc_zone_page_state);
608
609	void dec_zone_page_state(struct page *page, enum zone_stat_item item)
610	{
611	mod_zone_state(zone: page_zone(page), item, delta: -1, overstep_mode: -1);
612	}
613	EXPORT_SYMBOL(dec_zone_page_state);
614
615	static inline void mod_node_state(struct pglist_data *pgdat,
616	enum node_stat_item item, int delta, int overstep_mode)
617	{
618	struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
619	s8 __percpu *p = pcp->vm_node_stat_diff + item;
620	long n, t, z;
621	s8 o;
622
623	if (vmstat_item_in_bytes(idx: item)) {
624	/*
625	* Only cgroups use subpage accounting right now; at
626	* the global level, these items still change in
627	* multiples of whole pages. Store them as pages
628	* internally to keep the per-cpu counters compact.
629	*/
630	VM_WARN_ON_ONCE(delta & (PAGE_SIZE - 1));
631	delta >>= PAGE_SHIFT;
632	}
633
634	o = this_cpu_read(*p);
635	do {
636	z = 0; /* overflow to node counters */
637
638	/*
639	* The fetching of the stat_threshold is racy. We may apply
640	* a counter threshold to the wrong the cpu if we get
641	* rescheduled while executing here. However, the next
642	* counter update will apply the threshold again and
643	* therefore bring the counter under the threshold again.
644	*
645	* Most of the time the thresholds are the same anyways
646	* for all cpus in a node.
647	*/
648	t = this_cpu_read(pcp->stat_threshold);
649
650	n = delta + (long)o;
651
652	if (abs(n) > t) {
653	int os = overstep_mode * (t >> 1) ;
654
655	/* Overflow must be added to node counters */
656	z = n + os;
657	n = -os;
658	}
659	} while (!this_cpu_try_cmpxchg(*p, &o, n));
660
661	if (z)
662	node_page_state_add(x: z, pgdat, item);
663	}
664
665	void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
666	long delta)
667	{
668	mod_node_state(pgdat, item, delta, overstep_mode: 0);
669	}
670	EXPORT_SYMBOL(mod_node_page_state);
671
672	void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
673	{
674	mod_node_state(pgdat, item, delta: 1, overstep_mode: 1);
675	}
676
677	void inc_node_page_state(struct page *page, enum node_stat_item item)
678	{
679	mod_node_state(pgdat: page_pgdat(page), item, delta: 1, overstep_mode: 1);
680	}
681	EXPORT_SYMBOL(inc_node_page_state);
682
683	void dec_node_page_state(struct page *page, enum node_stat_item item)
684	{
685	mod_node_state(pgdat: page_pgdat(page), item, delta: -1, overstep_mode: -1);
686	}
687	EXPORT_SYMBOL(dec_node_page_state);
688	#else
689	/*
690	* Use interrupt disable to serialize counter updates
691	*/
692	void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
693	long delta)
694	{
695	unsigned long flags;
696
697	local_irq_save(flags);
698	__mod_zone_page_state(zone, item, delta);
699	local_irq_restore(flags);
700	}
701	EXPORT_SYMBOL(mod_zone_page_state);
702
703	void inc_zone_page_state(struct page *page, enum zone_stat_item item)
704	{
705	unsigned long flags;
706	struct zone *zone;
707
708	zone = page_zone(page);
709	local_irq_save(flags);
710	__inc_zone_state(zone, item);
711	local_irq_restore(flags);
712	}
713	EXPORT_SYMBOL(inc_zone_page_state);
714
715	void dec_zone_page_state(struct page *page, enum zone_stat_item item)
716	{
717	unsigned long flags;
718
719	local_irq_save(flags);
720	__dec_zone_page_state(page, item);
721	local_irq_restore(flags);
722	}
723	EXPORT_SYMBOL(dec_zone_page_state);
724
725	void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
726	{
727	unsigned long flags;
728
729	local_irq_save(flags);
730	__inc_node_state(pgdat, item);
731	local_irq_restore(flags);
732	}
733	EXPORT_SYMBOL(inc_node_state);
734
735	void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
736	long delta)
737	{
738	unsigned long flags;
739
740	local_irq_save(flags);
741	__mod_node_page_state(pgdat, item, delta);
742	local_irq_restore(flags);
743	}
744	EXPORT_SYMBOL(mod_node_page_state);
745
746	void inc_node_page_state(struct page *page, enum node_stat_item item)
747	{
748	unsigned long flags;
749	struct pglist_data *pgdat;
750
751	pgdat = page_pgdat(page);
752	local_irq_save(flags);
753	__inc_node_state(pgdat, item);
754	local_irq_restore(flags);
755	}
756	EXPORT_SYMBOL(inc_node_page_state);
757
758	void dec_node_page_state(struct page *page, enum node_stat_item item)
759	{
760	unsigned long flags;
761
762	local_irq_save(flags);
763	__dec_node_page_state(page, item);
764	local_irq_restore(flags);
765	}
766	EXPORT_SYMBOL(dec_node_page_state);
767	#endif
768
769	/*
770	* Fold a differential into the global counters.
771	* Returns the number of counters updated.
772	*/
773	static int fold_diff(int *zone_diff, int *node_diff)
774	{
775	int i;
776	int changes = 0;
777
778	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
779	if (zone_diff[i]) {
780	atomic_long_add(i: zone_diff[i], v: &vm_zone_stat[i]);
781	changes++;
782	}
783
784	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
785	if (node_diff[i]) {
786	atomic_long_add(i: node_diff[i], v: &vm_node_stat[i]);
787	changes++;
788	}
789	return changes;
790	}
791
792	/*
793	* Update the zone counters for the current cpu.
794	*
795	* Note that refresh_cpu_vm_stats strives to only access
796	* node local memory. The per cpu pagesets on remote zones are placed
797	* in the memory local to the processor using that pageset. So the
798	* loop over all zones will access a series of cachelines local to
799	* the processor.
800	*
801	* The call to zone_page_state_add updates the cachelines with the
802	* statistics in the remote zone struct as well as the global cachelines
803	* with the global counters. These could cause remote node cache line
804	* bouncing and will have to be only done when necessary.
805	*
806	* The function returns the number of global counters updated.
807	*/
808	static int refresh_cpu_vm_stats(bool do_pagesets)
809	{
810	struct pglist_data *pgdat;
811	struct zone *zone;
812	int i;
813	int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
814	int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
815	int changes = 0;
816
817	for_each_populated_zone(zone) {
818	struct per_cpu_zonestat __percpu *pzstats = zone->per_cpu_zonestats;
819	struct per_cpu_pages __percpu *pcp = zone->per_cpu_pageset;
820
821	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
822	int v;
823
824	v = this_cpu_xchg(pzstats->vm_stat_diff[i], 0);
825	if (v) {
826
827	atomic_long_add(i: v, v: &zone->vm_stat[i]);
828	global_zone_diff[i] += v;
829	#ifdef CONFIG_NUMA
830	/* 3 seconds idle till flush */
831	__this_cpu_write(pcp->expire, 3);
832	#endif
833	}
834	}
835
836	if (do_pagesets) {
837	cond_resched();
838
839	changes += decay_pcp_high(zone, this_cpu_ptr(pcp));
840	#ifdef CONFIG_NUMA
841	/*
842	* Deal with draining the remote pageset of this
843	* processor
844	*
845	* Check if there are pages remaining in this pageset
846	* if not then there is nothing to expire.
847	*/
848	if (!__this_cpu_read(pcp->expire) ||
849	!__this_cpu_read(pcp->count))
850	continue;
851
852	/*
853	* We never drain zones local to this processor.
854	*/
855	if (zone_to_nid(zone) == numa_node_id()) {
856	__this_cpu_write(pcp->expire, 0);
857	continue;
858	}
859
860	if (__this_cpu_dec_return(pcp->expire)) {
861	changes++;
862	continue;
863	}
864
865	if (__this_cpu_read(pcp->count)) {
866	drain_zone_pages(zone, this_cpu_ptr(pcp));
867	changes++;
868	}
869	#endif
870	}
871	}
872
873	for_each_online_pgdat(pgdat) {
874	struct per_cpu_nodestat __percpu *p = pgdat->per_cpu_nodestats;
875
876	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
877	int v;
878
879	v = this_cpu_xchg(p->vm_node_stat_diff[i], 0);
880	if (v) {
881	atomic_long_add(i: v, v: &pgdat->vm_stat[i]);
882	global_node_diff[i] += v;
883	}
884	}
885	}
886
887	changes += fold_diff(zone_diff: global_zone_diff, node_diff: global_node_diff);
888	return changes;
889	}
890
891	/*
892	* Fold the data for an offline cpu into the global array.
893	* There cannot be any access by the offline cpu and therefore
894	* synchronization is simplified.
895	*/
896	void cpu_vm_stats_fold(int cpu)
897	{
898	struct pglist_data *pgdat;
899	struct zone *zone;
900	int i;
901	int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
902	int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
903
904	for_each_populated_zone(zone) {
905	struct per_cpu_zonestat *pzstats;
906
907	pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
908
909	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
910	if (pzstats->vm_stat_diff[i]) {
911	int v;
912
913	v = pzstats->vm_stat_diff[i];
914	pzstats->vm_stat_diff[i] = 0;
915	atomic_long_add(i: v, v: &zone->vm_stat[i]);
916	global_zone_diff[i] += v;
917	}
918	}
919	#ifdef CONFIG_NUMA
920	for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++) {
921	if (pzstats->vm_numa_event[i]) {
922	unsigned long v;
923
924	v = pzstats->vm_numa_event[i];
925	pzstats->vm_numa_event[i] = 0;
926	zone_numa_event_add(x: v, zone, item: i);
927	}
928	}
929	#endif
930	}
931
932	for_each_online_pgdat(pgdat) {
933	struct per_cpu_nodestat *p;
934
935	p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
936
937	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
938	if (p->vm_node_stat_diff[i]) {
939	int v;
940
941	v = p->vm_node_stat_diff[i];
942	p->vm_node_stat_diff[i] = 0;
943	atomic_long_add(i: v, v: &pgdat->vm_stat[i]);
944	global_node_diff[i] += v;
945	}
946	}
947
948	fold_diff(zone_diff: global_zone_diff, node_diff: global_node_diff);
949	}
950
951	/*
952	* this is only called if !populated_zone(zone), which implies no other users of
953	* pset->vm_stat_diff[] exist.
954	*/
955	void drain_zonestat(struct zone *zone, struct per_cpu_zonestat *pzstats)
956	{
957	unsigned long v;
958	int i;
959
960	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
961	if (pzstats->vm_stat_diff[i]) {
962	v = pzstats->vm_stat_diff[i];
963	pzstats->vm_stat_diff[i] = 0;
964	zone_page_state_add(x: v, zone, item: i);
965	}
966	}
967
968	#ifdef CONFIG_NUMA
969	for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++) {
970	if (pzstats->vm_numa_event[i]) {
971	v = pzstats->vm_numa_event[i];
972	pzstats->vm_numa_event[i] = 0;
973	zone_numa_event_add(x: v, zone, item: i);
974	}
975	}
976	#endif
977	}
978	#endif
979
980	#ifdef CONFIG_NUMA
981	/*
982	* Determine the per node value of a stat item. This function
983	* is called frequently in a NUMA machine, so try to be as
984	* frugal as possible.
985	*/
986	unsigned long sum_zone_node_page_state(int node,
987	enum zone_stat_item item)
988	{
989	struct zone *zones = NODE_DATA(node)->node_zones;
990	int i;
991	unsigned long count = 0;
992
993	for (i = 0; i < MAX_NR_ZONES; i++)
994	count += zone_page_state(zone: zones + i, item);
995
996	return count;
997	}
998
999	/* Determine the per node value of a numa stat item. */
1000	unsigned long sum_zone_numa_event_state(int node,
1001	enum numa_stat_item item)
1002	{
1003	struct zone *zones = NODE_DATA(node)->node_zones;
1004	unsigned long count = 0;
1005	int i;
1006
1007	for (i = 0; i < MAX_NR_ZONES; i++)
1008	count += zone_numa_event_state(zone: zones + i, item);
1009
1010	return count;
1011	}
1012
1013	/*
1014	* Determine the per node value of a stat item.
1015	*/
1016	unsigned long node_page_state_pages(struct pglist_data *pgdat,
1017	enum node_stat_item item)
1018	{
1019	long x = atomic_long_read(v: &pgdat->vm_stat[item]);
1020	#ifdef CONFIG_SMP
1021	if (x < 0)
1022	x = 0;
1023	#endif
1024	return x;
1025	}
1026
1027	unsigned long node_page_state(struct pglist_data *pgdat,
1028	enum node_stat_item item)
1029	{
1030	VM_WARN_ON_ONCE(vmstat_item_in_bytes(item));
1031
1032	return node_page_state_pages(pgdat, item);
1033	}
1034	#endif
1035
1036	#ifdef CONFIG_COMPACTION
1037
1038	struct contig_page_info {
1039	unsigned long free_pages;
1040	unsigned long free_blocks_total;
1041	unsigned long free_blocks_suitable;
1042	};
1043
1044	/*
1045	* Calculate the number of free pages in a zone, how many contiguous
1046	* pages are free and how many are large enough to satisfy an allocation of
1047	* the target size. Note that this function makes no attempt to estimate
1048	* how many suitable free blocks there *might* be if MOVABLE pages were
1049	* migrated. Calculating that is possible, but expensive and can be
1050	* figured out from userspace
1051	*/
1052	static void fill_contig_page_info(struct zone *zone,
1053	unsigned int suitable_order,
1054	struct contig_page_info *info)
1055	{
1056	unsigned int order;
1057
1058	info->free_pages = 0;
1059	info->free_blocks_total = 0;
1060	info->free_blocks_suitable = 0;
1061
1062	for (order = 0; order <= MAX_ORDER; order++) {
1063	unsigned long blocks;
1064
1065	/*
1066	* Count number of free blocks.
1067	*
1068	* Access to nr_free is lockless as nr_free is used only for
1069	* diagnostic purposes. Use data_race to avoid KCSAN warning.
1070	*/
1071	blocks = data_race(zone->free_area[order].nr_free);
1072	info->free_blocks_total += blocks;
1073
1074	/* Count free base pages */
1075	info->free_pages += blocks << order;
1076
1077	/* Count the suitable free blocks */
1078	if (order >= suitable_order)
1079	info->free_blocks_suitable += blocks <<
1080	(order - suitable_order);
1081	}
1082	}
1083
1084	/*
1085	* A fragmentation index only makes sense if an allocation of a requested
1086	* size would fail. If that is true, the fragmentation index indicates
1087	* whether external fragmentation or a lack of memory was the problem.
1088	* The value can be used to determine if page reclaim or compaction
1089	* should be used
1090	*/
1091	static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
1092	{
1093	unsigned long requested = 1UL << order;
1094
1095	if (WARN_ON_ONCE(order > MAX_ORDER))
1096	return 0;
1097
1098	if (!info->free_blocks_total)
1099	return 0;
1100
1101	/* Fragmentation index only makes sense when a request would fail */
1102	if (info->free_blocks_suitable)
1103	return -1000;
1104
1105	/*
1106	* Index is between 0 and 1 so return within 3 decimal places
1107	*
1108	* 0 => allocation would fail due to lack of memory
1109	* 1 => allocation would fail due to fragmentation
1110	*/
1111	return 1000 - div_u64( dividend: (1000+(div_u64(dividend: info->free_pages * 1000ULL, divisor: requested))), divisor: info->free_blocks_total);
1112	}
1113
1114	/*
1115	* Calculates external fragmentation within a zone wrt the given order.
1116	* It is defined as the percentage of pages found in blocks of size
1117	* less than 1 << order. It returns values in range [0, 100].
1118	*/
1119	unsigned int extfrag_for_order(struct zone *zone, unsigned int order)
1120	{
1121	struct contig_page_info info;
1122
1123	fill_contig_page_info(zone, suitable_order: order, info: &info);
1124	if (info.free_pages == 0)
1125	return 0;
1126
1127	return div_u64(dividend: (info.free_pages -
1128	(info.free_blocks_suitable << order)) * 100,
1129	divisor: info.free_pages);
1130	}
1131
1132	/* Same as __fragmentation index but allocs contig_page_info on stack */
1133	int fragmentation_index(struct zone *zone, unsigned int order)
1134	{
1135	struct contig_page_info info;
1136
1137	fill_contig_page_info(zone, suitable_order: order, info: &info);
1138	return __fragmentation_index(order, info: &info);
1139	}
1140	#endif
1141
1142	#if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || \
1143	defined(CONFIG_NUMA) || defined(CONFIG_MEMCG)
1144	#ifdef CONFIG_ZONE_DMA
1145	#define TEXT_FOR_DMA(xx) xx "_dma",
1146	#else
1147	#define TEXT_FOR_DMA(xx)
1148	#endif
1149
1150	#ifdef CONFIG_ZONE_DMA32
1151	#define TEXT_FOR_DMA32(xx) xx "_dma32",
1152	#else
1153	#define TEXT_FOR_DMA32(xx)
1154	#endif
1155
1156	#ifdef CONFIG_HIGHMEM
1157	#define TEXT_FOR_HIGHMEM(xx) xx "_high",
1158	#else
1159	#define TEXT_FOR_HIGHMEM(xx)
1160	#endif
1161
1162	#ifdef CONFIG_ZONE_DEVICE
1163	#define TEXT_FOR_DEVICE(xx) xx "_device",
1164	#else
1165	#define TEXT_FOR_DEVICE(xx)
1166	#endif
1167
1168	#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
1169	TEXT_FOR_HIGHMEM(xx) xx "_movable", \
1170	TEXT_FOR_DEVICE(xx)
1171
1172	const char * const vmstat_text[] = {
1173	/* enum zone_stat_item counters */
1174	"nr_free_pages",
1175	"nr_zone_inactive_anon",
1176	"nr_zone_active_anon",
1177	"nr_zone_inactive_file",
1178	"nr_zone_active_file",
1179	"nr_zone_unevictable",
1180	"nr_zone_write_pending",
1181	"nr_mlock",
1182	"nr_bounce",
1183	#if IS_ENABLED(CONFIG_ZSMALLOC)
1184	"nr_zspages",
1185	#endif
1186	"nr_free_cma",
1187	#ifdef CONFIG_UNACCEPTED_MEMORY
1188	"nr_unaccepted",
1189	#endif
1190
1191	/* enum numa_stat_item counters */
1192	#ifdef CONFIG_NUMA
1193	"numa_hit",
1194	"numa_miss",
1195	"numa_foreign",
1196	"numa_interleave",
1197	"numa_local",
1198	"numa_other",
1199	#endif
1200
1201	/* enum node_stat_item counters */
1202	"nr_inactive_anon",
1203	"nr_active_anon",
1204	"nr_inactive_file",
1205	"nr_active_file",
1206	"nr_unevictable",
1207	"nr_slab_reclaimable",
1208	"nr_slab_unreclaimable",
1209	"nr_isolated_anon",
1210	"nr_isolated_file",
1211	"workingset_nodes",
1212	"workingset_refault_anon",
1213	"workingset_refault_file",
1214	"workingset_activate_anon",
1215	"workingset_activate_file",
1216	"workingset_restore_anon",
1217	"workingset_restore_file",
1218	"workingset_nodereclaim",
1219	"nr_anon_pages",
1220	"nr_mapped",
1221	"nr_file_pages",
1222	"nr_dirty",
1223	"nr_writeback",
1224	"nr_writeback_temp",
1225	"nr_shmem",
1226	"nr_shmem_hugepages",
1227	"nr_shmem_pmdmapped",
1228	"nr_file_hugepages",
1229	"nr_file_pmdmapped",
1230	"nr_anon_transparent_hugepages",
1231	"nr_vmscan_write",
1232	"nr_vmscan_immediate_reclaim",
1233	"nr_dirtied",
1234	"nr_written",
1235	"nr_throttled_written",
1236	"nr_kernel_misc_reclaimable",
1237	"nr_foll_pin_acquired",
1238	"nr_foll_pin_released",
1239	"nr_kernel_stack",
1240	#if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
1241	"nr_shadow_call_stack",
1242	#endif
1243	"nr_page_table_pages",
1244	"nr_sec_page_table_pages",
1245	#ifdef CONFIG_SWAP
1246	"nr_swapcached",
1247	#endif
1248	#ifdef CONFIG_NUMA_BALANCING
1249	"pgpromote_success",
1250	"pgpromote_candidate",
1251	#endif
1252
1253	/* enum writeback_stat_item counters */
1254	"nr_dirty_threshold",
1255	"nr_dirty_background_threshold",
1256
1257	#if defined(CONFIG_VM_EVENT_COUNTERS) || defined(CONFIG_MEMCG)
1258	/* enum vm_event_item counters */
1259	"pgpgin",
1260	"pgpgout",
1261	"pswpin",
1262	"pswpout",
1263
1264	TEXTS_FOR_ZONES("pgalloc")
1265	TEXTS_FOR_ZONES("allocstall")
1266	TEXTS_FOR_ZONES("pgskip")
1267
1268	"pgfree",
1269	"pgactivate",
1270	"pgdeactivate",
1271	"pglazyfree",
1272
1273	"pgfault",
1274	"pgmajfault",
1275	"pglazyfreed",
1276
1277	"pgrefill",
1278	"pgreuse",
1279	"pgsteal_kswapd",
1280	"pgsteal_direct",
1281	"pgsteal_khugepaged",
1282	"pgdemote_kswapd",
1283	"pgdemote_direct",
1284	"pgdemote_khugepaged",
1285	"pgscan_kswapd",
1286	"pgscan_direct",
1287	"pgscan_khugepaged",
1288	"pgscan_direct_throttle",
1289	"pgscan_anon",
1290	"pgscan_file",
1291	"pgsteal_anon",
1292	"pgsteal_file",
1293
1294	#ifdef CONFIG_NUMA
1295	"zone_reclaim_failed",
1296	#endif
1297	"pginodesteal",
1298	"slabs_scanned",
1299	"kswapd_inodesteal",
1300	"kswapd_low_wmark_hit_quickly",
1301	"kswapd_high_wmark_hit_quickly",
1302	"pageoutrun",
1303
1304	"pgrotated",
1305
1306	"drop_pagecache",
1307	"drop_slab",
1308	"oom_kill",
1309
1310	#ifdef CONFIG_NUMA_BALANCING
1311	"numa_pte_updates",
1312	"numa_huge_pte_updates",
1313	"numa_hint_faults",
1314	"numa_hint_faults_local",
1315	"numa_pages_migrated",
1316	#endif
1317	#ifdef CONFIG_MIGRATION
1318	"pgmigrate_success",
1319	"pgmigrate_fail",
1320	"thp_migration_success",
1321	"thp_migration_fail",
1322	"thp_migration_split",
1323	#endif
1324	#ifdef CONFIG_COMPACTION
1325	"compact_migrate_scanned",
1326	"compact_free_scanned",
1327	"compact_isolated",
1328	"compact_stall",
1329	"compact_fail",
1330	"compact_success",
1331	"compact_daemon_wake",
1332	"compact_daemon_migrate_scanned",
1333	"compact_daemon_free_scanned",
1334	#endif
1335
1336	#ifdef CONFIG_HUGETLB_PAGE
1337	"htlb_buddy_alloc_success",
1338	"htlb_buddy_alloc_fail",
1339	#endif
1340	#ifdef CONFIG_CMA
1341	"cma_alloc_success",
1342	"cma_alloc_fail",
1343	#endif
1344	"unevictable_pgs_culled",
1345	"unevictable_pgs_scanned",
1346	"unevictable_pgs_rescued",
1347	"unevictable_pgs_mlocked",
1348	"unevictable_pgs_munlocked",
1349	"unevictable_pgs_cleared",
1350	"unevictable_pgs_stranded",
1351
1352	#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1353	"thp_fault_alloc",
1354	"thp_fault_fallback",
1355	"thp_fault_fallback_charge",
1356	"thp_collapse_alloc",
1357	"thp_collapse_alloc_failed",
1358	"thp_file_alloc",
1359	"thp_file_fallback",
1360	"thp_file_fallback_charge",
1361	"thp_file_mapped",
1362	"thp_split_page",
1363	"thp_split_page_failed",
1364	"thp_deferred_split_page",
1365	"thp_split_pmd",
1366	"thp_scan_exceed_none_pte",
1367	"thp_scan_exceed_swap_pte",
1368	"thp_scan_exceed_share_pte",
1369	#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
1370	"thp_split_pud",
1371	#endif
1372	"thp_zero_page_alloc",
1373	"thp_zero_page_alloc_failed",
1374	"thp_swpout",
1375	"thp_swpout_fallback",
1376	#endif
1377	#ifdef CONFIG_MEMORY_BALLOON
1378	"balloon_inflate",
1379	"balloon_deflate",
1380	#ifdef CONFIG_BALLOON_COMPACTION
1381	"balloon_migrate",
1382	#endif
1383	#endif /* CONFIG_MEMORY_BALLOON */
1384	#ifdef CONFIG_DEBUG_TLBFLUSH
1385	"nr_tlb_remote_flush",
1386	"nr_tlb_remote_flush_received",
1387	"nr_tlb_local_flush_all",
1388	"nr_tlb_local_flush_one",
1389	#endif /* CONFIG_DEBUG_TLBFLUSH */
1390
1391	#ifdef CONFIG_SWAP
1392	"swap_ra",
1393	"swap_ra_hit",
1394	#ifdef CONFIG_KSM
1395	"ksm_swpin_copy",
1396	#endif
1397	#endif
1398	#ifdef CONFIG_KSM
1399	"cow_ksm",
1400	#endif
1401	#ifdef CONFIG_ZSWAP
1402	"zswpin",
1403	"zswpout",
1404	#endif
1405	#ifdef CONFIG_X86
1406	"direct_map_level2_splits",
1407	"direct_map_level3_splits",
1408	#endif
1409	#ifdef CONFIG_PER_VMA_LOCK_STATS
1410	"vma_lock_success",
1411	"vma_lock_abort",
1412	"vma_lock_retry",
1413	"vma_lock_miss",
1414	#endif
1415	#endif /* CONFIG_VM_EVENT_COUNTERS || CONFIG_MEMCG */
1416	};
1417	#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA || CONFIG_MEMCG */
1418
1419	#if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
1420	defined(CONFIG_PROC_FS)
1421	static void *frag_start(struct seq_file *m, loff_t *pos)
1422	{
1423	pg_data_t *pgdat;
1424	loff_t node = *pos;
1425
1426	for (pgdat = first_online_pgdat();
1427	pgdat && node;
1428	pgdat = next_online_pgdat(pgdat))
1429	--node;
1430
1431	return pgdat;
1432	}
1433
1434	static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
1435	{
1436	pg_data_t *pgdat = (pg_data_t *)arg;
1437
1438	(*pos)++;
1439	return next_online_pgdat(pgdat);
1440	}
1441
1442	static void frag_stop(struct seq_file *m, void *arg)
1443	{
1444	}
1445
1446	/*
1447	* Walk zones in a node and print using a callback.
1448	* If @assert_populated is true, only use callback for zones that are populated.
1449	*/
1450	static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
1451	bool assert_populated, bool nolock,
1452	void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
1453	{
1454	struct zone *zone;
1455	struct zone *node_zones = pgdat->node_zones;
1456	unsigned long flags;
1457
1458	for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
1459	if (assert_populated && !populated_zone(zone))
1460	continue;
1461
1462	if (!nolock)
1463	spin_lock_irqsave(&zone->lock, flags);
1464	print(m, pgdat, zone);
1465	if (!nolock)
1466	spin_unlock_irqrestore(lock: &zone->lock, flags);
1467	}
1468	}
1469	#endif
1470
1471	#ifdef CONFIG_PROC_FS
1472	static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
1473	struct zone *zone)
1474	{
1475	int order;
1476
1477	seq_printf(m, fmt: "Node %d, zone %8s ", pgdat->node_id, zone->name);
1478	for (order = 0; order <= MAX_ORDER; ++order)
1479	/*
1480	* Access to nr_free is lockless as nr_free is used only for
1481	* printing purposes. Use data_race to avoid KCSAN warning.
1482	*/
1483	seq_printf(m, fmt: "%6lu ", data_race(zone->free_area[order].nr_free));
1484	seq_putc(m, c: '\n');
1485	}
1486
1487	/*
1488	* This walks the free areas for each zone.
1489	*/
1490	static int frag_show(struct seq_file *m, void *arg)
1491	{
1492	pg_data_t *pgdat = (pg_data_t *)arg;
1493	walk_zones_in_node(m, pgdat, assert_populated: true, nolock: false, print: frag_show_print);
1494	return 0;
1495	}
1496
1497	static void pagetypeinfo_showfree_print(struct seq_file *m,
1498	pg_data_t *pgdat, struct zone *zone)
1499	{
1500	int order, mtype;
1501
1502	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
1503	seq_printf(m, fmt: "Node %4d, zone %8s, type %12s ",
1504	pgdat->node_id,
1505	zone->name,
1506	migratetype_names[mtype]);
1507	for (order = 0; order <= MAX_ORDER; ++order) {
1508	unsigned long freecount = 0;
1509	struct free_area *area;
1510	struct list_head *curr;
1511	bool overflow = false;
1512
1513	area = &(zone->free_area[order]);
1514
1515	list_for_each(curr, &area->free_list[mtype]) {
1516	/*
1517	* Cap the free_list iteration because it might
1518	* be really large and we are under a spinlock
1519	* so a long time spent here could trigger a
1520	* hard lockup detector. Anyway this is a
1521	* debugging tool so knowing there is a handful
1522	* of pages of this order should be more than
1523	* sufficient.
1524	*/
1525	if (++freecount >= 100000) {
1526	overflow = true;
1527	break;
1528	}
1529	}
1530	seq_printf(m, fmt: "%s%6lu ", overflow ? ">" : "", freecount);
1531	spin_unlock_irq(lock: &zone->lock);
1532	cond_resched();
1533	spin_lock_irq(lock: &zone->lock);
1534	}
1535	seq_putc(m, c: '\n');
1536	}
1537	}
1538
1539	/* Print out the free pages at each order for each migatetype */
1540	static void pagetypeinfo_showfree(struct seq_file *m, void *arg)
1541	{
1542	int order;
1543	pg_data_t *pgdat = (pg_data_t *)arg;
1544
1545	/* Print header */
1546	seq_printf(m, fmt: "%-43s ", "Free pages count per migrate type at order");
1547	for (order = 0; order <= MAX_ORDER; ++order)
1548	seq_printf(m, fmt: "%6d ", order);
1549	seq_putc(m, c: '\n');
1550
1551	walk_zones_in_node(m, pgdat, assert_populated: true, nolock: false, print: pagetypeinfo_showfree_print);
1552	}
1553
1554	static void pagetypeinfo_showblockcount_print(struct seq_file *m,
1555	pg_data_t *pgdat, struct zone *zone)
1556	{
1557	int mtype;
1558	unsigned long pfn;
1559	unsigned long start_pfn = zone->zone_start_pfn;
1560	unsigned long end_pfn = zone_end_pfn(zone);
1561	unsigned long count[MIGRATE_TYPES] = { 0, };
1562
1563	for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
1564	struct page *page;
1565
1566	page = pfn_to_online_page(pfn);
1567	if (!page)
1568	continue;
1569
1570	if (page_zone(page) != zone)
1571	continue;
1572
1573	mtype = get_pageblock_migratetype(page);
1574
1575	if (mtype < MIGRATE_TYPES)
1576	count[mtype]++;
1577	}
1578
1579	/* Print counts */
1580	seq_printf(m, fmt: "Node %d, zone %8s ", pgdat->node_id, zone->name);
1581	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1582	seq_printf(m, fmt: "%12lu ", count[mtype]);
1583	seq_putc(m, c: '\n');
1584	}
1585
1586	/* Print out the number of pageblocks for each migratetype */
1587	static void pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1588	{
1589	int mtype;
1590	pg_data_t *pgdat = (pg_data_t *)arg;
1591
1592	seq_printf(m, fmt: "\n%-23s", "Number of blocks type ");
1593	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1594	seq_printf(m, fmt: "%12s ", migratetype_names[mtype]);
1595	seq_putc(m, c: '\n');
1596	walk_zones_in_node(m, pgdat, assert_populated: true, nolock: false,
1597	print: pagetypeinfo_showblockcount_print);
1598	}
1599
1600	/*
1601	* Print out the number of pageblocks for each migratetype that contain pages
1602	* of other types. This gives an indication of how well fallbacks are being
1603	* contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1604	* to determine what is going on
1605	*/
1606	static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1607	{
1608	#ifdef CONFIG_PAGE_OWNER
1609	int mtype;
1610
1611	if (!static_branch_unlikely(&page_owner_inited))
1612	return;
1613
1614	drain_all_pages(NULL);
1615
1616	seq_printf(m, fmt: "\n%-23s", "Number of mixed blocks ");
1617	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1618	seq_printf(m, fmt: "%12s ", migratetype_names[mtype]);
1619	seq_putc(m, c: '\n');
1620
1621	walk_zones_in_node(m, pgdat, assert_populated: true, nolock: true,
1622	print: pagetypeinfo_showmixedcount_print);
1623	#endif /* CONFIG_PAGE_OWNER */
1624	}
1625
1626	/*
1627	* This prints out statistics in relation to grouping pages by mobility.
1628	* It is expensive to collect so do not constantly read the file.
1629	*/
1630	static int pagetypeinfo_show(struct seq_file *m, void *arg)
1631	{
1632	pg_data_t *pgdat = (pg_data_t *)arg;
1633
1634	/* check memoryless node */
1635	if (!node_state(node: pgdat->node_id, state: N_MEMORY))
1636	return 0;
1637
1638	seq_printf(m, fmt: "Page block order: %d\n", pageblock_order);
1639	seq_printf(m, fmt: "Pages per block: %lu\n", pageblock_nr_pages);
1640	seq_putc(m, c: '\n');
1641	pagetypeinfo_showfree(m, arg: pgdat);
1642	pagetypeinfo_showblockcount(m, arg: pgdat);
1643	pagetypeinfo_showmixedcount(m, pgdat);
1644
1645	return 0;
1646	}
1647
1648	static const struct seq_operations fragmentation_op = {
1649	.start = frag_start,
1650	.next = frag_next,
1651	.stop = frag_stop,
1652	.show = frag_show,
1653	};
1654
1655	static const struct seq_operations pagetypeinfo_op = {
1656	.start = frag_start,
1657	.next = frag_next,
1658	.stop = frag_stop,
1659	.show = pagetypeinfo_show,
1660	};
1661
1662	static bool is_zone_first_populated(pg_data_t *pgdat, struct zone *zone)
1663	{
1664	int zid;
1665
1666	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1667	struct zone *compare = &pgdat->node_zones[zid];
1668
1669	if (populated_zone(zone: compare))
1670	return zone == compare;
1671	}
1672
1673	return false;
1674	}
1675
1676	static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1677	struct zone *zone)
1678	{
1679	int i;
1680	seq_printf(m, fmt: "Node %d, zone %8s", pgdat->node_id, zone->name);
1681	if (is_zone_first_populated(pgdat, zone)) {
1682	seq_printf(m, fmt: "\n per-node stats");
1683	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
1684	unsigned long pages = node_page_state_pages(pgdat, item: i);
1685
1686	if (vmstat_item_print_in_thp(item: i))
1687	pages /= HPAGE_PMD_NR;
1688	seq_printf(m, fmt: "\n %-12s %lu", node_stat_name(item: i),
1689	pages);
1690	}
1691	}
1692	seq_printf(m,
1693	fmt: "\n pages free %lu"
1694	"\n boost %lu"
1695	"\n min %lu"
1696	"\n low %lu"
1697	"\n high %lu"
1698	"\n spanned %lu"
1699	"\n present %lu"
1700	"\n managed %lu"
1701	"\n cma %lu",
1702	zone_page_state(zone, item: NR_FREE_PAGES),
1703	zone->watermark_boost,
1704	min_wmark_pages(zone),
1705	low_wmark_pages(zone),
1706	high_wmark_pages(zone),
1707	zone->spanned_pages,
1708	zone->present_pages,
1709	zone_managed_pages(zone),
1710	zone_cma_pages(zone));
1711
1712	seq_printf(m,
1713	fmt: "\n protection: (%ld",
1714	zone->lowmem_reserve[0]);
1715	for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1716	seq_printf(m, fmt: ", %ld", zone->lowmem_reserve[i]);
1717	seq_putc(m, c: ')');
1718
1719	/* If unpopulated, no other information is useful */
1720	if (!populated_zone(zone)) {
1721	seq_putc(m, c: '\n');
1722	return;
1723	}
1724
1725	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1726	seq_printf(m, fmt: "\n %-12s %lu", zone_stat_name(item: i),
1727	zone_page_state(zone, item: i));
1728
1729	#ifdef CONFIG_NUMA
1730	for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
1731	seq_printf(m, fmt: "\n %-12s %lu", numa_stat_name(item: i),
1732	zone_numa_event_state(zone, item: i));
1733	#endif
1734
1735	seq_printf(m, fmt: "\n pagesets");
1736	for_each_online_cpu(i) {
1737	struct per_cpu_pages *pcp;
1738	struct per_cpu_zonestat __maybe_unused *pzstats;
1739
1740	pcp = per_cpu_ptr(zone->per_cpu_pageset, i);
1741	seq_printf(m,
1742	fmt: "\n cpu: %i"
1743	"\n count: %i"
1744	"\n high: %i"
1745	"\n batch: %i",
1746	i,
1747	pcp->count,
1748	pcp->high,
1749	pcp->batch);
1750	#ifdef CONFIG_SMP
1751	pzstats = per_cpu_ptr(zone->per_cpu_zonestats, i);
1752	seq_printf(m, fmt: "\n vm stats threshold: %d",
1753	pzstats->stat_threshold);
1754	#endif
1755	}
1756	seq_printf(m,
1757	fmt: "\n node_unreclaimable: %u"
1758	"\n start_pfn: %lu",
1759	pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES,
1760	zone->zone_start_pfn);
1761	seq_putc(m, c: '\n');
1762	}
1763
1764	/*
1765	* Output information about zones in @pgdat. All zones are printed regardless
1766	* of whether they are populated or not: lowmem_reserve_ratio operates on the
1767	* set of all zones and userspace would not be aware of such zones if they are
1768	* suppressed here (zoneinfo displays the effect of lowmem_reserve_ratio).
1769	*/
1770	static int zoneinfo_show(struct seq_file *m, void *arg)
1771	{
1772	pg_data_t *pgdat = (pg_data_t *)arg;
1773	walk_zones_in_node(m, pgdat, assert_populated: false, nolock: false, print: zoneinfo_show_print);
1774	return 0;
1775	}
1776
1777	static const struct seq_operations zoneinfo_op = {
1778	.start = frag_start, /* iterate over all zones. The same as in
1779	* fragmentation. */
1780	.next = frag_next,
1781	.stop = frag_stop,
1782	.show = zoneinfo_show,
1783	};
1784
1785	#define NR_VMSTAT_ITEMS (NR_VM_ZONE_STAT_ITEMS + \
1786	NR_VM_NUMA_EVENT_ITEMS + \
1787	NR_VM_NODE_STAT_ITEMS + \
1788	NR_VM_WRITEBACK_STAT_ITEMS + \
1789	(IS_ENABLED(CONFIG_VM_EVENT_COUNTERS) ? \
1790	NR_VM_EVENT_ITEMS : 0))
1791
1792	static void *vmstat_start(struct seq_file *m, loff_t *pos)
1793	{
1794	unsigned long *v;
1795	int i;
1796
1797	if (*pos >= NR_VMSTAT_ITEMS)
1798	return NULL;
1799
1800	BUILD_BUG_ON(ARRAY_SIZE(vmstat_text) < NR_VMSTAT_ITEMS);
1801	fold_vm_numa_events();
1802	v = kmalloc_array(NR_VMSTAT_ITEMS, size: sizeof(unsigned long), GFP_KERNEL);
1803	m->private = v;
1804	if (!v)
1805	return ERR_PTR(error: -ENOMEM);
1806	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1807	v[i] = global_zone_page_state(item: i);
1808	v += NR_VM_ZONE_STAT_ITEMS;
1809
1810	#ifdef CONFIG_NUMA
1811	for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
1812	v[i] = global_numa_event_state(item: i);
1813	v += NR_VM_NUMA_EVENT_ITEMS;
1814	#endif
1815
1816	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
1817	v[i] = global_node_page_state_pages(item: i);
1818	if (vmstat_item_print_in_thp(item: i))
1819	v[i] /= HPAGE_PMD_NR;
1820	}
1821	v += NR_VM_NODE_STAT_ITEMS;
1822
1823	global_dirty_limits(pbackground: v + NR_DIRTY_BG_THRESHOLD,
1824	pdirty: v + NR_DIRTY_THRESHOLD);
1825	v += NR_VM_WRITEBACK_STAT_ITEMS;
1826
1827	#ifdef CONFIG_VM_EVENT_COUNTERS
1828	all_vm_events(v);
1829	v[PGPGIN] /= 2; /* sectors -> kbytes */
1830	v[PGPGOUT] /= 2;
1831	#endif
1832	return (unsigned long *)m->private + *pos;
1833	}
1834
1835	static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1836	{
1837	(*pos)++;
1838	if (*pos >= NR_VMSTAT_ITEMS)
1839	return NULL;
1840	return (unsigned long *)m->private + *pos;
1841	}
1842
1843	static int vmstat_show(struct seq_file *m, void *arg)
1844	{
1845	unsigned long *l = arg;
1846	unsigned long off = l - (unsigned long *)m->private;
1847
1848	seq_puts(m, s: vmstat_text[off]);
1849	seq_put_decimal_ull(m, delimiter: " ", num: *l);
1850	seq_putc(m, c: '\n');
1851
1852	if (off == NR_VMSTAT_ITEMS - 1) {
1853	/*
1854	* We've come to the end - add any deprecated counters to avoid
1855	* breaking userspace which might depend on them being present.
1856	*/
1857	seq_puts(m, s: "nr_unstable 0\n");
1858	}
1859	return 0;
1860	}
1861
1862	static void vmstat_stop(struct seq_file *m, void *arg)
1863	{
1864	kfree(objp: m->private);
1865	m->private = NULL;
1866	}
1867
1868	static const struct seq_operations vmstat_op = {
1869	.start = vmstat_start,
1870	.next = vmstat_next,
1871	.stop = vmstat_stop,
1872	.show = vmstat_show,
1873	};
1874	#endif /* CONFIG_PROC_FS */
1875
1876	#ifdef CONFIG_SMP
1877	static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1878	int sysctl_stat_interval __read_mostly = HZ;
1879
1880	#ifdef CONFIG_PROC_FS
1881	static void refresh_vm_stats(struct work_struct *work)
1882	{
1883	refresh_cpu_vm_stats(do_pagesets: true);
1884	}
1885
1886	int vmstat_refresh(struct ctl_table *table, int write,
1887	void *buffer, size_t *lenp, loff_t *ppos)
1888	{
1889	long val;
1890	int err;
1891	int i;
1892
1893	/*
1894	* The regular update, every sysctl_stat_interval, may come later
1895	* than expected: leaving a significant amount in per_cpu buckets.
1896	* This is particularly misleading when checking a quantity of HUGE
1897	* pages, immediately after running a test. /proc/sys/vm/stat_refresh,
1898	* which can equally be echo'ed to or cat'ted from (by root),
1899	* can be used to update the stats just before reading them.
1900	*
1901	* Oh, and since global_zone_page_state() etc. are so careful to hide
1902	* transiently negative values, report an error here if any of
1903	* the stats is negative, so we know to go looking for imbalance.
1904	*/
1905	err = schedule_on_each_cpu(func: refresh_vm_stats);
1906	if (err)
1907	return err;
1908	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
1909	/*
1910	* Skip checking stats known to go negative occasionally.
1911	*/
1912	switch (i) {
1913	case NR_ZONE_WRITE_PENDING:
1914	case NR_FREE_CMA_PAGES:
1915	continue;
1916	}
1917	val = atomic_long_read(v: &vm_zone_stat[i]);
1918	if (val < 0) {
1919	pr_warn("%s: %s %ld\n",
1920	__func__, zone_stat_name(i), val);
1921	}
1922	}
1923	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
1924	/*
1925	* Skip checking stats known to go negative occasionally.
1926	*/
1927	switch (i) {
1928	case NR_WRITEBACK:
1929	continue;
1930	}
1931	val = atomic_long_read(v: &vm_node_stat[i]);
1932	if (val < 0) {
1933	pr_warn("%s: %s %ld\n",
1934	__func__, node_stat_name(i), val);
1935	}
1936	}
1937	if (write)
1938	*ppos += *lenp;
1939	else
1940	*lenp = 0;
1941	return 0;
1942	}
1943	#endif /* CONFIG_PROC_FS */
1944
1945	static void vmstat_update(struct work_struct *w)
1946	{
1947	if (refresh_cpu_vm_stats(do_pagesets: true)) {
1948	/*
1949	* Counters were updated so we expect more updates
1950	* to occur in the future. Keep on running the
1951	* update worker thread.
1952	*/
1953	queue_delayed_work_on(smp_processor_id(), wq: mm_percpu_wq,
1954	this_cpu_ptr(&vmstat_work),
1955	delay: round_jiffies_relative(j: sysctl_stat_interval));
1956	}
1957	}
1958
1959	/*
1960	* Check if the diffs for a certain cpu indicate that
1961	* an update is needed.
1962	*/
1963	static bool need_update(int cpu)
1964	{
1965	pg_data_t *last_pgdat = NULL;
1966	struct zone *zone;
1967
1968	for_each_populated_zone(zone) {
1969	struct per_cpu_zonestat *pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
1970	struct per_cpu_nodestat *n;
1971
1972	/*
1973	* The fast way of checking if there are any vmstat diffs.
1974	*/
1975	if (memchr_inv(p: pzstats->vm_stat_diff, c: 0, size: sizeof(pzstats->vm_stat_diff)))
1976	return true;
1977
1978	if (last_pgdat == zone->zone_pgdat)
1979	continue;
1980	last_pgdat = zone->zone_pgdat;
1981	n = per_cpu_ptr(zone->zone_pgdat->per_cpu_nodestats, cpu);
1982	if (memchr_inv(p: n->vm_node_stat_diff, c: 0, size: sizeof(n->vm_node_stat_diff)))
1983	return true;
1984	}
1985	return false;
1986	}
1987
1988	/*
1989	* Switch off vmstat processing and then fold all the remaining differentials
1990	* until the diffs stay at zero. The function is used by NOHZ and can only be
1991	* invoked when tick processing is not active.
1992	*/
1993	void quiet_vmstat(void)
1994	{
1995	if (system_state != SYSTEM_RUNNING)
1996	return;
1997
1998	if (!delayed_work_pending(this_cpu_ptr(&vmstat_work)))
1999	return;
2000
2001	if (!need_update(smp_processor_id()))
2002	return;
2003
2004	/*
2005	* Just refresh counters and do not care about the pending delayed
2006	* vmstat_update. It doesn't fire that often to matter and canceling
2007	* it would be too expensive from this path.
2008	* vmstat_shepherd will take care about that for us.
2009	*/
2010	refresh_cpu_vm_stats(do_pagesets: false);
2011	}
2012
2013	/*
2014	* Shepherd worker thread that checks the
2015	* differentials of processors that have their worker
2016	* threads for vm statistics updates disabled because of
2017	* inactivity.
2018	*/
2019	static void vmstat_shepherd(struct work_struct *w);
2020
2021	static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
2022
2023	static void vmstat_shepherd(struct work_struct *w)
2024	{
2025	int cpu;
2026
2027	cpus_read_lock();
2028	/* Check processors whose vmstat worker threads have been disabled */
2029	for_each_online_cpu(cpu) {
2030	struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
2031
2032	/*
2033	* In kernel users of vmstat counters either require the precise value and
2034	* they are using zone_page_state_snapshot interface or they can live with
2035	* an imprecision as the regular flushing can happen at arbitrary time and
2036	* cumulative error can grow (see calculate_normal_threshold).
2037	*
2038	* From that POV the regular flushing can be postponed for CPUs that have
2039	* been isolated from the kernel interference without critical
2040	* infrastructure ever noticing. Skip regular flushing from vmstat_shepherd
2041	* for all isolated CPUs to avoid interference with the isolated workload.
2042	*/
2043	if (cpu_is_isolated(cpu))
2044	continue;
2045
2046	if (!delayed_work_pending(dw) && need_update(cpu))
2047	queue_delayed_work_on(cpu, wq: mm_percpu_wq, work: dw, delay: 0);
2048
2049	cond_resched();
2050	}
2051	cpus_read_unlock();
2052
2053	schedule_delayed_work(dwork: &shepherd,
2054	delay: round_jiffies_relative(j: sysctl_stat_interval));
2055	}
2056
2057	static void __init start_shepherd_timer(void)
2058	{
2059	int cpu;
2060
2061	for_each_possible_cpu(cpu)
2062	INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
2063	vmstat_update);
2064
2065	schedule_delayed_work(dwork: &shepherd,
2066	delay: round_jiffies_relative(j: sysctl_stat_interval));
2067	}
2068
2069	static void __init init_cpu_node_state(void)
2070	{
2071	int node;
2072
2073	for_each_online_node(node) {
2074	if (!cpumask_empty(srcp: cpumask_of_node(node)))
2075	node_set_state(node, state: N_CPU);
2076	}
2077	}
2078
2079	static int vmstat_cpu_online(unsigned int cpu)
2080	{
2081	refresh_zone_stat_thresholds();
2082
2083	if (!node_state(cpu_to_node(cpu), state: N_CPU)) {
2084	node_set_state(cpu_to_node(cpu), state: N_CPU);
2085	}
2086
2087	return 0;
2088	}
2089
2090	static int vmstat_cpu_down_prep(unsigned int cpu)
2091	{
2092	cancel_delayed_work_sync(dwork: &per_cpu(vmstat_work, cpu));
2093	return 0;
2094	}
2095
2096	static int vmstat_cpu_dead(unsigned int cpu)
2097	{
2098	const struct cpumask *node_cpus;
2099	int node;
2100
2101	node = cpu_to_node(cpu);
2102
2103	refresh_zone_stat_thresholds();
2104	node_cpus = cpumask_of_node(node);
2105	if (!cpumask_empty(srcp: node_cpus))
2106	return 0;
2107
2108	node_clear_state(node, state: N_CPU);
2109
2110	return 0;
2111	}
2112
2113	#endif
2114
2115	struct workqueue_struct *mm_percpu_wq;
2116
2117	void __init init_mm_internals(void)
2118	{
2119	int ret __maybe_unused;
2120
2121	mm_percpu_wq = alloc_workqueue(fmt: "mm_percpu_wq", flags: WQ_MEM_RECLAIM, max_active: 0);
2122
2123	#ifdef CONFIG_SMP
2124	ret = cpuhp_setup_state_nocalls(state: CPUHP_MM_VMSTAT_DEAD, name: "mm/vmstat:dead",
2125	NULL, teardown: vmstat_cpu_dead);
2126	if (ret < 0)
2127	pr_err("vmstat: failed to register 'dead' hotplug state\n");
2128
2129	ret = cpuhp_setup_state_nocalls(state: CPUHP_AP_ONLINE_DYN, name: "mm/vmstat:online",
2130	startup: vmstat_cpu_online,
2131	teardown: vmstat_cpu_down_prep);
2132	if (ret < 0)
2133	pr_err("vmstat: failed to register 'online' hotplug state\n");
2134
2135	cpus_read_lock();
2136	init_cpu_node_state();
2137	cpus_read_unlock();
2138
2139	start_shepherd_timer();
2140	#endif
2141	#ifdef CONFIG_PROC_FS
2142	proc_create_seq("buddyinfo", 0444, NULL, &fragmentation_op);
2143	proc_create_seq("pagetypeinfo", 0400, NULL, &pagetypeinfo_op);
2144	proc_create_seq("vmstat", 0444, NULL, &vmstat_op);
2145	proc_create_seq("zoneinfo", 0444, NULL, &zoneinfo_op);
2146	#endif
2147	}
2148
2149	#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
2150
2151	/*
2152	* Return an index indicating how much of the available free memory is
2153	* unusable for an allocation of the requested size.
2154	*/
2155	static int unusable_free_index(unsigned int order,
2156	struct contig_page_info *info)
2157	{
2158	/* No free memory is interpreted as all free memory is unusable */
2159	if (info->free_pages == 0)
2160	return 1000;
2161
2162	/*
2163	* Index should be a value between 0 and 1. Return a value to 3
2164	* decimal places.
2165	*
2166	* 0 => no fragmentation
2167	* 1 => high fragmentation
2168	*/
2169	return div_u64(dividend: (info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, divisor: info->free_pages);
2170
2171	}
2172
2173	static void unusable_show_print(struct seq_file *m,
2174	pg_data_t *pgdat, struct zone *zone)
2175	{
2176	unsigned int order;
2177	int index;
2178	struct contig_page_info info;
2179
2180	seq_printf(m, fmt: "Node %d, zone %8s ",
2181	pgdat->node_id,
2182	zone->name);
2183	for (order = 0; order <= MAX_ORDER; ++order) {
2184	fill_contig_page_info(zone, suitable_order: order, info: &info);
2185	index = unusable_free_index(order, info: &info);
2186	seq_printf(m, fmt: "%d.%03d ", index / 1000, index % 1000);
2187	}
2188
2189	seq_putc(m, c: '\n');
2190	}
2191
2192	/*
2193	* Display unusable free space index
2194	*
2195	* The unusable free space index measures how much of the available free
2196	* memory cannot be used to satisfy an allocation of a given size and is a
2197	* value between 0 and 1. The higher the value, the more of free memory is
2198	* unusable and by implication, the worse the external fragmentation is. This
2199	* can be expressed as a percentage by multiplying by 100.
2200	*/
2201	static int unusable_show(struct seq_file *m, void *arg)
2202	{
2203	pg_data_t *pgdat = (pg_data_t *)arg;
2204
2205	/* check memoryless node */
2206	if (!node_state(node: pgdat->node_id, state: N_MEMORY))
2207	return 0;
2208
2209	walk_zones_in_node(m, pgdat, assert_populated: true, nolock: false, print: unusable_show_print);
2210
2211	return 0;
2212	}
2213
2214	static const struct seq_operations unusable_sops = {
2215	.start = frag_start,
2216	.next = frag_next,
2217	.stop = frag_stop,
2218	.show = unusable_show,
2219	};
2220
2221	DEFINE_SEQ_ATTRIBUTE(unusable);
2222
2223	static void extfrag_show_print(struct seq_file *m,
2224	pg_data_t *pgdat, struct zone *zone)
2225	{
2226	unsigned int order;
2227	int index;
2228
2229	/* Alloc on stack as interrupts are disabled for zone walk */
2230	struct contig_page_info info;
2231
2232	seq_printf(m, fmt: "Node %d, zone %8s ",
2233	pgdat->node_id,
2234	zone->name);
2235	for (order = 0; order <= MAX_ORDER; ++order) {
2236	fill_contig_page_info(zone, suitable_order: order, info: &info);
2237	index = __fragmentation_index(order, info: &info);
2238	seq_printf(m, fmt: "%2d.%03d ", index / 1000, index % 1000);
2239	}
2240
2241	seq_putc(m, c: '\n');
2242	}
2243
2244	/*
2245	* Display fragmentation index for orders that allocations would fail for
2246	*/
2247	static int extfrag_show(struct seq_file *m, void *arg)
2248	{
2249	pg_data_t *pgdat = (pg_data_t *)arg;
2250
2251	walk_zones_in_node(m, pgdat, assert_populated: true, nolock: false, print: extfrag_show_print);
2252
2253	return 0;
2254	}
2255
2256	static const struct seq_operations extfrag_sops = {
2257	.start = frag_start,
2258	.next = frag_next,
2259	.stop = frag_stop,
2260	.show = extfrag_show,
2261	};
2262
2263	DEFINE_SEQ_ATTRIBUTE(extfrag);
2264
2265	static int __init extfrag_debug_init(void)
2266	{
2267	struct dentry *extfrag_debug_root;
2268
2269	extfrag_debug_root = debugfs_create_dir(name: "extfrag", NULL);
2270
2271	debugfs_create_file(name: "unusable_index", mode: 0444, parent: extfrag_debug_root, NULL,
2272	fops: &unusable_fops);
2273
2274	debugfs_create_file(name: "extfrag_index", mode: 0444, parent: extfrag_debug_root, NULL,
2275	fops: &extfrag_fops);
2276
2277	return 0;
2278	}
2279
2280	module_init(extfrag_debug_init);
2281	#endif
2282