summaryrefslogtreecommitdiff
path: root/mm/vmstat.c (plain)
blob: b69fd4f0444a07a318b1b9cf5d2e4bbc87ca8e90
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_zone_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
90atomic_long_t vm_node_stat[NR_VM_NODE_STAT_ITEMS] __cacheline_aligned_in_smp;
91EXPORT_SYMBOL(vm_zone_stat);
92EXPORT_SYMBOL(vm_node_stat);
93
94#ifdef CONFIG_SMP
95
96int calculate_pressure_threshold(struct zone *zone)
97{
98 int threshold;
99 int watermark_distance;
100
101 /*
102 * As vmstats are not up to date, there is drift between the estimated
103 * and real values. For high thresholds and a high number of CPUs, it
104 * is possible for the min watermark to be breached while the estimated
105 * value looks fine. The pressure threshold is a reduced value such
106 * that even the maximum amount of drift will not accidentally breach
107 * the min watermark
108 */
109 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
110 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
111
112 /*
113 * Maximum threshold is 125
114 */
115 threshold = min(125, threshold);
116
117 return threshold;
118}
119
120int calculate_normal_threshold(struct zone *zone)
121{
122 int threshold;
123 int mem; /* memory in 128 MB units */
124
125 /*
126 * The threshold scales with the number of processors and the amount
127 * of memory per zone. More memory means that we can defer updates for
128 * longer, more processors could lead to more contention.
129 * fls() is used to have a cheap way of logarithmic scaling.
130 *
131 * Some sample thresholds:
132 *
133 * Threshold Processors (fls) Zonesize fls(mem+1)
134 * ------------------------------------------------------------------
135 * 8 1 1 0.9-1 GB 4
136 * 16 2 2 0.9-1 GB 4
137 * 20 2 2 1-2 GB 5
138 * 24 2 2 2-4 GB 6
139 * 28 2 2 4-8 GB 7
140 * 32 2 2 8-16 GB 8
141 * 4 2 2 <128M 1
142 * 30 4 3 2-4 GB 5
143 * 48 4 3 8-16 GB 8
144 * 32 8 4 1-2 GB 4
145 * 32 8 4 0.9-1GB 4
146 * 10 16 5 <128M 1
147 * 40 16 5 900M 4
148 * 70 64 7 2-4 GB 5
149 * 84 64 7 4-8 GB 6
150 * 108 512 9 4-8 GB 6
151 * 125 1024 10 8-16 GB 8
152 * 125 1024 10 16-32 GB 9
153 */
154
155 mem = zone->managed_pages >> (27 - PAGE_SHIFT);
156
157 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
158
159 /*
160 * Maximum threshold is 125
161 */
162 threshold = min(125, threshold);
163
164 return threshold;
165}
166
167/*
168 * Refresh the thresholds for each zone.
169 */
170void refresh_zone_stat_thresholds(void)
171{
172 struct pglist_data *pgdat;
173 struct zone *zone;
174 int cpu;
175 int threshold;
176
177 /* Zero current pgdat thresholds */
178 for_each_online_pgdat(pgdat) {
179 for_each_online_cpu(cpu) {
180 per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold = 0;
181 }
182 }
183
184 for_each_populated_zone(zone) {
185 struct pglist_data *pgdat = zone->zone_pgdat;
186 unsigned long max_drift, tolerate_drift;
187
188 threshold = calculate_normal_threshold(zone);
189
190 for_each_online_cpu(cpu) {
191 int pgdat_threshold;
192
193 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
194 = threshold;
195
196 /* Base nodestat threshold on the largest populated zone. */
197 pgdat_threshold = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold;
198 per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold
199 = max(threshold, pgdat_threshold);
200 }
201
202 /*
203 * Only set percpu_drift_mark if there is a danger that
204 * NR_FREE_PAGES reports the low watermark is ok when in fact
205 * the min watermark could be breached by an allocation
206 */
207 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
208 max_drift = num_online_cpus() * threshold;
209 if (max_drift > tolerate_drift)
210 zone->percpu_drift_mark = high_wmark_pages(zone) +
211 max_drift;
212 }
213}
214
215void set_pgdat_percpu_threshold(pg_data_t *pgdat,
216 int (*calculate_pressure)(struct zone *))
217{
218 struct zone *zone;
219 int cpu;
220 int threshold;
221 int i;
222
223 for (i = 0; i < pgdat->nr_zones; i++) {
224 zone = &pgdat->node_zones[i];
225 if (!zone->percpu_drift_mark)
226 continue;
227
228 threshold = (*calculate_pressure)(zone);
229 for_each_online_cpu(cpu)
230 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
231 = threshold;
232 }
233}
234
235/*
236 * For use when we know that interrupts are disabled,
237 * or when we know that preemption is disabled and that
238 * particular counter cannot be updated from interrupt context.
239 */
240void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
241 long delta)
242{
243 struct per_cpu_pageset __percpu *pcp = zone->pageset;
244 s8 __percpu *p = pcp->vm_stat_diff + item;
245 long x;
246 long t;
247
248 x = delta + __this_cpu_read(*p);
249
250 t = __this_cpu_read(pcp->stat_threshold);
251
252 if (unlikely(x > t || x < -t)) {
253 zone_page_state_add(x, zone, item);
254 x = 0;
255 }
256 __this_cpu_write(*p, x);
257}
258EXPORT_SYMBOL(__mod_zone_page_state);
259
260void __mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
261 long delta)
262{
263 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
264 s8 __percpu *p = pcp->vm_node_stat_diff + item;
265 long x;
266 long t;
267
268 x = delta + __this_cpu_read(*p);
269
270 t = __this_cpu_read(pcp->stat_threshold);
271
272 if (unlikely(x > t || x < -t)) {
273 node_page_state_add(x, pgdat, item);
274 x = 0;
275 }
276 __this_cpu_write(*p, x);
277}
278EXPORT_SYMBOL(__mod_node_page_state);
279
280/*
281 * Optimized increment and decrement functions.
282 *
283 * These are only for a single page and therefore can take a struct page *
284 * argument instead of struct zone *. This allows the inclusion of the code
285 * generated for page_zone(page) into the optimized functions.
286 *
287 * No overflow check is necessary and therefore the differential can be
288 * incremented or decremented in place which may allow the compilers to
289 * generate better code.
290 * The increment or decrement is known and therefore one boundary check can
291 * be omitted.
292 *
293 * NOTE: These functions are very performance sensitive. Change only
294 * with care.
295 *
296 * Some processors have inc/dec instructions that are atomic vs an interrupt.
297 * However, the code must first determine the differential location in a zone
298 * based on the processor number and then inc/dec the counter. There is no
299 * guarantee without disabling preemption that the processor will not change
300 * in between and therefore the atomicity vs. interrupt cannot be exploited
301 * in a useful way here.
302 */
303void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
304{
305 struct per_cpu_pageset __percpu *pcp = zone->pageset;
306 s8 __percpu *p = pcp->vm_stat_diff + item;
307 s8 v, t;
308
309 v = __this_cpu_inc_return(*p);
310 t = __this_cpu_read(pcp->stat_threshold);
311 if (unlikely(v > t)) {
312 s8 overstep = t >> 1;
313
314 zone_page_state_add(v + overstep, zone, item);
315 __this_cpu_write(*p, -overstep);
316 }
317}
318
319void __inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
320{
321 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
322 s8 __percpu *p = pcp->vm_node_stat_diff + item;
323 s8 v, t;
324
325 v = __this_cpu_inc_return(*p);
326 t = __this_cpu_read(pcp->stat_threshold);
327 if (unlikely(v > t)) {
328 s8 overstep = t >> 1;
329
330 node_page_state_add(v + overstep, pgdat, item);
331 __this_cpu_write(*p, -overstep);
332 }
333}
334
335void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
336{
337 __inc_zone_state(page_zone(page), item);
338}
339EXPORT_SYMBOL(__inc_zone_page_state);
340
341void __inc_node_page_state(struct page *page, enum node_stat_item item)
342{
343 __inc_node_state(page_pgdat(page), item);
344}
345EXPORT_SYMBOL(__inc_node_page_state);
346
347void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
348{
349 struct per_cpu_pageset __percpu *pcp = zone->pageset;
350 s8 __percpu *p = pcp->vm_stat_diff + item;
351 s8 v, t;
352
353 v = __this_cpu_dec_return(*p);
354 t = __this_cpu_read(pcp->stat_threshold);
355 if (unlikely(v < - t)) {
356 s8 overstep = t >> 1;
357
358 zone_page_state_add(v - overstep, zone, item);
359 __this_cpu_write(*p, overstep);
360 }
361}
362
363void __dec_node_state(struct pglist_data *pgdat, enum node_stat_item item)
364{
365 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
366 s8 __percpu *p = pcp->vm_node_stat_diff + item;
367 s8 v, t;
368
369 v = __this_cpu_dec_return(*p);
370 t = __this_cpu_read(pcp->stat_threshold);
371 if (unlikely(v < - t)) {
372 s8 overstep = t >> 1;
373
374 node_page_state_add(v - overstep, pgdat, item);
375 __this_cpu_write(*p, overstep);
376 }
377}
378
379void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
380{
381 __dec_zone_state(page_zone(page), item);
382}
383EXPORT_SYMBOL(__dec_zone_page_state);
384
385void __dec_node_page_state(struct page *page, enum node_stat_item item)
386{
387 __dec_node_state(page_pgdat(page), item);
388}
389EXPORT_SYMBOL(__dec_node_page_state);
390
391#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
392/*
393 * If we have cmpxchg_local support then we do not need to incur the overhead
394 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
395 *
396 * mod_state() modifies the zone counter state through atomic per cpu
397 * operations.
398 *
399 * Overstep mode specifies how overstep should handled:
400 * 0 No overstepping
401 * 1 Overstepping half of threshold
402 * -1 Overstepping minus half of threshold
403*/
404static inline void mod_zone_state(struct zone *zone,
405 enum zone_stat_item item, long delta, int overstep_mode)
406{
407 struct per_cpu_pageset __percpu *pcp = zone->pageset;
408 s8 __percpu *p = pcp->vm_stat_diff + item;
409 long o, n, t, z;
410
411 do {
412 z = 0; /* overflow to zone counters */
413
414 /*
415 * The fetching of the stat_threshold is racy. We may apply
416 * a counter threshold to the wrong the cpu if we get
417 * rescheduled while executing here. However, the next
418 * counter update will apply the threshold again and
419 * therefore bring the counter under the threshold again.
420 *
421 * Most of the time the thresholds are the same anyways
422 * for all cpus in a zone.
423 */
424 t = this_cpu_read(pcp->stat_threshold);
425
426 o = this_cpu_read(*p);
427 n = delta + o;
428
429 if (n > t || n < -t) {
430 int os = overstep_mode * (t >> 1) ;
431
432 /* Overflow must be added to zone counters */
433 z = n + os;
434 n = -os;
435 }
436 } while (this_cpu_cmpxchg(*p, o, n) != o);
437
438 if (z)
439 zone_page_state_add(z, zone, item);
440}
441
442void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
443 long delta)
444{
445 mod_zone_state(zone, item, delta, 0);
446}
447EXPORT_SYMBOL(mod_zone_page_state);
448
449void inc_zone_page_state(struct page *page, enum zone_stat_item item)
450{
451 mod_zone_state(page_zone(page), item, 1, 1);
452}
453EXPORT_SYMBOL(inc_zone_page_state);
454
455void dec_zone_page_state(struct page *page, enum zone_stat_item item)
456{
457 mod_zone_state(page_zone(page), item, -1, -1);
458}
459EXPORT_SYMBOL(dec_zone_page_state);
460
461static inline void mod_node_state(struct pglist_data *pgdat,
462 enum node_stat_item item, int delta, int overstep_mode)
463{
464 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
465 s8 __percpu *p = pcp->vm_node_stat_diff + item;
466 long o, n, t, z;
467
468 do {
469 z = 0; /* overflow to node counters */
470
471 /*
472 * The fetching of the stat_threshold is racy. We may apply
473 * a counter threshold to the wrong the cpu if we get
474 * rescheduled while executing here. However, the next
475 * counter update will apply the threshold again and
476 * therefore bring the counter under the threshold again.
477 *
478 * Most of the time the thresholds are the same anyways
479 * for all cpus in a node.
480 */
481 t = this_cpu_read(pcp->stat_threshold);
482
483 o = this_cpu_read(*p);
484 n = delta + o;
485
486 if (n > t || n < -t) {
487 int os = overstep_mode * (t >> 1) ;
488
489 /* Overflow must be added to node counters */
490 z = n + os;
491 n = -os;
492 }
493 } while (this_cpu_cmpxchg(*p, o, n) != o);
494
495 if (z)
496 node_page_state_add(z, pgdat, item);
497}
498
499void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
500 long delta)
501{
502 mod_node_state(pgdat, item, delta, 0);
503}
504EXPORT_SYMBOL(mod_node_page_state);
505
506void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
507{
508 mod_node_state(pgdat, item, 1, 1);
509}
510
511void inc_node_page_state(struct page *page, enum node_stat_item item)
512{
513 mod_node_state(page_pgdat(page), item, 1, 1);
514}
515EXPORT_SYMBOL(inc_node_page_state);
516
517void dec_node_page_state(struct page *page, enum node_stat_item item)
518{
519 mod_node_state(page_pgdat(page), item, -1, -1);
520}
521EXPORT_SYMBOL(dec_node_page_state);
522#else
523/*
524 * Use interrupt disable to serialize counter updates
525 */
526void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
527 long delta)
528{
529 unsigned long flags;
530
531 local_irq_save(flags);
532 __mod_zone_page_state(zone, item, delta);
533 local_irq_restore(flags);
534}
535EXPORT_SYMBOL(mod_zone_page_state);
536
537void inc_zone_page_state(struct page *page, enum zone_stat_item item)
538{
539 unsigned long flags;
540 struct zone *zone;
541
542 zone = page_zone(page);
543 local_irq_save(flags);
544 __inc_zone_state(zone, item);
545 local_irq_restore(flags);
546}
547EXPORT_SYMBOL(inc_zone_page_state);
548
549void dec_zone_page_state(struct page *page, enum zone_stat_item item)
550{
551 unsigned long flags;
552
553 local_irq_save(flags);
554 __dec_zone_page_state(page, item);
555 local_irq_restore(flags);
556}
557EXPORT_SYMBOL(dec_zone_page_state);
558
559void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
560{
561 unsigned long flags;
562
563 local_irq_save(flags);
564 __inc_node_state(pgdat, item);
565 local_irq_restore(flags);
566}
567EXPORT_SYMBOL(inc_node_state);
568
569void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
570 long delta)
571{
572 unsigned long flags;
573
574 local_irq_save(flags);
575 __mod_node_page_state(pgdat, item, delta);
576 local_irq_restore(flags);
577}
578EXPORT_SYMBOL(mod_node_page_state);
579
580void inc_node_page_state(struct page *page, enum node_stat_item item)
581{
582 unsigned long flags;
583 struct pglist_data *pgdat;
584
585 pgdat = page_pgdat(page);
586 local_irq_save(flags);
587 __inc_node_state(pgdat, item);
588 local_irq_restore(flags);
589}
590EXPORT_SYMBOL(inc_node_page_state);
591
592void dec_node_page_state(struct page *page, enum node_stat_item item)
593{
594 unsigned long flags;
595
596 local_irq_save(flags);
597 __dec_node_page_state(page, item);
598 local_irq_restore(flags);
599}
600EXPORT_SYMBOL(dec_node_page_state);
601#endif
602
603/*
604 * Fold a differential into the global counters.
605 * Returns the number of counters updated.
606 */
607static int fold_diff(int *zone_diff, int *node_diff)
608{
609 int i;
610 int changes = 0;
611
612 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
613 if (zone_diff[i]) {
614 atomic_long_add(zone_diff[i], &vm_zone_stat[i]);
615 changes++;
616 }
617
618 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
619 if (node_diff[i]) {
620 atomic_long_add(node_diff[i], &vm_node_stat[i]);
621 changes++;
622 }
623 return changes;
624}
625
626/*
627 * Update the zone counters for the current cpu.
628 *
629 * Note that refresh_cpu_vm_stats strives to only access
630 * node local memory. The per cpu pagesets on remote zones are placed
631 * in the memory local to the processor using that pageset. So the
632 * loop over all zones will access a series of cachelines local to
633 * the processor.
634 *
635 * The call to zone_page_state_add updates the cachelines with the
636 * statistics in the remote zone struct as well as the global cachelines
637 * with the global counters. These could cause remote node cache line
638 * bouncing and will have to be only done when necessary.
639 *
640 * The function returns the number of global counters updated.
641 */
642static int refresh_cpu_vm_stats(bool do_pagesets)
643{
644 struct pglist_data *pgdat;
645 struct zone *zone;
646 int i;
647 int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
648 int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
649 int changes = 0;
650
651 for_each_populated_zone(zone) {
652 struct per_cpu_pageset __percpu *p = zone->pageset;
653
654 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
655 int v;
656
657 v = this_cpu_xchg(p->vm_stat_diff[i], 0);
658 if (v) {
659
660 atomic_long_add(v, &zone->vm_stat[i]);
661 global_zone_diff[i] += v;
662#ifdef CONFIG_NUMA
663 /* 3 seconds idle till flush */
664 __this_cpu_write(p->expire, 3);
665#endif
666 }
667 }
668#ifdef CONFIG_NUMA
669 if (do_pagesets) {
670 cond_resched();
671 /*
672 * Deal with draining the remote pageset of this
673 * processor
674 *
675 * Check if there are pages remaining in this pageset
676 * if not then there is nothing to expire.
677 */
678 if (!__this_cpu_read(p->expire) ||
679 !__this_cpu_read(p->pcp.count))
680 continue;
681
682 /*
683 * We never drain zones local to this processor.
684 */
685 if (zone_to_nid(zone) == numa_node_id()) {
686 __this_cpu_write(p->expire, 0);
687 continue;
688 }
689
690 if (__this_cpu_dec_return(p->expire))
691 continue;
692
693 if (__this_cpu_read(p->pcp.count)) {
694 drain_zone_pages(zone, this_cpu_ptr(&p->pcp));
695 changes++;
696 }
697 }
698#endif
699 }
700
701 for_each_online_pgdat(pgdat) {
702 struct per_cpu_nodestat __percpu *p = pgdat->per_cpu_nodestats;
703
704 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
705 int v;
706
707 v = this_cpu_xchg(p->vm_node_stat_diff[i], 0);
708 if (v) {
709 atomic_long_add(v, &pgdat->vm_stat[i]);
710 global_node_diff[i] += v;
711 }
712 }
713 }
714
715 changes += fold_diff(global_zone_diff, global_node_diff);
716 return changes;
717}
718
719/*
720 * Fold the data for an offline cpu into the global array.
721 * There cannot be any access by the offline cpu and therefore
722 * synchronization is simplified.
723 */
724void cpu_vm_stats_fold(int cpu)
725{
726 struct pglist_data *pgdat;
727 struct zone *zone;
728 int i;
729 int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
730 int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
731
732 for_each_populated_zone(zone) {
733 struct per_cpu_pageset *p;
734
735 p = per_cpu_ptr(zone->pageset, cpu);
736
737 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
738 if (p->vm_stat_diff[i]) {
739 int v;
740
741 v = p->vm_stat_diff[i];
742 p->vm_stat_diff[i] = 0;
743 atomic_long_add(v, &zone->vm_stat[i]);
744 global_zone_diff[i] += v;
745 }
746 }
747
748 for_each_online_pgdat(pgdat) {
749 struct per_cpu_nodestat *p;
750
751 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
752
753 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
754 if (p->vm_node_stat_diff[i]) {
755 int v;
756
757 v = p->vm_node_stat_diff[i];
758 p->vm_node_stat_diff[i] = 0;
759 atomic_long_add(v, &pgdat->vm_stat[i]);
760 global_node_diff[i] += v;
761 }
762 }
763
764 fold_diff(global_zone_diff, global_node_diff);
765}
766
767/*
768 * this is only called if !populated_zone(zone), which implies no other users of
769 * pset->vm_stat_diff[] exsist.
770 */
771void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
772{
773 int i;
774
775 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
776 if (pset->vm_stat_diff[i]) {
777 int v = pset->vm_stat_diff[i];
778 pset->vm_stat_diff[i] = 0;
779 atomic_long_add(v, &zone->vm_stat[i]);
780 atomic_long_add(v, &vm_zone_stat[i]);
781 }
782}
783#endif
784
785#ifdef CONFIG_NUMA
786/*
787 * Determine the per node value of a stat item. This function
788 * is called frequently in a NUMA machine, so try to be as
789 * frugal as possible.
790 */
791unsigned long sum_zone_node_page_state(int node,
792 enum zone_stat_item item)
793{
794 struct zone *zones = NODE_DATA(node)->node_zones;
795 int i;
796 unsigned long count = 0;
797
798 for (i = 0; i < MAX_NR_ZONES; i++)
799 count += zone_page_state(zones + i, item);
800
801 return count;
802}
803
804/*
805 * Determine the per node value of a stat item.
806 */
807unsigned long node_page_state(struct pglist_data *pgdat,
808 enum node_stat_item item)
809{
810 long x = atomic_long_read(&pgdat->vm_stat[item]);
811#ifdef CONFIG_SMP
812 if (x < 0)
813 x = 0;
814#endif
815 return x;
816}
817#endif
818
819#ifdef CONFIG_COMPACTION
820
821struct contig_page_info {
822 unsigned long free_pages;
823 unsigned long free_blocks_total;
824 unsigned long free_blocks_suitable;
825};
826
827/*
828 * Calculate the number of free pages in a zone, how many contiguous
829 * pages are free and how many are large enough to satisfy an allocation of
830 * the target size. Note that this function makes no attempt to estimate
831 * how many suitable free blocks there *might* be if MOVABLE pages were
832 * migrated. Calculating that is possible, but expensive and can be
833 * figured out from userspace
834 */
835static void fill_contig_page_info(struct zone *zone,
836 unsigned int suitable_order,
837 struct contig_page_info *info)
838{
839 unsigned int order;
840
841 info->free_pages = 0;
842 info->free_blocks_total = 0;
843 info->free_blocks_suitable = 0;
844
845 for (order = 0; order < MAX_ORDER; order++) {
846 unsigned long blocks;
847
848 /* Count number of free blocks */
849 blocks = zone->free_area[order].nr_free;
850 info->free_blocks_total += blocks;
851
852 /* Count free base pages */
853 info->free_pages += blocks << order;
854
855 /* Count the suitable free blocks */
856 if (order >= suitable_order)
857 info->free_blocks_suitable += blocks <<
858 (order - suitable_order);
859 }
860}
861
862/*
863 * A fragmentation index only makes sense if an allocation of a requested
864 * size would fail. If that is true, the fragmentation index indicates
865 * whether external fragmentation or a lack of memory was the problem.
866 * The value can be used to determine if page reclaim or compaction
867 * should be used
868 */
869static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
870{
871 unsigned long requested = 1UL << order;
872
873 if (!info->free_blocks_total)
874 return 0;
875
876 /* Fragmentation index only makes sense when a request would fail */
877 if (info->free_blocks_suitable)
878 return -1000;
879
880 /*
881 * Index is between 0 and 1 so return within 3 decimal places
882 *
883 * 0 => allocation would fail due to lack of memory
884 * 1 => allocation would fail due to fragmentation
885 */
886 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
887}
888
889/* Same as __fragmentation index but allocs contig_page_info on stack */
890int fragmentation_index(struct zone *zone, unsigned int order)
891{
892 struct contig_page_info info;
893
894 fill_contig_page_info(zone, order, &info);
895 return __fragmentation_index(order, &info);
896}
897#endif
898
899#if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
900#ifdef CONFIG_ZONE_DMA
901#define TEXT_FOR_DMA(xx) xx "_dma",
902#else
903#define TEXT_FOR_DMA(xx)
904#endif
905
906#ifdef CONFIG_ZONE_DMA32
907#define TEXT_FOR_DMA32(xx) xx "_dma32",
908#else
909#define TEXT_FOR_DMA32(xx)
910#endif
911
912#ifdef CONFIG_HIGHMEM
913#define TEXT_FOR_HIGHMEM(xx) xx "_high",
914#else
915#define TEXT_FOR_HIGHMEM(xx)
916#endif
917
918#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
919 TEXT_FOR_HIGHMEM(xx) xx "_movable",
920
921const char * const vmstat_text[] = {
922 /* enum zone_stat_item countes */
923 "nr_free_pages",
924 "nr_zone_inactive_anon",
925 "nr_zone_active_anon",
926 "nr_zone_inactive_file",
927 "nr_zone_active_file",
928 "nr_zone_unevictable",
929 "nr_zone_write_pending",
930 "nr_mlock",
931 "nr_slab_reclaimable",
932 "nr_slab_unreclaimable",
933 "nr_page_table_pages",
934 "nr_kernel_stack",
935 "nr_overhead",
936 "nr_bounce",
937#if IS_ENABLED(CONFIG_ZSMALLOC)
938 "nr_zspages",
939#endif
940#ifdef CONFIG_NUMA
941 "numa_hit",
942 "numa_miss",
943 "numa_foreign",
944 "numa_interleave",
945 "numa_local",
946 "numa_other",
947#endif
948 "nr_free_cma",
949
950 /* Node-based counters */
951 "nr_inactive_anon",
952 "nr_active_anon",
953 "nr_inactive_file",
954 "nr_active_file",
955 "nr_unevictable",
956 "nr_isolated_anon",
957 "nr_isolated_file",
958 "nr_pages_scanned",
959 "workingset_refault",
960 "workingset_activate",
961 "workingset_restore",
962 "workingset_nodereclaim",
963 "nr_anon_pages",
964 "nr_mapped",
965 "nr_file_pages",
966 "nr_dirty",
967 "nr_writeback",
968 "nr_writeback_temp",
969 "nr_shmem",
970 "nr_shmem_hugepages",
971 "nr_shmem_pmdmapped",
972 "nr_anon_transparent_hugepages",
973 "nr_unstable",
974 "nr_vmscan_write",
975 "nr_vmscan_immediate_reclaim",
976 "nr_dirtied",
977 "nr_written",
978
979 /* enum writeback_stat_item counters */
980 "nr_dirty_threshold",
981 "nr_dirty_background_threshold",
982
983#ifdef CONFIG_VM_EVENT_COUNTERS
984 /* enum vm_event_item counters */
985 "pgpgin",
986 "pgpgout",
987 "pswpin",
988 "pswpout",
989
990 TEXTS_FOR_ZONES("pgalloc")
991 TEXTS_FOR_ZONES("allocstall")
992 TEXTS_FOR_ZONES("pgskip")
993
994 "pgfree",
995 "pgactivate",
996 "pgdeactivate",
997
998 "pgfault",
999 "pgmajfault",
1000 "pglazyfreed",
1001
1002 "pgrefill",
1003 "pgsteal_kswapd",
1004 "pgsteal_direct",
1005 "pgscan_kswapd",
1006 "pgscan_direct",
1007 "pgscan_direct_throttle",
1008
1009#ifdef CONFIG_NUMA
1010 "zone_reclaim_failed",
1011#endif
1012 "pginodesteal",
1013 "slabs_scanned",
1014 "kswapd_inodesteal",
1015 "kswapd_low_wmark_hit_quickly",
1016 "kswapd_high_wmark_hit_quickly",
1017 "pageoutrun",
1018
1019 "pgrotated",
1020
1021 "drop_pagecache",
1022 "drop_slab",
1023
1024#ifdef CONFIG_NUMA_BALANCING
1025 "numa_pte_updates",
1026 "numa_huge_pte_updates",
1027 "numa_hint_faults",
1028 "numa_hint_faults_local",
1029 "numa_pages_migrated",
1030#endif
1031#ifdef CONFIG_MIGRATION
1032 "pgmigrate_success",
1033 "pgmigrate_fail",
1034#endif
1035#ifdef CONFIG_COMPACTION
1036 "compact_migrate_scanned",
1037 "compact_free_scanned",
1038 "compact_isolated",
1039 "compact_stall",
1040 "compact_fail",
1041 "compact_success",
1042 "compact_daemon_wake",
1043#endif
1044
1045#ifdef CONFIG_HUGETLB_PAGE
1046 "htlb_buddy_alloc_success",
1047 "htlb_buddy_alloc_fail",
1048#endif
1049 "unevictable_pgs_culled",
1050 "unevictable_pgs_scanned",
1051 "unevictable_pgs_rescued",
1052 "unevictable_pgs_mlocked",
1053 "unevictable_pgs_munlocked",
1054 "unevictable_pgs_cleared",
1055 "unevictable_pgs_stranded",
1056
1057#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1058 "thp_fault_alloc",
1059 "thp_fault_fallback",
1060 "thp_collapse_alloc",
1061 "thp_collapse_alloc_failed",
1062 "thp_file_alloc",
1063 "thp_file_mapped",
1064 "thp_split_page",
1065 "thp_split_page_failed",
1066 "thp_deferred_split_page",
1067 "thp_split_pmd",
1068 "thp_zero_page_alloc",
1069 "thp_zero_page_alloc_failed",
1070#endif
1071#ifdef CONFIG_MEMORY_BALLOON
1072 "balloon_inflate",
1073 "balloon_deflate",
1074#ifdef CONFIG_BALLOON_COMPACTION
1075 "balloon_migrate",
1076#endif
1077#endif /* CONFIG_MEMORY_BALLOON */
1078#ifdef CONFIG_DEBUG_TLBFLUSH
1079 "nr_tlb_remote_flush",
1080 "nr_tlb_remote_flush_received",
1081 "nr_tlb_local_flush_all",
1082 "nr_tlb_local_flush_one",
1083#endif /* CONFIG_DEBUG_TLBFLUSH */
1084
1085#ifdef CONFIG_DEBUG_VM_VMACACHE
1086 "vmacache_find_calls",
1087 "vmacache_find_hits",
1088#endif
1089#endif /* CONFIG_VM_EVENTS_COUNTERS */
1090};
1091#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
1092
1093
1094#if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
1095 defined(CONFIG_PROC_FS)
1096static void *frag_start(struct seq_file *m, loff_t *pos)
1097{
1098 pg_data_t *pgdat;
1099 loff_t node = *pos;
1100
1101 for (pgdat = first_online_pgdat();
1102 pgdat && node;
1103 pgdat = next_online_pgdat(pgdat))
1104 --node;
1105
1106 return pgdat;
1107}
1108
1109static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
1110{
1111 pg_data_t *pgdat = (pg_data_t *)arg;
1112
1113 (*pos)++;
1114 return next_online_pgdat(pgdat);
1115}
1116
1117static void frag_stop(struct seq_file *m, void *arg)
1118{
1119}
1120
1121/* Walk all the zones in a node and print using a callback */
1122static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
1123 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
1124{
1125 struct zone *zone;
1126 struct zone *node_zones = pgdat->node_zones;
1127 unsigned long flags;
1128
1129 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
1130 if (!populated_zone(zone))
1131 continue;
1132
1133 spin_lock_irqsave(&zone->lock, flags);
1134 print(m, pgdat, zone);
1135 spin_unlock_irqrestore(&zone->lock, flags);
1136 }
1137}
1138#endif
1139
1140#ifdef CONFIG_PROC_FS
1141static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
1142 struct zone *zone)
1143{
1144 int order;
1145
1146 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1147 for (order = 0; order < MAX_ORDER; ++order)
1148 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
1149 seq_putc(m, '\n');
1150}
1151
1152/*
1153 * This walks the free areas for each zone.
1154 */
1155static int frag_show(struct seq_file *m, void *arg)
1156{
1157 pg_data_t *pgdat = (pg_data_t *)arg;
1158 walk_zones_in_node(m, pgdat, frag_show_print);
1159 return 0;
1160}
1161
1162static void pagetypeinfo_showfree_print(struct seq_file *m,
1163 pg_data_t *pgdat, struct zone *zone)
1164{
1165 int order, mtype;
1166
1167 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
1168 seq_printf(m, "Node %4d, zone %8s, type %12s ",
1169 pgdat->node_id,
1170 zone->name,
1171 migratetype_names[mtype]);
1172 for (order = 0; order < MAX_ORDER; ++order) {
1173 unsigned long freecount = 0;
1174 struct free_area *area;
1175 struct list_head *curr;
1176
1177 area = &(zone->free_area[order]);
1178
1179 list_for_each(curr, &area->free_list[mtype])
1180 freecount++;
1181 seq_printf(m, "%6lu ", freecount);
1182 }
1183 seq_putc(m, '\n');
1184 }
1185}
1186
1187/* Print out the free pages at each order for each migatetype */
1188static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
1189{
1190 int order;
1191 pg_data_t *pgdat = (pg_data_t *)arg;
1192
1193 /* Print header */
1194 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
1195 for (order = 0; order < MAX_ORDER; ++order)
1196 seq_printf(m, "%6d ", order);
1197 seq_putc(m, '\n');
1198
1199 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
1200
1201 return 0;
1202}
1203
1204static void pagetypeinfo_showblockcount_print(struct seq_file *m,
1205 pg_data_t *pgdat, struct zone *zone)
1206{
1207 int mtype;
1208 unsigned long pfn;
1209 unsigned long start_pfn = zone->zone_start_pfn;
1210 unsigned long end_pfn = zone_end_pfn(zone);
1211 unsigned long count[MIGRATE_TYPES] = { 0, };
1212
1213 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
1214 struct page *page;
1215
1216 if (!pfn_valid(pfn))
1217 continue;
1218
1219 page = pfn_to_page(pfn);
1220
1221 /* Watch for unexpected holes punched in the memmap */
1222 if (!memmap_valid_within(pfn, page, zone))
1223 continue;
1224
1225 if (page_zone(page) != zone)
1226 continue;
1227
1228 mtype = get_pageblock_migratetype(page);
1229
1230 if (mtype < MIGRATE_TYPES)
1231 count[mtype]++;
1232 }
1233
1234 /* Print counts */
1235 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1236 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1237 seq_printf(m, "%12lu ", count[mtype]);
1238 seq_putc(m, '\n');
1239}
1240
1241/* Print out the free pages at each order for each migratetype */
1242static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1243{
1244 int mtype;
1245 pg_data_t *pgdat = (pg_data_t *)arg;
1246
1247 seq_printf(m, "\n%-23s", "Number of blocks type ");
1248 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1249 seq_printf(m, "%12s ", migratetype_names[mtype]);
1250 seq_putc(m, '\n');
1251 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
1252
1253 return 0;
1254}
1255
1256/*
1257 * Print out the number of pageblocks for each migratetype that contain pages
1258 * of other types. This gives an indication of how well fallbacks are being
1259 * contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1260 * to determine what is going on
1261 */
1262static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1263{
1264#ifdef CONFIG_PAGE_OWNER
1265 int mtype;
1266
1267 if (!static_branch_unlikely(&page_owner_inited))
1268 return;
1269
1270 drain_all_pages(NULL);
1271
1272 seq_printf(m, "\n%-23s", "Number of mixed blocks ");
1273 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1274 seq_printf(m, "%12s ", migratetype_names[mtype]);
1275 seq_putc(m, '\n');
1276
1277 walk_zones_in_node(m, pgdat, pagetypeinfo_showmixedcount_print);
1278#endif /* CONFIG_PAGE_OWNER */
1279}
1280
1281/*
1282 * This prints out statistics in relation to grouping pages by mobility.
1283 * It is expensive to collect so do not constantly read the file.
1284 */
1285static int pagetypeinfo_show(struct seq_file *m, void *arg)
1286{
1287 pg_data_t *pgdat = (pg_data_t *)arg;
1288
1289 /* check memoryless node */
1290 if (!node_state(pgdat->node_id, N_MEMORY))
1291 return 0;
1292
1293 seq_printf(m, "Page block order: %d\n", pageblock_order);
1294 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
1295 seq_putc(m, '\n');
1296 pagetypeinfo_showfree(m, pgdat);
1297 pagetypeinfo_showblockcount(m, pgdat);
1298 pagetypeinfo_showmixedcount(m, pgdat);
1299
1300 return 0;
1301}
1302
1303static const struct seq_operations fragmentation_op = {
1304 .start = frag_start,
1305 .next = frag_next,
1306 .stop = frag_stop,
1307 .show = frag_show,
1308};
1309
1310static int fragmentation_open(struct inode *inode, struct file *file)
1311{
1312 return seq_open(file, &fragmentation_op);
1313}
1314
1315static const struct file_operations fragmentation_file_operations = {
1316 .open = fragmentation_open,
1317 .read = seq_read,
1318 .llseek = seq_lseek,
1319 .release = seq_release,
1320};
1321
1322static const struct seq_operations pagetypeinfo_op = {
1323 .start = frag_start,
1324 .next = frag_next,
1325 .stop = frag_stop,
1326 .show = pagetypeinfo_show,
1327};
1328
1329static int pagetypeinfo_open(struct inode *inode, struct file *file)
1330{
1331 return seq_open(file, &pagetypeinfo_op);
1332}
1333
1334static const struct file_operations pagetypeinfo_file_ops = {
1335 .open = pagetypeinfo_open,
1336 .read = seq_read,
1337 .llseek = seq_lseek,
1338 .release = seq_release,
1339};
1340
1341static bool is_zone_first_populated(pg_data_t *pgdat, struct zone *zone)
1342{
1343 int zid;
1344
1345 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1346 struct zone *compare = &pgdat->node_zones[zid];
1347
1348 if (populated_zone(compare))
1349 return zone == compare;
1350 }
1351
1352 return false;
1353}
1354
1355static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1356 struct zone *zone)
1357{
1358 int i;
1359 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1360 if (is_zone_first_populated(pgdat, zone)) {
1361 seq_printf(m, "\n per-node stats");
1362 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
1363 seq_printf(m, "\n %-12s %lu",
1364 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
1365 node_page_state(pgdat, i));
1366 }
1367 }
1368 seq_printf(m,
1369 "\n pages free %lu"
1370 "\n min %lu"
1371 "\n low %lu"
1372 "\n high %lu"
1373 "\n node_scanned %lu"
1374 "\n spanned %lu"
1375 "\n present %lu"
1376 "\n managed %lu",
1377 zone_page_state(zone, NR_FREE_PAGES),
1378 min_wmark_pages(zone),
1379 low_wmark_pages(zone),
1380 high_wmark_pages(zone),
1381 node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED),
1382 zone->spanned_pages,
1383 zone->present_pages,
1384 zone->managed_pages);
1385
1386 seq_printf(m,
1387 "\n protection: (%ld",
1388 zone->lowmem_reserve[0]);
1389 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1390 seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
1391 seq_putc(m, ')');
1392
1393 /* If unpopulated, no other information is useful */
1394 if (!populated_zone(zone)) {
1395 seq_putc(m, '\n');
1396 return;
1397 }
1398
1399 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1400 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1401 zone_page_state(zone, i));
1402
1403 seq_printf(m, "\n pagesets");
1404 for_each_online_cpu(i) {
1405 struct per_cpu_pageset *pageset;
1406
1407 pageset = per_cpu_ptr(zone->pageset, i);
1408 seq_printf(m,
1409 "\n cpu: %i"
1410 "\n count: %i"
1411 "\n high: %i"
1412 "\n batch: %i",
1413 i,
1414 pageset->pcp.count,
1415 pageset->pcp.high,
1416 pageset->pcp.batch);
1417#ifdef CONFIG_SMP
1418 seq_printf(m, "\n vm stats threshold: %d",
1419 pageset->stat_threshold);
1420#endif
1421 }
1422 seq_printf(m,
1423 "\n node_unreclaimable: %u"
1424 "\n start_pfn: %lu"
1425 "\n node_inactive_ratio: %u",
1426 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES,
1427 zone->zone_start_pfn,
1428 zone->zone_pgdat->inactive_ratio);
1429 seq_putc(m, '\n');
1430}
1431
1432/*
1433 * Output information about zones in @pgdat.
1434 */
1435static int zoneinfo_show(struct seq_file *m, void *arg)
1436{
1437 pg_data_t *pgdat = (pg_data_t *)arg;
1438 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
1439 return 0;
1440}
1441
1442static const struct seq_operations zoneinfo_op = {
1443 .start = frag_start, /* iterate over all zones. The same as in
1444 * fragmentation. */
1445 .next = frag_next,
1446 .stop = frag_stop,
1447 .show = zoneinfo_show,
1448};
1449
1450static int zoneinfo_open(struct inode *inode, struct file *file)
1451{
1452 return seq_open(file, &zoneinfo_op);
1453}
1454
1455static const struct file_operations proc_zoneinfo_file_operations = {
1456 .open = zoneinfo_open,
1457 .read = seq_read,
1458 .llseek = seq_lseek,
1459 .release = seq_release,
1460};
1461
1462enum writeback_stat_item {
1463 NR_DIRTY_THRESHOLD,
1464 NR_DIRTY_BG_THRESHOLD,
1465 NR_VM_WRITEBACK_STAT_ITEMS,
1466};
1467
1468static void *vmstat_start(struct seq_file *m, loff_t *pos)
1469{
1470 unsigned long *v;
1471 int i, stat_items_size;
1472
1473 if (*pos >= ARRAY_SIZE(vmstat_text))
1474 return NULL;
1475 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1476 NR_VM_NODE_STAT_ITEMS * sizeof(unsigned long) +
1477 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
1478
1479#ifdef CONFIG_VM_EVENT_COUNTERS
1480 stat_items_size += sizeof(struct vm_event_state);
1481#endif
1482
1483 v = kmalloc(stat_items_size, GFP_KERNEL);
1484 m->private = v;
1485 if (!v)
1486 return ERR_PTR(-ENOMEM);
1487 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1488 v[i] = global_page_state(i);
1489 v += NR_VM_ZONE_STAT_ITEMS;
1490
1491 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
1492 v[i] = global_node_page_state(i);
1493 v += NR_VM_NODE_STAT_ITEMS;
1494
1495 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1496 v + NR_DIRTY_THRESHOLD);
1497 v += NR_VM_WRITEBACK_STAT_ITEMS;
1498
1499#ifdef CONFIG_VM_EVENT_COUNTERS
1500 all_vm_events(v);
1501 v[PGPGIN] /= 2; /* sectors -> kbytes */
1502 v[PGPGOUT] /= 2;
1503#endif
1504 return (unsigned long *)m->private + *pos;
1505}
1506
1507static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1508{
1509 (*pos)++;
1510 if (*pos >= ARRAY_SIZE(vmstat_text))
1511 return NULL;
1512 return (unsigned long *)m->private + *pos;
1513}
1514
1515static int vmstat_show(struct seq_file *m, void *arg)
1516{
1517 unsigned long *l = arg;
1518 unsigned long off = l - (unsigned long *)m->private;
1519
1520 seq_puts(m, vmstat_text[off]);
1521 seq_put_decimal_ull(m, " ", *l);
1522 seq_putc(m, '\n');
1523 return 0;
1524}
1525
1526static void vmstat_stop(struct seq_file *m, void *arg)
1527{
1528 kfree(m->private);
1529 m->private = NULL;
1530}
1531
1532static const struct seq_operations vmstat_op = {
1533 .start = vmstat_start,
1534 .next = vmstat_next,
1535 .stop = vmstat_stop,
1536 .show = vmstat_show,
1537};
1538
1539static int vmstat_open(struct inode *inode, struct file *file)
1540{
1541 return seq_open(file, &vmstat_op);
1542}
1543
1544static const struct file_operations proc_vmstat_file_operations = {
1545 .open = vmstat_open,
1546 .read = seq_read,
1547 .llseek = seq_lseek,
1548 .release = seq_release,
1549};
1550#endif /* CONFIG_PROC_FS */
1551
1552#ifdef CONFIG_SMP
1553static struct workqueue_struct *vmstat_wq;
1554static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1555int sysctl_stat_interval __read_mostly = HZ;
1556
1557#ifdef CONFIG_PROC_FS
1558static void refresh_vm_stats(struct work_struct *work)
1559{
1560 refresh_cpu_vm_stats(true);
1561}
1562
1563int vmstat_refresh(struct ctl_table *table, int write,
1564 void __user *buffer, size_t *lenp, loff_t *ppos)
1565{
1566 long val;
1567 int err;
1568 int i;
1569
1570 /*
1571 * The regular update, every sysctl_stat_interval, may come later
1572 * than expected: leaving a significant amount in per_cpu buckets.
1573 * This is particularly misleading when checking a quantity of HUGE
1574 * pages, immediately after running a test. /proc/sys/vm/stat_refresh,
1575 * which can equally be echo'ed to or cat'ted from (by root),
1576 * can be used to update the stats just before reading them.
1577 *
1578 * Oh, and since global_page_state() etc. are so careful to hide
1579 * transiently negative values, report an error here if any of
1580 * the stats is negative, so we know to go looking for imbalance.
1581 */
1582 err = schedule_on_each_cpu(refresh_vm_stats);
1583 if (err)
1584 return err;
1585 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
1586 val = atomic_long_read(&vm_zone_stat[i]);
1587 if (val < 0) {
1588 switch (i) {
1589 case NR_PAGES_SCANNED:
1590 /*
1591 * This is often seen to go negative in
1592 * recent kernels, but not to go permanently
1593 * negative. Whilst it would be nicer not to
1594 * have exceptions, rooting them out would be
1595 * another task, of rather low priority.
1596 */
1597 break;
1598 default:
1599 pr_warn("%s: %s %ld\n",
1600 __func__, vmstat_text[i], val);
1601 err = -EINVAL;
1602 break;
1603 }
1604 }
1605 }
1606 if (err)
1607 return err;
1608 if (write)
1609 *ppos += *lenp;
1610 else
1611 *lenp = 0;
1612 return 0;
1613}
1614#endif /* CONFIG_PROC_FS */
1615
1616static void vmstat_update(struct work_struct *w)
1617{
1618 if (refresh_cpu_vm_stats(true)) {
1619 /*
1620 * Counters were updated so we expect more updates
1621 * to occur in the future. Keep on running the
1622 * update worker thread.
1623 */
1624 queue_delayed_work_on(smp_processor_id(), vmstat_wq,
1625 this_cpu_ptr(&vmstat_work),
1626 round_jiffies_relative(sysctl_stat_interval));
1627 }
1628}
1629
1630/*
1631 * Switch off vmstat processing and then fold all the remaining differentials
1632 * until the diffs stay at zero. The function is used by NOHZ and can only be
1633 * invoked when tick processing is not active.
1634 */
1635/*
1636 * Check if the diffs for a certain cpu indicate that
1637 * an update is needed.
1638 */
1639static bool need_update(int cpu)
1640{
1641 struct zone *zone;
1642
1643 for_each_populated_zone(zone) {
1644 struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu);
1645
1646 BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1);
1647 /*
1648 * The fast way of checking if there are any vmstat diffs.
1649 * This works because the diffs are byte sized items.
1650 */
1651 if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS))
1652 return true;
1653
1654 }
1655 return false;
1656}
1657
1658/*
1659 * Switch off vmstat processing and then fold all the remaining differentials
1660 * until the diffs stay at zero. The function is used by NOHZ and can only be
1661 * invoked when tick processing is not active.
1662 */
1663void quiet_vmstat(void)
1664{
1665 if (system_state != SYSTEM_RUNNING)
1666 return;
1667
1668 if (!delayed_work_pending(this_cpu_ptr(&vmstat_work)))
1669 return;
1670
1671 if (!need_update(smp_processor_id()))
1672 return;
1673
1674 /*
1675 * Just refresh counters and do not care about the pending delayed
1676 * vmstat_update. It doesn't fire that often to matter and canceling
1677 * it would be too expensive from this path.
1678 * vmstat_shepherd will take care about that for us.
1679 */
1680 refresh_cpu_vm_stats(false);
1681}
1682
1683/*
1684 * Shepherd worker thread that checks the
1685 * differentials of processors that have their worker
1686 * threads for vm statistics updates disabled because of
1687 * inactivity.
1688 */
1689static void vmstat_shepherd(struct work_struct *w);
1690
1691static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
1692
1693static void vmstat_shepherd(struct work_struct *w)
1694{
1695 int cpu;
1696
1697 get_online_cpus();
1698 /* Check processors whose vmstat worker threads have been disabled */
1699 for_each_online_cpu(cpu) {
1700 struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
1701
1702 if (!delayed_work_pending(dw) && need_update(cpu))
1703 queue_delayed_work_on(cpu, vmstat_wq, dw, 0);
1704 }
1705 put_online_cpus();
1706
1707 schedule_delayed_work(&shepherd,
1708 round_jiffies_relative(sysctl_stat_interval));
1709}
1710
1711static void __init start_shepherd_timer(void)
1712{
1713 int cpu;
1714
1715 for_each_possible_cpu(cpu)
1716 INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
1717 vmstat_update);
1718
1719 vmstat_wq = alloc_workqueue("vmstat", WQ_FREEZABLE|WQ_MEM_RECLAIM, 0);
1720 schedule_delayed_work(&shepherd,
1721 round_jiffies_relative(sysctl_stat_interval));
1722}
1723
1724static void __init init_cpu_node_state(void)
1725{
1726 int cpu;
1727
1728 get_online_cpus();
1729 for_each_online_cpu(cpu)
1730 node_set_state(cpu_to_node(cpu), N_CPU);
1731 put_online_cpus();
1732}
1733
1734static void vmstat_cpu_dead(int node)
1735{
1736 int cpu;
1737
1738 get_online_cpus();
1739 for_each_online_cpu(cpu)
1740 if (cpu_to_node(cpu) == node)
1741 goto end;
1742
1743 node_clear_state(node, N_CPU);
1744end:
1745 put_online_cpus();
1746}
1747
1748/*
1749 * Use the cpu notifier to insure that the thresholds are recalculated
1750 * when necessary.
1751 */
1752static int vmstat_cpuup_callback(struct notifier_block *nfb,
1753 unsigned long action,
1754 void *hcpu)
1755{
1756 long cpu = (long)hcpu;
1757
1758 switch (action) {
1759 case CPU_ONLINE:
1760 case CPU_ONLINE_FROZEN:
1761 refresh_zone_stat_thresholds();
1762 node_set_state(cpu_to_node(cpu), N_CPU);
1763 break;
1764 case CPU_DOWN_PREPARE:
1765 case CPU_DOWN_PREPARE_FROZEN:
1766 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1767 break;
1768 case CPU_DOWN_FAILED:
1769 case CPU_DOWN_FAILED_FROZEN:
1770 break;
1771 case CPU_DEAD:
1772 case CPU_DEAD_FROZEN:
1773 refresh_zone_stat_thresholds();
1774 vmstat_cpu_dead(cpu_to_node(cpu));
1775 break;
1776 default:
1777 break;
1778 }
1779 return NOTIFY_OK;
1780}
1781
1782static struct notifier_block vmstat_notifier =
1783 { &vmstat_cpuup_callback, NULL, 0 };
1784#endif
1785
1786static int __init setup_vmstat(void)
1787{
1788#ifdef CONFIG_SMP
1789 cpu_notifier_register_begin();
1790 __register_cpu_notifier(&vmstat_notifier);
1791 init_cpu_node_state();
1792
1793 start_shepherd_timer();
1794 cpu_notifier_register_done();
1795#endif
1796#ifdef CONFIG_PROC_FS
1797 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1798 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1799 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1800 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1801#endif
1802 return 0;
1803}
1804module_init(setup_vmstat)
1805
1806#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1807
1808/*
1809 * Return an index indicating how much of the available free memory is
1810 * unusable for an allocation of the requested size.
1811 */
1812static int unusable_free_index(unsigned int order,
1813 struct contig_page_info *info)
1814{
1815 /* No free memory is interpreted as all free memory is unusable */
1816 if (info->free_pages == 0)
1817 return 1000;
1818
1819 /*
1820 * Index should be a value between 0 and 1. Return a value to 3
1821 * decimal places.
1822 *
1823 * 0 => no fragmentation
1824 * 1 => high fragmentation
1825 */
1826 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1827
1828}
1829
1830static void unusable_show_print(struct seq_file *m,
1831 pg_data_t *pgdat, struct zone *zone)
1832{
1833 unsigned int order;
1834 int index;
1835 struct contig_page_info info;
1836
1837 seq_printf(m, "Node %d, zone %8s ",
1838 pgdat->node_id,
1839 zone->name);
1840 for (order = 0; order < MAX_ORDER; ++order) {
1841 fill_contig_page_info(zone, order, &info);
1842 index = unusable_free_index(order, &info);
1843 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1844 }
1845
1846 seq_putc(m, '\n');
1847}
1848
1849/*
1850 * Display unusable free space index
1851 *
1852 * The unusable free space index measures how much of the available free
1853 * memory cannot be used to satisfy an allocation of a given size and is a
1854 * value between 0 and 1. The higher the value, the more of free memory is
1855 * unusable and by implication, the worse the external fragmentation is. This
1856 * can be expressed as a percentage by multiplying by 100.
1857 */
1858static int unusable_show(struct seq_file *m, void *arg)
1859{
1860 pg_data_t *pgdat = (pg_data_t *)arg;
1861
1862 /* check memoryless node */
1863 if (!node_state(pgdat->node_id, N_MEMORY))
1864 return 0;
1865
1866 walk_zones_in_node(m, pgdat, unusable_show_print);
1867
1868 return 0;
1869}
1870
1871static const struct seq_operations unusable_op = {
1872 .start = frag_start,
1873 .next = frag_next,
1874 .stop = frag_stop,
1875 .show = unusable_show,
1876};
1877
1878static int unusable_open(struct inode *inode, struct file *file)
1879{
1880 return seq_open(file, &unusable_op);
1881}
1882
1883static const struct file_operations unusable_file_ops = {
1884 .open = unusable_open,
1885 .read = seq_read,
1886 .llseek = seq_lseek,
1887 .release = seq_release,
1888};
1889
1890static void extfrag_show_print(struct seq_file *m,
1891 pg_data_t *pgdat, struct zone *zone)
1892{
1893 unsigned int order;
1894 int index;
1895
1896 /* Alloc on stack as interrupts are disabled for zone walk */
1897 struct contig_page_info info;
1898
1899 seq_printf(m, "Node %d, zone %8s ",
1900 pgdat->node_id,
1901 zone->name);
1902 for (order = 0; order < MAX_ORDER; ++order) {
1903 fill_contig_page_info(zone, order, &info);
1904 index = __fragmentation_index(order, &info);
1905 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1906 }
1907
1908 seq_putc(m, '\n');
1909}
1910
1911/*
1912 * Display fragmentation index for orders that allocations would fail for
1913 */
1914static int extfrag_show(struct seq_file *m, void *arg)
1915{
1916 pg_data_t *pgdat = (pg_data_t *)arg;
1917
1918 walk_zones_in_node(m, pgdat, extfrag_show_print);
1919
1920 return 0;
1921}
1922
1923static const struct seq_operations extfrag_op = {
1924 .start = frag_start,
1925 .next = frag_next,
1926 .stop = frag_stop,
1927 .show = extfrag_show,
1928};
1929
1930static int extfrag_open(struct inode *inode, struct file *file)
1931{
1932 return seq_open(file, &extfrag_op);
1933}
1934
1935static const struct file_operations extfrag_file_ops = {
1936 .open = extfrag_open,
1937 .read = seq_read,
1938 .llseek = seq_lseek,
1939 .release = seq_release,
1940};
1941
1942static int __init extfrag_debug_init(void)
1943{
1944 struct dentry *extfrag_debug_root;
1945
1946 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1947 if (!extfrag_debug_root)
1948 return -ENOMEM;
1949
1950 if (!debugfs_create_file("unusable_index", 0444,
1951 extfrag_debug_root, NULL, &unusable_file_ops))
1952 goto fail;
1953
1954 if (!debugfs_create_file("extfrag_index", 0444,
1955 extfrag_debug_root, NULL, &extfrag_file_ops))
1956 goto fail;
1957
1958 return 0;
1959fail:
1960 debugfs_remove_recursive(extfrag_debug_root);
1961 return -ENOMEM;
1962}
1963
1964module_init(extfrag_debug_init);
1965#endif
1966
generated by cgit at 2024-05-18 17:06:20 (GMT)