]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - mm/vmstat.c
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc
[mirror_ubuntu-bionic-kernel.git] / mm / vmstat.c
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
34 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
35 EXPORT_PER_CPU_SYMBOL(vm_event_states);
36
37 static 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 */
57 void all_vm_events(unsigned long *ret)
58 {
59 get_online_cpus();
60 sum_vm_events(ret);
61 put_online_cpus();
62 }
63 EXPORT_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 */
71 void 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 */
89 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
90 EXPORT_SYMBOL(vm_stat);
91
92 #ifdef CONFIG_SMP
93
94 int calculate_pressure_threshold(struct zone *zone)
95 {
96 int threshold;
97 int watermark_distance;
98
99 /*
100 * As vmstats are not up to date, there is drift between the estimated
101 * and real values. For high thresholds and a high number of CPUs, it
102 * is possible for the min watermark to be breached while the estimated
103 * value looks fine. The pressure threshold is a reduced value such
104 * that even the maximum amount of drift will not accidentally breach
105 * the min watermark
106 */
107 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
108 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
109
110 /*
111 * Maximum threshold is 125
112 */
113 threshold = min(125, threshold);
114
115 return threshold;
116 }
117
118 int calculate_normal_threshold(struct zone *zone)
119 {
120 int threshold;
121 int mem; /* memory in 128 MB units */
122
123 /*
124 * The threshold scales with the number of processors and the amount
125 * of memory per zone. More memory means that we can defer updates for
126 * longer, more processors could lead to more contention.
127 * fls() is used to have a cheap way of logarithmic scaling.
128 *
129 * Some sample thresholds:
130 *
131 * Threshold Processors (fls) Zonesize fls(mem+1)
132 * ------------------------------------------------------------------
133 * 8 1 1 0.9-1 GB 4
134 * 16 2 2 0.9-1 GB 4
135 * 20 2 2 1-2 GB 5
136 * 24 2 2 2-4 GB 6
137 * 28 2 2 4-8 GB 7
138 * 32 2 2 8-16 GB 8
139 * 4 2 2 <128M 1
140 * 30 4 3 2-4 GB 5
141 * 48 4 3 8-16 GB 8
142 * 32 8 4 1-2 GB 4
143 * 32 8 4 0.9-1GB 4
144 * 10 16 5 <128M 1
145 * 40 16 5 900M 4
146 * 70 64 7 2-4 GB 5
147 * 84 64 7 4-8 GB 6
148 * 108 512 9 4-8 GB 6
149 * 125 1024 10 8-16 GB 8
150 * 125 1024 10 16-32 GB 9
151 */
152
153 mem = zone->managed_pages >> (27 - PAGE_SHIFT);
154
155 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
156
157 /*
158 * Maximum threshold is 125
159 */
160 threshold = min(125, threshold);
161
162 return threshold;
163 }
164
165 /*
166 * Refresh the thresholds for each zone.
167 */
168 void refresh_zone_stat_thresholds(void)
169 {
170 struct zone *zone;
171 int cpu;
172 int threshold;
173
174 for_each_populated_zone(zone) {
175 unsigned long max_drift, tolerate_drift;
176
177 threshold = calculate_normal_threshold(zone);
178
179 for_each_online_cpu(cpu)
180 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
181 = threshold;
182
183 /*
184 * Only set percpu_drift_mark if there is a danger that
185 * NR_FREE_PAGES reports the low watermark is ok when in fact
186 * the min watermark could be breached by an allocation
187 */
188 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
189 max_drift = num_online_cpus() * threshold;
190 if (max_drift > tolerate_drift)
191 zone->percpu_drift_mark = high_wmark_pages(zone) +
192 max_drift;
193 }
194 }
195
196 void set_pgdat_percpu_threshold(pg_data_t *pgdat,
197 int (*calculate_pressure)(struct zone *))
198 {
199 struct zone *zone;
200 int cpu;
201 int threshold;
202 int i;
203
204 for (i = 0; i < pgdat->nr_zones; i++) {
205 zone = &pgdat->node_zones[i];
206 if (!zone->percpu_drift_mark)
207 continue;
208
209 threshold = (*calculate_pressure)(zone);
210 for_each_online_cpu(cpu)
211 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
212 = threshold;
213 }
214 }
215
216 /*
217 * For use when we know that interrupts are disabled,
218 * or when we know that preemption is disabled and that
219 * particular counter cannot be updated from interrupt context.
220 */
221 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
222 long delta)
223 {
224 struct per_cpu_pageset __percpu *pcp = zone->pageset;
225 s8 __percpu *p = pcp->vm_stat_diff + item;
226 long x;
227 long t;
228
229 x = delta + __this_cpu_read(*p);
230
231 t = __this_cpu_read(pcp->stat_threshold);
232
233 if (unlikely(x > t || x < -t)) {
234 zone_page_state_add(x, zone, item);
235 x = 0;
236 }
237 __this_cpu_write(*p, x);
238 }
239 EXPORT_SYMBOL(__mod_zone_page_state);
240
241 /*
242 * Optimized increment and decrement functions.
243 *
244 * These are only for a single page and therefore can take a struct page *
245 * argument instead of struct zone *. This allows the inclusion of the code
246 * generated for page_zone(page) into the optimized functions.
247 *
248 * No overflow check is necessary and therefore the differential can be
249 * incremented or decremented in place which may allow the compilers to
250 * generate better code.
251 * The increment or decrement is known and therefore one boundary check can
252 * be omitted.
253 *
254 * NOTE: These functions are very performance sensitive. Change only
255 * with care.
256 *
257 * Some processors have inc/dec instructions that are atomic vs an interrupt.
258 * However, the code must first determine the differential location in a zone
259 * based on the processor number and then inc/dec the counter. There is no
260 * guarantee without disabling preemption that the processor will not change
261 * in between and therefore the atomicity vs. interrupt cannot be exploited
262 * in a useful way here.
263 */
264 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
265 {
266 struct per_cpu_pageset __percpu *pcp = zone->pageset;
267 s8 __percpu *p = pcp->vm_stat_diff + item;
268 s8 v, t;
269
270 v = __this_cpu_inc_return(*p);
271 t = __this_cpu_read(pcp->stat_threshold);
272 if (unlikely(v > t)) {
273 s8 overstep = t >> 1;
274
275 zone_page_state_add(v + overstep, zone, item);
276 __this_cpu_write(*p, -overstep);
277 }
278 }
279
280 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
281 {
282 __inc_zone_state(page_zone(page), item);
283 }
284 EXPORT_SYMBOL(__inc_zone_page_state);
285
286 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
287 {
288 struct per_cpu_pageset __percpu *pcp = zone->pageset;
289 s8 __percpu *p = pcp->vm_stat_diff + item;
290 s8 v, t;
291
292 v = __this_cpu_dec_return(*p);
293 t = __this_cpu_read(pcp->stat_threshold);
294 if (unlikely(v < - t)) {
295 s8 overstep = t >> 1;
296
297 zone_page_state_add(v - overstep, zone, item);
298 __this_cpu_write(*p, overstep);
299 }
300 }
301
302 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
303 {
304 __dec_zone_state(page_zone(page), item);
305 }
306 EXPORT_SYMBOL(__dec_zone_page_state);
307
308 #ifdef CONFIG_HAVE_CMPXCHG_LOCAL
309 /*
310 * If we have cmpxchg_local support then we do not need to incur the overhead
311 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
312 *
313 * mod_state() modifies the zone counter state through atomic per cpu
314 * operations.
315 *
316 * Overstep mode specifies how overstep should handled:
317 * 0 No overstepping
318 * 1 Overstepping half of threshold
319 * -1 Overstepping minus half of threshold
320 */
321 static inline void mod_state(struct zone *zone, enum zone_stat_item item,
322 long delta, int overstep_mode)
323 {
324 struct per_cpu_pageset __percpu *pcp = zone->pageset;
325 s8 __percpu *p = pcp->vm_stat_diff + item;
326 long o, n, t, z;
327
328 do {
329 z = 0; /* overflow to zone counters */
330
331 /*
332 * The fetching of the stat_threshold is racy. We may apply
333 * a counter threshold to the wrong the cpu if we get
334 * rescheduled while executing here. However, the next
335 * counter update will apply the threshold again and
336 * therefore bring the counter under the threshold again.
337 *
338 * Most of the time the thresholds are the same anyways
339 * for all cpus in a zone.
340 */
341 t = this_cpu_read(pcp->stat_threshold);
342
343 o = this_cpu_read(*p);
344 n = delta + o;
345
346 if (n > t || n < -t) {
347 int os = overstep_mode * (t >> 1) ;
348
349 /* Overflow must be added to zone counters */
350 z = n + os;
351 n = -os;
352 }
353 } while (this_cpu_cmpxchg(*p, o, n) != o);
354
355 if (z)
356 zone_page_state_add(z, zone, item);
357 }
358
359 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
360 long delta)
361 {
362 mod_state(zone, item, delta, 0);
363 }
364 EXPORT_SYMBOL(mod_zone_page_state);
365
366 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
367 {
368 mod_state(zone, item, 1, 1);
369 }
370
371 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
372 {
373 mod_state(page_zone(page), item, 1, 1);
374 }
375 EXPORT_SYMBOL(inc_zone_page_state);
376
377 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
378 {
379 mod_state(page_zone(page), item, -1, -1);
380 }
381 EXPORT_SYMBOL(dec_zone_page_state);
382 #else
383 /*
384 * Use interrupt disable to serialize counter updates
385 */
386 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
387 long delta)
388 {
389 unsigned long flags;
390
391 local_irq_save(flags);
392 __mod_zone_page_state(zone, item, delta);
393 local_irq_restore(flags);
394 }
395 EXPORT_SYMBOL(mod_zone_page_state);
396
397 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
398 {
399 unsigned long flags;
400
401 local_irq_save(flags);
402 __inc_zone_state(zone, item);
403 local_irq_restore(flags);
404 }
405
406 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
407 {
408 unsigned long flags;
409 struct zone *zone;
410
411 zone = page_zone(page);
412 local_irq_save(flags);
413 __inc_zone_state(zone, item);
414 local_irq_restore(flags);
415 }
416 EXPORT_SYMBOL(inc_zone_page_state);
417
418 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
419 {
420 unsigned long flags;
421
422 local_irq_save(flags);
423 __dec_zone_page_state(page, item);
424 local_irq_restore(flags);
425 }
426 EXPORT_SYMBOL(dec_zone_page_state);
427 #endif
428
429
430 /*
431 * Fold a differential into the global counters.
432 * Returns the number of counters updated.
433 */
434 static int fold_diff(int *diff)
435 {
436 int i;
437 int changes = 0;
438
439 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
440 if (diff[i]) {
441 atomic_long_add(diff[i], &vm_stat[i]);
442 changes++;
443 }
444 return changes;
445 }
446
447 /*
448 * Update the zone counters for the current cpu.
449 *
450 * Note that refresh_cpu_vm_stats strives to only access
451 * node local memory. The per cpu pagesets on remote zones are placed
452 * in the memory local to the processor using that pageset. So the
453 * loop over all zones will access a series of cachelines local to
454 * the processor.
455 *
456 * The call to zone_page_state_add updates the cachelines with the
457 * statistics in the remote zone struct as well as the global cachelines
458 * with the global counters. These could cause remote node cache line
459 * bouncing and will have to be only done when necessary.
460 *
461 * The function returns the number of global counters updated.
462 */
463 static int refresh_cpu_vm_stats(bool do_pagesets)
464 {
465 struct zone *zone;
466 int i;
467 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
468 int changes = 0;
469
470 for_each_populated_zone(zone) {
471 struct per_cpu_pageset __percpu *p = zone->pageset;
472
473 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
474 int v;
475
476 v = this_cpu_xchg(p->vm_stat_diff[i], 0);
477 if (v) {
478
479 atomic_long_add(v, &zone->vm_stat[i]);
480 global_diff[i] += v;
481 #ifdef CONFIG_NUMA
482 /* 3 seconds idle till flush */
483 __this_cpu_write(p->expire, 3);
484 #endif
485 }
486 }
487 #ifdef CONFIG_NUMA
488 if (do_pagesets) {
489 cond_resched();
490 /*
491 * Deal with draining the remote pageset of this
492 * processor
493 *
494 * Check if there are pages remaining in this pageset
495 * if not then there is nothing to expire.
496 */
497 if (!__this_cpu_read(p->expire) ||
498 !__this_cpu_read(p->pcp.count))
499 continue;
500
501 /*
502 * We never drain zones local to this processor.
503 */
504 if (zone_to_nid(zone) == numa_node_id()) {
505 __this_cpu_write(p->expire, 0);
506 continue;
507 }
508
509 if (__this_cpu_dec_return(p->expire))
510 continue;
511
512 if (__this_cpu_read(p->pcp.count)) {
513 drain_zone_pages(zone, this_cpu_ptr(&p->pcp));
514 changes++;
515 }
516 }
517 #endif
518 }
519 changes += fold_diff(global_diff);
520 return changes;
521 }
522
523 /*
524 * Fold the data for an offline cpu into the global array.
525 * There cannot be any access by the offline cpu and therefore
526 * synchronization is simplified.
527 */
528 void cpu_vm_stats_fold(int cpu)
529 {
530 struct zone *zone;
531 int i;
532 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
533
534 for_each_populated_zone(zone) {
535 struct per_cpu_pageset *p;
536
537 p = per_cpu_ptr(zone->pageset, cpu);
538
539 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
540 if (p->vm_stat_diff[i]) {
541 int v;
542
543 v = p->vm_stat_diff[i];
544 p->vm_stat_diff[i] = 0;
545 atomic_long_add(v, &zone->vm_stat[i]);
546 global_diff[i] += v;
547 }
548 }
549
550 fold_diff(global_diff);
551 }
552
553 /*
554 * this is only called if !populated_zone(zone), which implies no other users of
555 * pset->vm_stat_diff[] exsist.
556 */
557 void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
558 {
559 int i;
560
561 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
562 if (pset->vm_stat_diff[i]) {
563 int v = pset->vm_stat_diff[i];
564 pset->vm_stat_diff[i] = 0;
565 atomic_long_add(v, &zone->vm_stat[i]);
566 atomic_long_add(v, &vm_stat[i]);
567 }
568 }
569 #endif
570
571 #ifdef CONFIG_NUMA
572 /*
573 * Determine the per node value of a stat item.
574 */
575 unsigned long node_page_state(int node, enum zone_stat_item item)
576 {
577 struct zone *zones = NODE_DATA(node)->node_zones;
578 int i;
579 unsigned long count = 0;
580
581 for (i = 0; i < MAX_NR_ZONES; i++)
582 count += zone_page_state(zones + i, item);
583
584 return count;
585 }
586
587 #endif
588
589 #ifdef CONFIG_COMPACTION
590
591 struct contig_page_info {
592 unsigned long free_pages;
593 unsigned long free_blocks_total;
594 unsigned long free_blocks_suitable;
595 };
596
597 /*
598 * Calculate the number of free pages in a zone, how many contiguous
599 * pages are free and how many are large enough to satisfy an allocation of
600 * the target size. Note that this function makes no attempt to estimate
601 * how many suitable free blocks there *might* be if MOVABLE pages were
602 * migrated. Calculating that is possible, but expensive and can be
603 * figured out from userspace
604 */
605 static void fill_contig_page_info(struct zone *zone,
606 unsigned int suitable_order,
607 struct contig_page_info *info)
608 {
609 unsigned int order;
610
611 info->free_pages = 0;
612 info->free_blocks_total = 0;
613 info->free_blocks_suitable = 0;
614
615 for (order = 0; order < MAX_ORDER; order++) {
616 unsigned long blocks;
617
618 /* Count number of free blocks */
619 blocks = zone->free_area[order].nr_free;
620 info->free_blocks_total += blocks;
621
622 /* Count free base pages */
623 info->free_pages += blocks << order;
624
625 /* Count the suitable free blocks */
626 if (order >= suitable_order)
627 info->free_blocks_suitable += blocks <<
628 (order - suitable_order);
629 }
630 }
631
632 /*
633 * A fragmentation index only makes sense if an allocation of a requested
634 * size would fail. If that is true, the fragmentation index indicates
635 * whether external fragmentation or a lack of memory was the problem.
636 * The value can be used to determine if page reclaim or compaction
637 * should be used
638 */
639 static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
640 {
641 unsigned long requested = 1UL << order;
642
643 if (!info->free_blocks_total)
644 return 0;
645
646 /* Fragmentation index only makes sense when a request would fail */
647 if (info->free_blocks_suitable)
648 return -1000;
649
650 /*
651 * Index is between 0 and 1 so return within 3 decimal places
652 *
653 * 0 => allocation would fail due to lack of memory
654 * 1 => allocation would fail due to fragmentation
655 */
656 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
657 }
658
659 /* Same as __fragmentation index but allocs contig_page_info on stack */
660 int fragmentation_index(struct zone *zone, unsigned int order)
661 {
662 struct contig_page_info info;
663
664 fill_contig_page_info(zone, order, &info);
665 return __fragmentation_index(order, &info);
666 }
667 #endif
668
669 #if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
670 #ifdef CONFIG_ZONE_DMA
671 #define TEXT_FOR_DMA(xx) xx "_dma",
672 #else
673 #define TEXT_FOR_DMA(xx)
674 #endif
675
676 #ifdef CONFIG_ZONE_DMA32
677 #define TEXT_FOR_DMA32(xx) xx "_dma32",
678 #else
679 #define TEXT_FOR_DMA32(xx)
680 #endif
681
682 #ifdef CONFIG_HIGHMEM
683 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
684 #else
685 #define TEXT_FOR_HIGHMEM(xx)
686 #endif
687
688 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
689 TEXT_FOR_HIGHMEM(xx) xx "_movable",
690
691 const char * const vmstat_text[] = {
692 /* enum zone_stat_item countes */
693 "nr_free_pages",
694 "nr_alloc_batch",
695 "nr_inactive_anon",
696 "nr_active_anon",
697 "nr_inactive_file",
698 "nr_active_file",
699 "nr_unevictable",
700 "nr_mlock",
701 "nr_anon_pages",
702 "nr_mapped",
703 "nr_file_pages",
704 "nr_dirty",
705 "nr_writeback",
706 "nr_slab_reclaimable",
707 "nr_slab_unreclaimable",
708 "nr_page_table_pages",
709 "nr_kernel_stack",
710 "nr_unstable",
711 "nr_bounce",
712 "nr_vmscan_write",
713 "nr_vmscan_immediate_reclaim",
714 "nr_writeback_temp",
715 "nr_isolated_anon",
716 "nr_isolated_file",
717 "nr_shmem",
718 "nr_dirtied",
719 "nr_written",
720 "nr_pages_scanned",
721
722 #ifdef CONFIG_NUMA
723 "numa_hit",
724 "numa_miss",
725 "numa_foreign",
726 "numa_interleave",
727 "numa_local",
728 "numa_other",
729 #endif
730 "workingset_refault",
731 "workingset_activate",
732 "workingset_nodereclaim",
733 "nr_anon_transparent_hugepages",
734 "nr_free_cma",
735
736 /* enum writeback_stat_item counters */
737 "nr_dirty_threshold",
738 "nr_dirty_background_threshold",
739
740 #ifdef CONFIG_VM_EVENT_COUNTERS
741 /* enum vm_event_item counters */
742 "pgpgin",
743 "pgpgout",
744 "pswpin",
745 "pswpout",
746
747 TEXTS_FOR_ZONES("pgalloc")
748
749 "pgfree",
750 "pgactivate",
751 "pgdeactivate",
752
753 "pgfault",
754 "pgmajfault",
755 "pglazyfreed",
756
757 TEXTS_FOR_ZONES("pgrefill")
758 TEXTS_FOR_ZONES("pgsteal_kswapd")
759 TEXTS_FOR_ZONES("pgsteal_direct")
760 TEXTS_FOR_ZONES("pgscan_kswapd")
761 TEXTS_FOR_ZONES("pgscan_direct")
762 "pgscan_direct_throttle",
763
764 #ifdef CONFIG_NUMA
765 "zone_reclaim_failed",
766 #endif
767 "pginodesteal",
768 "slabs_scanned",
769 "kswapd_inodesteal",
770 "kswapd_low_wmark_hit_quickly",
771 "kswapd_high_wmark_hit_quickly",
772 "pageoutrun",
773 "allocstall",
774
775 "pgrotated",
776
777 "drop_pagecache",
778 "drop_slab",
779
780 #ifdef CONFIG_NUMA_BALANCING
781 "numa_pte_updates",
782 "numa_huge_pte_updates",
783 "numa_hint_faults",
784 "numa_hint_faults_local",
785 "numa_pages_migrated",
786 #endif
787 #ifdef CONFIG_MIGRATION
788 "pgmigrate_success",
789 "pgmigrate_fail",
790 #endif
791 #ifdef CONFIG_COMPACTION
792 "compact_migrate_scanned",
793 "compact_free_scanned",
794 "compact_isolated",
795 "compact_stall",
796 "compact_fail",
797 "compact_success",
798 "compact_daemon_wake",
799 #endif
800
801 #ifdef CONFIG_HUGETLB_PAGE
802 "htlb_buddy_alloc_success",
803 "htlb_buddy_alloc_fail",
804 #endif
805 "unevictable_pgs_culled",
806 "unevictable_pgs_scanned",
807 "unevictable_pgs_rescued",
808 "unevictable_pgs_mlocked",
809 "unevictable_pgs_munlocked",
810 "unevictable_pgs_cleared",
811 "unevictable_pgs_stranded",
812
813 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
814 "thp_fault_alloc",
815 "thp_fault_fallback",
816 "thp_collapse_alloc",
817 "thp_collapse_alloc_failed",
818 "thp_split_page",
819 "thp_split_page_failed",
820 "thp_deferred_split_page",
821 "thp_split_pmd",
822 "thp_zero_page_alloc",
823 "thp_zero_page_alloc_failed",
824 #endif
825 #ifdef CONFIG_MEMORY_BALLOON
826 "balloon_inflate",
827 "balloon_deflate",
828 #ifdef CONFIG_BALLOON_COMPACTION
829 "balloon_migrate",
830 #endif
831 #endif /* CONFIG_MEMORY_BALLOON */
832 #ifdef CONFIG_DEBUG_TLBFLUSH
833 #ifdef CONFIG_SMP
834 "nr_tlb_remote_flush",
835 "nr_tlb_remote_flush_received",
836 #endif /* CONFIG_SMP */
837 "nr_tlb_local_flush_all",
838 "nr_tlb_local_flush_one",
839 #endif /* CONFIG_DEBUG_TLBFLUSH */
840
841 #ifdef CONFIG_DEBUG_VM_VMACACHE
842 "vmacache_find_calls",
843 "vmacache_find_hits",
844 "vmacache_full_flushes",
845 #endif
846 #endif /* CONFIG_VM_EVENTS_COUNTERS */
847 };
848 #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
849
850
851 #if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
852 defined(CONFIG_PROC_FS)
853 static void *frag_start(struct seq_file *m, loff_t *pos)
854 {
855 pg_data_t *pgdat;
856 loff_t node = *pos;
857
858 for (pgdat = first_online_pgdat();
859 pgdat && node;
860 pgdat = next_online_pgdat(pgdat))
861 --node;
862
863 return pgdat;
864 }
865
866 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
867 {
868 pg_data_t *pgdat = (pg_data_t *)arg;
869
870 (*pos)++;
871 return next_online_pgdat(pgdat);
872 }
873
874 static void frag_stop(struct seq_file *m, void *arg)
875 {
876 }
877
878 /* Walk all the zones in a node and print using a callback */
879 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
880 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
881 {
882 struct zone *zone;
883 struct zone *node_zones = pgdat->node_zones;
884 unsigned long flags;
885
886 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
887 if (!populated_zone(zone))
888 continue;
889
890 spin_lock_irqsave(&zone->lock, flags);
891 print(m, pgdat, zone);
892 spin_unlock_irqrestore(&zone->lock, flags);
893 }
894 }
895 #endif
896
897 #ifdef CONFIG_PROC_FS
898 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
899 struct zone *zone)
900 {
901 int order;
902
903 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
904 for (order = 0; order < MAX_ORDER; ++order)
905 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
906 seq_putc(m, '\n');
907 }
908
909 /*
910 * This walks the free areas for each zone.
911 */
912 static int frag_show(struct seq_file *m, void *arg)
913 {
914 pg_data_t *pgdat = (pg_data_t *)arg;
915 walk_zones_in_node(m, pgdat, frag_show_print);
916 return 0;
917 }
918
919 static void pagetypeinfo_showfree_print(struct seq_file *m,
920 pg_data_t *pgdat, struct zone *zone)
921 {
922 int order, mtype;
923
924 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
925 seq_printf(m, "Node %4d, zone %8s, type %12s ",
926 pgdat->node_id,
927 zone->name,
928 migratetype_names[mtype]);
929 for (order = 0; order < MAX_ORDER; ++order) {
930 unsigned long freecount = 0;
931 struct free_area *area;
932 struct list_head *curr;
933
934 area = &(zone->free_area[order]);
935
936 list_for_each(curr, &area->free_list[mtype])
937 freecount++;
938 seq_printf(m, "%6lu ", freecount);
939 }
940 seq_putc(m, '\n');
941 }
942 }
943
944 /* Print out the free pages at each order for each migatetype */
945 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
946 {
947 int order;
948 pg_data_t *pgdat = (pg_data_t *)arg;
949
950 /* Print header */
951 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
952 for (order = 0; order < MAX_ORDER; ++order)
953 seq_printf(m, "%6d ", order);
954 seq_putc(m, '\n');
955
956 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
957
958 return 0;
959 }
960
961 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
962 pg_data_t *pgdat, struct zone *zone)
963 {
964 int mtype;
965 unsigned long pfn;
966 unsigned long start_pfn = zone->zone_start_pfn;
967 unsigned long end_pfn = zone_end_pfn(zone);
968 unsigned long count[MIGRATE_TYPES] = { 0, };
969
970 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
971 struct page *page;
972
973 if (!pfn_valid(pfn))
974 continue;
975
976 page = pfn_to_page(pfn);
977
978 /* Watch for unexpected holes punched in the memmap */
979 if (!memmap_valid_within(pfn, page, zone))
980 continue;
981
982 if (page_zone(page) != zone)
983 continue;
984
985 mtype = get_pageblock_migratetype(page);
986
987 if (mtype < MIGRATE_TYPES)
988 count[mtype]++;
989 }
990
991 /* Print counts */
992 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
993 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
994 seq_printf(m, "%12lu ", count[mtype]);
995 seq_putc(m, '\n');
996 }
997
998 /* Print out the free pages at each order for each migratetype */
999 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1000 {
1001 int mtype;
1002 pg_data_t *pgdat = (pg_data_t *)arg;
1003
1004 seq_printf(m, "\n%-23s", "Number of blocks type ");
1005 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1006 seq_printf(m, "%12s ", migratetype_names[mtype]);
1007 seq_putc(m, '\n');
1008 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
1009
1010 return 0;
1011 }
1012
1013 #ifdef CONFIG_PAGE_OWNER
1014 static void pagetypeinfo_showmixedcount_print(struct seq_file *m,
1015 pg_data_t *pgdat,
1016 struct zone *zone)
1017 {
1018 struct page *page;
1019 struct page_ext *page_ext;
1020 unsigned long pfn = zone->zone_start_pfn, block_end_pfn;
1021 unsigned long end_pfn = pfn + zone->spanned_pages;
1022 unsigned long count[MIGRATE_TYPES] = { 0, };
1023 int pageblock_mt, page_mt;
1024 int i;
1025
1026 /* Scan block by block. First and last block may be incomplete */
1027 pfn = zone->zone_start_pfn;
1028
1029 /*
1030 * Walk the zone in pageblock_nr_pages steps. If a page block spans
1031 * a zone boundary, it will be double counted between zones. This does
1032 * not matter as the mixed block count will still be correct
1033 */
1034 for (; pfn < end_pfn; ) {
1035 if (!pfn_valid(pfn)) {
1036 pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
1037 continue;
1038 }
1039
1040 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
1041 block_end_pfn = min(block_end_pfn, end_pfn);
1042
1043 page = pfn_to_page(pfn);
1044 pageblock_mt = get_pageblock_migratetype(page);
1045
1046 for (; pfn < block_end_pfn; pfn++) {
1047 if (!pfn_valid_within(pfn))
1048 continue;
1049
1050 page = pfn_to_page(pfn);
1051
1052 if (page_zone(page) != zone)
1053 continue;
1054
1055 if (PageBuddy(page)) {
1056 pfn += (1UL << page_order(page)) - 1;
1057 continue;
1058 }
1059
1060 if (PageReserved(page))
1061 continue;
1062
1063 page_ext = lookup_page_ext(page);
1064
1065 if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
1066 continue;
1067
1068 page_mt = gfpflags_to_migratetype(page_ext->gfp_mask);
1069 if (pageblock_mt != page_mt) {
1070 if (is_migrate_cma(pageblock_mt))
1071 count[MIGRATE_MOVABLE]++;
1072 else
1073 count[pageblock_mt]++;
1074
1075 pfn = block_end_pfn;
1076 break;
1077 }
1078 pfn += (1UL << page_ext->order) - 1;
1079 }
1080 }
1081
1082 /* Print counts */
1083 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1084 for (i = 0; i < MIGRATE_TYPES; i++)
1085 seq_printf(m, "%12lu ", count[i]);
1086 seq_putc(m, '\n');
1087 }
1088 #endif /* CONFIG_PAGE_OWNER */
1089
1090 /*
1091 * Print out the number of pageblocks for each migratetype that contain pages
1092 * of other types. This gives an indication of how well fallbacks are being
1093 * contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1094 * to determine what is going on
1095 */
1096 static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1097 {
1098 #ifdef CONFIG_PAGE_OWNER
1099 int mtype;
1100
1101 if (!static_branch_unlikely(&page_owner_inited))
1102 return;
1103
1104 drain_all_pages(NULL);
1105
1106 seq_printf(m, "\n%-23s", "Number of mixed blocks ");
1107 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1108 seq_printf(m, "%12s ", migratetype_names[mtype]);
1109 seq_putc(m, '\n');
1110
1111 walk_zones_in_node(m, pgdat, pagetypeinfo_showmixedcount_print);
1112 #endif /* CONFIG_PAGE_OWNER */
1113 }
1114
1115 /*
1116 * This prints out statistics in relation to grouping pages by mobility.
1117 * It is expensive to collect so do not constantly read the file.
1118 */
1119 static int pagetypeinfo_show(struct seq_file *m, void *arg)
1120 {
1121 pg_data_t *pgdat = (pg_data_t *)arg;
1122
1123 /* check memoryless node */
1124 if (!node_state(pgdat->node_id, N_MEMORY))
1125 return 0;
1126
1127 seq_printf(m, "Page block order: %d\n", pageblock_order);
1128 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
1129 seq_putc(m, '\n');
1130 pagetypeinfo_showfree(m, pgdat);
1131 pagetypeinfo_showblockcount(m, pgdat);
1132 pagetypeinfo_showmixedcount(m, pgdat);
1133
1134 return 0;
1135 }
1136
1137 static const struct seq_operations fragmentation_op = {
1138 .start = frag_start,
1139 .next = frag_next,
1140 .stop = frag_stop,
1141 .show = frag_show,
1142 };
1143
1144 static int fragmentation_open(struct inode *inode, struct file *file)
1145 {
1146 return seq_open(file, &fragmentation_op);
1147 }
1148
1149 static const struct file_operations fragmentation_file_operations = {
1150 .open = fragmentation_open,
1151 .read = seq_read,
1152 .llseek = seq_lseek,
1153 .release = seq_release,
1154 };
1155
1156 static const struct seq_operations pagetypeinfo_op = {
1157 .start = frag_start,
1158 .next = frag_next,
1159 .stop = frag_stop,
1160 .show = pagetypeinfo_show,
1161 };
1162
1163 static int pagetypeinfo_open(struct inode *inode, struct file *file)
1164 {
1165 return seq_open(file, &pagetypeinfo_op);
1166 }
1167
1168 static const struct file_operations pagetypeinfo_file_ops = {
1169 .open = pagetypeinfo_open,
1170 .read = seq_read,
1171 .llseek = seq_lseek,
1172 .release = seq_release,
1173 };
1174
1175 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1176 struct zone *zone)
1177 {
1178 int i;
1179 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1180 seq_printf(m,
1181 "\n pages free %lu"
1182 "\n min %lu"
1183 "\n low %lu"
1184 "\n high %lu"
1185 "\n scanned %lu"
1186 "\n spanned %lu"
1187 "\n present %lu"
1188 "\n managed %lu",
1189 zone_page_state(zone, NR_FREE_PAGES),
1190 min_wmark_pages(zone),
1191 low_wmark_pages(zone),
1192 high_wmark_pages(zone),
1193 zone_page_state(zone, NR_PAGES_SCANNED),
1194 zone->spanned_pages,
1195 zone->present_pages,
1196 zone->managed_pages);
1197
1198 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1199 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1200 zone_page_state(zone, i));
1201
1202 seq_printf(m,
1203 "\n protection: (%ld",
1204 zone->lowmem_reserve[0]);
1205 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1206 seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
1207 seq_printf(m,
1208 ")"
1209 "\n pagesets");
1210 for_each_online_cpu(i) {
1211 struct per_cpu_pageset *pageset;
1212
1213 pageset = per_cpu_ptr(zone->pageset, i);
1214 seq_printf(m,
1215 "\n cpu: %i"
1216 "\n count: %i"
1217 "\n high: %i"
1218 "\n batch: %i",
1219 i,
1220 pageset->pcp.count,
1221 pageset->pcp.high,
1222 pageset->pcp.batch);
1223 #ifdef CONFIG_SMP
1224 seq_printf(m, "\n vm stats threshold: %d",
1225 pageset->stat_threshold);
1226 #endif
1227 }
1228 seq_printf(m,
1229 "\n all_unreclaimable: %u"
1230 "\n start_pfn: %lu"
1231 "\n inactive_ratio: %u",
1232 !zone_reclaimable(zone),
1233 zone->zone_start_pfn,
1234 zone->inactive_ratio);
1235 seq_putc(m, '\n');
1236 }
1237
1238 /*
1239 * Output information about zones in @pgdat.
1240 */
1241 static int zoneinfo_show(struct seq_file *m, void *arg)
1242 {
1243 pg_data_t *pgdat = (pg_data_t *)arg;
1244 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
1245 return 0;
1246 }
1247
1248 static const struct seq_operations zoneinfo_op = {
1249 .start = frag_start, /* iterate over all zones. The same as in
1250 * fragmentation. */
1251 .next = frag_next,
1252 .stop = frag_stop,
1253 .show = zoneinfo_show,
1254 };
1255
1256 static int zoneinfo_open(struct inode *inode, struct file *file)
1257 {
1258 return seq_open(file, &zoneinfo_op);
1259 }
1260
1261 static const struct file_operations proc_zoneinfo_file_operations = {
1262 .open = zoneinfo_open,
1263 .read = seq_read,
1264 .llseek = seq_lseek,
1265 .release = seq_release,
1266 };
1267
1268 enum writeback_stat_item {
1269 NR_DIRTY_THRESHOLD,
1270 NR_DIRTY_BG_THRESHOLD,
1271 NR_VM_WRITEBACK_STAT_ITEMS,
1272 };
1273
1274 static void *vmstat_start(struct seq_file *m, loff_t *pos)
1275 {
1276 unsigned long *v;
1277 int i, stat_items_size;
1278
1279 if (*pos >= ARRAY_SIZE(vmstat_text))
1280 return NULL;
1281 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1282 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
1283
1284 #ifdef CONFIG_VM_EVENT_COUNTERS
1285 stat_items_size += sizeof(struct vm_event_state);
1286 #endif
1287
1288 v = kmalloc(stat_items_size, GFP_KERNEL);
1289 m->private = v;
1290 if (!v)
1291 return ERR_PTR(-ENOMEM);
1292 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1293 v[i] = global_page_state(i);
1294 v += NR_VM_ZONE_STAT_ITEMS;
1295
1296 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1297 v + NR_DIRTY_THRESHOLD);
1298 v += NR_VM_WRITEBACK_STAT_ITEMS;
1299
1300 #ifdef CONFIG_VM_EVENT_COUNTERS
1301 all_vm_events(v);
1302 v[PGPGIN] /= 2; /* sectors -> kbytes */
1303 v[PGPGOUT] /= 2;
1304 #endif
1305 return (unsigned long *)m->private + *pos;
1306 }
1307
1308 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1309 {
1310 (*pos)++;
1311 if (*pos >= ARRAY_SIZE(vmstat_text))
1312 return NULL;
1313 return (unsigned long *)m->private + *pos;
1314 }
1315
1316 static int vmstat_show(struct seq_file *m, void *arg)
1317 {
1318 unsigned long *l = arg;
1319 unsigned long off = l - (unsigned long *)m->private;
1320
1321 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1322 return 0;
1323 }
1324
1325 static void vmstat_stop(struct seq_file *m, void *arg)
1326 {
1327 kfree(m->private);
1328 m->private = NULL;
1329 }
1330
1331 static const struct seq_operations vmstat_op = {
1332 .start = vmstat_start,
1333 .next = vmstat_next,
1334 .stop = vmstat_stop,
1335 .show = vmstat_show,
1336 };
1337
1338 static int vmstat_open(struct inode *inode, struct file *file)
1339 {
1340 return seq_open(file, &vmstat_op);
1341 }
1342
1343 static const struct file_operations proc_vmstat_file_operations = {
1344 .open = vmstat_open,
1345 .read = seq_read,
1346 .llseek = seq_lseek,
1347 .release = seq_release,
1348 };
1349 #endif /* CONFIG_PROC_FS */
1350
1351 #ifdef CONFIG_SMP
1352 static struct workqueue_struct *vmstat_wq;
1353 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1354 int sysctl_stat_interval __read_mostly = HZ;
1355
1356 #ifdef CONFIG_PROC_FS
1357 static void refresh_vm_stats(struct work_struct *work)
1358 {
1359 refresh_cpu_vm_stats(true);
1360 }
1361
1362 int vmstat_refresh(struct ctl_table *table, int write,
1363 void __user *buffer, size_t *lenp, loff_t *ppos)
1364 {
1365 long val;
1366 int err;
1367 int i;
1368
1369 /*
1370 * The regular update, every sysctl_stat_interval, may come later
1371 * than expected: leaving a significant amount in per_cpu buckets.
1372 * This is particularly misleading when checking a quantity of HUGE
1373 * pages, immediately after running a test. /proc/sys/vm/stat_refresh,
1374 * which can equally be echo'ed to or cat'ted from (by root),
1375 * can be used to update the stats just before reading them.
1376 *
1377 * Oh, and since global_page_state() etc. are so careful to hide
1378 * transiently negative values, report an error here if any of
1379 * the stats is negative, so we know to go looking for imbalance.
1380 */
1381 err = schedule_on_each_cpu(refresh_vm_stats);
1382 if (err)
1383 return err;
1384 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
1385 val = atomic_long_read(&vm_stat[i]);
1386 if (val < 0) {
1387 switch (i) {
1388 case NR_ALLOC_BATCH:
1389 case NR_PAGES_SCANNED:
1390 /*
1391 * These are often seen to go negative in
1392 * recent kernels, but not to go permanently
1393 * negative. Whilst it would be nicer not to
1394 * have exceptions, rooting them out would be
1395 * another task, of rather low priority.
1396 */
1397 break;
1398 default:
1399 pr_warn("%s: %s %ld\n",
1400 __func__, vmstat_text[i], val);
1401 err = -EINVAL;
1402 break;
1403 }
1404 }
1405 }
1406 if (err)
1407 return err;
1408 if (write)
1409 *ppos += *lenp;
1410 else
1411 *lenp = 0;
1412 return 0;
1413 }
1414 #endif /* CONFIG_PROC_FS */
1415
1416 static void vmstat_update(struct work_struct *w)
1417 {
1418 if (refresh_cpu_vm_stats(true)) {
1419 /*
1420 * Counters were updated so we expect more updates
1421 * to occur in the future. Keep on running the
1422 * update worker thread.
1423 */
1424 queue_delayed_work_on(smp_processor_id(), vmstat_wq,
1425 this_cpu_ptr(&vmstat_work),
1426 round_jiffies_relative(sysctl_stat_interval));
1427 }
1428 }
1429
1430 /*
1431 * Switch off vmstat processing and then fold all the remaining differentials
1432 * until the diffs stay at zero. The function is used by NOHZ and can only be
1433 * invoked when tick processing is not active.
1434 */
1435 /*
1436 * Check if the diffs for a certain cpu indicate that
1437 * an update is needed.
1438 */
1439 static bool need_update(int cpu)
1440 {
1441 struct zone *zone;
1442
1443 for_each_populated_zone(zone) {
1444 struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu);
1445
1446 BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1);
1447 /*
1448 * The fast way of checking if there are any vmstat diffs.
1449 * This works because the diffs are byte sized items.
1450 */
1451 if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS))
1452 return true;
1453
1454 }
1455 return false;
1456 }
1457
1458 /*
1459 * Switch off vmstat processing and then fold all the remaining differentials
1460 * until the diffs stay at zero. The function is used by NOHZ and can only be
1461 * invoked when tick processing is not active.
1462 */
1463 void quiet_vmstat(void)
1464 {
1465 if (system_state != SYSTEM_RUNNING)
1466 return;
1467
1468 if (!delayed_work_pending(this_cpu_ptr(&vmstat_work)))
1469 return;
1470
1471 if (!need_update(smp_processor_id()))
1472 return;
1473
1474 /*
1475 * Just refresh counters and do not care about the pending delayed
1476 * vmstat_update. It doesn't fire that often to matter and canceling
1477 * it would be too expensive from this path.
1478 * vmstat_shepherd will take care about that for us.
1479 */
1480 refresh_cpu_vm_stats(false);
1481 }
1482
1483 /*
1484 * Shepherd worker thread that checks the
1485 * differentials of processors that have their worker
1486 * threads for vm statistics updates disabled because of
1487 * inactivity.
1488 */
1489 static void vmstat_shepherd(struct work_struct *w);
1490
1491 static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
1492
1493 static void vmstat_shepherd(struct work_struct *w)
1494 {
1495 int cpu;
1496
1497 get_online_cpus();
1498 /* Check processors whose vmstat worker threads have been disabled */
1499 for_each_online_cpu(cpu) {
1500 struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
1501
1502 if (!delayed_work_pending(dw) && need_update(cpu))
1503 queue_delayed_work_on(cpu, vmstat_wq, dw, 0);
1504 }
1505 put_online_cpus();
1506
1507 schedule_delayed_work(&shepherd,
1508 round_jiffies_relative(sysctl_stat_interval));
1509 }
1510
1511 static void __init start_shepherd_timer(void)
1512 {
1513 int cpu;
1514
1515 for_each_possible_cpu(cpu)
1516 INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
1517 vmstat_update);
1518
1519 vmstat_wq = alloc_workqueue("vmstat", WQ_FREEZABLE|WQ_MEM_RECLAIM, 0);
1520 schedule_delayed_work(&shepherd,
1521 round_jiffies_relative(sysctl_stat_interval));
1522 }
1523
1524 static void vmstat_cpu_dead(int node)
1525 {
1526 int cpu;
1527
1528 get_online_cpus();
1529 for_each_online_cpu(cpu)
1530 if (cpu_to_node(cpu) == node)
1531 goto end;
1532
1533 node_clear_state(node, N_CPU);
1534 end:
1535 put_online_cpus();
1536 }
1537
1538 /*
1539 * Use the cpu notifier to insure that the thresholds are recalculated
1540 * when necessary.
1541 */
1542 static int vmstat_cpuup_callback(struct notifier_block *nfb,
1543 unsigned long action,
1544 void *hcpu)
1545 {
1546 long cpu = (long)hcpu;
1547
1548 switch (action) {
1549 case CPU_ONLINE:
1550 case CPU_ONLINE_FROZEN:
1551 refresh_zone_stat_thresholds();
1552 node_set_state(cpu_to_node(cpu), N_CPU);
1553 break;
1554 case CPU_DOWN_PREPARE:
1555 case CPU_DOWN_PREPARE_FROZEN:
1556 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1557 break;
1558 case CPU_DOWN_FAILED:
1559 case CPU_DOWN_FAILED_FROZEN:
1560 break;
1561 case CPU_DEAD:
1562 case CPU_DEAD_FROZEN:
1563 refresh_zone_stat_thresholds();
1564 vmstat_cpu_dead(cpu_to_node(cpu));
1565 break;
1566 default:
1567 break;
1568 }
1569 return NOTIFY_OK;
1570 }
1571
1572 static struct notifier_block vmstat_notifier =
1573 { &vmstat_cpuup_callback, NULL, 0 };
1574 #endif
1575
1576 static int __init setup_vmstat(void)
1577 {
1578 #ifdef CONFIG_SMP
1579 cpu_notifier_register_begin();
1580 __register_cpu_notifier(&vmstat_notifier);
1581
1582 start_shepherd_timer();
1583 cpu_notifier_register_done();
1584 #endif
1585 #ifdef CONFIG_PROC_FS
1586 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1587 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1588 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1589 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1590 #endif
1591 return 0;
1592 }
1593 module_init(setup_vmstat)
1594
1595 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1596
1597 /*
1598 * Return an index indicating how much of the available free memory is
1599 * unusable for an allocation of the requested size.
1600 */
1601 static int unusable_free_index(unsigned int order,
1602 struct contig_page_info *info)
1603 {
1604 /* No free memory is interpreted as all free memory is unusable */
1605 if (info->free_pages == 0)
1606 return 1000;
1607
1608 /*
1609 * Index should be a value between 0 and 1. Return a value to 3
1610 * decimal places.
1611 *
1612 * 0 => no fragmentation
1613 * 1 => high fragmentation
1614 */
1615 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1616
1617 }
1618
1619 static void unusable_show_print(struct seq_file *m,
1620 pg_data_t *pgdat, struct zone *zone)
1621 {
1622 unsigned int order;
1623 int index;
1624 struct contig_page_info info;
1625
1626 seq_printf(m, "Node %d, zone %8s ",
1627 pgdat->node_id,
1628 zone->name);
1629 for (order = 0; order < MAX_ORDER; ++order) {
1630 fill_contig_page_info(zone, order, &info);
1631 index = unusable_free_index(order, &info);
1632 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1633 }
1634
1635 seq_putc(m, '\n');
1636 }
1637
1638 /*
1639 * Display unusable free space index
1640 *
1641 * The unusable free space index measures how much of the available free
1642 * memory cannot be used to satisfy an allocation of a given size and is a
1643 * value between 0 and 1. The higher the value, the more of free memory is
1644 * unusable and by implication, the worse the external fragmentation is. This
1645 * can be expressed as a percentage by multiplying by 100.
1646 */
1647 static int unusable_show(struct seq_file *m, void *arg)
1648 {
1649 pg_data_t *pgdat = (pg_data_t *)arg;
1650
1651 /* check memoryless node */
1652 if (!node_state(pgdat->node_id, N_MEMORY))
1653 return 0;
1654
1655 walk_zones_in_node(m, pgdat, unusable_show_print);
1656
1657 return 0;
1658 }
1659
1660 static const struct seq_operations unusable_op = {
1661 .start = frag_start,
1662 .next = frag_next,
1663 .stop = frag_stop,
1664 .show = unusable_show,
1665 };
1666
1667 static int unusable_open(struct inode *inode, struct file *file)
1668 {
1669 return seq_open(file, &unusable_op);
1670 }
1671
1672 static const struct file_operations unusable_file_ops = {
1673 .open = unusable_open,
1674 .read = seq_read,
1675 .llseek = seq_lseek,
1676 .release = seq_release,
1677 };
1678
1679 static void extfrag_show_print(struct seq_file *m,
1680 pg_data_t *pgdat, struct zone *zone)
1681 {
1682 unsigned int order;
1683 int index;
1684
1685 /* Alloc on stack as interrupts are disabled for zone walk */
1686 struct contig_page_info info;
1687
1688 seq_printf(m, "Node %d, zone %8s ",
1689 pgdat->node_id,
1690 zone->name);
1691 for (order = 0; order < MAX_ORDER; ++order) {
1692 fill_contig_page_info(zone, order, &info);
1693 index = __fragmentation_index(order, &info);
1694 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1695 }
1696
1697 seq_putc(m, '\n');
1698 }
1699
1700 /*
1701 * Display fragmentation index for orders that allocations would fail for
1702 */
1703 static int extfrag_show(struct seq_file *m, void *arg)
1704 {
1705 pg_data_t *pgdat = (pg_data_t *)arg;
1706
1707 walk_zones_in_node(m, pgdat, extfrag_show_print);
1708
1709 return 0;
1710 }
1711
1712 static const struct seq_operations extfrag_op = {
1713 .start = frag_start,
1714 .next = frag_next,
1715 .stop = frag_stop,
1716 .show = extfrag_show,
1717 };
1718
1719 static int extfrag_open(struct inode *inode, struct file *file)
1720 {
1721 return seq_open(file, &extfrag_op);
1722 }
1723
1724 static const struct file_operations extfrag_file_ops = {
1725 .open = extfrag_open,
1726 .read = seq_read,
1727 .llseek = seq_lseek,
1728 .release = seq_release,
1729 };
1730
1731 static int __init extfrag_debug_init(void)
1732 {
1733 struct dentry *extfrag_debug_root;
1734
1735 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1736 if (!extfrag_debug_root)
1737 return -ENOMEM;
1738
1739 if (!debugfs_create_file("unusable_index", 0444,
1740 extfrag_debug_root, NULL, &unusable_file_ops))
1741 goto fail;
1742
1743 if (!debugfs_create_file("extfrag_index", 0444,
1744 extfrag_debug_root, NULL, &extfrag_file_ops))
1745 goto fail;
1746
1747 return 0;
1748 fail:
1749 debugfs_remove_recursive(extfrag_debug_root);
1750 return -ENOMEM;
1751 }
1752
1753 module_init(extfrag_debug_init);
1754 #endif