xref: /linux/mm/vmstat.c (revision 42fda66387daa53538ae13a2c858396aaf037158)
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  */
11 
12 #include <linux/mm.h>
13 #include <linux/err.h>
14 #include <linux/module.h>
15 #include <linux/cpu.h>
16 #include <linux/sched.h>
17 
18 #ifdef CONFIG_VM_EVENT_COUNTERS
19 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
20 EXPORT_PER_CPU_SYMBOL(vm_event_states);
21 
22 static void sum_vm_events(unsigned long *ret, cpumask_t *cpumask)
23 {
24 	int cpu = 0;
25 	int i;
26 
27 	memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
28 
29 	cpu = first_cpu(*cpumask);
30 	while (cpu < NR_CPUS) {
31 		struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
32 
33 		cpu = next_cpu(cpu, *cpumask);
34 
35 		if (cpu < NR_CPUS)
36 			prefetch(&per_cpu(vm_event_states, cpu));
37 
38 
39 		for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
40 			ret[i] += this->event[i];
41 	}
42 }
43 
44 /*
45  * Accumulate the vm event counters across all CPUs.
46  * The result is unavoidably approximate - it can change
47  * during and after execution of this function.
48 */
49 void all_vm_events(unsigned long *ret)
50 {
51 	sum_vm_events(ret, &cpu_online_map);
52 }
53 EXPORT_SYMBOL_GPL(all_vm_events);
54 
55 #ifdef CONFIG_HOTPLUG
56 /*
57  * Fold the foreign cpu events into our own.
58  *
59  * This is adding to the events on one processor
60  * but keeps the global counts constant.
61  */
62 void vm_events_fold_cpu(int cpu)
63 {
64 	struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
65 	int i;
66 
67 	for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
68 		count_vm_events(i, fold_state->event[i]);
69 		fold_state->event[i] = 0;
70 	}
71 }
72 #endif /* CONFIG_HOTPLUG */
73 
74 #endif /* CONFIG_VM_EVENT_COUNTERS */
75 
76 /*
77  * Manage combined zone based / global counters
78  *
79  * vm_stat contains the global counters
80  */
81 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
82 EXPORT_SYMBOL(vm_stat);
83 
84 #ifdef CONFIG_SMP
85 
86 static int calculate_threshold(struct zone *zone)
87 {
88 	int threshold;
89 	int mem;	/* memory in 128 MB units */
90 
91 	/*
92 	 * The threshold scales with the number of processors and the amount
93 	 * of memory per zone. More memory means that we can defer updates for
94 	 * longer, more processors could lead to more contention.
95  	 * fls() is used to have a cheap way of logarithmic scaling.
96 	 *
97 	 * Some sample thresholds:
98 	 *
99 	 * Threshold	Processors	(fls)	Zonesize	fls(mem+1)
100 	 * ------------------------------------------------------------------
101 	 * 8		1		1	0.9-1 GB	4
102 	 * 16		2		2	0.9-1 GB	4
103 	 * 20 		2		2	1-2 GB		5
104 	 * 24		2		2	2-4 GB		6
105 	 * 28		2		2	4-8 GB		7
106 	 * 32		2		2	8-16 GB		8
107 	 * 4		2		2	<128M		1
108 	 * 30		4		3	2-4 GB		5
109 	 * 48		4		3	8-16 GB		8
110 	 * 32		8		4	1-2 GB		4
111 	 * 32		8		4	0.9-1GB		4
112 	 * 10		16		5	<128M		1
113 	 * 40		16		5	900M		4
114 	 * 70		64		7	2-4 GB		5
115 	 * 84		64		7	4-8 GB		6
116 	 * 108		512		9	4-8 GB		6
117 	 * 125		1024		10	8-16 GB		8
118 	 * 125		1024		10	16-32 GB	9
119 	 */
120 
121 	mem = zone->present_pages >> (27 - PAGE_SHIFT);
122 
123 	threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
124 
125 	/*
126 	 * Maximum threshold is 125
127 	 */
128 	threshold = min(125, threshold);
129 
130 	return threshold;
131 }
132 
133 /*
134  * Refresh the thresholds for each zone.
135  */
136 static void refresh_zone_stat_thresholds(void)
137 {
138 	struct zone *zone;
139 	int cpu;
140 	int threshold;
141 
142 	for_each_zone(zone) {
143 
144 		if (!zone->present_pages)
145 			continue;
146 
147 		threshold = calculate_threshold(zone);
148 
149 		for_each_online_cpu(cpu)
150 			zone_pcp(zone, cpu)->stat_threshold = threshold;
151 	}
152 }
153 
154 /*
155  * For use when we know that interrupts are disabled.
156  */
157 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
158 				int delta)
159 {
160 	struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
161 	s8 *p = pcp->vm_stat_diff + item;
162 	long x;
163 
164 	x = delta + *p;
165 
166 	if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
167 		zone_page_state_add(x, zone, item);
168 		x = 0;
169 	}
170 	*p = x;
171 }
172 EXPORT_SYMBOL(__mod_zone_page_state);
173 
174 /*
175  * For an unknown interrupt state
176  */
177 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
178 					int delta)
179 {
180 	unsigned long flags;
181 
182 	local_irq_save(flags);
183 	__mod_zone_page_state(zone, item, delta);
184 	local_irq_restore(flags);
185 }
186 EXPORT_SYMBOL(mod_zone_page_state);
187 
188 /*
189  * Optimized increment and decrement functions.
190  *
191  * These are only for a single page and therefore can take a struct page *
192  * argument instead of struct zone *. This allows the inclusion of the code
193  * generated for page_zone(page) into the optimized functions.
194  *
195  * No overflow check is necessary and therefore the differential can be
196  * incremented or decremented in place which may allow the compilers to
197  * generate better code.
198  * The increment or decrement is known and therefore one boundary check can
199  * be omitted.
200  *
201  * NOTE: These functions are very performance sensitive. Change only
202  * with care.
203  *
204  * Some processors have inc/dec instructions that are atomic vs an interrupt.
205  * However, the code must first determine the differential location in a zone
206  * based on the processor number and then inc/dec the counter. There is no
207  * guarantee without disabling preemption that the processor will not change
208  * in between and therefore the atomicity vs. interrupt cannot be exploited
209  * in a useful way here.
210  */
211 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
212 {
213 	struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
214 	s8 *p = pcp->vm_stat_diff + item;
215 
216 	(*p)++;
217 
218 	if (unlikely(*p > pcp->stat_threshold)) {
219 		int overstep = pcp->stat_threshold / 2;
220 
221 		zone_page_state_add(*p + overstep, zone, item);
222 		*p = -overstep;
223 	}
224 }
225 
226 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
227 {
228 	__inc_zone_state(page_zone(page), item);
229 }
230 EXPORT_SYMBOL(__inc_zone_page_state);
231 
232 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
233 {
234 	struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
235 	s8 *p = pcp->vm_stat_diff + item;
236 
237 	(*p)--;
238 
239 	if (unlikely(*p < - pcp->stat_threshold)) {
240 		int overstep = pcp->stat_threshold / 2;
241 
242 		zone_page_state_add(*p - overstep, zone, item);
243 		*p = overstep;
244 	}
245 }
246 
247 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
248 {
249 	__dec_zone_state(page_zone(page), item);
250 }
251 EXPORT_SYMBOL(__dec_zone_page_state);
252 
253 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
254 {
255 	unsigned long flags;
256 
257 	local_irq_save(flags);
258 	__inc_zone_state(zone, item);
259 	local_irq_restore(flags);
260 }
261 
262 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
263 {
264 	unsigned long flags;
265 	struct zone *zone;
266 
267 	zone = page_zone(page);
268 	local_irq_save(flags);
269 	__inc_zone_state(zone, item);
270 	local_irq_restore(flags);
271 }
272 EXPORT_SYMBOL(inc_zone_page_state);
273 
274 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
275 {
276 	unsigned long flags;
277 
278 	local_irq_save(flags);
279 	__dec_zone_page_state(page, item);
280 	local_irq_restore(flags);
281 }
282 EXPORT_SYMBOL(dec_zone_page_state);
283 
284 /*
285  * Update the zone counters for one cpu.
286  *
287  * Note that refresh_cpu_vm_stats strives to only access
288  * node local memory. The per cpu pagesets on remote zones are placed
289  * in the memory local to the processor using that pageset. So the
290  * loop over all zones will access a series of cachelines local to
291  * the processor.
292  *
293  * The call to zone_page_state_add updates the cachelines with the
294  * statistics in the remote zone struct as well as the global cachelines
295  * with the global counters. These could cause remote node cache line
296  * bouncing and will have to be only done when necessary.
297  */
298 void refresh_cpu_vm_stats(int cpu)
299 {
300 	struct zone *zone;
301 	int i;
302 	unsigned long flags;
303 
304 	for_each_zone(zone) {
305 		struct per_cpu_pageset *p;
306 
307 		if (!populated_zone(zone))
308 			continue;
309 
310 		p = zone_pcp(zone, cpu);
311 
312 		for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
313 			if (p->vm_stat_diff[i]) {
314 				local_irq_save(flags);
315 				zone_page_state_add(p->vm_stat_diff[i],
316 					zone, i);
317 				p->vm_stat_diff[i] = 0;
318 #ifdef CONFIG_NUMA
319 				/* 3 seconds idle till flush */
320 				p->expire = 3;
321 #endif
322 				local_irq_restore(flags);
323 			}
324 #ifdef CONFIG_NUMA
325 		/*
326 		 * Deal with draining the remote pageset of this
327 		 * processor
328 		 *
329 		 * Check if there are pages remaining in this pageset
330 		 * if not then there is nothing to expire.
331 		 */
332 		if (!p->expire || (!p->pcp[0].count && !p->pcp[1].count))
333 			continue;
334 
335 		/*
336 		 * We never drain zones local to this processor.
337 		 */
338 		if (zone_to_nid(zone) == numa_node_id()) {
339 			p->expire = 0;
340 			continue;
341 		}
342 
343 		p->expire--;
344 		if (p->expire)
345 			continue;
346 
347 		if (p->pcp[0].count)
348 			drain_zone_pages(zone, p->pcp + 0);
349 
350 		if (p->pcp[1].count)
351 			drain_zone_pages(zone, p->pcp + 1);
352 #endif
353 	}
354 }
355 
356 #endif
357 
358 #ifdef CONFIG_NUMA
359 /*
360  * zonelist = the list of zones passed to the allocator
361  * z 	    = the zone from which the allocation occurred.
362  *
363  * Must be called with interrupts disabled.
364  */
365 void zone_statistics(struct zonelist *zonelist, struct zone *z)
366 {
367 	if (z->zone_pgdat == zonelist->zones[0]->zone_pgdat) {
368 		__inc_zone_state(z, NUMA_HIT);
369 	} else {
370 		__inc_zone_state(z, NUMA_MISS);
371 		__inc_zone_state(zonelist->zones[0], NUMA_FOREIGN);
372 	}
373 	if (z->node == numa_node_id())
374 		__inc_zone_state(z, NUMA_LOCAL);
375 	else
376 		__inc_zone_state(z, NUMA_OTHER);
377 }
378 #endif
379 
380 #ifdef CONFIG_PROC_FS
381 
382 #include <linux/seq_file.h>
383 
384 static char * const migratetype_names[MIGRATE_TYPES] = {
385 	"Unmovable",
386 	"Reclaimable",
387 	"Movable",
388 	"Reserve",
389 };
390 
391 static void *frag_start(struct seq_file *m, loff_t *pos)
392 {
393 	pg_data_t *pgdat;
394 	loff_t node = *pos;
395 	for (pgdat = first_online_pgdat();
396 	     pgdat && node;
397 	     pgdat = next_online_pgdat(pgdat))
398 		--node;
399 
400 	return pgdat;
401 }
402 
403 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
404 {
405 	pg_data_t *pgdat = (pg_data_t *)arg;
406 
407 	(*pos)++;
408 	return next_online_pgdat(pgdat);
409 }
410 
411 static void frag_stop(struct seq_file *m, void *arg)
412 {
413 }
414 
415 /* Walk all the zones in a node and print using a callback */
416 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
417 		void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
418 {
419 	struct zone *zone;
420 	struct zone *node_zones = pgdat->node_zones;
421 	unsigned long flags;
422 
423 	for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
424 		if (!populated_zone(zone))
425 			continue;
426 
427 		spin_lock_irqsave(&zone->lock, flags);
428 		print(m, pgdat, zone);
429 		spin_unlock_irqrestore(&zone->lock, flags);
430 	}
431 }
432 
433 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
434 						struct zone *zone)
435 {
436 	int order;
437 
438 	seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
439 	for (order = 0; order < MAX_ORDER; ++order)
440 		seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
441 	seq_putc(m, '\n');
442 }
443 
444 /*
445  * This walks the free areas for each zone.
446  */
447 static int frag_show(struct seq_file *m, void *arg)
448 {
449 	pg_data_t *pgdat = (pg_data_t *)arg;
450 	walk_zones_in_node(m, pgdat, frag_show_print);
451 	return 0;
452 }
453 
454 static void pagetypeinfo_showfree_print(struct seq_file *m,
455 					pg_data_t *pgdat, struct zone *zone)
456 {
457 	int order, mtype;
458 
459 	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
460 		seq_printf(m, "Node %4d, zone %8s, type %12s ",
461 					pgdat->node_id,
462 					zone->name,
463 					migratetype_names[mtype]);
464 		for (order = 0; order < MAX_ORDER; ++order) {
465 			unsigned long freecount = 0;
466 			struct free_area *area;
467 			struct list_head *curr;
468 
469 			area = &(zone->free_area[order]);
470 
471 			list_for_each(curr, &area->free_list[mtype])
472 				freecount++;
473 			seq_printf(m, "%6lu ", freecount);
474 		}
475 		seq_putc(m, '\n');
476 	}
477 }
478 
479 /* Print out the free pages at each order for each migatetype */
480 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
481 {
482 	int order;
483 	pg_data_t *pgdat = (pg_data_t *)arg;
484 
485 	/* Print header */
486 	seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
487 	for (order = 0; order < MAX_ORDER; ++order)
488 		seq_printf(m, "%6d ", order);
489 	seq_putc(m, '\n');
490 
491 	walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
492 
493 	return 0;
494 }
495 
496 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
497 					pg_data_t *pgdat, struct zone *zone)
498 {
499 	int mtype;
500 	unsigned long pfn;
501 	unsigned long start_pfn = zone->zone_start_pfn;
502 	unsigned long end_pfn = start_pfn + zone->spanned_pages;
503 	unsigned long count[MIGRATE_TYPES] = { 0, };
504 
505 	for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
506 		struct page *page;
507 
508 		if (!pfn_valid(pfn))
509 			continue;
510 
511 		page = pfn_to_page(pfn);
512 		mtype = get_pageblock_migratetype(page);
513 
514 		count[mtype]++;
515 	}
516 
517 	/* Print counts */
518 	seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
519 	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
520 		seq_printf(m, "%12lu ", count[mtype]);
521 	seq_putc(m, '\n');
522 }
523 
524 /* Print out the free pages at each order for each migratetype */
525 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
526 {
527 	int mtype;
528 	pg_data_t *pgdat = (pg_data_t *)arg;
529 
530 	seq_printf(m, "\n%-23s", "Number of blocks type ");
531 	for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
532 		seq_printf(m, "%12s ", migratetype_names[mtype]);
533 	seq_putc(m, '\n');
534 	walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
535 
536 	return 0;
537 }
538 
539 /*
540  * This prints out statistics in relation to grouping pages by mobility.
541  * It is expensive to collect so do not constantly read the file.
542  */
543 static int pagetypeinfo_show(struct seq_file *m, void *arg)
544 {
545 	pg_data_t *pgdat = (pg_data_t *)arg;
546 
547 	seq_printf(m, "Page block order: %d\n", pageblock_order);
548 	seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
549 	seq_putc(m, '\n');
550 	pagetypeinfo_showfree(m, pgdat);
551 	pagetypeinfo_showblockcount(m, pgdat);
552 
553 	return 0;
554 }
555 
556 const struct seq_operations fragmentation_op = {
557 	.start	= frag_start,
558 	.next	= frag_next,
559 	.stop	= frag_stop,
560 	.show	= frag_show,
561 };
562 
563 const struct seq_operations pagetypeinfo_op = {
564 	.start	= frag_start,
565 	.next	= frag_next,
566 	.stop	= frag_stop,
567 	.show	= pagetypeinfo_show,
568 };
569 
570 #ifdef CONFIG_ZONE_DMA
571 #define TEXT_FOR_DMA(xx) xx "_dma",
572 #else
573 #define TEXT_FOR_DMA(xx)
574 #endif
575 
576 #ifdef CONFIG_ZONE_DMA32
577 #define TEXT_FOR_DMA32(xx) xx "_dma32",
578 #else
579 #define TEXT_FOR_DMA32(xx)
580 #endif
581 
582 #ifdef CONFIG_HIGHMEM
583 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
584 #else
585 #define TEXT_FOR_HIGHMEM(xx)
586 #endif
587 
588 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
589 					TEXT_FOR_HIGHMEM(xx) xx "_movable",
590 
591 static const char * const vmstat_text[] = {
592 	/* Zoned VM counters */
593 	"nr_free_pages",
594 	"nr_inactive",
595 	"nr_active",
596 	"nr_anon_pages",
597 	"nr_mapped",
598 	"nr_file_pages",
599 	"nr_dirty",
600 	"nr_writeback",
601 	"nr_slab_reclaimable",
602 	"nr_slab_unreclaimable",
603 	"nr_page_table_pages",
604 	"nr_unstable",
605 	"nr_bounce",
606 	"nr_vmscan_write",
607 
608 #ifdef CONFIG_NUMA
609 	"numa_hit",
610 	"numa_miss",
611 	"numa_foreign",
612 	"numa_interleave",
613 	"numa_local",
614 	"numa_other",
615 #endif
616 
617 #ifdef CONFIG_VM_EVENT_COUNTERS
618 	"pgpgin",
619 	"pgpgout",
620 	"pswpin",
621 	"pswpout",
622 
623 	TEXTS_FOR_ZONES("pgalloc")
624 
625 	"pgfree",
626 	"pgactivate",
627 	"pgdeactivate",
628 
629 	"pgfault",
630 	"pgmajfault",
631 
632 	TEXTS_FOR_ZONES("pgrefill")
633 	TEXTS_FOR_ZONES("pgsteal")
634 	TEXTS_FOR_ZONES("pgscan_kswapd")
635 	TEXTS_FOR_ZONES("pgscan_direct")
636 
637 	"pginodesteal",
638 	"slabs_scanned",
639 	"kswapd_steal",
640 	"kswapd_inodesteal",
641 	"pageoutrun",
642 	"allocstall",
643 
644 	"pgrotated",
645 #endif
646 };
647 
648 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
649 							struct zone *zone)
650 {
651 	int i;
652 	seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
653 	seq_printf(m,
654 		   "\n  pages free     %lu"
655 		   "\n        min      %lu"
656 		   "\n        low      %lu"
657 		   "\n        high     %lu"
658 		   "\n        scanned  %lu (a: %lu i: %lu)"
659 		   "\n        spanned  %lu"
660 		   "\n        present  %lu",
661 		   zone_page_state(zone, NR_FREE_PAGES),
662 		   zone->pages_min,
663 		   zone->pages_low,
664 		   zone->pages_high,
665 		   zone->pages_scanned,
666 		   zone->nr_scan_active, zone->nr_scan_inactive,
667 		   zone->spanned_pages,
668 		   zone->present_pages);
669 
670 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
671 		seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
672 				zone_page_state(zone, i));
673 
674 	seq_printf(m,
675 		   "\n        protection: (%lu",
676 		   zone->lowmem_reserve[0]);
677 	for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
678 		seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
679 	seq_printf(m,
680 		   ")"
681 		   "\n  pagesets");
682 	for_each_online_cpu(i) {
683 		struct per_cpu_pageset *pageset;
684 		int j;
685 
686 		pageset = zone_pcp(zone, i);
687 		for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
688 			seq_printf(m,
689 				   "\n    cpu: %i pcp: %i"
690 				   "\n              count: %i"
691 				   "\n              high:  %i"
692 				   "\n              batch: %i",
693 				   i, j,
694 				   pageset->pcp[j].count,
695 				   pageset->pcp[j].high,
696 				   pageset->pcp[j].batch);
697 			}
698 #ifdef CONFIG_SMP
699 		seq_printf(m, "\n  vm stats threshold: %d",
700 				pageset->stat_threshold);
701 #endif
702 	}
703 	seq_printf(m,
704 		   "\n  all_unreclaimable: %u"
705 		   "\n  prev_priority:     %i"
706 		   "\n  start_pfn:         %lu",
707 		   zone->all_unreclaimable,
708 		   zone->prev_priority,
709 		   zone->zone_start_pfn);
710 	seq_putc(m, '\n');
711 }
712 
713 /*
714  * Output information about zones in @pgdat.
715  */
716 static int zoneinfo_show(struct seq_file *m, void *arg)
717 {
718 	pg_data_t *pgdat = (pg_data_t *)arg;
719 	walk_zones_in_node(m, pgdat, zoneinfo_show_print);
720 	return 0;
721 }
722 
723 const struct seq_operations zoneinfo_op = {
724 	.start	= frag_start, /* iterate over all zones. The same as in
725 			       * fragmentation. */
726 	.next	= frag_next,
727 	.stop	= frag_stop,
728 	.show	= zoneinfo_show,
729 };
730 
731 static void *vmstat_start(struct seq_file *m, loff_t *pos)
732 {
733 	unsigned long *v;
734 #ifdef CONFIG_VM_EVENT_COUNTERS
735 	unsigned long *e;
736 #endif
737 	int i;
738 
739 	if (*pos >= ARRAY_SIZE(vmstat_text))
740 		return NULL;
741 
742 #ifdef CONFIG_VM_EVENT_COUNTERS
743 	v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
744 			+ sizeof(struct vm_event_state), GFP_KERNEL);
745 #else
746 	v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
747 			GFP_KERNEL);
748 #endif
749 	m->private = v;
750 	if (!v)
751 		return ERR_PTR(-ENOMEM);
752 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
753 		v[i] = global_page_state(i);
754 #ifdef CONFIG_VM_EVENT_COUNTERS
755 	e = v + NR_VM_ZONE_STAT_ITEMS;
756 	all_vm_events(e);
757 	e[PGPGIN] /= 2;		/* sectors -> kbytes */
758 	e[PGPGOUT] /= 2;
759 #endif
760 	return v + *pos;
761 }
762 
763 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
764 {
765 	(*pos)++;
766 	if (*pos >= ARRAY_SIZE(vmstat_text))
767 		return NULL;
768 	return (unsigned long *)m->private + *pos;
769 }
770 
771 static int vmstat_show(struct seq_file *m, void *arg)
772 {
773 	unsigned long *l = arg;
774 	unsigned long off = l - (unsigned long *)m->private;
775 
776 	seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
777 	return 0;
778 }
779 
780 static void vmstat_stop(struct seq_file *m, void *arg)
781 {
782 	kfree(m->private);
783 	m->private = NULL;
784 }
785 
786 const struct seq_operations vmstat_op = {
787 	.start	= vmstat_start,
788 	.next	= vmstat_next,
789 	.stop	= vmstat_stop,
790 	.show	= vmstat_show,
791 };
792 
793 #endif /* CONFIG_PROC_FS */
794 
795 #ifdef CONFIG_SMP
796 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
797 int sysctl_stat_interval __read_mostly = HZ;
798 
799 static void vmstat_update(struct work_struct *w)
800 {
801 	refresh_cpu_vm_stats(smp_processor_id());
802 	schedule_delayed_work(&__get_cpu_var(vmstat_work),
803 		sysctl_stat_interval);
804 }
805 
806 static void __devinit start_cpu_timer(int cpu)
807 {
808 	struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
809 
810 	INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
811 	schedule_delayed_work_on(cpu, vmstat_work, HZ + cpu);
812 }
813 
814 /*
815  * Use the cpu notifier to insure that the thresholds are recalculated
816  * when necessary.
817  */
818 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
819 		unsigned long action,
820 		void *hcpu)
821 {
822 	long cpu = (long)hcpu;
823 
824 	switch (action) {
825 	case CPU_ONLINE:
826 	case CPU_ONLINE_FROZEN:
827 		start_cpu_timer(cpu);
828 		break;
829 	case CPU_DOWN_PREPARE:
830 	case CPU_DOWN_PREPARE_FROZEN:
831 		cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
832 		per_cpu(vmstat_work, cpu).work.func = NULL;
833 		break;
834 	case CPU_DOWN_FAILED:
835 	case CPU_DOWN_FAILED_FROZEN:
836 		start_cpu_timer(cpu);
837 		break;
838 	case CPU_DEAD:
839 	case CPU_DEAD_FROZEN:
840 		refresh_zone_stat_thresholds();
841 		break;
842 	default:
843 		break;
844 	}
845 	return NOTIFY_OK;
846 }
847 
848 static struct notifier_block __cpuinitdata vmstat_notifier =
849 	{ &vmstat_cpuup_callback, NULL, 0 };
850 
851 static int __init setup_vmstat(void)
852 {
853 	int cpu;
854 
855 	refresh_zone_stat_thresholds();
856 	register_cpu_notifier(&vmstat_notifier);
857 
858 	for_each_online_cpu(cpu)
859 		start_cpu_timer(cpu);
860 	return 0;
861 }
862 module_init(setup_vmstat)
863 #endif
864