4 * Manages VM statistics
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Copyright (C) 2006 Silicon Graphics, Inc.,
9 * Christoph Lameter <christoph@lameter.com>
12 #include <linux/config.h>
14 #include <linux/module.h>
17 * Accumulate the page_state information across all CPUs.
18 * The result is unavoidably approximate - it can change
19 * during and after execution of this function.
21 DEFINE_PER_CPU(struct page_state
, page_states
) = {0};
23 static void __get_page_state(struct page_state
*ret
, int nr
, cpumask_t
*cpumask
)
27 memset(ret
, 0, nr
* sizeof(unsigned long));
28 cpus_and(*cpumask
, *cpumask
, cpu_online_map
);
30 for_each_cpu_mask(cpu
, *cpumask
) {
36 in
= (unsigned long *)&per_cpu(page_states
, cpu
);
38 next_cpu
= next_cpu(cpu
, *cpumask
);
39 if (likely(next_cpu
< NR_CPUS
))
40 prefetch(&per_cpu(page_states
, next_cpu
));
42 out
= (unsigned long *)ret
;
43 for (off
= 0; off
< nr
; off
++)
48 void get_page_state_node(struct page_state
*ret
, int node
)
51 cpumask_t mask
= node_to_cpumask(node
);
53 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
54 nr
/= sizeof(unsigned long);
56 __get_page_state(ret
, nr
+1, &mask
);
59 void get_page_state(struct page_state
*ret
)
62 cpumask_t mask
= CPU_MASK_ALL
;
64 nr
= offsetof(struct page_state
, GET_PAGE_STATE_LAST
);
65 nr
/= sizeof(unsigned long);
67 __get_page_state(ret
, nr
+ 1, &mask
);
70 void get_full_page_state(struct page_state
*ret
)
72 cpumask_t mask
= CPU_MASK_ALL
;
74 __get_page_state(ret
, sizeof(*ret
) / sizeof(unsigned long), &mask
);
77 unsigned long read_page_state_offset(unsigned long offset
)
79 unsigned long ret
= 0;
82 for_each_online_cpu(cpu
) {
85 in
= (unsigned long)&per_cpu(page_states
, cpu
) + offset
;
86 ret
+= *((unsigned long *)in
);
91 void __mod_page_state_offset(unsigned long offset
, unsigned long delta
)
95 ptr
= &__get_cpu_var(page_states
);
96 *(unsigned long *)(ptr
+ offset
) += delta
;
98 EXPORT_SYMBOL(__mod_page_state_offset
);
100 void mod_page_state_offset(unsigned long offset
, unsigned long delta
)
105 local_irq_save(flags
);
106 ptr
= &__get_cpu_var(page_states
);
107 *(unsigned long *)(ptr
+ offset
) += delta
;
108 local_irq_restore(flags
);
110 EXPORT_SYMBOL(mod_page_state_offset
);
112 void __get_zone_counts(unsigned long *active
, unsigned long *inactive
,
113 unsigned long *free
, struct pglist_data
*pgdat
)
115 struct zone
*zones
= pgdat
->node_zones
;
121 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
122 *active
+= zones
[i
].nr_active
;
123 *inactive
+= zones
[i
].nr_inactive
;
124 *free
+= zones
[i
].free_pages
;
128 void get_zone_counts(unsigned long *active
,
129 unsigned long *inactive
, unsigned long *free
)
131 struct pglist_data
*pgdat
;
136 for_each_online_pgdat(pgdat
) {
137 unsigned long l
, m
, n
;
138 __get_zone_counts(&l
, &m
, &n
, pgdat
);
146 * Manage combined zone based / global counters
148 * vm_stat contains the global counters
150 atomic_long_t vm_stat
[NR_VM_ZONE_STAT_ITEMS
];
151 EXPORT_SYMBOL(vm_stat
);
155 #define STAT_THRESHOLD 32
158 * Determine pointer to currently valid differential byte given a zone and
161 * Preemption must be off
163 static inline s8
*diff_pointer(struct zone
*zone
, enum zone_stat_item item
)
165 return &zone_pcp(zone
, smp_processor_id())->vm_stat_diff
[item
];
169 * For use when we know that interrupts are disabled.
171 void __mod_zone_page_state(struct zone
*zone
, enum zone_stat_item item
,
177 p
= diff_pointer(zone
, item
);
180 if (unlikely(x
> STAT_THRESHOLD
|| x
< -STAT_THRESHOLD
)) {
181 zone_page_state_add(x
, zone
, item
);
187 EXPORT_SYMBOL(__mod_zone_page_state
);
190 * For an unknown interrupt state
192 void mod_zone_page_state(struct zone
*zone
, enum zone_stat_item item
,
197 local_irq_save(flags
);
198 __mod_zone_page_state(zone
, item
, delta
);
199 local_irq_restore(flags
);
201 EXPORT_SYMBOL(mod_zone_page_state
);
204 * Optimized increment and decrement functions.
206 * These are only for a single page and therefore can take a struct page *
207 * argument instead of struct zone *. This allows the inclusion of the code
208 * generated for page_zone(page) into the optimized functions.
210 * No overflow check is necessary and therefore the differential can be
211 * incremented or decremented in place which may allow the compilers to
212 * generate better code.
214 * The increment or decrement is known and therefore one boundary check can
217 * Some processors have inc/dec instructions that are atomic vs an interrupt.
218 * However, the code must first determine the differential location in a zone
219 * based on the processor number and then inc/dec the counter. There is no
220 * guarantee without disabling preemption that the processor will not change
221 * in between and therefore the atomicity vs. interrupt cannot be exploited
222 * in a useful way here.
224 void __inc_zone_page_state(struct page
*page
, enum zone_stat_item item
)
226 struct zone
*zone
= page_zone(page
);
227 s8
*p
= diff_pointer(zone
, item
);
231 if (unlikely(*p
> STAT_THRESHOLD
)) {
232 zone_page_state_add(*p
, zone
, item
);
236 EXPORT_SYMBOL(__inc_zone_page_state
);
238 void __dec_zone_page_state(struct page
*page
, enum zone_stat_item item
)
240 struct zone
*zone
= page_zone(page
);
241 s8
*p
= diff_pointer(zone
, item
);
245 if (unlikely(*p
< -STAT_THRESHOLD
)) {
246 zone_page_state_add(*p
, zone
, item
);
250 EXPORT_SYMBOL(__dec_zone_page_state
);
252 void inc_zone_page_state(struct page
*page
, enum zone_stat_item item
)
258 zone
= page_zone(page
);
259 local_irq_save(flags
);
260 p
= diff_pointer(zone
, item
);
264 if (unlikely(*p
> STAT_THRESHOLD
)) {
265 zone_page_state_add(*p
, zone
, item
);
268 local_irq_restore(flags
);
270 EXPORT_SYMBOL(inc_zone_page_state
);
272 void dec_zone_page_state(struct page
*page
, enum zone_stat_item item
)
278 zone
= page_zone(page
);
279 local_irq_save(flags
);
280 p
= diff_pointer(zone
, item
);
284 if (unlikely(*p
< -STAT_THRESHOLD
)) {
285 zone_page_state_add(*p
, zone
, item
);
288 local_irq_restore(flags
);
290 EXPORT_SYMBOL(dec_zone_page_state
);
293 * Update the zone counters for one cpu.
295 void refresh_cpu_vm_stats(int cpu
)
301 for_each_zone(zone
) {
302 struct per_cpu_pageset
*pcp
;
304 pcp
= zone_pcp(zone
, cpu
);
306 for (i
= 0; i
< NR_VM_ZONE_STAT_ITEMS
; i
++)
307 if (pcp
->vm_stat_diff
[i
]) {
308 local_irq_save(flags
);
309 zone_page_state_add(pcp
->vm_stat_diff
[i
],
311 pcp
->vm_stat_diff
[i
] = 0;
312 local_irq_restore(flags
);
317 static void __refresh_cpu_vm_stats(void *dummy
)
319 refresh_cpu_vm_stats(smp_processor_id());
323 * Consolidate all counters.
325 * Note that the result is less inaccurate but still inaccurate
326 * if concurrent processes are allowed to run.
328 void refresh_vm_stats(void)
330 on_each_cpu(__refresh_cpu_vm_stats
, NULL
, 0, 1);
332 EXPORT_SYMBOL(refresh_vm_stats
);
336 #ifdef CONFIG_PROC_FS
338 #include <linux/seq_file.h>
340 static void *frag_start(struct seq_file
*m
, loff_t
*pos
)
344 for (pgdat
= first_online_pgdat();
346 pgdat
= next_online_pgdat(pgdat
))
352 static void *frag_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
354 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
357 return next_online_pgdat(pgdat
);
360 static void frag_stop(struct seq_file
*m
, void *arg
)
365 * This walks the free areas for each zone.
367 static int frag_show(struct seq_file
*m
, void *arg
)
369 pg_data_t
*pgdat
= (pg_data_t
*)arg
;
371 struct zone
*node_zones
= pgdat
->node_zones
;
375 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
376 if (!populated_zone(zone
))
379 spin_lock_irqsave(&zone
->lock
, flags
);
380 seq_printf(m
, "Node %d, zone %8s ", pgdat
->node_id
, zone
->name
);
381 for (order
= 0; order
< MAX_ORDER
; ++order
)
382 seq_printf(m
, "%6lu ", zone
->free_area
[order
].nr_free
);
383 spin_unlock_irqrestore(&zone
->lock
, flags
);
389 struct seq_operations fragmentation_op
= {
396 static char *vmstat_text
[] = {
397 /* Zoned VM counters */
402 "nr_page_table_pages",
436 "pgscan_kswapd_high",
437 "pgscan_kswapd_normal",
438 "pgscan_kswapd_dma32",
441 "pgscan_direct_high",
442 "pgscan_direct_normal",
443 "pgscan_direct_dma32",
458 * Output information about zones in @pgdat.
460 static int zoneinfo_show(struct seq_file
*m
, void *arg
)
462 pg_data_t
*pgdat
= arg
;
464 struct zone
*node_zones
= pgdat
->node_zones
;
467 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; zone
++) {
470 if (!populated_zone(zone
))
473 spin_lock_irqsave(&zone
->lock
, flags
);
474 seq_printf(m
, "Node %d, zone %8s", pgdat
->node_id
, zone
->name
);
482 "\n scanned %lu (a: %lu i: %lu)"
492 zone
->nr_scan_active
, zone
->nr_scan_inactive
,
494 zone
->present_pages
);
496 for (i
= 0; i
< NR_VM_ZONE_STAT_ITEMS
; i
++)
497 seq_printf(m
, "\n %-12s %lu", vmstat_text
[i
],
498 zone_page_state(zone
, i
));
501 "\n protection: (%lu",
502 zone
->lowmem_reserve
[0]);
503 for (i
= 1; i
< ARRAY_SIZE(zone
->lowmem_reserve
); i
++)
504 seq_printf(m
, ", %lu", zone
->lowmem_reserve
[i
]);
508 for_each_online_cpu(i
) {
509 struct per_cpu_pageset
*pageset
;
512 pageset
= zone_pcp(zone
, i
);
513 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
514 if (pageset
->pcp
[j
].count
)
517 if (j
== ARRAY_SIZE(pageset
->pcp
))
519 for (j
= 0; j
< ARRAY_SIZE(pageset
->pcp
); j
++) {
526 pageset
->pcp
[j
].count
,
527 pageset
->pcp
[j
].high
,
528 pageset
->pcp
[j
].batch
);
534 "\n numa_foreign: %lu"
535 "\n interleave_hit: %lu"
537 "\n other_node: %lu",
540 pageset
->numa_foreign
,
541 pageset
->interleave_hit
,
543 pageset
->other_node
);
547 "\n all_unreclaimable: %u"
548 "\n prev_priority: %i"
549 "\n temp_priority: %i"
551 zone
->all_unreclaimable
,
554 zone
->zone_start_pfn
);
555 spin_unlock_irqrestore(&zone
->lock
, flags
);
561 struct seq_operations zoneinfo_op
= {
562 .start
= frag_start
, /* iterate over all zones. The same as in
566 .show
= zoneinfo_show
,
569 static void *vmstat_start(struct seq_file
*m
, loff_t
*pos
)
572 struct page_state
*ps
;
575 if (*pos
>= ARRAY_SIZE(vmstat_text
))
578 v
= kmalloc(NR_VM_ZONE_STAT_ITEMS
* sizeof(unsigned long)
579 + sizeof(*ps
), GFP_KERNEL
);
582 return ERR_PTR(-ENOMEM
);
583 for (i
= 0; i
< NR_VM_ZONE_STAT_ITEMS
; i
++)
584 v
[i
] = global_page_state(i
);
585 ps
= (struct page_state
*)(v
+ NR_VM_ZONE_STAT_ITEMS
);
586 get_full_page_state(ps
);
587 ps
->pgpgin
/= 2; /* sectors -> kbytes */
592 static void *vmstat_next(struct seq_file
*m
, void *arg
, loff_t
*pos
)
595 if (*pos
>= ARRAY_SIZE(vmstat_text
))
597 return (unsigned long *)m
->private + *pos
;
600 static int vmstat_show(struct seq_file
*m
, void *arg
)
602 unsigned long *l
= arg
;
603 unsigned long off
= l
- (unsigned long *)m
->private;
605 seq_printf(m
, "%s %lu\n", vmstat_text
[off
], *l
);
609 static void vmstat_stop(struct seq_file
*m
, void *arg
)
615 struct seq_operations vmstat_op
= {
616 .start
= vmstat_start
,
622 #endif /* CONFIG_PROC_FS */