page allocator: use allocation flags as an index to the zone watermark
[linux-2.6.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  */
11 #include <linux/fs.h>
12 #include <linux/mm.h>
13 #include <linux/err.h>
14 #include <linux/module.h>
15 #include <linux/cpu.h>
16 #include <linux/vmstat.h>
17 #include <linux/sched.h>
18
19 #ifdef CONFIG_VM_EVENT_COUNTERS
20 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
21 EXPORT_PER_CPU_SYMBOL(vm_event_states);
22
23 static void sum_vm_events(unsigned long *ret, const struct cpumask *cpumask)
24 {
25         int cpu;
26         int i;
27
28         memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
29
30         for_each_cpu(cpu, cpumask) {
31                 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
32
33                 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
34                         ret[i] += this->event[i];
35         }
36 }
37
38 /*
39  * Accumulate the vm event counters across all CPUs.
40  * The result is unavoidably approximate - it can change
41  * during and after execution of this function.
42 */
43 void all_vm_events(unsigned long *ret)
44 {
45         get_online_cpus();
46         sum_vm_events(ret, cpu_online_mask);
47         put_online_cpus();
48 }
49 EXPORT_SYMBOL_GPL(all_vm_events);
50
51 #ifdef CONFIG_HOTPLUG
52 /*
53  * Fold the foreign cpu events into our own.
54  *
55  * This is adding to the events on one processor
56  * but keeps the global counts constant.
57  */
58 void vm_events_fold_cpu(int cpu)
59 {
60         struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
61         int i;
62
63         for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
64                 count_vm_events(i, fold_state->event[i]);
65                 fold_state->event[i] = 0;
66         }
67 }
68 #endif /* CONFIG_HOTPLUG */
69
70 #endif /* CONFIG_VM_EVENT_COUNTERS */
71
72 /*
73  * Manage combined zone based / global counters
74  *
75  * vm_stat contains the global counters
76  */
77 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
78 EXPORT_SYMBOL(vm_stat);
79
80 #ifdef CONFIG_SMP
81
82 static int calculate_threshold(struct zone *zone)
83 {
84         int threshold;
85         int mem;        /* memory in 128 MB units */
86
87         /*
88          * The threshold scales with the number of processors and the amount
89          * of memory per zone. More memory means that we can defer updates for
90          * longer, more processors could lead to more contention.
91          * fls() is used to have a cheap way of logarithmic scaling.
92          *
93          * Some sample thresholds:
94          *
95          * Threshold    Processors      (fls)   Zonesize        fls(mem+1)
96          * ------------------------------------------------------------------
97          * 8            1               1       0.9-1 GB        4
98          * 16           2               2       0.9-1 GB        4
99          * 20           2               2       1-2 GB          5
100          * 24           2               2       2-4 GB          6
101          * 28           2               2       4-8 GB          7
102          * 32           2               2       8-16 GB         8
103          * 4            2               2       <128M           1
104          * 30           4               3       2-4 GB          5
105          * 48           4               3       8-16 GB         8
106          * 32           8               4       1-2 GB          4
107          * 32           8               4       0.9-1GB         4
108          * 10           16              5       <128M           1
109          * 40           16              5       900M            4
110          * 70           64              7       2-4 GB          5
111          * 84           64              7       4-8 GB          6
112          * 108          512             9       4-8 GB          6
113          * 125          1024            10      8-16 GB         8
114          * 125          1024            10      16-32 GB        9
115          */
116
117         mem = zone->present_pages >> (27 - PAGE_SHIFT);
118
119         threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
120
121         /*
122          * Maximum threshold is 125
123          */
124         threshold = min(125, threshold);
125
126         return threshold;
127 }
128
129 /*
130  * Refresh the thresholds for each zone.
131  */
132 static void refresh_zone_stat_thresholds(void)
133 {
134         struct zone *zone;
135         int cpu;
136         int threshold;
137
138         for_each_populated_zone(zone) {
139                 threshold = calculate_threshold(zone);
140
141                 for_each_online_cpu(cpu)
142                         zone_pcp(zone, cpu)->stat_threshold = threshold;
143         }
144 }
145
146 /*
147  * For use when we know that interrupts are disabled.
148  */
149 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
150                                 int delta)
151 {
152         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
153         s8 *p = pcp->vm_stat_diff + item;
154         long x;
155
156         x = delta + *p;
157
158         if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
159                 zone_page_state_add(x, zone, item);
160                 x = 0;
161         }
162         *p = x;
163 }
164 EXPORT_SYMBOL(__mod_zone_page_state);
165
166 /*
167  * For an unknown interrupt state
168  */
169 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
170                                         int delta)
171 {
172         unsigned long flags;
173
174         local_irq_save(flags);
175         __mod_zone_page_state(zone, item, delta);
176         local_irq_restore(flags);
177 }
178 EXPORT_SYMBOL(mod_zone_page_state);
179
180 /*
181  * Optimized increment and decrement functions.
182  *
183  * These are only for a single page and therefore can take a struct page *
184  * argument instead of struct zone *. This allows the inclusion of the code
185  * generated for page_zone(page) into the optimized functions.
186  *
187  * No overflow check is necessary and therefore the differential can be
188  * incremented or decremented in place which may allow the compilers to
189  * generate better code.
190  * The increment or decrement is known and therefore one boundary check can
191  * be omitted.
192  *
193  * NOTE: These functions are very performance sensitive. Change only
194  * with care.
195  *
196  * Some processors have inc/dec instructions that are atomic vs an interrupt.
197  * However, the code must first determine the differential location in a zone
198  * based on the processor number and then inc/dec the counter. There is no
199  * guarantee without disabling preemption that the processor will not change
200  * in between and therefore the atomicity vs. interrupt cannot be exploited
201  * in a useful way here.
202  */
203 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
204 {
205         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
206         s8 *p = pcp->vm_stat_diff + item;
207
208         (*p)++;
209
210         if (unlikely(*p > pcp->stat_threshold)) {
211                 int overstep = pcp->stat_threshold / 2;
212
213                 zone_page_state_add(*p + overstep, zone, item);
214                 *p = -overstep;
215         }
216 }
217
218 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
219 {
220         __inc_zone_state(page_zone(page), item);
221 }
222 EXPORT_SYMBOL(__inc_zone_page_state);
223
224 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
225 {
226         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
227         s8 *p = pcp->vm_stat_diff + item;
228
229         (*p)--;
230
231         if (unlikely(*p < - pcp->stat_threshold)) {
232                 int overstep = pcp->stat_threshold / 2;
233
234                 zone_page_state_add(*p - overstep, zone, item);
235                 *p = overstep;
236         }
237 }
238
239 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
240 {
241         __dec_zone_state(page_zone(page), item);
242 }
243 EXPORT_SYMBOL(__dec_zone_page_state);
244
245 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
246 {
247         unsigned long flags;
248
249         local_irq_save(flags);
250         __inc_zone_state(zone, item);
251         local_irq_restore(flags);
252 }
253
254 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
255 {
256         unsigned long flags;
257         struct zone *zone;
258
259         zone = page_zone(page);
260         local_irq_save(flags);
261         __inc_zone_state(zone, item);
262         local_irq_restore(flags);
263 }
264 EXPORT_SYMBOL(inc_zone_page_state);
265
266 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
267 {
268         unsigned long flags;
269
270         local_irq_save(flags);
271         __dec_zone_page_state(page, item);
272         local_irq_restore(flags);
273 }
274 EXPORT_SYMBOL(dec_zone_page_state);
275
276 /*
277  * Update the zone counters for one cpu.
278  *
279  * The cpu specified must be either the current cpu or a processor that
280  * is not online. If it is the current cpu then the execution thread must
281  * be pinned to the current cpu.
282  *
283  * Note that refresh_cpu_vm_stats strives to only access
284  * node local memory. The per cpu pagesets on remote zones are placed
285  * in the memory local to the processor using that pageset. So the
286  * loop over all zones will access a series of cachelines local to
287  * the processor.
288  *
289  * The call to zone_page_state_add updates the cachelines with the
290  * statistics in the remote zone struct as well as the global cachelines
291  * with the global counters. These could cause remote node cache line
292  * bouncing and will have to be only done when necessary.
293  */
294 void refresh_cpu_vm_stats(int cpu)
295 {
296         struct zone *zone;
297         int i;
298         int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
299
300         for_each_populated_zone(zone) {
301                 struct per_cpu_pageset *p;
302
303                 p = zone_pcp(zone, cpu);
304
305                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
306                         if (p->vm_stat_diff[i]) {
307                                 unsigned long flags;
308                                 int v;
309
310                                 local_irq_save(flags);
311                                 v = p->vm_stat_diff[i];
312                                 p->vm_stat_diff[i] = 0;
313                                 local_irq_restore(flags);
314                                 atomic_long_add(v, &zone->vm_stat[i]);
315                                 global_diff[i] += v;
316 #ifdef CONFIG_NUMA
317                                 /* 3 seconds idle till flush */
318                                 p->expire = 3;
319 #endif
320                         }
321                 cond_resched();
322 #ifdef CONFIG_NUMA
323                 /*
324                  * Deal with draining the remote pageset of this
325                  * processor
326                  *
327                  * Check if there are pages remaining in this pageset
328                  * if not then there is nothing to expire.
329                  */
330                 if (!p->expire || !p->pcp.count)
331                         continue;
332
333                 /*
334                  * We never drain zones local to this processor.
335                  */
336                 if (zone_to_nid(zone) == numa_node_id()) {
337                         p->expire = 0;
338                         continue;
339                 }
340
341                 p->expire--;
342                 if (p->expire)
343                         continue;
344
345                 if (p->pcp.count)
346                         drain_zone_pages(zone, &p->pcp);
347 #endif
348         }
349
350         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
351                 if (global_diff[i])
352                         atomic_long_add(global_diff[i], &vm_stat[i]);
353 }
354
355 #endif
356
357 #ifdef CONFIG_NUMA
358 /*
359  * zonelist = the list of zones passed to the allocator
360  * z        = the zone from which the allocation occurred.
361  *
362  * Must be called with interrupts disabled.
363  */
364 void zone_statistics(struct zone *preferred_zone, struct zone *z)
365 {
366         if (z->zone_pgdat == preferred_zone->zone_pgdat) {
367                 __inc_zone_state(z, NUMA_HIT);
368         } else {
369                 __inc_zone_state(z, NUMA_MISS);
370                 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
371         }
372         if (z->node == numa_node_id())
373                 __inc_zone_state(z, NUMA_LOCAL);
374         else
375                 __inc_zone_state(z, NUMA_OTHER);
376 }
377 #endif
378
379 #ifdef CONFIG_PROC_FS
380 #include <linux/proc_fs.h>
381 #include <linux/seq_file.h>
382
383 static char * const migratetype_names[MIGRATE_TYPES] = {
384         "Unmovable",
385         "Reclaimable",
386         "Movable",
387         "Reserve",
388         "Isolate",
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
513                 /* Watch for unexpected holes punched in the memmap */
514                 if (!memmap_valid_within(pfn, page, zone))
515                         continue;
516
517                 mtype = get_pageblock_migratetype(page);
518
519                 if (mtype < MIGRATE_TYPES)
520                         count[mtype]++;
521         }
522
523         /* Print counts */
524         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
525         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
526                 seq_printf(m, "%12lu ", count[mtype]);
527         seq_putc(m, '\n');
528 }
529
530 /* Print out the free pages at each order for each migratetype */
531 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
532 {
533         int mtype;
534         pg_data_t *pgdat = (pg_data_t *)arg;
535
536         seq_printf(m, "\n%-23s", "Number of blocks type ");
537         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
538                 seq_printf(m, "%12s ", migratetype_names[mtype]);
539         seq_putc(m, '\n');
540         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
541
542         return 0;
543 }
544
545 /*
546  * This prints out statistics in relation to grouping pages by mobility.
547  * It is expensive to collect so do not constantly read the file.
548  */
549 static int pagetypeinfo_show(struct seq_file *m, void *arg)
550 {
551         pg_data_t *pgdat = (pg_data_t *)arg;
552
553         /* check memoryless node */
554         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
555                 return 0;
556
557         seq_printf(m, "Page block order: %d\n", pageblock_order);
558         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
559         seq_putc(m, '\n');
560         pagetypeinfo_showfree(m, pgdat);
561         pagetypeinfo_showblockcount(m, pgdat);
562
563         return 0;
564 }
565
566 static const struct seq_operations fragmentation_op = {
567         .start  = frag_start,
568         .next   = frag_next,
569         .stop   = frag_stop,
570         .show   = frag_show,
571 };
572
573 static int fragmentation_open(struct inode *inode, struct file *file)
574 {
575         return seq_open(file, &fragmentation_op);
576 }
577
578 static const struct file_operations fragmentation_file_operations = {
579         .open           = fragmentation_open,
580         .read           = seq_read,
581         .llseek         = seq_lseek,
582         .release        = seq_release,
583 };
584
585 static const struct seq_operations pagetypeinfo_op = {
586         .start  = frag_start,
587         .next   = frag_next,
588         .stop   = frag_stop,
589         .show   = pagetypeinfo_show,
590 };
591
592 static int pagetypeinfo_open(struct inode *inode, struct file *file)
593 {
594         return seq_open(file, &pagetypeinfo_op);
595 }
596
597 static const struct file_operations pagetypeinfo_file_ops = {
598         .open           = pagetypeinfo_open,
599         .read           = seq_read,
600         .llseek         = seq_lseek,
601         .release        = seq_release,
602 };
603
604 #ifdef CONFIG_ZONE_DMA
605 #define TEXT_FOR_DMA(xx) xx "_dma",
606 #else
607 #define TEXT_FOR_DMA(xx)
608 #endif
609
610 #ifdef CONFIG_ZONE_DMA32
611 #define TEXT_FOR_DMA32(xx) xx "_dma32",
612 #else
613 #define TEXT_FOR_DMA32(xx)
614 #endif
615
616 #ifdef CONFIG_HIGHMEM
617 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
618 #else
619 #define TEXT_FOR_HIGHMEM(xx)
620 #endif
621
622 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
623                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
624
625 static const char * const vmstat_text[] = {
626         /* Zoned VM counters */
627         "nr_free_pages",
628         "nr_inactive_anon",
629         "nr_active_anon",
630         "nr_inactive_file",
631         "nr_active_file",
632 #ifdef CONFIG_UNEVICTABLE_LRU
633         "nr_unevictable",
634         "nr_mlock",
635 #endif
636         "nr_anon_pages",
637         "nr_mapped",
638         "nr_file_pages",
639         "nr_dirty",
640         "nr_writeback",
641         "nr_slab_reclaimable",
642         "nr_slab_unreclaimable",
643         "nr_page_table_pages",
644         "nr_unstable",
645         "nr_bounce",
646         "nr_vmscan_write",
647         "nr_writeback_temp",
648
649 #ifdef CONFIG_NUMA
650         "numa_hit",
651         "numa_miss",
652         "numa_foreign",
653         "numa_interleave",
654         "numa_local",
655         "numa_other",
656 #endif
657
658 #ifdef CONFIG_VM_EVENT_COUNTERS
659         "pgpgin",
660         "pgpgout",
661         "pswpin",
662         "pswpout",
663
664         TEXTS_FOR_ZONES("pgalloc")
665
666         "pgfree",
667         "pgactivate",
668         "pgdeactivate",
669
670         "pgfault",
671         "pgmajfault",
672
673         TEXTS_FOR_ZONES("pgrefill")
674         TEXTS_FOR_ZONES("pgsteal")
675         TEXTS_FOR_ZONES("pgscan_kswapd")
676         TEXTS_FOR_ZONES("pgscan_direct")
677
678         "pginodesteal",
679         "slabs_scanned",
680         "kswapd_steal",
681         "kswapd_inodesteal",
682         "pageoutrun",
683         "allocstall",
684
685         "pgrotated",
686 #ifdef CONFIG_HUGETLB_PAGE
687         "htlb_buddy_alloc_success",
688         "htlb_buddy_alloc_fail",
689 #endif
690 #ifdef CONFIG_UNEVICTABLE_LRU
691         "unevictable_pgs_culled",
692         "unevictable_pgs_scanned",
693         "unevictable_pgs_rescued",
694         "unevictable_pgs_mlocked",
695         "unevictable_pgs_munlocked",
696         "unevictable_pgs_cleared",
697         "unevictable_pgs_stranded",
698         "unevictable_pgs_mlockfreed",
699 #endif
700 #endif
701 };
702
703 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
704                                                         struct zone *zone)
705 {
706         int i;
707         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
708         seq_printf(m,
709                    "\n  pages free     %lu"
710                    "\n        min      %lu"
711                    "\n        low      %lu"
712                    "\n        high     %lu"
713                    "\n        scanned  %lu (aa: %lu ia: %lu af: %lu if: %lu)"
714                    "\n        spanned  %lu"
715                    "\n        present  %lu",
716                    zone_page_state(zone, NR_FREE_PAGES),
717                    min_wmark_pages(zone),
718                    low_wmark_pages(zone),
719                    high_wmark_pages(zone),
720                    zone->pages_scanned,
721                    zone->lru[LRU_ACTIVE_ANON].nr_scan,
722                    zone->lru[LRU_INACTIVE_ANON].nr_scan,
723                    zone->lru[LRU_ACTIVE_FILE].nr_scan,
724                    zone->lru[LRU_INACTIVE_FILE].nr_scan,
725                    zone->spanned_pages,
726                    zone->present_pages);
727
728         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
729                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
730                                 zone_page_state(zone, i));
731
732         seq_printf(m,
733                    "\n        protection: (%lu",
734                    zone->lowmem_reserve[0]);
735         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
736                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
737         seq_printf(m,
738                    ")"
739                    "\n  pagesets");
740         for_each_online_cpu(i) {
741                 struct per_cpu_pageset *pageset;
742
743                 pageset = zone_pcp(zone, i);
744                 seq_printf(m,
745                            "\n    cpu: %i"
746                            "\n              count: %i"
747                            "\n              high:  %i"
748                            "\n              batch: %i",
749                            i,
750                            pageset->pcp.count,
751                            pageset->pcp.high,
752                            pageset->pcp.batch);
753 #ifdef CONFIG_SMP
754                 seq_printf(m, "\n  vm stats threshold: %d",
755                                 pageset->stat_threshold);
756 #endif
757         }
758         seq_printf(m,
759                    "\n  all_unreclaimable: %u"
760                    "\n  prev_priority:     %i"
761                    "\n  start_pfn:         %lu"
762                    "\n  inactive_ratio:    %u",
763                            zone_is_all_unreclaimable(zone),
764                    zone->prev_priority,
765                    zone->zone_start_pfn,
766                    zone->inactive_ratio);
767         seq_putc(m, '\n');
768 }
769
770 /*
771  * Output information about zones in @pgdat.
772  */
773 static int zoneinfo_show(struct seq_file *m, void *arg)
774 {
775         pg_data_t *pgdat = (pg_data_t *)arg;
776         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
777         return 0;
778 }
779
780 static const struct seq_operations zoneinfo_op = {
781         .start  = frag_start, /* iterate over all zones. The same as in
782                                * fragmentation. */
783         .next   = frag_next,
784         .stop   = frag_stop,
785         .show   = zoneinfo_show,
786 };
787
788 static int zoneinfo_open(struct inode *inode, struct file *file)
789 {
790         return seq_open(file, &zoneinfo_op);
791 }
792
793 static const struct file_operations proc_zoneinfo_file_operations = {
794         .open           = zoneinfo_open,
795         .read           = seq_read,
796         .llseek         = seq_lseek,
797         .release        = seq_release,
798 };
799
800 static void *vmstat_start(struct seq_file *m, loff_t *pos)
801 {
802         unsigned long *v;
803 #ifdef CONFIG_VM_EVENT_COUNTERS
804         unsigned long *e;
805 #endif
806         int i;
807
808         if (*pos >= ARRAY_SIZE(vmstat_text))
809                 return NULL;
810
811 #ifdef CONFIG_VM_EVENT_COUNTERS
812         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
813                         + sizeof(struct vm_event_state), GFP_KERNEL);
814 #else
815         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
816                         GFP_KERNEL);
817 #endif
818         m->private = v;
819         if (!v)
820                 return ERR_PTR(-ENOMEM);
821         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
822                 v[i] = global_page_state(i);
823 #ifdef CONFIG_VM_EVENT_COUNTERS
824         e = v + NR_VM_ZONE_STAT_ITEMS;
825         all_vm_events(e);
826         e[PGPGIN] /= 2;         /* sectors -> kbytes */
827         e[PGPGOUT] /= 2;
828 #endif
829         return v + *pos;
830 }
831
832 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
833 {
834         (*pos)++;
835         if (*pos >= ARRAY_SIZE(vmstat_text))
836                 return NULL;
837         return (unsigned long *)m->private + *pos;
838 }
839
840 static int vmstat_show(struct seq_file *m, void *arg)
841 {
842         unsigned long *l = arg;
843         unsigned long off = l - (unsigned long *)m->private;
844
845         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
846         return 0;
847 }
848
849 static void vmstat_stop(struct seq_file *m, void *arg)
850 {
851         kfree(m->private);
852         m->private = NULL;
853 }
854
855 static const struct seq_operations vmstat_op = {
856         .start  = vmstat_start,
857         .next   = vmstat_next,
858         .stop   = vmstat_stop,
859         .show   = vmstat_show,
860 };
861
862 static int vmstat_open(struct inode *inode, struct file *file)
863 {
864         return seq_open(file, &vmstat_op);
865 }
866
867 static const struct file_operations proc_vmstat_file_operations = {
868         .open           = vmstat_open,
869         .read           = seq_read,
870         .llseek         = seq_lseek,
871         .release        = seq_release,
872 };
873 #endif /* CONFIG_PROC_FS */
874
875 #ifdef CONFIG_SMP
876 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
877 int sysctl_stat_interval __read_mostly = HZ;
878
879 static void vmstat_update(struct work_struct *w)
880 {
881         refresh_cpu_vm_stats(smp_processor_id());
882         schedule_delayed_work(&__get_cpu_var(vmstat_work),
883                 round_jiffies_relative(sysctl_stat_interval));
884 }
885
886 static void __cpuinit start_cpu_timer(int cpu)
887 {
888         struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
889
890         INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
891         schedule_delayed_work_on(cpu, vmstat_work,
892                                  __round_jiffies_relative(HZ, cpu));
893 }
894
895 /*
896  * Use the cpu notifier to insure that the thresholds are recalculated
897  * when necessary.
898  */
899 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
900                 unsigned long action,
901                 void *hcpu)
902 {
903         long cpu = (long)hcpu;
904
905         switch (action) {
906         case CPU_ONLINE:
907         case CPU_ONLINE_FROZEN:
908                 start_cpu_timer(cpu);
909                 break;
910         case CPU_DOWN_PREPARE:
911         case CPU_DOWN_PREPARE_FROZEN:
912                 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
913                 per_cpu(vmstat_work, cpu).work.func = NULL;
914                 break;
915         case CPU_DOWN_FAILED:
916         case CPU_DOWN_FAILED_FROZEN:
917                 start_cpu_timer(cpu);
918                 break;
919         case CPU_DEAD:
920         case CPU_DEAD_FROZEN:
921                 refresh_zone_stat_thresholds();
922                 break;
923         default:
924                 break;
925         }
926         return NOTIFY_OK;
927 }
928
929 static struct notifier_block __cpuinitdata vmstat_notifier =
930         { &vmstat_cpuup_callback, NULL, 0 };
931 #endif
932
933 static int __init setup_vmstat(void)
934 {
935 #ifdef CONFIG_SMP
936         int cpu;
937
938         refresh_zone_stat_thresholds();
939         register_cpu_notifier(&vmstat_notifier);
940
941         for_each_online_cpu(cpu)
942                 start_cpu_timer(cpu);
943 #endif
944 #ifdef CONFIG_PROC_FS
945         proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
946         proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
947         proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
948         proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
949 #endif
950         return 0;
951 }
952 module_init(setup_vmstat)