KVM: Inject asynchronous page fault into a PV guest if page is swapped out.
[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/slab.h>
16 #include <linux/cpu.h>
17 #include <linux/vmstat.h>
18 #include <linux/sched.h>
19 #include <linux/math64.h>
20 #include <linux/writeback.h>
21 #include <linux/compaction.h>
22
23 #ifdef CONFIG_VM_EVENT_COUNTERS
24 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
25 EXPORT_PER_CPU_SYMBOL(vm_event_states);
26
27 static void sum_vm_events(unsigned long *ret)
28 {
29         int cpu;
30         int i;
31
32         memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
33
34         for_each_online_cpu(cpu) {
35                 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
36
37                 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
38                         ret[i] += this->event[i];
39         }
40 }
41
42 /*
43  * Accumulate the vm event counters across all CPUs.
44  * The result is unavoidably approximate - it can change
45  * during and after execution of this function.
46 */
47 void all_vm_events(unsigned long *ret)
48 {
49         get_online_cpus();
50         sum_vm_events(ret);
51         put_online_cpus();
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_populated_zone(zone) {
143                 unsigned long max_drift, tolerate_drift;
144
145                 threshold = calculate_threshold(zone);
146
147                 for_each_online_cpu(cpu)
148                         per_cpu_ptr(zone->pageset, cpu)->stat_threshold
149                                                         = threshold;
150
151                 /*
152                  * Only set percpu_drift_mark if there is a danger that
153                  * NR_FREE_PAGES reports the low watermark is ok when in fact
154                  * the min watermark could be breached by an allocation
155                  */
156                 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
157                 max_drift = num_online_cpus() * threshold;
158                 if (max_drift > tolerate_drift)
159                         zone->percpu_drift_mark = high_wmark_pages(zone) +
160                                         max_drift;
161         }
162 }
163
164 /*
165  * For use when we know that interrupts are disabled.
166  */
167 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
168                                 int delta)
169 {
170         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
171
172         s8 *p = pcp->vm_stat_diff + item;
173         long x;
174
175         x = delta + *p;
176
177         if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
178                 zone_page_state_add(x, zone, item);
179                 x = 0;
180         }
181         *p = x;
182 }
183 EXPORT_SYMBOL(__mod_zone_page_state);
184
185 /*
186  * For an unknown interrupt state
187  */
188 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
189                                         int delta)
190 {
191         unsigned long flags;
192
193         local_irq_save(flags);
194         __mod_zone_page_state(zone, item, delta);
195         local_irq_restore(flags);
196 }
197 EXPORT_SYMBOL(mod_zone_page_state);
198
199 /*
200  * Optimized increment and decrement functions.
201  *
202  * These are only for a single page and therefore can take a struct page *
203  * argument instead of struct zone *. This allows the inclusion of the code
204  * generated for page_zone(page) into the optimized functions.
205  *
206  * No overflow check is necessary and therefore the differential can be
207  * incremented or decremented in place which may allow the compilers to
208  * generate better code.
209  * The increment or decrement is known and therefore one boundary check can
210  * be omitted.
211  *
212  * NOTE: These functions are very performance sensitive. Change only
213  * with care.
214  *
215  * Some processors have inc/dec instructions that are atomic vs an interrupt.
216  * However, the code must first determine the differential location in a zone
217  * based on the processor number and then inc/dec the counter. There is no
218  * guarantee without disabling preemption that the processor will not change
219  * in between and therefore the atomicity vs. interrupt cannot be exploited
220  * in a useful way here.
221  */
222 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
223 {
224         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
225         s8 *p = pcp->vm_stat_diff + item;
226
227         (*p)++;
228
229         if (unlikely(*p > pcp->stat_threshold)) {
230                 int overstep = pcp->stat_threshold / 2;
231
232                 zone_page_state_add(*p + overstep, zone, item);
233                 *p = -overstep;
234         }
235 }
236
237 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
238 {
239         __inc_zone_state(page_zone(page), item);
240 }
241 EXPORT_SYMBOL(__inc_zone_page_state);
242
243 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
244 {
245         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
246         s8 *p = pcp->vm_stat_diff + item;
247
248         (*p)--;
249
250         if (unlikely(*p < - pcp->stat_threshold)) {
251                 int overstep = pcp->stat_threshold / 2;
252
253                 zone_page_state_add(*p - overstep, zone, item);
254                 *p = overstep;
255         }
256 }
257
258 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
259 {
260         __dec_zone_state(page_zone(page), item);
261 }
262 EXPORT_SYMBOL(__dec_zone_page_state);
263
264 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
265 {
266         unsigned long flags;
267
268         local_irq_save(flags);
269         __inc_zone_state(zone, item);
270         local_irq_restore(flags);
271 }
272
273 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
274 {
275         unsigned long flags;
276         struct zone *zone;
277
278         zone = page_zone(page);
279         local_irq_save(flags);
280         __inc_zone_state(zone, item);
281         local_irq_restore(flags);
282 }
283 EXPORT_SYMBOL(inc_zone_page_state);
284
285 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
286 {
287         unsigned long flags;
288
289         local_irq_save(flags);
290         __dec_zone_page_state(page, item);
291         local_irq_restore(flags);
292 }
293 EXPORT_SYMBOL(dec_zone_page_state);
294
295 /*
296  * Update the zone counters for one cpu.
297  *
298  * The cpu specified must be either the current cpu or a processor that
299  * is not online. If it is the current cpu then the execution thread must
300  * be pinned to the current cpu.
301  *
302  * Note that refresh_cpu_vm_stats strives to only access
303  * node local memory. The per cpu pagesets on remote zones are placed
304  * in the memory local to the processor using that pageset. So the
305  * loop over all zones will access a series of cachelines local to
306  * the processor.
307  *
308  * The call to zone_page_state_add updates the cachelines with the
309  * statistics in the remote zone struct as well as the global cachelines
310  * with the global counters. These could cause remote node cache line
311  * bouncing and will have to be only done when necessary.
312  */
313 void refresh_cpu_vm_stats(int cpu)
314 {
315         struct zone *zone;
316         int i;
317         int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
318
319         for_each_populated_zone(zone) {
320                 struct per_cpu_pageset *p;
321
322                 p = per_cpu_ptr(zone->pageset, cpu);
323
324                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
325                         if (p->vm_stat_diff[i]) {
326                                 unsigned long flags;
327                                 int v;
328
329                                 local_irq_save(flags);
330                                 v = p->vm_stat_diff[i];
331                                 p->vm_stat_diff[i] = 0;
332                                 local_irq_restore(flags);
333                                 atomic_long_add(v, &zone->vm_stat[i]);
334                                 global_diff[i] += v;
335 #ifdef CONFIG_NUMA
336                                 /* 3 seconds idle till flush */
337                                 p->expire = 3;
338 #endif
339                         }
340                 cond_resched();
341 #ifdef CONFIG_NUMA
342                 /*
343                  * Deal with draining the remote pageset of this
344                  * processor
345                  *
346                  * Check if there are pages remaining in this pageset
347                  * if not then there is nothing to expire.
348                  */
349                 if (!p->expire || !p->pcp.count)
350                         continue;
351
352                 /*
353                  * We never drain zones local to this processor.
354                  */
355                 if (zone_to_nid(zone) == numa_node_id()) {
356                         p->expire = 0;
357                         continue;
358                 }
359
360                 p->expire--;
361                 if (p->expire)
362                         continue;
363
364                 if (p->pcp.count)
365                         drain_zone_pages(zone, &p->pcp);
366 #endif
367         }
368
369         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
370                 if (global_diff[i])
371                         atomic_long_add(global_diff[i], &vm_stat[i]);
372 }
373
374 #endif
375
376 #ifdef CONFIG_NUMA
377 /*
378  * zonelist = the list of zones passed to the allocator
379  * z        = the zone from which the allocation occurred.
380  *
381  * Must be called with interrupts disabled.
382  */
383 void zone_statistics(struct zone *preferred_zone, struct zone *z)
384 {
385         if (z->zone_pgdat == preferred_zone->zone_pgdat) {
386                 __inc_zone_state(z, NUMA_HIT);
387         } else {
388                 __inc_zone_state(z, NUMA_MISS);
389                 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
390         }
391         if (z->node == numa_node_id())
392                 __inc_zone_state(z, NUMA_LOCAL);
393         else
394                 __inc_zone_state(z, NUMA_OTHER);
395 }
396 #endif
397
398 #ifdef CONFIG_COMPACTION
399
400 struct contig_page_info {
401         unsigned long free_pages;
402         unsigned long free_blocks_total;
403         unsigned long free_blocks_suitable;
404 };
405
406 /*
407  * Calculate the number of free pages in a zone, how many contiguous
408  * pages are free and how many are large enough to satisfy an allocation of
409  * the target size. Note that this function makes no attempt to estimate
410  * how many suitable free blocks there *might* be if MOVABLE pages were
411  * migrated. Calculating that is possible, but expensive and can be
412  * figured out from userspace
413  */
414 static void fill_contig_page_info(struct zone *zone,
415                                 unsigned int suitable_order,
416                                 struct contig_page_info *info)
417 {
418         unsigned int order;
419
420         info->free_pages = 0;
421         info->free_blocks_total = 0;
422         info->free_blocks_suitable = 0;
423
424         for (order = 0; order < MAX_ORDER; order++) {
425                 unsigned long blocks;
426
427                 /* Count number of free blocks */
428                 blocks = zone->free_area[order].nr_free;
429                 info->free_blocks_total += blocks;
430
431                 /* Count free base pages */
432                 info->free_pages += blocks << order;
433
434                 /* Count the suitable free blocks */
435                 if (order >= suitable_order)
436                         info->free_blocks_suitable += blocks <<
437                                                 (order - suitable_order);
438         }
439 }
440
441 /*
442  * A fragmentation index only makes sense if an allocation of a requested
443  * size would fail. If that is true, the fragmentation index indicates
444  * whether external fragmentation or a lack of memory was the problem.
445  * The value can be used to determine if page reclaim or compaction
446  * should be used
447  */
448 static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
449 {
450         unsigned long requested = 1UL << order;
451
452         if (!info->free_blocks_total)
453                 return 0;
454
455         /* Fragmentation index only makes sense when a request would fail */
456         if (info->free_blocks_suitable)
457                 return -1000;
458
459         /*
460          * Index is between 0 and 1 so return within 3 decimal places
461          *
462          * 0 => allocation would fail due to lack of memory
463          * 1 => allocation would fail due to fragmentation
464          */
465         return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
466 }
467
468 /* Same as __fragmentation index but allocs contig_page_info on stack */
469 int fragmentation_index(struct zone *zone, unsigned int order)
470 {
471         struct contig_page_info info;
472
473         fill_contig_page_info(zone, order, &info);
474         return __fragmentation_index(order, &info);
475 }
476 #endif
477
478 #if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
479 #include <linux/proc_fs.h>
480 #include <linux/seq_file.h>
481
482 static char * const migratetype_names[MIGRATE_TYPES] = {
483         "Unmovable",
484         "Reclaimable",
485         "Movable",
486         "Reserve",
487         "Isolate",
488 };
489
490 static void *frag_start(struct seq_file *m, loff_t *pos)
491 {
492         pg_data_t *pgdat;
493         loff_t node = *pos;
494         for (pgdat = first_online_pgdat();
495              pgdat && node;
496              pgdat = next_online_pgdat(pgdat))
497                 --node;
498
499         return pgdat;
500 }
501
502 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
503 {
504         pg_data_t *pgdat = (pg_data_t *)arg;
505
506         (*pos)++;
507         return next_online_pgdat(pgdat);
508 }
509
510 static void frag_stop(struct seq_file *m, void *arg)
511 {
512 }
513
514 /* Walk all the zones in a node and print using a callback */
515 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
516                 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
517 {
518         struct zone *zone;
519         struct zone *node_zones = pgdat->node_zones;
520         unsigned long flags;
521
522         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
523                 if (!populated_zone(zone))
524                         continue;
525
526                 spin_lock_irqsave(&zone->lock, flags);
527                 print(m, pgdat, zone);
528                 spin_unlock_irqrestore(&zone->lock, flags);
529         }
530 }
531 #endif
532
533 #ifdef CONFIG_PROC_FS
534 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
535                                                 struct zone *zone)
536 {
537         int order;
538
539         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
540         for (order = 0; order < MAX_ORDER; ++order)
541                 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
542         seq_putc(m, '\n');
543 }
544
545 /*
546  * This walks the free areas for each zone.
547  */
548 static int frag_show(struct seq_file *m, void *arg)
549 {
550         pg_data_t *pgdat = (pg_data_t *)arg;
551         walk_zones_in_node(m, pgdat, frag_show_print);
552         return 0;
553 }
554
555 static void pagetypeinfo_showfree_print(struct seq_file *m,
556                                         pg_data_t *pgdat, struct zone *zone)
557 {
558         int order, mtype;
559
560         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
561                 seq_printf(m, "Node %4d, zone %8s, type %12s ",
562                                         pgdat->node_id,
563                                         zone->name,
564                                         migratetype_names[mtype]);
565                 for (order = 0; order < MAX_ORDER; ++order) {
566                         unsigned long freecount = 0;
567                         struct free_area *area;
568                         struct list_head *curr;
569
570                         area = &(zone->free_area[order]);
571
572                         list_for_each(curr, &area->free_list[mtype])
573                                 freecount++;
574                         seq_printf(m, "%6lu ", freecount);
575                 }
576                 seq_putc(m, '\n');
577         }
578 }
579
580 /* Print out the free pages at each order for each migatetype */
581 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
582 {
583         int order;
584         pg_data_t *pgdat = (pg_data_t *)arg;
585
586         /* Print header */
587         seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
588         for (order = 0; order < MAX_ORDER; ++order)
589                 seq_printf(m, "%6d ", order);
590         seq_putc(m, '\n');
591
592         walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
593
594         return 0;
595 }
596
597 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
598                                         pg_data_t *pgdat, struct zone *zone)
599 {
600         int mtype;
601         unsigned long pfn;
602         unsigned long start_pfn = zone->zone_start_pfn;
603         unsigned long end_pfn = start_pfn + zone->spanned_pages;
604         unsigned long count[MIGRATE_TYPES] = { 0, };
605
606         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
607                 struct page *page;
608
609                 if (!pfn_valid(pfn))
610                         continue;
611
612                 page = pfn_to_page(pfn);
613
614                 /* Watch for unexpected holes punched in the memmap */
615                 if (!memmap_valid_within(pfn, page, zone))
616                         continue;
617
618                 mtype = get_pageblock_migratetype(page);
619
620                 if (mtype < MIGRATE_TYPES)
621                         count[mtype]++;
622         }
623
624         /* Print counts */
625         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
626         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
627                 seq_printf(m, "%12lu ", count[mtype]);
628         seq_putc(m, '\n');
629 }
630
631 /* Print out the free pages at each order for each migratetype */
632 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
633 {
634         int mtype;
635         pg_data_t *pgdat = (pg_data_t *)arg;
636
637         seq_printf(m, "\n%-23s", "Number of blocks type ");
638         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
639                 seq_printf(m, "%12s ", migratetype_names[mtype]);
640         seq_putc(m, '\n');
641         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
642
643         return 0;
644 }
645
646 /*
647  * This prints out statistics in relation to grouping pages by mobility.
648  * It is expensive to collect so do not constantly read the file.
649  */
650 static int pagetypeinfo_show(struct seq_file *m, void *arg)
651 {
652         pg_data_t *pgdat = (pg_data_t *)arg;
653
654         /* check memoryless node */
655         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
656                 return 0;
657
658         seq_printf(m, "Page block order: %d\n", pageblock_order);
659         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
660         seq_putc(m, '\n');
661         pagetypeinfo_showfree(m, pgdat);
662         pagetypeinfo_showblockcount(m, pgdat);
663
664         return 0;
665 }
666
667 static const struct seq_operations fragmentation_op = {
668         .start  = frag_start,
669         .next   = frag_next,
670         .stop   = frag_stop,
671         .show   = frag_show,
672 };
673
674 static int fragmentation_open(struct inode *inode, struct file *file)
675 {
676         return seq_open(file, &fragmentation_op);
677 }
678
679 static const struct file_operations fragmentation_file_operations = {
680         .open           = fragmentation_open,
681         .read           = seq_read,
682         .llseek         = seq_lseek,
683         .release        = seq_release,
684 };
685
686 static const struct seq_operations pagetypeinfo_op = {
687         .start  = frag_start,
688         .next   = frag_next,
689         .stop   = frag_stop,
690         .show   = pagetypeinfo_show,
691 };
692
693 static int pagetypeinfo_open(struct inode *inode, struct file *file)
694 {
695         return seq_open(file, &pagetypeinfo_op);
696 }
697
698 static const struct file_operations pagetypeinfo_file_ops = {
699         .open           = pagetypeinfo_open,
700         .read           = seq_read,
701         .llseek         = seq_lseek,
702         .release        = seq_release,
703 };
704
705 #ifdef CONFIG_ZONE_DMA
706 #define TEXT_FOR_DMA(xx) xx "_dma",
707 #else
708 #define TEXT_FOR_DMA(xx)
709 #endif
710
711 #ifdef CONFIG_ZONE_DMA32
712 #define TEXT_FOR_DMA32(xx) xx "_dma32",
713 #else
714 #define TEXT_FOR_DMA32(xx)
715 #endif
716
717 #ifdef CONFIG_HIGHMEM
718 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
719 #else
720 #define TEXT_FOR_HIGHMEM(xx)
721 #endif
722
723 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
724                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
725
726 static const char * const vmstat_text[] = {
727         /* Zoned VM counters */
728         "nr_free_pages",
729         "nr_inactive_anon",
730         "nr_active_anon",
731         "nr_inactive_file",
732         "nr_active_file",
733         "nr_unevictable",
734         "nr_mlock",
735         "nr_anon_pages",
736         "nr_mapped",
737         "nr_file_pages",
738         "nr_dirty",
739         "nr_writeback",
740         "nr_slab_reclaimable",
741         "nr_slab_unreclaimable",
742         "nr_page_table_pages",
743         "nr_kernel_stack",
744         "nr_unstable",
745         "nr_bounce",
746         "nr_vmscan_write",
747         "nr_writeback_temp",
748         "nr_isolated_anon",
749         "nr_isolated_file",
750         "nr_shmem",
751         "nr_dirtied",
752         "nr_written",
753
754 #ifdef CONFIG_NUMA
755         "numa_hit",
756         "numa_miss",
757         "numa_foreign",
758         "numa_interleave",
759         "numa_local",
760         "numa_other",
761 #endif
762         "nr_dirty_threshold",
763         "nr_dirty_background_threshold",
764
765 #ifdef CONFIG_VM_EVENT_COUNTERS
766         "pgpgin",
767         "pgpgout",
768         "pswpin",
769         "pswpout",
770
771         TEXTS_FOR_ZONES("pgalloc")
772
773         "pgfree",
774         "pgactivate",
775         "pgdeactivate",
776
777         "pgfault",
778         "pgmajfault",
779
780         TEXTS_FOR_ZONES("pgrefill")
781         TEXTS_FOR_ZONES("pgsteal")
782         TEXTS_FOR_ZONES("pgscan_kswapd")
783         TEXTS_FOR_ZONES("pgscan_direct")
784
785 #ifdef CONFIG_NUMA
786         "zone_reclaim_failed",
787 #endif
788         "pginodesteal",
789         "slabs_scanned",
790         "kswapd_steal",
791         "kswapd_inodesteal",
792         "kswapd_low_wmark_hit_quickly",
793         "kswapd_high_wmark_hit_quickly",
794         "kswapd_skip_congestion_wait",
795         "pageoutrun",
796         "allocstall",
797
798         "pgrotated",
799
800 #ifdef CONFIG_COMPACTION
801         "compact_blocks_moved",
802         "compact_pages_moved",
803         "compact_pagemigrate_failed",
804         "compact_stall",
805         "compact_fail",
806         "compact_success",
807 #endif
808
809 #ifdef CONFIG_HUGETLB_PAGE
810         "htlb_buddy_alloc_success",
811         "htlb_buddy_alloc_fail",
812 #endif
813         "unevictable_pgs_culled",
814         "unevictable_pgs_scanned",
815         "unevictable_pgs_rescued",
816         "unevictable_pgs_mlocked",
817         "unevictable_pgs_munlocked",
818         "unevictable_pgs_cleared",
819         "unevictable_pgs_stranded",
820         "unevictable_pgs_mlockfreed",
821 #endif
822 };
823
824 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
825                                                         struct zone *zone)
826 {
827         int i;
828         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
829         seq_printf(m,
830                    "\n  pages free     %lu"
831                    "\n        min      %lu"
832                    "\n        low      %lu"
833                    "\n        high     %lu"
834                    "\n        scanned  %lu"
835                    "\n        spanned  %lu"
836                    "\n        present  %lu",
837                    zone_nr_free_pages(zone),
838                    min_wmark_pages(zone),
839                    low_wmark_pages(zone),
840                    high_wmark_pages(zone),
841                    zone->pages_scanned,
842                    zone->spanned_pages,
843                    zone->present_pages);
844
845         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
846                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
847                                 zone_page_state(zone, i));
848
849         seq_printf(m,
850                    "\n        protection: (%lu",
851                    zone->lowmem_reserve[0]);
852         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
853                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
854         seq_printf(m,
855                    ")"
856                    "\n  pagesets");
857         for_each_online_cpu(i) {
858                 struct per_cpu_pageset *pageset;
859
860                 pageset = per_cpu_ptr(zone->pageset, i);
861                 seq_printf(m,
862                            "\n    cpu: %i"
863                            "\n              count: %i"
864                            "\n              high:  %i"
865                            "\n              batch: %i",
866                            i,
867                            pageset->pcp.count,
868                            pageset->pcp.high,
869                            pageset->pcp.batch);
870 #ifdef CONFIG_SMP
871                 seq_printf(m, "\n  vm stats threshold: %d",
872                                 pageset->stat_threshold);
873 #endif
874         }
875         seq_printf(m,
876                    "\n  all_unreclaimable: %u"
877                    "\n  start_pfn:         %lu"
878                    "\n  inactive_ratio:    %u",
879                    zone->all_unreclaimable,
880                    zone->zone_start_pfn,
881                    zone->inactive_ratio);
882         seq_putc(m, '\n');
883 }
884
885 /*
886  * Output information about zones in @pgdat.
887  */
888 static int zoneinfo_show(struct seq_file *m, void *arg)
889 {
890         pg_data_t *pgdat = (pg_data_t *)arg;
891         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
892         return 0;
893 }
894
895 static const struct seq_operations zoneinfo_op = {
896         .start  = frag_start, /* iterate over all zones. The same as in
897                                * fragmentation. */
898         .next   = frag_next,
899         .stop   = frag_stop,
900         .show   = zoneinfo_show,
901 };
902
903 static int zoneinfo_open(struct inode *inode, struct file *file)
904 {
905         return seq_open(file, &zoneinfo_op);
906 }
907
908 static const struct file_operations proc_zoneinfo_file_operations = {
909         .open           = zoneinfo_open,
910         .read           = seq_read,
911         .llseek         = seq_lseek,
912         .release        = seq_release,
913 };
914
915 enum writeback_stat_item {
916         NR_DIRTY_THRESHOLD,
917         NR_DIRTY_BG_THRESHOLD,
918         NR_VM_WRITEBACK_STAT_ITEMS,
919 };
920
921 static void *vmstat_start(struct seq_file *m, loff_t *pos)
922 {
923         unsigned long *v;
924         int i, stat_items_size;
925
926         if (*pos >= ARRAY_SIZE(vmstat_text))
927                 return NULL;
928         stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
929                           NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
930
931 #ifdef CONFIG_VM_EVENT_COUNTERS
932         stat_items_size += sizeof(struct vm_event_state);
933 #endif
934
935         v = kmalloc(stat_items_size, GFP_KERNEL);
936         m->private = v;
937         if (!v)
938                 return ERR_PTR(-ENOMEM);
939         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
940                 v[i] = global_page_state(i);
941         v += NR_VM_ZONE_STAT_ITEMS;
942
943         global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
944                             v + NR_DIRTY_THRESHOLD);
945         v += NR_VM_WRITEBACK_STAT_ITEMS;
946
947 #ifdef CONFIG_VM_EVENT_COUNTERS
948         all_vm_events(v);
949         v[PGPGIN] /= 2;         /* sectors -> kbytes */
950         v[PGPGOUT] /= 2;
951 #endif
952         return (unsigned long *)m->private + *pos;
953 }
954
955 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
956 {
957         (*pos)++;
958         if (*pos >= ARRAY_SIZE(vmstat_text))
959                 return NULL;
960         return (unsigned long *)m->private + *pos;
961 }
962
963 static int vmstat_show(struct seq_file *m, void *arg)
964 {
965         unsigned long *l = arg;
966         unsigned long off = l - (unsigned long *)m->private;
967
968         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
969         return 0;
970 }
971
972 static void vmstat_stop(struct seq_file *m, void *arg)
973 {
974         kfree(m->private);
975         m->private = NULL;
976 }
977
978 static const struct seq_operations vmstat_op = {
979         .start  = vmstat_start,
980         .next   = vmstat_next,
981         .stop   = vmstat_stop,
982         .show   = vmstat_show,
983 };
984
985 static int vmstat_open(struct inode *inode, struct file *file)
986 {
987         return seq_open(file, &vmstat_op);
988 }
989
990 static const struct file_operations proc_vmstat_file_operations = {
991         .open           = vmstat_open,
992         .read           = seq_read,
993         .llseek         = seq_lseek,
994         .release        = seq_release,
995 };
996 #endif /* CONFIG_PROC_FS */
997
998 #ifdef CONFIG_SMP
999 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1000 int sysctl_stat_interval __read_mostly = HZ;
1001
1002 static void vmstat_update(struct work_struct *w)
1003 {
1004         refresh_cpu_vm_stats(smp_processor_id());
1005         schedule_delayed_work(&__get_cpu_var(vmstat_work),
1006                 round_jiffies_relative(sysctl_stat_interval));
1007 }
1008
1009 static void __cpuinit start_cpu_timer(int cpu)
1010 {
1011         struct delayed_work *work = &per_cpu(vmstat_work, cpu);
1012
1013         INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
1014         schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
1015 }
1016
1017 /*
1018  * Use the cpu notifier to insure that the thresholds are recalculated
1019  * when necessary.
1020  */
1021 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
1022                 unsigned long action,
1023                 void *hcpu)
1024 {
1025         long cpu = (long)hcpu;
1026
1027         switch (action) {
1028         case CPU_ONLINE:
1029         case CPU_ONLINE_FROZEN:
1030                 refresh_zone_stat_thresholds();
1031                 start_cpu_timer(cpu);
1032                 node_set_state(cpu_to_node(cpu), N_CPU);
1033                 break;
1034         case CPU_DOWN_PREPARE:
1035         case CPU_DOWN_PREPARE_FROZEN:
1036                 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
1037                 per_cpu(vmstat_work, cpu).work.func = NULL;
1038                 break;
1039         case CPU_DOWN_FAILED:
1040         case CPU_DOWN_FAILED_FROZEN:
1041                 start_cpu_timer(cpu);
1042                 break;
1043         case CPU_DEAD:
1044         case CPU_DEAD_FROZEN:
1045                 refresh_zone_stat_thresholds();
1046                 break;
1047         default:
1048                 break;
1049         }
1050         return NOTIFY_OK;
1051 }
1052
1053 static struct notifier_block __cpuinitdata vmstat_notifier =
1054         { &vmstat_cpuup_callback, NULL, 0 };
1055 #endif
1056
1057 static int __init setup_vmstat(void)
1058 {
1059 #ifdef CONFIG_SMP
1060         int cpu;
1061
1062         refresh_zone_stat_thresholds();
1063         register_cpu_notifier(&vmstat_notifier);
1064
1065         for_each_online_cpu(cpu)
1066                 start_cpu_timer(cpu);
1067 #endif
1068 #ifdef CONFIG_PROC_FS
1069         proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1070         proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1071         proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1072         proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1073 #endif
1074         return 0;
1075 }
1076 module_init(setup_vmstat)
1077
1078 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1079 #include <linux/debugfs.h>
1080
1081 static struct dentry *extfrag_debug_root;
1082
1083 /*
1084  * Return an index indicating how much of the available free memory is
1085  * unusable for an allocation of the requested size.
1086  */
1087 static int unusable_free_index(unsigned int order,
1088                                 struct contig_page_info *info)
1089 {
1090         /* No free memory is interpreted as all free memory is unusable */
1091         if (info->free_pages == 0)
1092                 return 1000;
1093
1094         /*
1095          * Index should be a value between 0 and 1. Return a value to 3
1096          * decimal places.
1097          *
1098          * 0 => no fragmentation
1099          * 1 => high fragmentation
1100          */
1101         return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1102
1103 }
1104
1105 static void unusable_show_print(struct seq_file *m,
1106                                         pg_data_t *pgdat, struct zone *zone)
1107 {
1108         unsigned int order;
1109         int index;
1110         struct contig_page_info info;
1111
1112         seq_printf(m, "Node %d, zone %8s ",
1113                                 pgdat->node_id,
1114                                 zone->name);
1115         for (order = 0; order < MAX_ORDER; ++order) {
1116                 fill_contig_page_info(zone, order, &info);
1117                 index = unusable_free_index(order, &info);
1118                 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1119         }
1120
1121         seq_putc(m, '\n');
1122 }
1123
1124 /*
1125  * Display unusable free space index
1126  *
1127  * The unusable free space index measures how much of the available free
1128  * memory cannot be used to satisfy an allocation of a given size and is a
1129  * value between 0 and 1. The higher the value, the more of free memory is
1130  * unusable and by implication, the worse the external fragmentation is. This
1131  * can be expressed as a percentage by multiplying by 100.
1132  */
1133 static int unusable_show(struct seq_file *m, void *arg)
1134 {
1135         pg_data_t *pgdat = (pg_data_t *)arg;
1136
1137         /* check memoryless node */
1138         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
1139                 return 0;
1140
1141         walk_zones_in_node(m, pgdat, unusable_show_print);
1142
1143         return 0;
1144 }
1145
1146 static const struct seq_operations unusable_op = {
1147         .start  = frag_start,
1148         .next   = frag_next,
1149         .stop   = frag_stop,
1150         .show   = unusable_show,
1151 };
1152
1153 static int unusable_open(struct inode *inode, struct file *file)
1154 {
1155         return seq_open(file, &unusable_op);
1156 }
1157
1158 static const struct file_operations unusable_file_ops = {
1159         .open           = unusable_open,
1160         .read           = seq_read,
1161         .llseek         = seq_lseek,
1162         .release        = seq_release,
1163 };
1164
1165 static void extfrag_show_print(struct seq_file *m,
1166                                         pg_data_t *pgdat, struct zone *zone)
1167 {
1168         unsigned int order;
1169         int index;
1170
1171         /* Alloc on stack as interrupts are disabled for zone walk */
1172         struct contig_page_info info;
1173
1174         seq_printf(m, "Node %d, zone %8s ",
1175                                 pgdat->node_id,
1176                                 zone->name);
1177         for (order = 0; order < MAX_ORDER; ++order) {
1178                 fill_contig_page_info(zone, order, &info);
1179                 index = __fragmentation_index(order, &info);
1180                 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1181         }
1182
1183         seq_putc(m, '\n');
1184 }
1185
1186 /*
1187  * Display fragmentation index for orders that allocations would fail for
1188  */
1189 static int extfrag_show(struct seq_file *m, void *arg)
1190 {
1191         pg_data_t *pgdat = (pg_data_t *)arg;
1192
1193         walk_zones_in_node(m, pgdat, extfrag_show_print);
1194
1195         return 0;
1196 }
1197
1198 static const struct seq_operations extfrag_op = {
1199         .start  = frag_start,
1200         .next   = frag_next,
1201         .stop   = frag_stop,
1202         .show   = extfrag_show,
1203 };
1204
1205 static int extfrag_open(struct inode *inode, struct file *file)
1206 {
1207         return seq_open(file, &extfrag_op);
1208 }
1209
1210 static const struct file_operations extfrag_file_ops = {
1211         .open           = extfrag_open,
1212         .read           = seq_read,
1213         .llseek         = seq_lseek,
1214         .release        = seq_release,
1215 };
1216
1217 static int __init extfrag_debug_init(void)
1218 {
1219         extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1220         if (!extfrag_debug_root)
1221                 return -ENOMEM;
1222
1223         if (!debugfs_create_file("unusable_index", 0444,
1224                         extfrag_debug_root, NULL, &unusable_file_ops))
1225                 return -ENOMEM;
1226
1227         if (!debugfs_create_file("extfrag_index", 0444,
1228                         extfrag_debug_root, NULL, &extfrag_file_ops))
1229                 return -ENOMEM;
1230
1231         return 0;
1232 }
1233
1234 module_init(extfrag_debug_init);
1235 #endif