mm: arch: rename in_gate_area_no_task to in_gate_area_no_mm
[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 int calculate_pressure_threshold(struct zone *zone)
87 {
88         int threshold;
89         int watermark_distance;
90
91         /*
92          * As vmstats are not up to date, there is drift between the estimated
93          * and real values. For high thresholds and a high number of CPUs, it
94          * is possible for the min watermark to be breached while the estimated
95          * value looks fine. The pressure threshold is a reduced value such
96          * that even the maximum amount of drift will not accidentally breach
97          * the min watermark
98          */
99         watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
100         threshold = max(1, (int)(watermark_distance / num_online_cpus()));
101
102         /*
103          * Maximum threshold is 125
104          */
105         threshold = min(125, threshold);
106
107         return threshold;
108 }
109
110 int calculate_normal_threshold(struct zone *zone)
111 {
112         int threshold;
113         int mem;        /* memory in 128 MB units */
114
115         /*
116          * The threshold scales with the number of processors and the amount
117          * of memory per zone. More memory means that we can defer updates for
118          * longer, more processors could lead to more contention.
119          * fls() is used to have a cheap way of logarithmic scaling.
120          *
121          * Some sample thresholds:
122          *
123          * Threshold    Processors      (fls)   Zonesize        fls(mem+1)
124          * ------------------------------------------------------------------
125          * 8            1               1       0.9-1 GB        4
126          * 16           2               2       0.9-1 GB        4
127          * 20           2               2       1-2 GB          5
128          * 24           2               2       2-4 GB          6
129          * 28           2               2       4-8 GB          7
130          * 32           2               2       8-16 GB         8
131          * 4            2               2       <128M           1
132          * 30           4               3       2-4 GB          5
133          * 48           4               3       8-16 GB         8
134          * 32           8               4       1-2 GB          4
135          * 32           8               4       0.9-1GB         4
136          * 10           16              5       <128M           1
137          * 40           16              5       900M            4
138          * 70           64              7       2-4 GB          5
139          * 84           64              7       4-8 GB          6
140          * 108          512             9       4-8 GB          6
141          * 125          1024            10      8-16 GB         8
142          * 125          1024            10      16-32 GB        9
143          */
144
145         mem = zone->present_pages >> (27 - PAGE_SHIFT);
146
147         threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
148
149         /*
150          * Maximum threshold is 125
151          */
152         threshold = min(125, threshold);
153
154         return threshold;
155 }
156
157 /*
158  * Refresh the thresholds for each zone.
159  */
160 static void refresh_zone_stat_thresholds(void)
161 {
162         struct zone *zone;
163         int cpu;
164         int threshold;
165
166         for_each_populated_zone(zone) {
167                 unsigned long max_drift, tolerate_drift;
168
169                 threshold = calculate_normal_threshold(zone);
170
171                 for_each_online_cpu(cpu)
172                         per_cpu_ptr(zone->pageset, cpu)->stat_threshold
173                                                         = threshold;
174
175                 /*
176                  * Only set percpu_drift_mark if there is a danger that
177                  * NR_FREE_PAGES reports the low watermark is ok when in fact
178                  * the min watermark could be breached by an allocation
179                  */
180                 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
181                 max_drift = num_online_cpus() * threshold;
182                 if (max_drift > tolerate_drift)
183                         zone->percpu_drift_mark = high_wmark_pages(zone) +
184                                         max_drift;
185         }
186 }
187
188 void set_pgdat_percpu_threshold(pg_data_t *pgdat,
189                                 int (*calculate_pressure)(struct zone *))
190 {
191         struct zone *zone;
192         int cpu;
193         int threshold;
194         int i;
195
196         for (i = 0; i < pgdat->nr_zones; i++) {
197                 zone = &pgdat->node_zones[i];
198                 if (!zone->percpu_drift_mark)
199                         continue;
200
201                 threshold = (*calculate_pressure)(zone);
202                 for_each_possible_cpu(cpu)
203                         per_cpu_ptr(zone->pageset, cpu)->stat_threshold
204                                                         = threshold;
205         }
206 }
207
208 /*
209  * For use when we know that interrupts are disabled.
210  */
211 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
212                                 int delta)
213 {
214         struct per_cpu_pageset __percpu *pcp = zone->pageset;
215         s8 __percpu *p = pcp->vm_stat_diff + item;
216         long x;
217         long t;
218
219         x = delta + __this_cpu_read(*p);
220
221         t = __this_cpu_read(pcp->stat_threshold);
222
223         if (unlikely(x > t || x < -t)) {
224                 zone_page_state_add(x, zone, item);
225                 x = 0;
226         }
227         __this_cpu_write(*p, x);
228 }
229 EXPORT_SYMBOL(__mod_zone_page_state);
230
231 /*
232  * Optimized increment and decrement functions.
233  *
234  * These are only for a single page and therefore can take a struct page *
235  * argument instead of struct zone *. This allows the inclusion of the code
236  * generated for page_zone(page) into the optimized functions.
237  *
238  * No overflow check is necessary and therefore the differential can be
239  * incremented or decremented in place which may allow the compilers to
240  * generate better code.
241  * The increment or decrement is known and therefore one boundary check can
242  * be omitted.
243  *
244  * NOTE: These functions are very performance sensitive. Change only
245  * with care.
246  *
247  * Some processors have inc/dec instructions that are atomic vs an interrupt.
248  * However, the code must first determine the differential location in a zone
249  * based on the processor number and then inc/dec the counter. There is no
250  * guarantee without disabling preemption that the processor will not change
251  * in between and therefore the atomicity vs. interrupt cannot be exploited
252  * in a useful way here.
253  */
254 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
255 {
256         struct per_cpu_pageset __percpu *pcp = zone->pageset;
257         s8 __percpu *p = pcp->vm_stat_diff + item;
258         s8 v, t;
259
260         v = __this_cpu_inc_return(*p);
261         t = __this_cpu_read(pcp->stat_threshold);
262         if (unlikely(v > t)) {
263                 s8 overstep = t >> 1;
264
265                 zone_page_state_add(v + overstep, zone, item);
266                 __this_cpu_write(*p, -overstep);
267         }
268 }
269
270 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
271 {
272         __inc_zone_state(page_zone(page), item);
273 }
274 EXPORT_SYMBOL(__inc_zone_page_state);
275
276 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
277 {
278         struct per_cpu_pageset __percpu *pcp = zone->pageset;
279         s8 __percpu *p = pcp->vm_stat_diff + item;
280         s8 v, t;
281
282         v = __this_cpu_dec_return(*p);
283         t = __this_cpu_read(pcp->stat_threshold);
284         if (unlikely(v < - t)) {
285                 s8 overstep = t >> 1;
286
287                 zone_page_state_add(v - overstep, zone, item);
288                 __this_cpu_write(*p, overstep);
289         }
290 }
291
292 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
293 {
294         __dec_zone_state(page_zone(page), item);
295 }
296 EXPORT_SYMBOL(__dec_zone_page_state);
297
298 #ifdef CONFIG_CMPXCHG_LOCAL
299 /*
300  * If we have cmpxchg_local support then we do not need to incur the overhead
301  * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
302  *
303  * mod_state() modifies the zone counter state through atomic per cpu
304  * operations.
305  *
306  * Overstep mode specifies how overstep should handled:
307  *     0       No overstepping
308  *     1       Overstepping half of threshold
309  *     -1      Overstepping minus half of threshold
310 */
311 static inline void mod_state(struct zone *zone,
312        enum zone_stat_item item, int delta, int overstep_mode)
313 {
314         struct per_cpu_pageset __percpu *pcp = zone->pageset;
315         s8 __percpu *p = pcp->vm_stat_diff + item;
316         long o, n, t, z;
317
318         do {
319                 z = 0;  /* overflow to zone counters */
320
321                 /*
322                  * The fetching of the stat_threshold is racy. We may apply
323                  * a counter threshold to the wrong the cpu if we get
324                  * rescheduled while executing here. However, the following
325                  * will apply the threshold again and therefore bring the
326                  * counter under the threshold.
327                  */
328                 t = this_cpu_read(pcp->stat_threshold);
329
330                 o = this_cpu_read(*p);
331                 n = delta + o;
332
333                 if (n > t || n < -t) {
334                         int os = overstep_mode * (t >> 1) ;
335
336                         /* Overflow must be added to zone counters */
337                         z = n + os;
338                         n = -os;
339                 }
340         } while (this_cpu_cmpxchg(*p, o, n) != o);
341
342         if (z)
343                 zone_page_state_add(z, zone, item);
344 }
345
346 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
347                                         int delta)
348 {
349         mod_state(zone, item, delta, 0);
350 }
351 EXPORT_SYMBOL(mod_zone_page_state);
352
353 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
354 {
355         mod_state(zone, item, 1, 1);
356 }
357
358 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
359 {
360         mod_state(page_zone(page), item, 1, 1);
361 }
362 EXPORT_SYMBOL(inc_zone_page_state);
363
364 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
365 {
366         mod_state(page_zone(page), item, -1, -1);
367 }
368 EXPORT_SYMBOL(dec_zone_page_state);
369 #else
370 /*
371  * Use interrupt disable to serialize counter updates
372  */
373 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
374                                         int delta)
375 {
376         unsigned long flags;
377
378         local_irq_save(flags);
379         __mod_zone_page_state(zone, item, delta);
380         local_irq_restore(flags);
381 }
382 EXPORT_SYMBOL(mod_zone_page_state);
383
384 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
385 {
386         unsigned long flags;
387
388         local_irq_save(flags);
389         __inc_zone_state(zone, item);
390         local_irq_restore(flags);
391 }
392
393 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
394 {
395         unsigned long flags;
396         struct zone *zone;
397
398         zone = page_zone(page);
399         local_irq_save(flags);
400         __inc_zone_state(zone, item);
401         local_irq_restore(flags);
402 }
403 EXPORT_SYMBOL(inc_zone_page_state);
404
405 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
406 {
407         unsigned long flags;
408
409         local_irq_save(flags);
410         __dec_zone_page_state(page, item);
411         local_irq_restore(flags);
412 }
413 EXPORT_SYMBOL(dec_zone_page_state);
414 #endif
415
416 /*
417  * Update the zone counters for one cpu.
418  *
419  * The cpu specified must be either the current cpu or a processor that
420  * is not online. If it is the current cpu then the execution thread must
421  * be pinned to the current cpu.
422  *
423  * Note that refresh_cpu_vm_stats strives to only access
424  * node local memory. The per cpu pagesets on remote zones are placed
425  * in the memory local to the processor using that pageset. So the
426  * loop over all zones will access a series of cachelines local to
427  * the processor.
428  *
429  * The call to zone_page_state_add updates the cachelines with the
430  * statistics in the remote zone struct as well as the global cachelines
431  * with the global counters. These could cause remote node cache line
432  * bouncing and will have to be only done when necessary.
433  */
434 void refresh_cpu_vm_stats(int cpu)
435 {
436         struct zone *zone;
437         int i;
438         int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
439
440         for_each_populated_zone(zone) {
441                 struct per_cpu_pageset *p;
442
443                 p = per_cpu_ptr(zone->pageset, cpu);
444
445                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
446                         if (p->vm_stat_diff[i]) {
447                                 unsigned long flags;
448                                 int v;
449
450                                 local_irq_save(flags);
451                                 v = p->vm_stat_diff[i];
452                                 p->vm_stat_diff[i] = 0;
453                                 local_irq_restore(flags);
454                                 atomic_long_add(v, &zone->vm_stat[i]);
455                                 global_diff[i] += v;
456 #ifdef CONFIG_NUMA
457                                 /* 3 seconds idle till flush */
458                                 p->expire = 3;
459 #endif
460                         }
461                 cond_resched();
462 #ifdef CONFIG_NUMA
463                 /*
464                  * Deal with draining the remote pageset of this
465                  * processor
466                  *
467                  * Check if there are pages remaining in this pageset
468                  * if not then there is nothing to expire.
469                  */
470                 if (!p->expire || !p->pcp.count)
471                         continue;
472
473                 /*
474                  * We never drain zones local to this processor.
475                  */
476                 if (zone_to_nid(zone) == numa_node_id()) {
477                         p->expire = 0;
478                         continue;
479                 }
480
481                 p->expire--;
482                 if (p->expire)
483                         continue;
484
485                 if (p->pcp.count)
486                         drain_zone_pages(zone, &p->pcp);
487 #endif
488         }
489
490         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
491                 if (global_diff[i])
492                         atomic_long_add(global_diff[i], &vm_stat[i]);
493 }
494
495 #endif
496
497 #ifdef CONFIG_NUMA
498 /*
499  * zonelist = the list of zones passed to the allocator
500  * z        = the zone from which the allocation occurred.
501  *
502  * Must be called with interrupts disabled.
503  */
504 void zone_statistics(struct zone *preferred_zone, struct zone *z)
505 {
506         if (z->zone_pgdat == preferred_zone->zone_pgdat) {
507                 __inc_zone_state(z, NUMA_HIT);
508         } else {
509                 __inc_zone_state(z, NUMA_MISS);
510                 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
511         }
512         if (z->node == numa_node_id())
513                 __inc_zone_state(z, NUMA_LOCAL);
514         else
515                 __inc_zone_state(z, NUMA_OTHER);
516 }
517 #endif
518
519 #ifdef CONFIG_COMPACTION
520
521 struct contig_page_info {
522         unsigned long free_pages;
523         unsigned long free_blocks_total;
524         unsigned long free_blocks_suitable;
525 };
526
527 /*
528  * Calculate the number of free pages in a zone, how many contiguous
529  * pages are free and how many are large enough to satisfy an allocation of
530  * the target size. Note that this function makes no attempt to estimate
531  * how many suitable free blocks there *might* be if MOVABLE pages were
532  * migrated. Calculating that is possible, but expensive and can be
533  * figured out from userspace
534  */
535 static void fill_contig_page_info(struct zone *zone,
536                                 unsigned int suitable_order,
537                                 struct contig_page_info *info)
538 {
539         unsigned int order;
540
541         info->free_pages = 0;
542         info->free_blocks_total = 0;
543         info->free_blocks_suitable = 0;
544
545         for (order = 0; order < MAX_ORDER; order++) {
546                 unsigned long blocks;
547
548                 /* Count number of free blocks */
549                 blocks = zone->free_area[order].nr_free;
550                 info->free_blocks_total += blocks;
551
552                 /* Count free base pages */
553                 info->free_pages += blocks << order;
554
555                 /* Count the suitable free blocks */
556                 if (order >= suitable_order)
557                         info->free_blocks_suitable += blocks <<
558                                                 (order - suitable_order);
559         }
560 }
561
562 /*
563  * A fragmentation index only makes sense if an allocation of a requested
564  * size would fail. If that is true, the fragmentation index indicates
565  * whether external fragmentation or a lack of memory was the problem.
566  * The value can be used to determine if page reclaim or compaction
567  * should be used
568  */
569 static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
570 {
571         unsigned long requested = 1UL << order;
572
573         if (!info->free_blocks_total)
574                 return 0;
575
576         /* Fragmentation index only makes sense when a request would fail */
577         if (info->free_blocks_suitable)
578                 return -1000;
579
580         /*
581          * Index is between 0 and 1 so return within 3 decimal places
582          *
583          * 0 => allocation would fail due to lack of memory
584          * 1 => allocation would fail due to fragmentation
585          */
586         return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
587 }
588
589 /* Same as __fragmentation index but allocs contig_page_info on stack */
590 int fragmentation_index(struct zone *zone, unsigned int order)
591 {
592         struct contig_page_info info;
593
594         fill_contig_page_info(zone, order, &info);
595         return __fragmentation_index(order, &info);
596 }
597 #endif
598
599 #if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
600 #include <linux/proc_fs.h>
601 #include <linux/seq_file.h>
602
603 static char * const migratetype_names[MIGRATE_TYPES] = {
604         "Unmovable",
605         "Reclaimable",
606         "Movable",
607         "Reserve",
608         "Isolate",
609 };
610
611 static void *frag_start(struct seq_file *m, loff_t *pos)
612 {
613         pg_data_t *pgdat;
614         loff_t node = *pos;
615         for (pgdat = first_online_pgdat();
616              pgdat && node;
617              pgdat = next_online_pgdat(pgdat))
618                 --node;
619
620         return pgdat;
621 }
622
623 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
624 {
625         pg_data_t *pgdat = (pg_data_t *)arg;
626
627         (*pos)++;
628         return next_online_pgdat(pgdat);
629 }
630
631 static void frag_stop(struct seq_file *m, void *arg)
632 {
633 }
634
635 /* Walk all the zones in a node and print using a callback */
636 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
637                 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
638 {
639         struct zone *zone;
640         struct zone *node_zones = pgdat->node_zones;
641         unsigned long flags;
642
643         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
644                 if (!populated_zone(zone))
645                         continue;
646
647                 spin_lock_irqsave(&zone->lock, flags);
648                 print(m, pgdat, zone);
649                 spin_unlock_irqrestore(&zone->lock, flags);
650         }
651 }
652 #endif
653
654 #ifdef CONFIG_PROC_FS
655 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
656                                                 struct zone *zone)
657 {
658         int order;
659
660         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
661         for (order = 0; order < MAX_ORDER; ++order)
662                 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
663         seq_putc(m, '\n');
664 }
665
666 /*
667  * This walks the free areas for each zone.
668  */
669 static int frag_show(struct seq_file *m, void *arg)
670 {
671         pg_data_t *pgdat = (pg_data_t *)arg;
672         walk_zones_in_node(m, pgdat, frag_show_print);
673         return 0;
674 }
675
676 static void pagetypeinfo_showfree_print(struct seq_file *m,
677                                         pg_data_t *pgdat, struct zone *zone)
678 {
679         int order, mtype;
680
681         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
682                 seq_printf(m, "Node %4d, zone %8s, type %12s ",
683                                         pgdat->node_id,
684                                         zone->name,
685                                         migratetype_names[mtype]);
686                 for (order = 0; order < MAX_ORDER; ++order) {
687                         unsigned long freecount = 0;
688                         struct free_area *area;
689                         struct list_head *curr;
690
691                         area = &(zone->free_area[order]);
692
693                         list_for_each(curr, &area->free_list[mtype])
694                                 freecount++;
695                         seq_printf(m, "%6lu ", freecount);
696                 }
697                 seq_putc(m, '\n');
698         }
699 }
700
701 /* Print out the free pages at each order for each migatetype */
702 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
703 {
704         int order;
705         pg_data_t *pgdat = (pg_data_t *)arg;
706
707         /* Print header */
708         seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
709         for (order = 0; order < MAX_ORDER; ++order)
710                 seq_printf(m, "%6d ", order);
711         seq_putc(m, '\n');
712
713         walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
714
715         return 0;
716 }
717
718 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
719                                         pg_data_t *pgdat, struct zone *zone)
720 {
721         int mtype;
722         unsigned long pfn;
723         unsigned long start_pfn = zone->zone_start_pfn;
724         unsigned long end_pfn = start_pfn + zone->spanned_pages;
725         unsigned long count[MIGRATE_TYPES] = { 0, };
726
727         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
728                 struct page *page;
729
730                 if (!pfn_valid(pfn))
731                         continue;
732
733                 page = pfn_to_page(pfn);
734
735                 /* Watch for unexpected holes punched in the memmap */
736                 if (!memmap_valid_within(pfn, page, zone))
737                         continue;
738
739                 mtype = get_pageblock_migratetype(page);
740
741                 if (mtype < MIGRATE_TYPES)
742                         count[mtype]++;
743         }
744
745         /* Print counts */
746         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
747         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
748                 seq_printf(m, "%12lu ", count[mtype]);
749         seq_putc(m, '\n');
750 }
751
752 /* Print out the free pages at each order for each migratetype */
753 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
754 {
755         int mtype;
756         pg_data_t *pgdat = (pg_data_t *)arg;
757
758         seq_printf(m, "\n%-23s", "Number of blocks type ");
759         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
760                 seq_printf(m, "%12s ", migratetype_names[mtype]);
761         seq_putc(m, '\n');
762         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
763
764         return 0;
765 }
766
767 /*
768  * This prints out statistics in relation to grouping pages by mobility.
769  * It is expensive to collect so do not constantly read the file.
770  */
771 static int pagetypeinfo_show(struct seq_file *m, void *arg)
772 {
773         pg_data_t *pgdat = (pg_data_t *)arg;
774
775         /* check memoryless node */
776         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
777                 return 0;
778
779         seq_printf(m, "Page block order: %d\n", pageblock_order);
780         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
781         seq_putc(m, '\n');
782         pagetypeinfo_showfree(m, pgdat);
783         pagetypeinfo_showblockcount(m, pgdat);
784
785         return 0;
786 }
787
788 static const struct seq_operations fragmentation_op = {
789         .start  = frag_start,
790         .next   = frag_next,
791         .stop   = frag_stop,
792         .show   = frag_show,
793 };
794
795 static int fragmentation_open(struct inode *inode, struct file *file)
796 {
797         return seq_open(file, &fragmentation_op);
798 }
799
800 static const struct file_operations fragmentation_file_operations = {
801         .open           = fragmentation_open,
802         .read           = seq_read,
803         .llseek         = seq_lseek,
804         .release        = seq_release,
805 };
806
807 static const struct seq_operations pagetypeinfo_op = {
808         .start  = frag_start,
809         .next   = frag_next,
810         .stop   = frag_stop,
811         .show   = pagetypeinfo_show,
812 };
813
814 static int pagetypeinfo_open(struct inode *inode, struct file *file)
815 {
816         return seq_open(file, &pagetypeinfo_op);
817 }
818
819 static const struct file_operations pagetypeinfo_file_ops = {
820         .open           = pagetypeinfo_open,
821         .read           = seq_read,
822         .llseek         = seq_lseek,
823         .release        = seq_release,
824 };
825
826 #ifdef CONFIG_ZONE_DMA
827 #define TEXT_FOR_DMA(xx) xx "_dma",
828 #else
829 #define TEXT_FOR_DMA(xx)
830 #endif
831
832 #ifdef CONFIG_ZONE_DMA32
833 #define TEXT_FOR_DMA32(xx) xx "_dma32",
834 #else
835 #define TEXT_FOR_DMA32(xx)
836 #endif
837
838 #ifdef CONFIG_HIGHMEM
839 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
840 #else
841 #define TEXT_FOR_HIGHMEM(xx)
842 #endif
843
844 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
845                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
846
847 static const char * const vmstat_text[] = {
848         /* Zoned VM counters */
849         "nr_free_pages",
850         "nr_inactive_anon",
851         "nr_active_anon",
852         "nr_inactive_file",
853         "nr_active_file",
854         "nr_unevictable",
855         "nr_mlock",
856         "nr_anon_pages",
857         "nr_mapped",
858         "nr_file_pages",
859         "nr_dirty",
860         "nr_writeback",
861         "nr_slab_reclaimable",
862         "nr_slab_unreclaimable",
863         "nr_page_table_pages",
864         "nr_kernel_stack",
865         "nr_unstable",
866         "nr_bounce",
867         "nr_vmscan_write",
868         "nr_writeback_temp",
869         "nr_isolated_anon",
870         "nr_isolated_file",
871         "nr_shmem",
872         "nr_dirtied",
873         "nr_written",
874
875 #ifdef CONFIG_NUMA
876         "numa_hit",
877         "numa_miss",
878         "numa_foreign",
879         "numa_interleave",
880         "numa_local",
881         "numa_other",
882 #endif
883         "nr_anon_transparent_hugepages",
884         "nr_dirty_threshold",
885         "nr_dirty_background_threshold",
886
887 #ifdef CONFIG_VM_EVENT_COUNTERS
888         "pgpgin",
889         "pgpgout",
890         "pswpin",
891         "pswpout",
892
893         TEXTS_FOR_ZONES("pgalloc")
894
895         "pgfree",
896         "pgactivate",
897         "pgdeactivate",
898
899         "pgfault",
900         "pgmajfault",
901
902         TEXTS_FOR_ZONES("pgrefill")
903         TEXTS_FOR_ZONES("pgsteal")
904         TEXTS_FOR_ZONES("pgscan_kswapd")
905         TEXTS_FOR_ZONES("pgscan_direct")
906
907 #ifdef CONFIG_NUMA
908         "zone_reclaim_failed",
909 #endif
910         "pginodesteal",
911         "slabs_scanned",
912         "kswapd_steal",
913         "kswapd_inodesteal",
914         "kswapd_low_wmark_hit_quickly",
915         "kswapd_high_wmark_hit_quickly",
916         "kswapd_skip_congestion_wait",
917         "pageoutrun",
918         "allocstall",
919
920         "pgrotated",
921
922 #ifdef CONFIG_COMPACTION
923         "compact_blocks_moved",
924         "compact_pages_moved",
925         "compact_pagemigrate_failed",
926         "compact_stall",
927         "compact_fail",
928         "compact_success",
929 #endif
930
931 #ifdef CONFIG_HUGETLB_PAGE
932         "htlb_buddy_alloc_success",
933         "htlb_buddy_alloc_fail",
934 #endif
935         "unevictable_pgs_culled",
936         "unevictable_pgs_scanned",
937         "unevictable_pgs_rescued",
938         "unevictable_pgs_mlocked",
939         "unevictable_pgs_munlocked",
940         "unevictable_pgs_cleared",
941         "unevictable_pgs_stranded",
942         "unevictable_pgs_mlockfreed",
943 #endif
944 };
945
946 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
947                                                         struct zone *zone)
948 {
949         int i;
950         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
951         seq_printf(m,
952                    "\n  pages free     %lu"
953                    "\n        min      %lu"
954                    "\n        low      %lu"
955                    "\n        high     %lu"
956                    "\n        scanned  %lu"
957                    "\n        spanned  %lu"
958                    "\n        present  %lu",
959                    zone_page_state(zone, NR_FREE_PAGES),
960                    min_wmark_pages(zone),
961                    low_wmark_pages(zone),
962                    high_wmark_pages(zone),
963                    zone->pages_scanned,
964                    zone->spanned_pages,
965                    zone->present_pages);
966
967         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
968                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
969                                 zone_page_state(zone, i));
970
971         seq_printf(m,
972                    "\n        protection: (%lu",
973                    zone->lowmem_reserve[0]);
974         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
975                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
976         seq_printf(m,
977                    ")"
978                    "\n  pagesets");
979         for_each_online_cpu(i) {
980                 struct per_cpu_pageset *pageset;
981
982                 pageset = per_cpu_ptr(zone->pageset, i);
983                 seq_printf(m,
984                            "\n    cpu: %i"
985                            "\n              count: %i"
986                            "\n              high:  %i"
987                            "\n              batch: %i",
988                            i,
989                            pageset->pcp.count,
990                            pageset->pcp.high,
991                            pageset->pcp.batch);
992 #ifdef CONFIG_SMP
993                 seq_printf(m, "\n  vm stats threshold: %d",
994                                 pageset->stat_threshold);
995 #endif
996         }
997         seq_printf(m,
998                    "\n  all_unreclaimable: %u"
999                    "\n  start_pfn:         %lu"
1000                    "\n  inactive_ratio:    %u",
1001                    zone->all_unreclaimable,
1002                    zone->zone_start_pfn,
1003                    zone->inactive_ratio);
1004         seq_putc(m, '\n');
1005 }
1006
1007 /*
1008  * Output information about zones in @pgdat.
1009  */
1010 static int zoneinfo_show(struct seq_file *m, void *arg)
1011 {
1012         pg_data_t *pgdat = (pg_data_t *)arg;
1013         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
1014         return 0;
1015 }
1016
1017 static const struct seq_operations zoneinfo_op = {
1018         .start  = frag_start, /* iterate over all zones. The same as in
1019                                * fragmentation. */
1020         .next   = frag_next,
1021         .stop   = frag_stop,
1022         .show   = zoneinfo_show,
1023 };
1024
1025 static int zoneinfo_open(struct inode *inode, struct file *file)
1026 {
1027         return seq_open(file, &zoneinfo_op);
1028 }
1029
1030 static const struct file_operations proc_zoneinfo_file_operations = {
1031         .open           = zoneinfo_open,
1032         .read           = seq_read,
1033         .llseek         = seq_lseek,
1034         .release        = seq_release,
1035 };
1036
1037 enum writeback_stat_item {
1038         NR_DIRTY_THRESHOLD,
1039         NR_DIRTY_BG_THRESHOLD,
1040         NR_VM_WRITEBACK_STAT_ITEMS,
1041 };
1042
1043 static void *vmstat_start(struct seq_file *m, loff_t *pos)
1044 {
1045         unsigned long *v;
1046         int i, stat_items_size;
1047
1048         if (*pos >= ARRAY_SIZE(vmstat_text))
1049                 return NULL;
1050         stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1051                           NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
1052
1053 #ifdef CONFIG_VM_EVENT_COUNTERS
1054         stat_items_size += sizeof(struct vm_event_state);
1055 #endif
1056
1057         v = kmalloc(stat_items_size, GFP_KERNEL);
1058         m->private = v;
1059         if (!v)
1060                 return ERR_PTR(-ENOMEM);
1061         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1062                 v[i] = global_page_state(i);
1063         v += NR_VM_ZONE_STAT_ITEMS;
1064
1065         global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1066                             v + NR_DIRTY_THRESHOLD);
1067         v += NR_VM_WRITEBACK_STAT_ITEMS;
1068
1069 #ifdef CONFIG_VM_EVENT_COUNTERS
1070         all_vm_events(v);
1071         v[PGPGIN] /= 2;         /* sectors -> kbytes */
1072         v[PGPGOUT] /= 2;
1073 #endif
1074         return (unsigned long *)m->private + *pos;
1075 }
1076
1077 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1078 {
1079         (*pos)++;
1080         if (*pos >= ARRAY_SIZE(vmstat_text))
1081                 return NULL;
1082         return (unsigned long *)m->private + *pos;
1083 }
1084
1085 static int vmstat_show(struct seq_file *m, void *arg)
1086 {
1087         unsigned long *l = arg;
1088         unsigned long off = l - (unsigned long *)m->private;
1089
1090         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1091         return 0;
1092 }
1093
1094 static void vmstat_stop(struct seq_file *m, void *arg)
1095 {
1096         kfree(m->private);
1097         m->private = NULL;
1098 }
1099
1100 static const struct seq_operations vmstat_op = {
1101         .start  = vmstat_start,
1102         .next   = vmstat_next,
1103         .stop   = vmstat_stop,
1104         .show   = vmstat_show,
1105 };
1106
1107 static int vmstat_open(struct inode *inode, struct file *file)
1108 {
1109         return seq_open(file, &vmstat_op);
1110 }
1111
1112 static const struct file_operations proc_vmstat_file_operations = {
1113         .open           = vmstat_open,
1114         .read           = seq_read,
1115         .llseek         = seq_lseek,
1116         .release        = seq_release,
1117 };
1118 #endif /* CONFIG_PROC_FS */
1119
1120 #ifdef CONFIG_SMP
1121 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1122 int sysctl_stat_interval __read_mostly = HZ;
1123
1124 static void vmstat_update(struct work_struct *w)
1125 {
1126         refresh_cpu_vm_stats(smp_processor_id());
1127         schedule_delayed_work(&__get_cpu_var(vmstat_work),
1128                 round_jiffies_relative(sysctl_stat_interval));
1129 }
1130
1131 static void __cpuinit start_cpu_timer(int cpu)
1132 {
1133         struct delayed_work *work = &per_cpu(vmstat_work, cpu);
1134
1135         INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
1136         schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
1137 }
1138
1139 /*
1140  * Use the cpu notifier to insure that the thresholds are recalculated
1141  * when necessary.
1142  */
1143 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
1144                 unsigned long action,
1145                 void *hcpu)
1146 {
1147         long cpu = (long)hcpu;
1148
1149         switch (action) {
1150         case CPU_ONLINE:
1151         case CPU_ONLINE_FROZEN:
1152                 refresh_zone_stat_thresholds();
1153                 start_cpu_timer(cpu);
1154                 node_set_state(cpu_to_node(cpu), N_CPU);
1155                 break;
1156         case CPU_DOWN_PREPARE:
1157         case CPU_DOWN_PREPARE_FROZEN:
1158                 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1159                 per_cpu(vmstat_work, cpu).work.func = NULL;
1160                 break;
1161         case CPU_DOWN_FAILED:
1162         case CPU_DOWN_FAILED_FROZEN:
1163                 start_cpu_timer(cpu);
1164                 break;
1165         case CPU_DEAD:
1166         case CPU_DEAD_FROZEN:
1167                 refresh_zone_stat_thresholds();
1168                 break;
1169         default:
1170                 break;
1171         }
1172         return NOTIFY_OK;
1173 }
1174
1175 static struct notifier_block __cpuinitdata vmstat_notifier =
1176         { &vmstat_cpuup_callback, NULL, 0 };
1177 #endif
1178
1179 static int __init setup_vmstat(void)
1180 {
1181 #ifdef CONFIG_SMP
1182         int cpu;
1183
1184         refresh_zone_stat_thresholds();
1185         register_cpu_notifier(&vmstat_notifier);
1186
1187         for_each_online_cpu(cpu)
1188                 start_cpu_timer(cpu);
1189 #endif
1190 #ifdef CONFIG_PROC_FS
1191         proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1192         proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1193         proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1194         proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1195 #endif
1196         return 0;
1197 }
1198 module_init(setup_vmstat)
1199
1200 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1201 #include <linux/debugfs.h>
1202
1203 static struct dentry *extfrag_debug_root;
1204
1205 /*
1206  * Return an index indicating how much of the available free memory is
1207  * unusable for an allocation of the requested size.
1208  */
1209 static int unusable_free_index(unsigned int order,
1210                                 struct contig_page_info *info)
1211 {
1212         /* No free memory is interpreted as all free memory is unusable */
1213         if (info->free_pages == 0)
1214                 return 1000;
1215
1216         /*
1217          * Index should be a value between 0 and 1. Return a value to 3
1218          * decimal places.
1219          *
1220          * 0 => no fragmentation
1221          * 1 => high fragmentation
1222          */
1223         return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1224
1225 }
1226
1227 static void unusable_show_print(struct seq_file *m,
1228                                         pg_data_t *pgdat, struct zone *zone)
1229 {
1230         unsigned int order;
1231         int index;
1232         struct contig_page_info info;
1233
1234         seq_printf(m, "Node %d, zone %8s ",
1235                                 pgdat->node_id,
1236                                 zone->name);
1237         for (order = 0; order < MAX_ORDER; ++order) {
1238                 fill_contig_page_info(zone, order, &info);
1239                 index = unusable_free_index(order, &info);
1240                 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1241         }
1242
1243         seq_putc(m, '\n');
1244 }
1245
1246 /*
1247  * Display unusable free space index
1248  *
1249  * The unusable free space index measures how much of the available free
1250  * memory cannot be used to satisfy an allocation of a given size and is a
1251  * value between 0 and 1. The higher the value, the more of free memory is
1252  * unusable and by implication, the worse the external fragmentation is. This
1253  * can be expressed as a percentage by multiplying by 100.
1254  */
1255 static int unusable_show(struct seq_file *m, void *arg)
1256 {
1257         pg_data_t *pgdat = (pg_data_t *)arg;
1258
1259         /* check memoryless node */
1260         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
1261                 return 0;
1262
1263         walk_zones_in_node(m, pgdat, unusable_show_print);
1264
1265         return 0;
1266 }
1267
1268 static const struct seq_operations unusable_op = {
1269         .start  = frag_start,
1270         .next   = frag_next,
1271         .stop   = frag_stop,
1272         .show   = unusable_show,
1273 };
1274
1275 static int unusable_open(struct inode *inode, struct file *file)
1276 {
1277         return seq_open(file, &unusable_op);
1278 }
1279
1280 static const struct file_operations unusable_file_ops = {
1281         .open           = unusable_open,
1282         .read           = seq_read,
1283         .llseek         = seq_lseek,
1284         .release        = seq_release,
1285 };
1286
1287 static void extfrag_show_print(struct seq_file *m,
1288                                         pg_data_t *pgdat, struct zone *zone)
1289 {
1290         unsigned int order;
1291         int index;
1292
1293         /* Alloc on stack as interrupts are disabled for zone walk */
1294         struct contig_page_info info;
1295
1296         seq_printf(m, "Node %d, zone %8s ",
1297                                 pgdat->node_id,
1298                                 zone->name);
1299         for (order = 0; order < MAX_ORDER; ++order) {
1300                 fill_contig_page_info(zone, order, &info);
1301                 index = __fragmentation_index(order, &info);
1302                 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1303         }
1304
1305         seq_putc(m, '\n');
1306 }
1307
1308 /*
1309  * Display fragmentation index for orders that allocations would fail for
1310  */
1311 static int extfrag_show(struct seq_file *m, void *arg)
1312 {
1313         pg_data_t *pgdat = (pg_data_t *)arg;
1314
1315         walk_zones_in_node(m, pgdat, extfrag_show_print);
1316
1317         return 0;
1318 }
1319
1320 static const struct seq_operations extfrag_op = {
1321         .start  = frag_start,
1322         .next   = frag_next,
1323         .stop   = frag_stop,
1324         .show   = extfrag_show,
1325 };
1326
1327 static int extfrag_open(struct inode *inode, struct file *file)
1328 {
1329         return seq_open(file, &extfrag_op);
1330 }
1331
1332 static const struct file_operations extfrag_file_ops = {
1333         .open           = extfrag_open,
1334         .read           = seq_read,
1335         .llseek         = seq_lseek,
1336         .release        = seq_release,
1337 };
1338
1339 static int __init extfrag_debug_init(void)
1340 {
1341         extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1342         if (!extfrag_debug_root)
1343                 return -ENOMEM;
1344
1345         if (!debugfs_create_file("unusable_index", 0444,
1346                         extfrag_debug_root, NULL, &unusable_file_ops))
1347                 return -ENOMEM;
1348
1349         if (!debugfs_create_file("extfrag_index", 0444,
1350                         extfrag_debug_root, NULL, &extfrag_file_ops))
1351                 return -ENOMEM;
1352
1353         return 0;
1354 }
1355
1356 module_init(extfrag_debug_init);
1357 #endif