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
  12 #include <linux/mm.h>
  13 #include <linux/err.h>
  14 #include <linux/module.h>
  15 #include <linux/cpu.h>
  16 #include <linux/sched.h>
  17
  18 #ifdef CONFIG_VM_EVENT_COUNTERS
  19 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
  20 EXPORT_PER_CPU_SYMBOL(vm_event_states);
  21
  22 static void sum_vm_events(unsigned long *ret, cpumask_t *cpumask)
  23 {
  24         int cpu = 0;
  25         int i;
  26
  27         memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
  28
  29         cpu = first_cpu(*cpumask);
  30         while (cpu < NR_CPUS) {
  31                 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
  32
  33                 cpu = next_cpu(cpu, *cpumask);
  34
  35                 if (cpu < NR_CPUS)
  36                         prefetch(&per_cpu(vm_event_states, cpu));
  37
  38
  39                 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
  40                         ret[i] += this->event[i];
  41         }
  42 }
  43
  44 /*
  45  * Accumulate the vm event counters across all CPUs.
  46  * The result is unavoidably approximate - it can change
  47  * during and after execution of this function.
  48 */
  49 void all_vm_events(unsigned long *ret)
  50 {
  51         sum_vm_events(ret, &cpu_online_map);
  52 }
  53 EXPORT_SYMBOL_GPL(all_vm_events);
  54
  55 #ifdef CONFIG_HOTPLUG
  56 /*
  57  * Fold the foreign cpu events into our own.
  58  *
  59  * This is adding to the events on one processor
  60  * but keeps the global counts constant.
  61  */
  62 void vm_events_fold_cpu(int cpu)
  63 {
  64         struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
  65         int i;
  66
  67         for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
  68                 count_vm_events(i, fold_state->event[i]);
  69                 fold_state->event[i] = 0;
  70         }
  71 }
  72 #endif /* CONFIG_HOTPLUG */
  73
  74 #endif /* CONFIG_VM_EVENT_COUNTERS */
  75
  76 /*
  77  * Manage combined zone based / global counters
  78  *
  79  * vm_stat contains the global counters
  80  */
  81 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
  82 EXPORT_SYMBOL(vm_stat);
  83
  84 #ifdef CONFIG_SMP
  85
  86 static int calculate_threshold(struct zone *zone)
  87 {
  88         int threshold;
  89         int mem;        /* memory in 128 MB units */
  90
  91         /*
  92          * The threshold scales with the number of processors and the amount
  93          * of memory per zone. More memory means that we can defer updates for
  94          * longer, more processors could lead to more contention.
  95          * fls() is used to have a cheap way of logarithmic scaling.
  96          *
  97          * Some sample thresholds:
  98          *
  99          * Threshold    Processors      (fls)   Zonesize        fls(mem+1)
 100          * ------------------------------------------------------------------
 101          * 8            1               1       0.9-1 GB        4
 102          * 16           2               2       0.9-1 GB        4
 103          * 20           2               2       1-2 GB          5
 104          * 24           2               2       2-4 GB          6
 105          * 28           2               2       4-8 GB          7
 106          * 32           2               2       8-16 GB         8
 107          * 4            2               2       <128M           1
 108          * 30           4               3       2-4 GB          5
 109          * 48           4               3       8-16 GB         8
 110          * 32           8               4       1-2 GB          4
 111          * 32           8               4       0.9-1GB         4
 112          * 10           16              5       <128M           1
 113          * 40           16              5       900M            4
 114          * 70           64              7       2-4 GB          5
 115          * 84           64              7       4-8 GB          6
 116          * 108          512             9       4-8 GB          6
 117          * 125          1024            10      8-16 GB         8
 118          * 125          1024            10      16-32 GB        9
 119          */
 120
 121         mem = zone->present_pages >> (27 - PAGE_SHIFT);
 122
 123         threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
 124
 125         /*
 126          * Maximum threshold is 125
 127          */
 128         threshold = min(125, threshold);
 129
 130         return threshold;
 131 }
 132
 133 /*
 134  * Refresh the thresholds for each zone.
 135  */
 136 static void refresh_zone_stat_thresholds(void)
 137 {
 138         struct zone *zone;
 139         int cpu;
 140         int threshold;
 141
 142         for_each_zone(zone) {
 143
 144                 if (!zone->present_pages)
 145                         continue;
 146
 147                 threshold = calculate_threshold(zone);
 148
 149                 for_each_online_cpu(cpu)
 150                         zone_pcp(zone, cpu)->stat_threshold = threshold;
 151         }
 152 }
 153
 154 /*
 155  * For use when we know that interrupts are disabled.
 156  */
 157 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 158                                 int delta)
 159 {
 160         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
 161         s8 *p = pcp->vm_stat_diff + item;
 162         long x;
 163
 164         x = delta + *p;
 165
 166         if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
 167                 zone_page_state_add(x, zone, item);
 168                 x = 0;
 169         }
 170         *p = x;
 171 }
 172 EXPORT_SYMBOL(__mod_zone_page_state);
 173
 174 /*
 175  * For an unknown interrupt state
 176  */
 177 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 178                                         int delta)
 179 {
 180         unsigned long flags;
 181
 182         local_irq_save(flags);
 183         __mod_zone_page_state(zone, item, delta);
 184         local_irq_restore(flags);
 185 }
 186 EXPORT_SYMBOL(mod_zone_page_state);
 187
 188 /*
 189  * Optimized increment and decrement functions.
 190  *
 191  * These are only for a single page and therefore can take a struct page *
 192  * argument instead of struct zone *. This allows the inclusion of the code
 193  * generated for page_zone(page) into the optimized functions.
 194  *
 195  * No overflow check is necessary and therefore the differential can be
 196  * incremented or decremented in place which may allow the compilers to
 197  * generate better code.
 198  * The increment or decrement is known and therefore one boundary check can
 199  * be omitted.
 200  *
 201  * NOTE: These functions are very performance sensitive. Change only
 202  * with care.
 203  *
 204  * Some processors have inc/dec instructions that are atomic vs an interrupt.
 205  * However, the code must first determine the differential location in a zone
 206  * based on the processor number and then inc/dec the counter. There is no
 207  * guarantee without disabling preemption that the processor will not change
 208  * in between and therefore the atomicity vs. interrupt cannot be exploited
 209  * in a useful way here.
 210  */
 211 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
 212 {
 213         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
 214         s8 *p = pcp->vm_stat_diff + item;
 215
 216         (*p)++;
 217
 218         if (unlikely(*p > pcp->stat_threshold)) {
 219                 int overstep = pcp->stat_threshold / 2;
 220
 221                 zone_page_state_add(*p + overstep, zone, item);
 222                 *p = -overstep;
 223         }
 224 }
 225
 226 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
 227 {
 228         __inc_zone_state(page_zone(page), item);
 229 }
 230 EXPORT_SYMBOL(__inc_zone_page_state);
 231
 232 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
 233 {
 234         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
 235         s8 *p = pcp->vm_stat_diff + item;
 236
 237         (*p)--;
 238
 239         if (unlikely(*p < - pcp->stat_threshold)) {
 240                 int overstep = pcp->stat_threshold / 2;
 241
 242                 zone_page_state_add(*p - overstep, zone, item);
 243                 *p = overstep;
 244         }
 245 }
 246
 247 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
 248 {
 249         __dec_zone_state(page_zone(page), item);
 250 }
 251 EXPORT_SYMBOL(__dec_zone_page_state);
 252
 253 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
 254 {
 255         unsigned long flags;
 256
 257         local_irq_save(flags);
 258         __inc_zone_state(zone, item);
 259         local_irq_restore(flags);
 260 }
 261
 262 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
 263 {
 264         unsigned long flags;
 265         struct zone *zone;
 266
 267         zone = page_zone(page);
 268         local_irq_save(flags);
 269         __inc_zone_state(zone, item);
 270         local_irq_restore(flags);
 271 }
 272 EXPORT_SYMBOL(inc_zone_page_state);
 273
 274 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
 275 {
 276         unsigned long flags;
 277
 278         local_irq_save(flags);
 279         __dec_zone_page_state(page, item);
 280         local_irq_restore(flags);
 281 }
 282 EXPORT_SYMBOL(dec_zone_page_state);
 283
 284 /*
 285  * Update the zone counters for one cpu.
 286  *
 287  * The cpu specified must be either the current cpu or a processor that
 288  * is not online. If it is the current cpu then the execution thread must
 289  * be pinned to the current cpu.
 290  *
 291  * Note that refresh_cpu_vm_stats strives to only access
 292  * node local memory. The per cpu pagesets on remote zones are placed
 293  * in the memory local to the processor using that pageset. So the
 294  * loop over all zones will access a series of cachelines local to
 295  * the processor.
 296  *
 297  * The call to zone_page_state_add updates the cachelines with the
 298  * statistics in the remote zone struct as well as the global cachelines
 299  * with the global counters. These could cause remote node cache line
 300  * bouncing and will have to be only done when necessary.
 301  */
 302 void refresh_cpu_vm_stats(int cpu)
 303 {
 304         struct zone *zone;
 305         int i;
 306         int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
 307
 308         for_each_zone(zone) {
 309                 struct per_cpu_pageset *p;
 310
 311                 if (!populated_zone(zone))
 312                         continue;
 313
 314                 p = zone_pcp(zone, cpu);
 315
 316                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 317                         if (p->vm_stat_diff[i]) {
 318                                 unsigned long flags;
 319                                 int v;
 320
 321                                 local_irq_save(flags);
 322                                 v = p->vm_stat_diff[i];
 323                                 p->vm_stat_diff[i] = 0;
 324                                 local_irq_restore(flags);
 325                                 atomic_long_add(v, &zone->vm_stat[i]);
 326                                 global_diff[i] += v;
 327 #ifdef CONFIG_NUMA
 328                                 /* 3 seconds idle till flush */
 329                                 p->expire = 3;
 330 #endif
 331                         }
 332 #ifdef CONFIG_NUMA
 333                 /*
 334                  * Deal with draining the remote pageset of this
 335                  * processor
 336                  *
 337                  * Check if there are pages remaining in this pageset
 338                  * if not then there is nothing to expire.
 339                  */
 340                 if (!p->expire || (!p->pcp[0].count && !p->pcp[1].count))
 341                         continue;
 342
 343                 /*
 344                  * We never drain zones local to this processor.
 345                  */
 346                 if (zone_to_nid(zone) == numa_node_id()) {
 347                         p->expire = 0;
 348                         continue;
 349                 }
 350
 351                 p->expire--;
 352                 if (p->expire)
 353                         continue;
 354
 355                 if (p->pcp[0].count)
 356                         drain_zone_pages(zone, p->pcp + 0);
 357
 358                 if (p->pcp[1].count)
 359                         drain_zone_pages(zone, p->pcp + 1);
 360 #endif
 361         }
 362
 363         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 364                 if (global_diff[i])
 365                         atomic_long_add(global_diff[i], &vm_stat[i]);
 366 }
 367
 368 #endif
 369
 370 #ifdef CONFIG_NUMA
 371 /*
 372  * zonelist = the list of zones passed to the allocator
 373  * z        = the zone from which the allocation occurred.
 374  *
 375  * Must be called with interrupts disabled.
 376  */
 377 void zone_statistics(struct zonelist *zonelist, struct zone *z)
 378 {
 379         if (z->zone_pgdat == zonelist->zones[0]->zone_pgdat) {
 380                 __inc_zone_state(z, NUMA_HIT);
 381         } else {
 382                 __inc_zone_state(z, NUMA_MISS);
 383                 __inc_zone_state(zonelist->zones[0], NUMA_FOREIGN);
 384         }
 385         if (z->node == numa_node_id())
 386                 __inc_zone_state(z, NUMA_LOCAL);
 387         else
 388                 __inc_zone_state(z, NUMA_OTHER);
 389 }
 390 #endif
 391
 392 #ifdef CONFIG_PROC_FS
 393
 394 #include <linux/seq_file.h>
 395
 396 static char * const migratetype_names[MIGRATE_TYPES] = {
 397         "Unmovable",
 398         "Reclaimable",
 399         "Movable",
 400         "Reserve",
 401 };
 402
 403 static void *frag_start(struct seq_file *m, loff_t *pos)
 404 {
 405         pg_data_t *pgdat;
 406         loff_t node = *pos;
 407         for (pgdat = first_online_pgdat();
 408              pgdat && node;
 409              pgdat = next_online_pgdat(pgdat))
 410                 --node;
 411
 412         return pgdat;
 413 }
 414
 415 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
 416 {
 417         pg_data_t *pgdat = (pg_data_t *)arg;
 418
 419         (*pos)++;
 420         return next_online_pgdat(pgdat);
 421 }
 422
 423 static void frag_stop(struct seq_file *m, void *arg)
 424 {
 425 }
 426
 427 /* Walk all the zones in a node and print using a callback */
 428 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
 429                 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
 430 {
 431         struct zone *zone;
 432         struct zone *node_zones = pgdat->node_zones;
 433         unsigned long flags;
 434
 435         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
 436                 if (!populated_zone(zone))
 437                         continue;
 438
 439                 spin_lock_irqsave(&zone->lock, flags);
 440                 print(m, pgdat, zone);
 441                 spin_unlock_irqrestore(&zone->lock, flags);
 442         }
 443 }
 444
 445 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
 446                                                 struct zone *zone)
 447 {
 448         int order;
 449
 450         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 451         for (order = 0; order < MAX_ORDER; ++order)
 452                 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
 453         seq_putc(m, '\n');
 454 }
 455
 456 /*
 457  * This walks the free areas for each zone.
 458  */
 459 static int frag_show(struct seq_file *m, void *arg)
 460 {
 461         pg_data_t *pgdat = (pg_data_t *)arg;
 462         walk_zones_in_node(m, pgdat, frag_show_print);
 463         return 0;
 464 }
 465
 466 static void pagetypeinfo_showfree_print(struct seq_file *m,
 467                                         pg_data_t *pgdat, struct zone *zone)
 468 {
 469         int order, mtype;
 470
 471         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
 472                 seq_printf(m, "Node %4d, zone %8s, type %12s ",
 473                                         pgdat->node_id,
 474                                         zone->name,
 475                                         migratetype_names[mtype]);
 476                 for (order = 0; order < MAX_ORDER; ++order) {
 477                         unsigned long freecount = 0;
 478                         struct free_area *area;
 479                         struct list_head *curr;
 480
 481                         area = &(zone->free_area[order]);
 482
 483                         list_for_each(curr, &area->free_list[mtype])
 484                                 freecount++;
 485                         seq_printf(m, "%6lu ", freecount);
 486                 }
 487                 seq_putc(m, '\n');
 488         }
 489 }
 490
 491 /* Print out the free pages at each order for each migatetype */
 492 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
 493 {
 494         int order;
 495         pg_data_t *pgdat = (pg_data_t *)arg;
 496
 497         /* Print header */
 498         seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
 499         for (order = 0; order < MAX_ORDER; ++order)
 500                 seq_printf(m, "%6d ", order);
 501         seq_putc(m, '\n');
 502
 503         walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
 504
 505         return 0;
 506 }
 507
 508 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
 509                                         pg_data_t *pgdat, struct zone *zone)
 510 {
 511         int mtype;
 512         unsigned long pfn;
 513         unsigned long start_pfn = zone->zone_start_pfn;
 514         unsigned long end_pfn = start_pfn + zone->spanned_pages;
 515         unsigned long count[MIGRATE_TYPES] = { 0, };
 516
 517         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
 518                 struct page *page;
 519
 520                 if (!pfn_valid(pfn))
 521                         continue;
 522
 523                 page = pfn_to_page(pfn);
 524                 mtype = get_pageblock_migratetype(page);
 525
 526                 count[mtype]++;
 527         }
 528
 529         /* Print counts */
 530         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 531         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 532                 seq_printf(m, "%12lu ", count[mtype]);
 533         seq_putc(m, '\n');
 534 }
 535
 536 /* Print out the free pages at each order for each migratetype */
 537 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
 538 {
 539         int mtype;
 540         pg_data_t *pgdat = (pg_data_t *)arg;
 541
 542         seq_printf(m, "\n%-23s", "Number of blocks type ");
 543         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 544                 seq_printf(m, "%12s ", migratetype_names[mtype]);
 545         seq_putc(m, '\n');
 546         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
 547
 548         return 0;
 549 }
 550
 551 /*
 552  * This prints out statistics in relation to grouping pages by mobility.
 553  * It is expensive to collect so do not constantly read the file.
 554  */
 555 static int pagetypeinfo_show(struct seq_file *m, void *arg)
 556 {
 557         pg_data_t *pgdat = (pg_data_t *)arg;
 558
 559         seq_printf(m, "Page block order: %d\n", pageblock_order);
 560         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
 561         seq_putc(m, '\n');
 562         pagetypeinfo_showfree(m, pgdat);
 563         pagetypeinfo_showblockcount(m, pgdat);
 564
 565         return 0;
 566 }
 567
 568 const struct seq_operations fragmentation_op = {
 569         .start  = frag_start,
 570         .next   = frag_next,
 571         .stop   = frag_stop,
 572         .show   = frag_show,
 573 };
 574
 575 const struct seq_operations pagetypeinfo_op = {
 576         .start  = frag_start,
 577         .next   = frag_next,
 578         .stop   = frag_stop,
 579         .show   = pagetypeinfo_show,
 580 };
 581
 582 #ifdef CONFIG_ZONE_DMA
 583 #define TEXT_FOR_DMA(xx) xx "_dma",
 584 #else
 585 #define TEXT_FOR_DMA(xx)
 586 #endif
 587
 588 #ifdef CONFIG_ZONE_DMA32
 589 #define TEXT_FOR_DMA32(xx) xx "_dma32",
 590 #else
 591 #define TEXT_FOR_DMA32(xx)
 592 #endif
 593
 594 #ifdef CONFIG_HIGHMEM
 595 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
 596 #else
 597 #define TEXT_FOR_HIGHMEM(xx)
 598 #endif
 599
 600 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
 601                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
 602
 603 static const char * const vmstat_text[] = {
 604         /* Zoned VM counters */
 605         "nr_free_pages",
 606         "nr_inactive",
 607         "nr_active",
 608         "nr_anon_pages",
 609         "nr_mapped",
 610         "nr_file_pages",
 611         "nr_dirty",
 612         "nr_writeback",
 613         "nr_slab_reclaimable",
 614         "nr_slab_unreclaimable",
 615         "nr_page_table_pages",
 616         "nr_unstable",
 617         "nr_bounce",
 618         "nr_vmscan_write",
 619
 620 #ifdef CONFIG_NUMA
 621         "numa_hit",
 622         "numa_miss",
 623         "numa_foreign",
 624         "numa_interleave",
 625         "numa_local",
 626         "numa_other",
 627 #endif
 628
 629 #ifdef CONFIG_VM_EVENT_COUNTERS
 630         "pgpgin",
 631         "pgpgout",
 632         "pswpin",
 633         "pswpout",
 634
 635         TEXTS_FOR_ZONES("pgalloc")
 636
 637         "pgfree",
 638         "pgactivate",
 639         "pgdeactivate",
 640
 641         "pgfault",
 642         "pgmajfault",
 643
 644         TEXTS_FOR_ZONES("pgrefill")
 645         TEXTS_FOR_ZONES("pgsteal")
 646         TEXTS_FOR_ZONES("pgscan_kswapd")
 647         TEXTS_FOR_ZONES("pgscan_direct")
 648
 649         "pginodesteal",
 650         "slabs_scanned",
 651         "kswapd_steal",
 652         "kswapd_inodesteal",
 653         "pageoutrun",
 654         "allocstall",
 655
 656         "pgrotated",
 657 #endif
 658 };
 659
 660 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
 661                                                         struct zone *zone)
 662 {
 663         int i;
 664         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
 665         seq_printf(m,
 666                    "\n  pages free     %lu"
 667                    "\n        min      %lu"
 668                    "\n        low      %lu"
 669                    "\n        high     %lu"
 670                    "\n        scanned  %lu (a: %lu i: %lu)"
 671                    "\n        spanned  %lu"
 672                    "\n        present  %lu",
 673                    zone_page_state(zone, NR_FREE_PAGES),
 674                    zone->pages_min,
 675                    zone->pages_low,
 676                    zone->pages_high,
 677                    zone->pages_scanned,
 678                    zone->nr_scan_active, zone->nr_scan_inactive,
 679                    zone->spanned_pages,
 680                    zone->present_pages);
 681
 682         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 683                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
 684                                 zone_page_state(zone, i));
 685
 686         seq_printf(m,
 687                    "\n        protection: (%lu",
 688                    zone->lowmem_reserve[0]);
 689         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
 690                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
 691         seq_printf(m,
 692                    ")"
 693                    "\n  pagesets");
 694         for_each_online_cpu(i) {
 695                 struct per_cpu_pageset *pageset;
 696                 int j;
 697
 698                 pageset = zone_pcp(zone, i);
 699                 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
 700                         seq_printf(m,
 701                                    "\n    cpu: %i pcp: %i"
 702                                    "\n              count: %i"
 703                                    "\n              high:  %i"
 704                                    "\n              batch: %i",
 705                                    i, j,
 706                                    pageset->pcp[j].count,
 707                                    pageset->pcp[j].high,
 708                                    pageset->pcp[j].batch);
 709                         }
 710 #ifdef CONFIG_SMP
 711                 seq_printf(m, "\n  vm stats threshold: %d",
 712                                 pageset->stat_threshold);
 713 #endif
 714         }
 715         seq_printf(m,
 716                    "\n  all_unreclaimable: %u"
 717                    "\n  prev_priority:     %i"
 718                    "\n  start_pfn:         %lu",
 719                            zone_is_all_unreclaimable(zone),
 720                    zone->prev_priority,
 721                    zone->zone_start_pfn);
 722         seq_putc(m, '\n');
 723 }
 724
 725 /*
 726  * Output information about zones in @pgdat.
 727  */
 728 static int zoneinfo_show(struct seq_file *m, void *arg)
 729 {
 730         pg_data_t *pgdat = (pg_data_t *)arg;
 731         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
 732         return 0;
 733 }
 734
 735 const struct seq_operations zoneinfo_op = {
 736         .start  = frag_start, /* iterate over all zones. The same as in
 737                                * fragmentation. */
 738         .next   = frag_next,
 739         .stop   = frag_stop,
 740         .show   = zoneinfo_show,
 741 };
 742
 743 static void *vmstat_start(struct seq_file *m, loff_t *pos)
 744 {
 745         unsigned long *v;
 746 #ifdef CONFIG_VM_EVENT_COUNTERS
 747         unsigned long *e;
 748 #endif
 749         int i;
 750
 751         if (*pos >= ARRAY_SIZE(vmstat_text))
 752                 return NULL;
 753
 754 #ifdef CONFIG_VM_EVENT_COUNTERS
 755         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
 756                         + sizeof(struct vm_event_state), GFP_KERNEL);
 757 #else
 758         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
 759                         GFP_KERNEL);
 760 #endif
 761         m->private = v;
 762         if (!v)
 763                 return ERR_PTR(-ENOMEM);
 764         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 765                 v[i] = global_page_state(i);
 766 #ifdef CONFIG_VM_EVENT_COUNTERS
 767         e = v + NR_VM_ZONE_STAT_ITEMS;
 768         all_vm_events(e);
 769         e[PGPGIN] /= 2;         /* sectors -> kbytes */
 770         e[PGPGOUT] /= 2;
 771 #endif
 772         return v + *pos;
 773 }
 774
 775 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
 776 {
 777         (*pos)++;
 778         if (*pos >= ARRAY_SIZE(vmstat_text))
 779                 return NULL;
 780         return (unsigned long *)m->private + *pos;
 781 }
 782
 783 static int vmstat_show(struct seq_file *m, void *arg)
 784 {
 785         unsigned long *l = arg;
 786         unsigned long off = l - (unsigned long *)m->private;
 787
 788         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
 789         return 0;
 790 }
 791
 792 static void vmstat_stop(struct seq_file *m, void *arg)
 793 {
 794         kfree(m->private);
 795         m->private = NULL;
 796 }
 797
 798 const struct seq_operations vmstat_op = {
 799         .start  = vmstat_start,
 800         .next   = vmstat_next,
 801         .stop   = vmstat_stop,
 802         .show   = vmstat_show,
 803 };
 804
 805 #endif /* CONFIG_PROC_FS */
 806
 807 #ifdef CONFIG_SMP
 808 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
 809 int sysctl_stat_interval __read_mostly = HZ;
 810
 811 static void vmstat_update(struct work_struct *w)
 812 {
 813         refresh_cpu_vm_stats(smp_processor_id());
 814         schedule_delayed_work(&__get_cpu_var(vmstat_work),
 815                 sysctl_stat_interval);
 816 }
 817
 818 static void __cpuinit start_cpu_timer(int cpu)
 819 {
 820         struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
 821
 822         INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
 823         schedule_delayed_work_on(cpu, vmstat_work, HZ + cpu);
 824 }
 825
 826 /*
 827  * Use the cpu notifier to insure that the thresholds are recalculated
 828  * when necessary.
 829  */
 830 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
 831                 unsigned long action,
 832                 void *hcpu)
 833 {
 834         long cpu = (long)hcpu;
 835
 836         switch (action) {
 837         case CPU_ONLINE:
 838         case CPU_ONLINE_FROZEN:
 839                 start_cpu_timer(cpu);
 840                 break;
 841         case CPU_DOWN_PREPARE:
 842         case CPU_DOWN_PREPARE_FROZEN:
 843                 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
 844                 per_cpu(vmstat_work, cpu).work.func = NULL;
 845                 break;
 846         case CPU_DOWN_FAILED:
 847         case CPU_DOWN_FAILED_FROZEN:
 848                 start_cpu_timer(cpu);
 849                 break;
 850         case CPU_DEAD:
 851         case CPU_DEAD_FROZEN:
 852                 refresh_zone_stat_thresholds();
 853                 break;
 854         default:
 855                 break;
 856         }
 857         return NOTIFY_OK;
 858 }
 859
 860 static struct notifier_block __cpuinitdata vmstat_notifier =
 861         { &vmstat_cpuup_callback, NULL, 0 };
 862
 863 static int __init setup_vmstat(void)
 864 {
 865         int cpu;
 866
 867         refresh_zone_stat_thresholds();
 868         register_cpu_notifier(&vmstat_notifier);
 869
 870         for_each_online_cpu(cpu)
 871                 start_cpu_timer(cpu);
 872         return 0;
 873 }
 874 module_init(setup_vmstat)
 875 #endif