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  * Note that refresh_cpu_vm_stats strives to only access
 288  * node local memory. The per cpu pagesets on remote zones are placed
 289  * in the memory local to the processor using that pageset. So the
 290  * loop over all zones will access a series of cachelines local to
 291  * the processor.
 292  *
 293  * The call to zone_page_state_add updates the cachelines with the
 294  * statistics in the remote zone struct as well as the global cachelines
 295  * with the global counters. These could cause remote node cache line
 296  * bouncing and will have to be only done when necessary.
 297  */
 298 void refresh_cpu_vm_stats(int cpu)
 299 {
 300         struct zone *zone;
 301         int i;
 302         unsigned long flags;
 303
 304         for_each_zone(zone) {
 305                 struct per_cpu_pageset *p;
 306
 307                 if (!populated_zone(zone))
 308                         continue;
 309
 310                 p = zone_pcp(zone, cpu);
 311
 312                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 313                         if (p->vm_stat_diff[i]) {
 314                                 local_irq_save(flags);
 315                                 zone_page_state_add(p->vm_stat_diff[i],
 316                                         zone, i);
 317                                 p->vm_stat_diff[i] = 0;
 318 #ifdef CONFIG_NUMA
 319                                 /* 3 seconds idle till flush */
 320                                 p->expire = 3;
 321 #endif
 322                                 local_irq_restore(flags);
 323                         }
 324 #ifdef CONFIG_NUMA
 325                 /*
 326                  * Deal with draining the remote pageset of this
 327                  * processor
 328                  *
 329                  * Check if there are pages remaining in this pageset
 330                  * if not then there is nothing to expire.
 331                  */
 332                 if (!p->expire || (!p->pcp[0].count && !p->pcp[1].count))
 333                         continue;
 334
 335                 /*
 336                  * We never drain zones local to this processor.
 337                  */
 338                 if (zone_to_nid(zone) == numa_node_id()) {
 339                         p->expire = 0;
 340                         continue;
 341                 }
 342
 343                 p->expire--;
 344                 if (p->expire)
 345                         continue;
 346
 347                 if (p->pcp[0].count)
 348                         drain_zone_pages(zone, p->pcp + 0);
 349
 350                 if (p->pcp[1].count)
 351                         drain_zone_pages(zone, p->pcp + 1);
 352 #endif
 353         }
 354 }
 355
 356 static void __refresh_cpu_vm_stats(void *dummy)
 357 {
 358         refresh_cpu_vm_stats(smp_processor_id());
 359 }
 360
 361 /*
 362  * Consolidate all counters.
 363  *
 364  * Note that the result is less inaccurate but still inaccurate
 365  * if concurrent processes are allowed to run.
 366  */
 367 void refresh_vm_stats(void)
 368 {
 369         on_each_cpu(__refresh_cpu_vm_stats, NULL, 0, 1);
 370 }
 371 EXPORT_SYMBOL(refresh_vm_stats);
 372
 373 #endif
 374
 375 #ifdef CONFIG_NUMA
 376 /*
 377  * zonelist = the list of zones passed to the allocator
 378  * z        = the zone from which the allocation occurred.
 379  *
 380  * Must be called with interrupts disabled.
 381  */
 382 void zone_statistics(struct zonelist *zonelist, struct zone *z)
 383 {
 384         if (z->zone_pgdat == zonelist->zones[0]->zone_pgdat) {
 385                 __inc_zone_state(z, NUMA_HIT);
 386         } else {
 387                 __inc_zone_state(z, NUMA_MISS);
 388                 __inc_zone_state(zonelist->zones[0], NUMA_FOREIGN);
 389         }
 390         if (z->node == numa_node_id())
 391                 __inc_zone_state(z, NUMA_LOCAL);
 392         else
 393                 __inc_zone_state(z, NUMA_OTHER);
 394 }
 395 #endif
 396
 397 #ifdef CONFIG_PROC_FS
 398
 399 #include <linux/seq_file.h>
 400
 401 static char * const migratetype_names[MIGRATE_TYPES] = {
 402         "Unmovable",
 403         "Reclaimable",
 404         "Movable",
 405         "Reserve",
 406 };
 407
 408 static void *frag_start(struct seq_file *m, loff_t *pos)
 409 {
 410         pg_data_t *pgdat;
 411         loff_t node = *pos;
 412         for (pgdat = first_online_pgdat();
 413              pgdat && node;
 414              pgdat = next_online_pgdat(pgdat))
 415                 --node;
 416
 417         return pgdat;
 418 }
 419
 420 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
 421 {
 422         pg_data_t *pgdat = (pg_data_t *)arg;
 423
 424         (*pos)++;
 425         return next_online_pgdat(pgdat);
 426 }
 427
 428 static void frag_stop(struct seq_file *m, void *arg)
 429 {
 430 }
 431
 432 /* Walk all the zones in a node and print using a callback */
 433 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
 434                 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
 435 {
 436         struct zone *zone;
 437         struct zone *node_zones = pgdat->node_zones;
 438         unsigned long flags;
 439
 440         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
 441                 if (!populated_zone(zone))
 442                         continue;
 443
 444                 spin_lock_irqsave(&zone->lock, flags);
 445                 print(m, pgdat, zone);
 446                 spin_unlock_irqrestore(&zone->lock, flags);
 447         }
 448 }
 449
 450 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
 451                                                 struct zone *zone)
 452 {
 453         int order;
 454
 455         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 456         for (order = 0; order < MAX_ORDER; ++order)
 457                 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
 458         seq_putc(m, '\n');
 459 }
 460
 461 /*
 462  * This walks the free areas for each zone.
 463  */
 464 static int frag_show(struct seq_file *m, void *arg)
 465 {
 466         pg_data_t *pgdat = (pg_data_t *)arg;
 467         walk_zones_in_node(m, pgdat, frag_show_print);
 468         return 0;
 469 }
 470
 471 static void pagetypeinfo_showfree_print(struct seq_file *m,
 472                                         pg_data_t *pgdat, struct zone *zone)
 473 {
 474         int order, mtype;
 475
 476         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
 477                 seq_printf(m, "Node %4d, zone %8s, type %12s ",
 478                                         pgdat->node_id,
 479                                         zone->name,
 480                                         migratetype_names[mtype]);
 481                 for (order = 0; order < MAX_ORDER; ++order) {
 482                         unsigned long freecount = 0;
 483                         struct free_area *area;
 484                         struct list_head *curr;
 485
 486                         area = &(zone->free_area[order]);
 487
 488                         list_for_each(curr, &area->free_list[mtype])
 489                                 freecount++;
 490                         seq_printf(m, "%6lu ", freecount);
 491                 }
 492                 seq_putc(m, '\n');
 493         }
 494 }
 495
 496 /* Print out the free pages at each order for each migatetype */
 497 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
 498 {
 499         int order;
 500         pg_data_t *pgdat = (pg_data_t *)arg;
 501
 502         /* Print header */
 503         seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
 504         for (order = 0; order < MAX_ORDER; ++order)
 505                 seq_printf(m, "%6d ", order);
 506         seq_putc(m, '\n');
 507
 508         walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
 509
 510         return 0;
 511 }
 512
 513 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
 514                                         pg_data_t *pgdat, struct zone *zone)
 515 {
 516         int mtype;
 517         unsigned long pfn;
 518         unsigned long start_pfn = zone->zone_start_pfn;
 519         unsigned long end_pfn = start_pfn + zone->spanned_pages;
 520         unsigned long count[MIGRATE_TYPES] = { 0, };
 521
 522         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
 523                 struct page *page;
 524
 525                 if (!pfn_valid(pfn))
 526                         continue;
 527
 528                 page = pfn_to_page(pfn);
 529                 mtype = get_pageblock_migratetype(page);
 530
 531                 count[mtype]++;
 532         }
 533
 534         /* Print counts */
 535         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 536         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 537                 seq_printf(m, "%12lu ", count[mtype]);
 538         seq_putc(m, '\n');
 539 }
 540
 541 /* Print out the free pages at each order for each migratetype */
 542 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
 543 {
 544         int mtype;
 545         pg_data_t *pgdat = (pg_data_t *)arg;
 546
 547         seq_printf(m, "\n%-23s", "Number of blocks type ");
 548         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 549                 seq_printf(m, "%12s ", migratetype_names[mtype]);
 550         seq_putc(m, '\n');
 551         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
 552
 553         return 0;
 554 }
 555
 556 /*
 557  * This prints out statistics in relation to grouping pages by mobility.
 558  * It is expensive to collect so do not constantly read the file.
 559  */
 560 static int pagetypeinfo_show(struct seq_file *m, void *arg)
 561 {
 562         pg_data_t *pgdat = (pg_data_t *)arg;
 563
 564         seq_printf(m, "Page block order: %d\n", pageblock_order);
 565         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
 566         seq_putc(m, '\n');
 567         pagetypeinfo_showfree(m, pgdat);
 568         pagetypeinfo_showblockcount(m, pgdat);
 569
 570         return 0;
 571 }
 572
 573 const struct seq_operations fragmentation_op = {
 574         .start  = frag_start,
 575         .next   = frag_next,
 576         .stop   = frag_stop,
 577         .show   = frag_show,
 578 };
 579
 580 const struct seq_operations pagetypeinfo_op = {
 581         .start  = frag_start,
 582         .next   = frag_next,
 583         .stop   = frag_stop,
 584         .show   = pagetypeinfo_show,
 585 };
 586
 587 #ifdef CONFIG_ZONE_DMA
 588 #define TEXT_FOR_DMA(xx) xx "_dma",
 589 #else
 590 #define TEXT_FOR_DMA(xx)
 591 #endif
 592
 593 #ifdef CONFIG_ZONE_DMA32
 594 #define TEXT_FOR_DMA32(xx) xx "_dma32",
 595 #else
 596 #define TEXT_FOR_DMA32(xx)
 597 #endif
 598
 599 #ifdef CONFIG_HIGHMEM
 600 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
 601 #else
 602 #define TEXT_FOR_HIGHMEM(xx)
 603 #endif
 604
 605 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
 606                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
 607
 608 static const char * const vmstat_text[] = {
 609         /* Zoned VM counters */
 610         "nr_free_pages",
 611         "nr_inactive",
 612         "nr_active",
 613         "nr_anon_pages",
 614         "nr_mapped",
 615         "nr_file_pages",
 616         "nr_dirty",
 617         "nr_writeback",
 618         "nr_slab_reclaimable",
 619         "nr_slab_unreclaimable",
 620         "nr_page_table_pages",
 621         "nr_unstable",
 622         "nr_bounce",
 623         "nr_vmscan_write",
 624
 625 #ifdef CONFIG_NUMA
 626         "numa_hit",
 627         "numa_miss",
 628         "numa_foreign",
 629         "numa_interleave",
 630         "numa_local",
 631         "numa_other",
 632 #endif
 633
 634 #ifdef CONFIG_VM_EVENT_COUNTERS
 635         "pgpgin",
 636         "pgpgout",
 637         "pswpin",
 638         "pswpout",
 639
 640         TEXTS_FOR_ZONES("pgalloc")
 641
 642         "pgfree",
 643         "pgactivate",
 644         "pgdeactivate",
 645
 646         "pgfault",
 647         "pgmajfault",
 648
 649         TEXTS_FOR_ZONES("pgrefill")
 650         TEXTS_FOR_ZONES("pgsteal")
 651         TEXTS_FOR_ZONES("pgscan_kswapd")
 652         TEXTS_FOR_ZONES("pgscan_direct")
 653
 654         "pginodesteal",
 655         "slabs_scanned",
 656         "kswapd_steal",
 657         "kswapd_inodesteal",
 658         "pageoutrun",
 659         "allocstall",
 660
 661         "pgrotated",
 662 #endif
 663 };
 664
 665 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
 666                                                         struct zone *zone)
 667 {
 668         int i;
 669         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
 670         seq_printf(m,
 671                    "\n  pages free     %lu"
 672                    "\n        min      %lu"
 673                    "\n        low      %lu"
 674                    "\n        high     %lu"
 675                    "\n        scanned  %lu (a: %lu i: %lu)"
 676                    "\n        spanned  %lu"
 677                    "\n        present  %lu",
 678                    zone_page_state(zone, NR_FREE_PAGES),
 679                    zone->pages_min,
 680                    zone->pages_low,
 681                    zone->pages_high,
 682                    zone->pages_scanned,
 683                    zone->nr_scan_active, zone->nr_scan_inactive,
 684                    zone->spanned_pages,
 685                    zone->present_pages);
 686
 687         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 688                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
 689                                 zone_page_state(zone, i));
 690
 691         seq_printf(m,
 692                    "\n        protection: (%lu",
 693                    zone->lowmem_reserve[0]);
 694         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
 695                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
 696         seq_printf(m,
 697                    ")"
 698                    "\n  pagesets");
 699         for_each_online_cpu(i) {
 700                 struct per_cpu_pageset *pageset;
 701                 int j;
 702
 703                 pageset = zone_pcp(zone, i);
 704                 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
 705                         seq_printf(m,
 706                                    "\n    cpu: %i pcp: %i"
 707                                    "\n              count: %i"
 708                                    "\n              high:  %i"
 709                                    "\n              batch: %i",
 710                                    i, j,
 711                                    pageset->pcp[j].count,
 712                                    pageset->pcp[j].high,
 713                                    pageset->pcp[j].batch);
 714                         }
 715 #ifdef CONFIG_SMP
 716                 seq_printf(m, "\n  vm stats threshold: %d",
 717                                 pageset->stat_threshold);
 718 #endif
 719         }
 720         seq_printf(m,
 721                    "\n  all_unreclaimable: %u"
 722                    "\n  prev_priority:     %i"
 723                    "\n  start_pfn:         %lu",
 724                    zone->all_unreclaimable,
 725                    zone->prev_priority,
 726                    zone->zone_start_pfn);
 727         seq_putc(m, '\n');
 728 }
 729
 730 /*
 731  * Output information about zones in @pgdat.
 732  */
 733 static int zoneinfo_show(struct seq_file *m, void *arg)
 734 {
 735         pg_data_t *pgdat = (pg_data_t *)arg;
 736         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
 737         return 0;
 738 }
 739
 740 const struct seq_operations zoneinfo_op = {
 741         .start  = frag_start, /* iterate over all zones. The same as in
 742                                * fragmentation. */
 743         .next   = frag_next,
 744         .stop   = frag_stop,
 745         .show   = zoneinfo_show,
 746 };
 747
 748 static void *vmstat_start(struct seq_file *m, loff_t *pos)
 749 {
 750         unsigned long *v;
 751 #ifdef CONFIG_VM_EVENT_COUNTERS
 752         unsigned long *e;
 753 #endif
 754         int i;
 755
 756         if (*pos >= ARRAY_SIZE(vmstat_text))
 757                 return NULL;
 758
 759 #ifdef CONFIG_VM_EVENT_COUNTERS
 760         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
 761                         + sizeof(struct vm_event_state), GFP_KERNEL);
 762 #else
 763         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
 764                         GFP_KERNEL);
 765 #endif
 766         m->private = v;
 767         if (!v)
 768                 return ERR_PTR(-ENOMEM);
 769         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 770                 v[i] = global_page_state(i);
 771 #ifdef CONFIG_VM_EVENT_COUNTERS
 772         e = v + NR_VM_ZONE_STAT_ITEMS;
 773         all_vm_events(e);
 774         e[PGPGIN] /= 2;         /* sectors -> kbytes */
 775         e[PGPGOUT] /= 2;
 776 #endif
 777         return v + *pos;
 778 }
 779
 780 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
 781 {
 782         (*pos)++;
 783         if (*pos >= ARRAY_SIZE(vmstat_text))
 784                 return NULL;
 785         return (unsigned long *)m->private + *pos;
 786 }
 787
 788 static int vmstat_show(struct seq_file *m, void *arg)
 789 {
 790         unsigned long *l = arg;
 791         unsigned long off = l - (unsigned long *)m->private;
 792
 793         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
 794         return 0;
 795 }
 796
 797 static void vmstat_stop(struct seq_file *m, void *arg)
 798 {
 799         kfree(m->private);
 800         m->private = NULL;
 801 }
 802
 803 const struct seq_operations vmstat_op = {
 804         .start  = vmstat_start,
 805         .next   = vmstat_next,
 806         .stop   = vmstat_stop,
 807         .show   = vmstat_show,
 808 };
 809
 810 #endif /* CONFIG_PROC_FS */
 811
 812 #ifdef CONFIG_SMP
 813 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
 814 int sysctl_stat_interval __read_mostly = HZ;
 815
 816 static void vmstat_update(struct work_struct *w)
 817 {
 818         refresh_cpu_vm_stats(smp_processor_id());
 819         schedule_delayed_work(&__get_cpu_var(vmstat_work),
 820                 sysctl_stat_interval);
 821 }
 822
 823 static void __devinit start_cpu_timer(int cpu)
 824 {
 825         struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
 826
 827         INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
 828         schedule_delayed_work_on(cpu, vmstat_work, HZ + cpu);
 829 }
 830
 831 /*
 832  * Use the cpu notifier to insure that the thresholds are recalculated
 833  * when necessary.
 834  */
 835 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
 836                 unsigned long action,
 837                 void *hcpu)
 838 {
 839         long cpu = (long)hcpu;
 840
 841         switch (action) {
 842         case CPU_ONLINE:
 843         case CPU_ONLINE_FROZEN:
 844                 start_cpu_timer(cpu);
 845                 break;
 846         case CPU_DOWN_PREPARE:
 847         case CPU_DOWN_PREPARE_FROZEN:
 848                 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
 849                 per_cpu(vmstat_work, cpu).work.func = NULL;
 850                 break;
 851         case CPU_DOWN_FAILED:
 852         case CPU_DOWN_FAILED_FROZEN:
 853                 start_cpu_timer(cpu);
 854                 break;
 855         case CPU_DEAD:
 856         case CPU_DEAD_FROZEN:
 857                 refresh_zone_stat_thresholds();
 858                 break;
 859         default:
 860                 break;
 861         }
 862         return NOTIFY_OK;
 863 }
 864
 865 static struct notifier_block __cpuinitdata vmstat_notifier =
 866         { &vmstat_cpuup_callback, NULL, 0 };
 867
 868 int __init setup_vmstat(void)
 869 {
 870         int cpu;
 871
 872         refresh_zone_stat_thresholds();
 873         register_cpu_notifier(&vmstat_notifier);
 874
 875         for_each_online_cpu(cpu)
 876                 start_cpu_timer(cpu);
 877         return 0;
 878 }
 879 module_init(setup_vmstat)
 880 #endif