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