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/cpu.h>
  16 #include <linux/vmstat.h>
  17 #include <linux/sched.h>
  18
  19 #ifdef CONFIG_VM_EVENT_COUNTERS
  20 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
  21 EXPORT_PER_CPU_SYMBOL(vm_event_states);
  22
  23 static void sum_vm_events(unsigned long *ret, const struct cpumask *cpumask)
  24 {
  25         int cpu;
  26         int i;
  27
  28         memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
  29
  30         for_each_cpu(cpu, cpumask) {
  31                 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
  32
  33                 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
  34                         ret[i] += this->event[i];
  35         }
  36 }
  37
  38 /*
  39  * Accumulate the vm event counters across all CPUs.
  40  * The result is unavoidably approximate - it can change
  41  * during and after execution of this function.
  42 */
  43 void all_vm_events(unsigned long *ret)
  44 {
  45         get_online_cpus();
  46         sum_vm_events(ret, cpu_online_mask);
  47         put_online_cpus();
  48 }
  49 EXPORT_SYMBOL_GPL(all_vm_events);
  50
  51 #ifdef CONFIG_HOTPLUG
  52 /*
  53  * Fold the foreign cpu events into our own.
  54  *
  55  * This is adding to the events on one processor
  56  * but keeps the global counts constant.
  57  */
  58 void vm_events_fold_cpu(int cpu)
  59 {
  60         struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
  61         int i;
  62
  63         for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
  64                 count_vm_events(i, fold_state->event[i]);
  65                 fold_state->event[i] = 0;
  66         }
  67 }
  68 #endif /* CONFIG_HOTPLUG */
  69
  70 #endif /* CONFIG_VM_EVENT_COUNTERS */
  71
  72 /*
  73  * Manage combined zone based / global counters
  74  *
  75  * vm_stat contains the global counters
  76  */
  77 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
  78 EXPORT_SYMBOL(vm_stat);
  79
  80 #ifdef CONFIG_SMP
  81
  82 static int calculate_threshold(struct zone *zone)
  83 {
  84         int threshold;
  85         int mem;        /* memory in 128 MB units */
  86
  87         /*
  88          * The threshold scales with the number of processors and the amount
  89          * of memory per zone. More memory means that we can defer updates for
  90          * longer, more processors could lead to more contention.
  91          * fls() is used to have a cheap way of logarithmic scaling.
  92          *
  93          * Some sample thresholds:
  94          *
  95          * Threshold    Processors      (fls)   Zonesize        fls(mem+1)
  96          * ------------------------------------------------------------------
  97          * 8            1               1       0.9-1 GB        4
  98          * 16           2               2       0.9-1 GB        4
  99          * 20           2               2       1-2 GB          5
 100          * 24           2               2       2-4 GB          6
 101          * 28           2               2       4-8 GB          7
 102          * 32           2               2       8-16 GB         8
 103          * 4            2               2       <128M           1
 104          * 30           4               3       2-4 GB          5
 105          * 48           4               3       8-16 GB         8
 106          * 32           8               4       1-2 GB          4
 107          * 32           8               4       0.9-1GB         4
 108          * 10           16              5       <128M           1
 109          * 40           16              5       900M            4
 110          * 70           64              7       2-4 GB          5
 111          * 84           64              7       4-8 GB          6
 112          * 108          512             9       4-8 GB          6
 113          * 125          1024            10      8-16 GB         8
 114          * 125          1024            10      16-32 GB        9
 115          */
 116
 117         mem = zone->present_pages >> (27 - PAGE_SHIFT);
 118
 119         threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
 120
 121         /*
 122          * Maximum threshold is 125
 123          */
 124         threshold = min(125, threshold);
 125
 126         return threshold;
 127 }
 128
 129 /*
 130  * Refresh the thresholds for each zone.
 131  */
 132 static void refresh_zone_stat_thresholds(void)
 133 {
 134         struct zone *zone;
 135         int cpu;
 136         int threshold;
 137
 138         for_each_populated_zone(zone) {
 139                 unsigned long max_drift, tolerate_drift;
 140
 141                 threshold = calculate_threshold(zone);
 142
 143                 for_each_online_cpu(cpu)
 144                         zone_pcp(zone, cpu)->stat_threshold = threshold;
 145
 146                 /*
 147                  * Only set percpu_drift_mark if there is a danger that
 148                  * NR_FREE_PAGES reports the low watermark is ok when in fact
 149                  * the min watermark could be breached by an allocation
 150                  */
 151                 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
 152                 max_drift = num_online_cpus() * threshold;
 153                 if (max_drift > tolerate_drift)
 154                         zone->percpu_drift_mark = high_wmark_pages(zone) +
 155                                         max_drift;
 156         }
 157 }
 158
 159 /*
 160  * For use when we know that interrupts are disabled.
 161  */
 162 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 163                                 int delta)
 164 {
 165         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
 166         s8 *p = pcp->vm_stat_diff + item;
 167         long x;
 168
 169         x = delta + *p;
 170
 171         if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
 172                 zone_page_state_add(x, zone, item);
 173                 x = 0;
 174         }
 175         *p = x;
 176 }
 177 EXPORT_SYMBOL(__mod_zone_page_state);
 178
 179 /*
 180  * For an unknown interrupt state
 181  */
 182 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 183                                         int delta)
 184 {
 185         unsigned long flags;
 186
 187         local_irq_save(flags);
 188         __mod_zone_page_state(zone, item, delta);
 189         local_irq_restore(flags);
 190 }
 191 EXPORT_SYMBOL(mod_zone_page_state);
 192
 193 /*
 194  * Optimized increment and decrement functions.
 195  *
 196  * These are only for a single page and therefore can take a struct page *
 197  * argument instead of struct zone *. This allows the inclusion of the code
 198  * generated for page_zone(page) into the optimized functions.
 199  *
 200  * No overflow check is necessary and therefore the differential can be
 201  * incremented or decremented in place which may allow the compilers to
 202  * generate better code.
 203  * The increment or decrement is known and therefore one boundary check can
 204  * be omitted.
 205  *
 206  * NOTE: These functions are very performance sensitive. Change only
 207  * with care.
 208  *
 209  * Some processors have inc/dec instructions that are atomic vs an interrupt.
 210  * However, the code must first determine the differential location in a zone
 211  * based on the processor number and then inc/dec the counter. There is no
 212  * guarantee without disabling preemption that the processor will not change
 213  * in between and therefore the atomicity vs. interrupt cannot be exploited
 214  * in a useful way here.
 215  */
 216 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
 217 {
 218         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
 219         s8 *p = pcp->vm_stat_diff + item;
 220
 221         (*p)++;
 222
 223         if (unlikely(*p > pcp->stat_threshold)) {
 224                 int overstep = pcp->stat_threshold / 2;
 225
 226                 zone_page_state_add(*p + overstep, zone, item);
 227                 *p = -overstep;
 228         }
 229 }
 230
 231 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
 232 {
 233         __inc_zone_state(page_zone(page), item);
 234 }
 235 EXPORT_SYMBOL(__inc_zone_page_state);
 236
 237 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
 238 {
 239         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
 240         s8 *p = pcp->vm_stat_diff + item;
 241
 242         (*p)--;
 243
 244         if (unlikely(*p < - pcp->stat_threshold)) {
 245                 int overstep = pcp->stat_threshold / 2;
 246
 247                 zone_page_state_add(*p - overstep, zone, item);
 248                 *p = overstep;
 249         }
 250 }
 251
 252 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
 253 {
 254         __dec_zone_state(page_zone(page), item);
 255 }
 256 EXPORT_SYMBOL(__dec_zone_page_state);
 257
 258 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
 259 {
 260         unsigned long flags;
 261
 262         local_irq_save(flags);
 263         __inc_zone_state(zone, item);
 264         local_irq_restore(flags);
 265 }
 266
 267 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
 268 {
 269         unsigned long flags;
 270         struct zone *zone;
 271
 272         zone = page_zone(page);
 273         local_irq_save(flags);
 274         __inc_zone_state(zone, item);
 275         local_irq_restore(flags);
 276 }
 277 EXPORT_SYMBOL(inc_zone_page_state);
 278
 279 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
 280 {
 281         unsigned long flags;
 282
 283         local_irq_save(flags);
 284         __dec_zone_page_state(page, item);
 285         local_irq_restore(flags);
 286 }
 287 EXPORT_SYMBOL(dec_zone_page_state);
 288
 289 /*
 290  * Update the zone counters for one cpu.
 291  *
 292  * The cpu specified must be either the current cpu or a processor that
 293  * is not online. If it is the current cpu then the execution thread must
 294  * be pinned to the current cpu.
 295  *
 296  * Note that refresh_cpu_vm_stats strives to only access
 297  * node local memory. The per cpu pagesets on remote zones are placed
 298  * in the memory local to the processor using that pageset. So the
 299  * loop over all zones will access a series of cachelines local to
 300  * the processor.
 301  *
 302  * The call to zone_page_state_add updates the cachelines with the
 303  * statistics in the remote zone struct as well as the global cachelines
 304  * with the global counters. These could cause remote node cache line
 305  * bouncing and will have to be only done when necessary.
 306  */
 307 void refresh_cpu_vm_stats(int cpu)
 308 {
 309         struct zone *zone;
 310         int i;
 311         int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
 312
 313         for_each_populated_zone(zone) {
 314                 struct per_cpu_pageset *p;
 315
 316                 p = zone_pcp(zone, cpu);
 317
 318                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 319                         if (p->vm_stat_diff[i]) {
 320                                 unsigned long flags;
 321                                 int v;
 322
 323                                 local_irq_save(flags);
 324                                 v = p->vm_stat_diff[i];
 325                                 p->vm_stat_diff[i] = 0;
 326                                 local_irq_restore(flags);
 327                                 atomic_long_add(v, &zone->vm_stat[i]);
 328                                 global_diff[i] += v;
 329 #ifdef CONFIG_NUMA
 330                                 /* 3 seconds idle till flush */
 331                                 p->expire = 3;
 332 #endif
 333                         }
 334                 cond_resched();
 335 #ifdef CONFIG_NUMA
 336                 /*
 337                  * Deal with draining the remote pageset of this
 338                  * processor
 339                  *
 340                  * Check if there are pages remaining in this pageset
 341                  * if not then there is nothing to expire.
 342                  */
 343                 if (!p->expire || !p->pcp.count)
 344                         continue;
 345
 346                 /*
 347                  * We never drain zones local to this processor.
 348                  */
 349                 if (zone_to_nid(zone) == numa_node_id()) {
 350                         p->expire = 0;
 351                         continue;
 352                 }
 353
 354                 p->expire--;
 355                 if (p->expire)
 356                         continue;
 357
 358                 if (p->pcp.count)
 359                         drain_zone_pages(zone, &p->pcp);
 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 zone *preferred_zone, struct zone *z)
 378 {
 379         if (z->zone_pgdat == preferred_zone->zone_pgdat) {
 380                 __inc_zone_state(z, NUMA_HIT);
 381         } else {
 382                 __inc_zone_state(z, NUMA_MISS);
 383                 __inc_zone_state(preferred_zone, 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 #include <linux/proc_fs.h>
 394 #include <linux/seq_file.h>
 395
 396 static char * const migratetype_names[MIGRATE_TYPES] = {
 397         "Unmovable",
 398         "Reclaimable",
 399         "Movable",
 400         "Reserve",
 401         "Isolate",
 402 };
 403
 404 static void *frag_start(struct seq_file *m, loff_t *pos)
 405 {
 406         pg_data_t *pgdat;
 407         loff_t node = *pos;
 408         for (pgdat = first_online_pgdat();
 409              pgdat && node;
 410              pgdat = next_online_pgdat(pgdat))
 411                 --node;
 412
 413         return pgdat;
 414 }
 415
 416 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
 417 {
 418         pg_data_t *pgdat = (pg_data_t *)arg;
 419
 420         (*pos)++;
 421         return next_online_pgdat(pgdat);
 422 }
 423
 424 static void frag_stop(struct seq_file *m, void *arg)
 425 {
 426 }
 427
 428 /* Walk all the zones in a node and print using a callback */
 429 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
 430                 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
 431 {
 432         struct zone *zone;
 433         struct zone *node_zones = pgdat->node_zones;
 434         unsigned long flags;
 435
 436         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
 437                 if (!populated_zone(zone))
 438                         continue;
 439
 440                 spin_lock_irqsave(&zone->lock, flags);
 441                 print(m, pgdat, zone);
 442                 spin_unlock_irqrestore(&zone->lock, flags);
 443         }
 444 }
 445
 446 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
 447                                                 struct zone *zone)
 448 {
 449         int order;
 450
 451         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 452         for (order = 0; order < MAX_ORDER; ++order)
 453                 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
 454         seq_putc(m, '\n');
 455 }
 456
 457 /*
 458  * This walks the free areas for each zone.
 459  */
 460 static int frag_show(struct seq_file *m, void *arg)
 461 {
 462         pg_data_t *pgdat = (pg_data_t *)arg;
 463         walk_zones_in_node(m, pgdat, frag_show_print);
 464         return 0;
 465 }
 466
 467 static void pagetypeinfo_showfree_print(struct seq_file *m,
 468                                         pg_data_t *pgdat, struct zone *zone)
 469 {
 470         int order, mtype;
 471
 472         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
 473                 seq_printf(m, "Node %4d, zone %8s, type %12s ",
 474                                         pgdat->node_id,
 475                                         zone->name,
 476                                         migratetype_names[mtype]);
 477                 for (order = 0; order < MAX_ORDER; ++order) {
 478                         unsigned long freecount = 0;
 479                         struct free_area *area;
 480                         struct list_head *curr;
 481
 482                         area = &(zone->free_area[order]);
 483
 484                         list_for_each(curr, &area->free_list[mtype])
 485                                 freecount++;
 486                         seq_printf(m, "%6lu ", freecount);
 487                 }
 488                 seq_putc(m, '\n');
 489         }
 490 }
 491
 492 /* Print out the free pages at each order for each migatetype */
 493 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
 494 {
 495         int order;
 496         pg_data_t *pgdat = (pg_data_t *)arg;
 497
 498         /* Print header */
 499         seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
 500         for (order = 0; order < MAX_ORDER; ++order)
 501                 seq_printf(m, "%6d ", order);
 502         seq_putc(m, '\n');
 503
 504         walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
 505
 506         return 0;
 507 }
 508
 509 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
 510                                         pg_data_t *pgdat, struct zone *zone)
 511 {
 512         int mtype;
 513         unsigned long pfn;
 514         unsigned long start_pfn = zone->zone_start_pfn;
 515         unsigned long end_pfn = start_pfn + zone->spanned_pages;
 516         unsigned long count[MIGRATE_TYPES] = { 0, };
 517
 518         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
 519                 struct page *page;
 520
 521                 if (!pfn_valid(pfn))
 522                         continue;
 523
 524                 page = pfn_to_page(pfn);
 525
 526                 /* Watch for unexpected holes punched in the memmap */
 527                 if (!memmap_valid_within(pfn, page, zone))
 528                         continue;
 529
 530                 mtype = get_pageblock_migratetype(page);
 531
 532                 if (mtype < MIGRATE_TYPES)
 533                         count[mtype]++;
 534         }
 535
 536         /* Print counts */
 537         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 538         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 539                 seq_printf(m, "%12lu ", count[mtype]);
 540         seq_putc(m, '\n');
 541 }
 542
 543 /* Print out the free pages at each order for each migratetype */
 544 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
 545 {
 546         int mtype;
 547         pg_data_t *pgdat = (pg_data_t *)arg;
 548
 549         seq_printf(m, "\n%-23s", "Number of blocks type ");
 550         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 551                 seq_printf(m, "%12s ", migratetype_names[mtype]);
 552         seq_putc(m, '\n');
 553         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
 554
 555         return 0;
 556 }
 557
 558 /*
 559  * This prints out statistics in relation to grouping pages by mobility.
 560  * It is expensive to collect so do not constantly read the file.
 561  */
 562 static int pagetypeinfo_show(struct seq_file *m, void *arg)
 563 {
 564         pg_data_t *pgdat = (pg_data_t *)arg;
 565
 566         /* check memoryless node */
 567         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
 568                 return 0;
 569
 570         seq_printf(m, "Page block order: %d\n", pageblock_order);
 571         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
 572         seq_putc(m, '\n');
 573         pagetypeinfo_showfree(m, pgdat);
 574         pagetypeinfo_showblockcount(m, pgdat);
 575
 576         return 0;
 577 }
 578
 579 static const struct seq_operations fragmentation_op = {
 580         .start  = frag_start,
 581         .next   = frag_next,
 582         .stop   = frag_stop,
 583         .show   = frag_show,
 584 };
 585
 586 static int fragmentation_open(struct inode *inode, struct file *file)
 587 {
 588         return seq_open(file, &fragmentation_op);
 589 }
 590
 591 static const struct file_operations fragmentation_file_operations = {
 592         .open           = fragmentation_open,
 593         .read           = seq_read,
 594         .llseek         = seq_lseek,
 595         .release        = seq_release,
 596 };
 597
 598 static const struct seq_operations pagetypeinfo_op = {
 599         .start  = frag_start,
 600         .next   = frag_next,
 601         .stop   = frag_stop,
 602         .show   = pagetypeinfo_show,
 603 };
 604
 605 static int pagetypeinfo_open(struct inode *inode, struct file *file)
 606 {
 607         return seq_open(file, &pagetypeinfo_op);
 608 }
 609
 610 static const struct file_operations pagetypeinfo_file_ops = {
 611         .open           = pagetypeinfo_open,
 612         .read           = seq_read,
 613         .llseek         = seq_lseek,
 614         .release        = seq_release,
 615 };
 616
 617 #ifdef CONFIG_ZONE_DMA
 618 #define TEXT_FOR_DMA(xx) xx "_dma",
 619 #else
 620 #define TEXT_FOR_DMA(xx)
 621 #endif
 622
 623 #ifdef CONFIG_ZONE_DMA32
 624 #define TEXT_FOR_DMA32(xx) xx "_dma32",
 625 #else
 626 #define TEXT_FOR_DMA32(xx)
 627 #endif
 628
 629 #ifdef CONFIG_HIGHMEM
 630 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
 631 #else
 632 #define TEXT_FOR_HIGHMEM(xx)
 633 #endif
 634
 635 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
 636                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
 637
 638 static const char * const vmstat_text[] = {
 639         /* Zoned VM counters */
 640         "nr_free_pages",
 641         "nr_inactive_anon",
 642         "nr_active_anon",
 643         "nr_inactive_file",
 644         "nr_active_file",
 645         "nr_unevictable",
 646         "nr_mlock",
 647         "nr_anon_pages",
 648         "nr_mapped",
 649         "nr_file_pages",
 650         "nr_dirty",
 651         "nr_writeback",
 652         "nr_slab_reclaimable",
 653         "nr_slab_unreclaimable",
 654         "nr_page_table_pages",
 655         "nr_kernel_stack",
 656         "nr_unstable",
 657         "nr_bounce",
 658         "nr_vmscan_write",
 659         "nr_writeback_temp",
 660         "nr_isolated_anon",
 661         "nr_isolated_file",
 662         "nr_shmem",
 663 #ifdef CONFIG_NUMA
 664         "numa_hit",
 665         "numa_miss",
 666         "numa_foreign",
 667         "numa_interleave",
 668         "numa_local",
 669         "numa_other",
 670 #endif
 671
 672 #ifdef CONFIG_VM_EVENT_COUNTERS
 673         "pgpgin",
 674         "pgpgout",
 675         "pswpin",
 676         "pswpout",
 677
 678         TEXTS_FOR_ZONES("pgalloc")
 679
 680         "pgfree",
 681         "pgactivate",
 682         "pgdeactivate",
 683
 684         "pgfault",
 685         "pgmajfault",
 686
 687         TEXTS_FOR_ZONES("pgrefill")
 688         TEXTS_FOR_ZONES("pgsteal")
 689         TEXTS_FOR_ZONES("pgscan_kswapd")
 690         TEXTS_FOR_ZONES("pgscan_direct")
 691
 692 #ifdef CONFIG_NUMA
 693         "zone_reclaim_failed",
 694 #endif
 695         "pginodesteal",
 696         "slabs_scanned",
 697         "kswapd_steal",
 698         "kswapd_inodesteal",
 699         "kswapd_low_wmark_hit_quickly",
 700         "kswapd_high_wmark_hit_quickly",
 701         "kswapd_skip_congestion_wait",
 702         "pageoutrun",
 703         "allocstall",
 704
 705         "pgrotated",
 706 #ifdef CONFIG_HUGETLB_PAGE
 707         "htlb_buddy_alloc_success",
 708         "htlb_buddy_alloc_fail",
 709 #endif
 710         "unevictable_pgs_culled",
 711         "unevictable_pgs_scanned",
 712         "unevictable_pgs_rescued",
 713         "unevictable_pgs_mlocked",
 714         "unevictable_pgs_munlocked",
 715         "unevictable_pgs_cleared",
 716         "unevictable_pgs_stranded",
 717         "unevictable_pgs_mlockfreed",
 718 #endif
 719 };
 720
 721 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
 722                                                         struct zone *zone)
 723 {
 724         int i;
 725         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
 726         seq_printf(m,
 727                    "\n  pages free     %lu"
 728                    "\n        min      %lu"
 729                    "\n        low      %lu"
 730                    "\n        high     %lu"
 731                    "\n        scanned  %lu"
 732                    "\n        spanned  %lu"
 733                    "\n        present  %lu",
 734                    zone_nr_free_pages(zone),
 735                    min_wmark_pages(zone),
 736                    low_wmark_pages(zone),
 737                    high_wmark_pages(zone),
 738                    zone->pages_scanned,
 739                    zone->spanned_pages,
 740                    zone->present_pages);
 741
 742         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 743                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
 744                                 zone_page_state(zone, i));
 745
 746         seq_printf(m,
 747                    "\n        protection: (%lu",
 748                    zone->lowmem_reserve[0]);
 749         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
 750                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
 751         seq_printf(m,
 752                    ")"
 753                    "\n  pagesets");
 754         for_each_online_cpu(i) {
 755                 struct per_cpu_pageset *pageset;
 756
 757                 pageset = zone_pcp(zone, i);
 758                 seq_printf(m,
 759                            "\n    cpu: %i"
 760                            "\n              count: %i"
 761                            "\n              high:  %i"
 762                            "\n              batch: %i",
 763                            i,
 764                            pageset->pcp.count,
 765                            pageset->pcp.high,
 766                            pageset->pcp.batch);
 767 #ifdef CONFIG_SMP
 768                 seq_printf(m, "\n  vm stats threshold: %d",
 769                                 pageset->stat_threshold);
 770 #endif
 771         }
 772         seq_printf(m,
 773                    "\n  all_unreclaimable: %u"
 774                    "\n  prev_priority:     %i"
 775                    "\n  start_pfn:         %lu"
 776                    "\n  inactive_ratio:    %u",
 777                            zone_is_all_unreclaimable(zone),
 778                    zone->prev_priority,
 779                    zone->zone_start_pfn,
 780                    zone->inactive_ratio);
 781         seq_putc(m, '\n');
 782 }
 783
 784 /*
 785  * Output information about zones in @pgdat.
 786  */
 787 static int zoneinfo_show(struct seq_file *m, void *arg)
 788 {
 789         pg_data_t *pgdat = (pg_data_t *)arg;
 790         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
 791         return 0;
 792 }
 793
 794 static const struct seq_operations zoneinfo_op = {
 795         .start  = frag_start, /* iterate over all zones. The same as in
 796                                * fragmentation. */
 797         .next   = frag_next,
 798         .stop   = frag_stop,
 799         .show   = zoneinfo_show,
 800 };
 801
 802 static int zoneinfo_open(struct inode *inode, struct file *file)
 803 {
 804         return seq_open(file, &zoneinfo_op);
 805 }
 806
 807 static const struct file_operations proc_zoneinfo_file_operations = {
 808         .open           = zoneinfo_open,
 809         .read           = seq_read,
 810         .llseek         = seq_lseek,
 811         .release        = seq_release,
 812 };
 813
 814 static void *vmstat_start(struct seq_file *m, loff_t *pos)
 815 {
 816         unsigned long *v;
 817 #ifdef CONFIG_VM_EVENT_COUNTERS
 818         unsigned long *e;
 819 #endif
 820         int i;
 821
 822         if (*pos >= ARRAY_SIZE(vmstat_text))
 823                 return NULL;
 824
 825 #ifdef CONFIG_VM_EVENT_COUNTERS
 826         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
 827                         + sizeof(struct vm_event_state), GFP_KERNEL);
 828 #else
 829         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
 830                         GFP_KERNEL);
 831 #endif
 832         m->private = v;
 833         if (!v)
 834                 return ERR_PTR(-ENOMEM);
 835         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 836                 v[i] = global_page_state(i);
 837 #ifdef CONFIG_VM_EVENT_COUNTERS
 838         e = v + NR_VM_ZONE_STAT_ITEMS;
 839         all_vm_events(e);
 840         e[PGPGIN] /= 2;         /* sectors -> kbytes */
 841         e[PGPGOUT] /= 2;
 842 #endif
 843         return v + *pos;
 844 }
 845
 846 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
 847 {
 848         (*pos)++;
 849         if (*pos >= ARRAY_SIZE(vmstat_text))
 850                 return NULL;
 851         return (unsigned long *)m->private + *pos;
 852 }
 853
 854 static int vmstat_show(struct seq_file *m, void *arg)
 855 {
 856         unsigned long *l = arg;
 857         unsigned long off = l - (unsigned long *)m->private;
 858
 859         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
 860         return 0;
 861 }
 862
 863 static void vmstat_stop(struct seq_file *m, void *arg)
 864 {
 865         kfree(m->private);
 866         m->private = NULL;
 867 }
 868
 869 static const struct seq_operations vmstat_op = {
 870         .start  = vmstat_start,
 871         .next   = vmstat_next,
 872         .stop   = vmstat_stop,
 873         .show   = vmstat_show,
 874 };
 875
 876 static int vmstat_open(struct inode *inode, struct file *file)
 877 {
 878         return seq_open(file, &vmstat_op);
 879 }
 880
 881 static const struct file_operations proc_vmstat_file_operations = {
 882         .open           = vmstat_open,
 883         .read           = seq_read,
 884         .llseek         = seq_lseek,
 885         .release        = seq_release,
 886 };
 887 #endif /* CONFIG_PROC_FS */
 888
 889 #ifdef CONFIG_SMP
 890 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
 891 int sysctl_stat_interval __read_mostly = HZ;
 892
 893 static void vmstat_update(struct work_struct *w)
 894 {
 895         refresh_cpu_vm_stats(smp_processor_id());
 896         schedule_delayed_work(&__get_cpu_var(vmstat_work),
 897                 round_jiffies_relative(sysctl_stat_interval));
 898 }
 899
 900 static void __cpuinit start_cpu_timer(int cpu)
 901 {
 902         struct delayed_work *work = &per_cpu(vmstat_work, cpu);
 903
 904         INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
 905         schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
 906 }
 907
 908 /*
 909  * Use the cpu notifier to insure that the thresholds are recalculated
 910  * when necessary.
 911  */
 912 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
 913                 unsigned long action,
 914                 void *hcpu)
 915 {
 916         long cpu = (long)hcpu;
 917
 918         switch (action) {
 919         case CPU_ONLINE:
 920         case CPU_ONLINE_FROZEN:
 921                 start_cpu_timer(cpu);
 922                 break;
 923         case CPU_DOWN_PREPARE:
 924         case CPU_DOWN_PREPARE_FROZEN:
 925                 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
 926                 per_cpu(vmstat_work, cpu).work.func = NULL;
 927                 break;
 928         case CPU_DOWN_FAILED:
 929         case CPU_DOWN_FAILED_FROZEN:
 930                 start_cpu_timer(cpu);
 931                 break;
 932         case CPU_DEAD:
 933         case CPU_DEAD_FROZEN:
 934                 refresh_zone_stat_thresholds();
 935                 break;
 936         default:
 937                 break;
 938         }
 939         return NOTIFY_OK;
 940 }
 941
 942 static struct notifier_block __cpuinitdata vmstat_notifier =
 943         { &vmstat_cpuup_callback, NULL, 0 };
 944 #endif
 945
 946 static int __init setup_vmstat(void)
 947 {
 948 #ifdef CONFIG_SMP
 949         int cpu;
 950
 951         refresh_zone_stat_thresholds();
 952         register_cpu_notifier(&vmstat_notifier);
 953
 954         for_each_online_cpu(cpu)
 955                 start_cpu_timer(cpu);
 956 #endif
 957 #ifdef CONFIG_PROC_FS
 958         proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
 959         proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
 960         proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
 961         proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
 962 #endif
 963         return 0;
 964 }
 965 module_init(setup_vmstat)