mm/vmstat.c

   1 /*
   2  *  linux/mm/vmstat.c
   3  *
   4  *  Manages VM statistics
   5  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   6  *
   7  *  zoned VM statistics
   8  *  Copyright (C) 2006 Silicon Graphics, Inc.,
   9  *              Christoph Lameter <christoph@lameter.com>
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/mm.h>
  13 #include <linux/err.h>
  14 #include <linux/module.h>
  15 #include <linux/slab.h>
  16 #include <linux/cpu.h>
  17 #include <linux/vmstat.h>
  18 #include <linux/sched.h>
  19
  20 #ifdef CONFIG_VM_EVENT_COUNTERS
  21 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
  22 EXPORT_PER_CPU_SYMBOL(vm_event_states);
  23
  24 static void sum_vm_events(unsigned long *ret, const struct cpumask *cpumask)
  25 {
  26         int cpu;
  27         int i;
  28
  29         memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
  30
  31         for_each_cpu(cpu, cpumask) {
  32                 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
  33
  34                 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
  35                         ret[i] += this->event[i];
  36         }
  37 }
  38
  39 /*
  40  * Accumulate the vm event counters across all CPUs.
  41  * The result is unavoidably approximate - it can change
  42  * during and after execution of this function.
  43 */
  44 void all_vm_events(unsigned long *ret)
  45 {
  46         get_online_cpus();
  47         sum_vm_events(ret, cpu_online_mask);
  48         put_online_cpus();
  49 }
  50 EXPORT_SYMBOL_GPL(all_vm_events);
  51
  52 #ifdef CONFIG_HOTPLUG
  53 /*
  54  * Fold the foreign cpu events into our own.
  55  *
  56  * This is adding to the events on one processor
  57  * but keeps the global counts constant.
  58  */
  59 void vm_events_fold_cpu(int cpu)
  60 {
  61         struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
  62         int i;
  63
  64         for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
  65                 count_vm_events(i, fold_state->event[i]);
  66                 fold_state->event[i] = 0;
  67         }
  68 }
  69 #endif /* CONFIG_HOTPLUG */
  70
  71 #endif /* CONFIG_VM_EVENT_COUNTERS */
  72
  73 /*
  74  * Manage combined zone based / global counters
  75  *
  76  * vm_stat contains the global counters
  77  */
  78 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
  79 EXPORT_SYMBOL(vm_stat);
  80
  81 #ifdef CONFIG_SMP
  82
  83 static int calculate_threshold(struct zone *zone)
  84 {
  85         int threshold;
  86         int mem;        /* memory in 128 MB units */
  87
  88         /*
  89          * The threshold scales with the number of processors and the amount
  90          * of memory per zone. More memory means that we can defer updates for
  91          * longer, more processors could lead to more contention.
  92          * fls() is used to have a cheap way of logarithmic scaling.
  93          *
  94          * Some sample thresholds:
  95          *
  96          * Threshold    Processors      (fls)   Zonesize        fls(mem+1)
  97          * ------------------------------------------------------------------
  98          * 8            1               1       0.9-1 GB        4
  99          * 16           2               2       0.9-1 GB        4
 100          * 20           2               2       1-2 GB          5
 101          * 24           2               2       2-4 GB          6
 102          * 28           2               2       4-8 GB          7
 103          * 32           2               2       8-16 GB         8
 104          * 4            2               2       <128M           1
 105          * 30           4               3       2-4 GB          5
 106          * 48           4               3       8-16 GB         8
 107          * 32           8               4       1-2 GB          4
 108          * 32           8               4       0.9-1GB         4
 109          * 10           16              5       <128M           1
 110          * 40           16              5       900M            4
 111          * 70           64              7       2-4 GB          5
 112          * 84           64              7       4-8 GB          6
 113          * 108          512             9       4-8 GB          6
 114          * 125          1024            10      8-16 GB         8
 115          * 125          1024            10      16-32 GB        9
 116          */
 117
 118         mem = zone->present_pages >> (27 - PAGE_SHIFT);
 119
 120         threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
 121
 122         /*
 123          * Maximum threshold is 125
 124          */
 125         threshold = min(125, threshold);
 126
 127         return threshold;
 128 }
 129
 130 /*
 131  * Refresh the thresholds for each zone.
 132  */
 133 static void refresh_zone_stat_thresholds(void)
 134 {
 135         struct zone *zone;
 136         int cpu;
 137         int threshold;
 138
 139         for_each_populated_zone(zone) {
 140                 unsigned long max_drift, tolerate_drift;
 141
 142                 threshold = calculate_threshold(zone);
 143
 144                 for_each_online_cpu(cpu)
 145                         per_cpu_ptr(zone->pageset, cpu)->stat_threshold
 146                                                         = threshold;
 147
 148                 /*
 149                  * Only set percpu_drift_mark if there is a danger that
 150                  * NR_FREE_PAGES reports the low watermark is ok when in fact
 151                  * the min watermark could be breached by an allocation
 152                  */
 153                 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
 154                 max_drift = num_online_cpus() * threshold;
 155                 if (max_drift > tolerate_drift)
 156                         zone->percpu_drift_mark = high_wmark_pages(zone) +
 157                                         max_drift;
 158         }
 159 }
 160
 161 /*
 162  * For use when we know that interrupts are disabled.
 163  */
 164 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 165                                 int delta)
 166 {
 167         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
 168
 169         s8 *p = pcp->vm_stat_diff + item;
 170         long x;
 171
 172         x = delta + *p;
 173
 174         if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
 175                 zone_page_state_add(x, zone, item);
 176                 x = 0;
 177         }
 178         *p = x;
 179 }
 180 EXPORT_SYMBOL(__mod_zone_page_state);
 181
 182 /*
 183  * For an unknown interrupt state
 184  */
 185 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 186                                         int delta)
 187 {
 188         unsigned long flags;
 189
 190         local_irq_save(flags);
 191         __mod_zone_page_state(zone, item, delta);
 192         local_irq_restore(flags);
 193 }
 194 EXPORT_SYMBOL(mod_zone_page_state);
 195
 196 /*
 197  * Optimized increment and decrement functions.
 198  *
 199  * These are only for a single page and therefore can take a struct page *
 200  * argument instead of struct zone *. This allows the inclusion of the code
 201  * generated for page_zone(page) into the optimized functions.
 202  *
 203  * No overflow check is necessary and therefore the differential can be
 204  * incremented or decremented in place which may allow the compilers to
 205  * generate better code.
 206  * The increment or decrement is known and therefore one boundary check can
 207  * be omitted.
 208  *
 209  * NOTE: These functions are very performance sensitive. Change only
 210  * with care.
 211  *
 212  * Some processors have inc/dec instructions that are atomic vs an interrupt.
 213  * However, the code must first determine the differential location in a zone
 214  * based on the processor number and then inc/dec the counter. There is no
 215  * guarantee without disabling preemption that the processor will not change
 216  * in between and therefore the atomicity vs. interrupt cannot be exploited
 217  * in a useful way here.
 218  */
 219 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
 220 {
 221         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
 222         s8 *p = pcp->vm_stat_diff + item;
 223
 224         (*p)++;
 225
 226         if (unlikely(*p > pcp->stat_threshold)) {
 227                 int overstep = pcp->stat_threshold / 2;
 228
 229                 zone_page_state_add(*p + overstep, zone, item);
 230                 *p = -overstep;
 231         }
 232 }
 233
 234 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
 235 {
 236         __inc_zone_state(page_zone(page), item);
 237 }
 238 EXPORT_SYMBOL(__inc_zone_page_state);
 239
 240 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
 241 {
 242         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
 243         s8 *p = pcp->vm_stat_diff + item;
 244
 245         (*p)--;
 246
 247         if (unlikely(*p < - pcp->stat_threshold)) {
 248                 int overstep = pcp->stat_threshold / 2;
 249
 250                 zone_page_state_add(*p - overstep, zone, item);
 251                 *p = overstep;
 252         }
 253 }
 254
 255 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
 256 {
 257         __dec_zone_state(page_zone(page), item);
 258 }
 259 EXPORT_SYMBOL(__dec_zone_page_state);
 260
 261 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
 262 {
 263         unsigned long flags;
 264
 265         local_irq_save(flags);
 266         __inc_zone_state(zone, item);
 267         local_irq_restore(flags);
 268 }
 269
 270 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
 271 {
 272         unsigned long flags;
 273         struct zone *zone;
 274
 275         zone = page_zone(page);
 276         local_irq_save(flags);
 277         __inc_zone_state(zone, item);
 278         local_irq_restore(flags);
 279 }
 280 EXPORT_SYMBOL(inc_zone_page_state);
 281
 282 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
 283 {
 284         unsigned long flags;
 285
 286         local_irq_save(flags);
 287         __dec_zone_page_state(page, item);
 288         local_irq_restore(flags);
 289 }
 290 EXPORT_SYMBOL(dec_zone_page_state);
 291
 292 /*
 293  * Update the zone counters for one cpu.
 294  *
 295  * The cpu specified must be either the current cpu or a processor that
 296  * is not online. If it is the current cpu then the execution thread must
 297  * be pinned to the current cpu.
 298  *
 299  * Note that refresh_cpu_vm_stats strives to only access
 300  * node local memory. The per cpu pagesets on remote zones are placed
 301  * in the memory local to the processor using that pageset. So the
 302  * loop over all zones will access a series of cachelines local to
 303  * the processor.
 304  *
 305  * The call to zone_page_state_add updates the cachelines with the
 306  * statistics in the remote zone struct as well as the global cachelines
 307  * with the global counters. These could cause remote node cache line
 308  * bouncing and will have to be only done when necessary.
 309  */
 310 void refresh_cpu_vm_stats(int cpu)
 311 {
 312         struct zone *zone;
 313         int i;
 314         int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
 315
 316         for_each_populated_zone(zone) {
 317                 struct per_cpu_pageset *p;
 318
 319                 p = per_cpu_ptr(zone->pageset, cpu);
 320
 321                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 322                         if (p->vm_stat_diff[i]) {
 323                                 unsigned long flags;
 324                                 int v;
 325
 326                                 local_irq_save(flags);
 327                                 v = p->vm_stat_diff[i];
 328                                 p->vm_stat_diff[i] = 0;
 329                                 local_irq_restore(flags);
 330                                 atomic_long_add(v, &zone->vm_stat[i]);
 331                                 global_diff[i] += v;
 332 #ifdef CONFIG_NUMA
 333                                 /* 3 seconds idle till flush */
 334                                 p->expire = 3;
 335 #endif
 336                         }
 337                 cond_resched();
 338 #ifdef CONFIG_NUMA
 339                 /*
 340                  * Deal with draining the remote pageset of this
 341                  * processor
 342                  *
 343                  * Check if there are pages remaining in this pageset
 344                  * if not then there is nothing to expire.
 345                  */
 346                 if (!p->expire || !p->pcp.count)
 347                         continue;
 348
 349                 /*
 350                  * We never drain zones local to this processor.
 351                  */
 352                 if (zone_to_nid(zone) == numa_node_id()) {
 353                         p->expire = 0;
 354                         continue;
 355                 }
 356
 357                 p->expire--;
 358                 if (p->expire)
 359                         continue;
 360
 361                 if (p->pcp.count)
 362                         drain_zone_pages(zone, &p->pcp);
 363 #endif
 364         }
 365
 366         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 367                 if (global_diff[i])
 368                         atomic_long_add(global_diff[i], &vm_stat[i]);
 369 }
 370
 371 #endif
 372
 373 #ifdef CONFIG_NUMA
 374 /*
 375  * zonelist = the list of zones passed to the allocator
 376  * z        = the zone from which the allocation occurred.
 377  *
 378  * Must be called with interrupts disabled.
 379  */
 380 void zone_statistics(struct zone *preferred_zone, struct zone *z)
 381 {
 382         if (z->zone_pgdat == preferred_zone->zone_pgdat) {
 383                 __inc_zone_state(z, NUMA_HIT);
 384         } else {
 385                 __inc_zone_state(z, NUMA_MISS);
 386                 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
 387         }
 388         if (z->node == numa_node_id())
 389                 __inc_zone_state(z, NUMA_LOCAL);
 390         else
 391                 __inc_zone_state(z, NUMA_OTHER);
 392 }
 393 #endif
 394
 395 #ifdef CONFIG_PROC_FS
 396 #include <linux/proc_fs.h>
 397 #include <linux/seq_file.h>
 398
 399 static char * const migratetype_names[MIGRATE_TYPES] = {
 400         "Unmovable",
 401         "Reclaimable",
 402         "Movable",
 403         "Reserve",
 404         "Isolate",
 405 };
 406
 407 static void *frag_start(struct seq_file *m, loff_t *pos)
 408 {
 409         pg_data_t *pgdat;
 410         loff_t node = *pos;
 411         for (pgdat = first_online_pgdat();
 412              pgdat && node;
 413              pgdat = next_online_pgdat(pgdat))
 414                 --node;
 415
 416         return pgdat;
 417 }
 418
 419 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
 420 {
 421         pg_data_t *pgdat = (pg_data_t *)arg;
 422
 423         (*pos)++;
 424         return next_online_pgdat(pgdat);
 425 }
 426
 427 static void frag_stop(struct seq_file *m, void *arg)
 428 {
 429 }
 430
 431 /* Walk all the zones in a node and print using a callback */
 432 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
 433                 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
 434 {
 435         struct zone *zone;
 436         struct zone *node_zones = pgdat->node_zones;
 437         unsigned long flags;
 438
 439         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
 440                 if (!populated_zone(zone))
 441                         continue;
 442
 443                 spin_lock_irqsave(&zone->lock, flags);
 444                 print(m, pgdat, zone);
 445                 spin_unlock_irqrestore(&zone->lock, flags);
 446         }
 447 }
 448
 449 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
 450                                                 struct zone *zone)
 451 {
 452         int order;
 453
 454         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 455         for (order = 0; order < MAX_ORDER; ++order)
 456                 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
 457         seq_putc(m, '\n');
 458 }
 459
 460 /*
 461  * This walks the free areas for each zone.
 462  */
 463 static int frag_show(struct seq_file *m, void *arg)
 464 {
 465         pg_data_t *pgdat = (pg_data_t *)arg;
 466         walk_zones_in_node(m, pgdat, frag_show_print);
 467         return 0;
 468 }
 469
 470 static void pagetypeinfo_showfree_print(struct seq_file *m,
 471                                         pg_data_t *pgdat, struct zone *zone)
 472 {
 473         int order, mtype;
 474
 475         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
 476                 seq_printf(m, "Node %4d, zone %8s, type %12s ",
 477                                         pgdat->node_id,
 478                                         zone->name,
 479                                         migratetype_names[mtype]);
 480                 for (order = 0; order < MAX_ORDER; ++order) {
 481                         unsigned long freecount = 0;
 482                         struct free_area *area;
 483                         struct list_head *curr;
 484
 485                         area = &(zone->free_area[order]);
 486
 487                         list_for_each(curr, &area->free_list[mtype])
 488                                 freecount++;
 489                         seq_printf(m, "%6lu ", freecount);
 490                 }
 491                 seq_putc(m, '\n');
 492         }
 493 }
 494
 495 /* Print out the free pages at each order for each migatetype */
 496 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
 497 {
 498         int order;
 499         pg_data_t *pgdat = (pg_data_t *)arg;
 500
 501         /* Print header */
 502         seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
 503         for (order = 0; order < MAX_ORDER; ++order)
 504                 seq_printf(m, "%6d ", order);
 505         seq_putc(m, '\n');
 506
 507         walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
 508
 509         return 0;
 510 }
 511
 512 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
 513                                         pg_data_t *pgdat, struct zone *zone)
 514 {
 515         int mtype;
 516         unsigned long pfn;
 517         unsigned long start_pfn = zone->zone_start_pfn;
 518         unsigned long end_pfn = start_pfn + zone->spanned_pages;
 519         unsigned long count[MIGRATE_TYPES] = { 0, };
 520
 521         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
 522                 struct page *page;
 523
 524                 if (!pfn_valid(pfn))
 525                         continue;
 526
 527                 page = pfn_to_page(pfn);
 528
 529                 /* Watch for unexpected holes punched in the memmap */
 530                 if (!memmap_valid_within(pfn, page, zone))
 531                         continue;
 532
 533                 mtype = get_pageblock_migratetype(page);
 534
 535                 if (mtype < MIGRATE_TYPES)
 536                         count[mtype]++;
 537         }
 538
 539         /* Print counts */
 540         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 541         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 542                 seq_printf(m, "%12lu ", count[mtype]);
 543         seq_putc(m, '\n');
 544 }
 545
 546 /* Print out the free pages at each order for each migratetype */
 547 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
 548 {
 549         int mtype;
 550         pg_data_t *pgdat = (pg_data_t *)arg;
 551
 552         seq_printf(m, "\n%-23s", "Number of blocks type ");
 553         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 554                 seq_printf(m, "%12s ", migratetype_names[mtype]);
 555         seq_putc(m, '\n');
 556         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
 557
 558         return 0;
 559 }
 560
 561 /*
 562  * This prints out statistics in relation to grouping pages by mobility.
 563  * It is expensive to collect so do not constantly read the file.
 564  */
 565 static int pagetypeinfo_show(struct seq_file *m, void *arg)
 566 {
 567         pg_data_t *pgdat = (pg_data_t *)arg;
 568
 569         /* check memoryless node */
 570         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
 571                 return 0;
 572
 573         seq_printf(m, "Page block order: %d\n", pageblock_order);
 574         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
 575         seq_putc(m, '\n');
 576         pagetypeinfo_showfree(m, pgdat);
 577         pagetypeinfo_showblockcount(m, pgdat);
 578
 579         return 0;
 580 }
 581
 582 static const struct seq_operations fragmentation_op = {
 583         .start  = frag_start,
 584         .next   = frag_next,
 585         .stop   = frag_stop,
 586         .show   = frag_show,
 587 };
 588
 589 static int fragmentation_open(struct inode *inode, struct file *file)
 590 {
 591         return seq_open(file, &fragmentation_op);
 592 }
 593
 594 static const struct file_operations fragmentation_file_operations = {
 595         .open           = fragmentation_open,
 596         .read           = seq_read,
 597         .llseek         = seq_lseek,
 598         .release        = seq_release,
 599 };
 600
 601 static const struct seq_operations pagetypeinfo_op = {
 602         .start  = frag_start,
 603         .next   = frag_next,
 604         .stop   = frag_stop,
 605         .show   = pagetypeinfo_show,
 606 };
 607
 608 static int pagetypeinfo_open(struct inode *inode, struct file *file)
 609 {
 610         return seq_open(file, &pagetypeinfo_op);
 611 }
 612
 613 static const struct file_operations pagetypeinfo_file_ops = {
 614         .open           = pagetypeinfo_open,
 615         .read           = seq_read,
 616         .llseek         = seq_lseek,
 617         .release        = seq_release,
 618 };
 619
 620 #ifdef CONFIG_ZONE_DMA
 621 #define TEXT_FOR_DMA(xx) xx "_dma",
 622 #else
 623 #define TEXT_FOR_DMA(xx)
 624 #endif
 625
 626 #ifdef CONFIG_ZONE_DMA32
 627 #define TEXT_FOR_DMA32(xx) xx "_dma32",
 628 #else
 629 #define TEXT_FOR_DMA32(xx)
 630 #endif
 631
 632 #ifdef CONFIG_HIGHMEM
 633 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
 634 #else
 635 #define TEXT_FOR_HIGHMEM(xx)
 636 #endif
 637
 638 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
 639                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
 640
 641 static const char * const vmstat_text[] = {
 642         /* Zoned VM counters */
 643         "nr_free_pages",
 644         "nr_inactive_anon",
 645         "nr_active_anon",
 646         "nr_inactive_file",
 647         "nr_active_file",
 648         "nr_unevictable",
 649         "nr_mlock",
 650         "nr_anon_pages",
 651         "nr_mapped",
 652         "nr_file_pages",
 653         "nr_dirty",
 654         "nr_writeback",
 655         "nr_slab_reclaimable",
 656         "nr_slab_unreclaimable",
 657         "nr_page_table_pages",
 658         "nr_kernel_stack",
 659         "nr_unstable",
 660         "nr_bounce",
 661         "nr_vmscan_write",
 662         "nr_writeback_temp",
 663         "nr_isolated_anon",
 664         "nr_isolated_file",
 665         "nr_shmem",
 666 #ifdef CONFIG_NUMA
 667         "numa_hit",
 668         "numa_miss",
 669         "numa_foreign",
 670         "numa_interleave",
 671         "numa_local",
 672         "numa_other",
 673 #endif
 674
 675 #ifdef CONFIG_VM_EVENT_COUNTERS
 676         "pgpgin",
 677         "pgpgout",
 678         "pswpin",
 679         "pswpout",
 680
 681         TEXTS_FOR_ZONES("pgalloc")
 682
 683         "pgfree",
 684         "pgactivate",
 685         "pgdeactivate",
 686
 687         "pgfault",
 688         "pgmajfault",
 689
 690         TEXTS_FOR_ZONES("pgrefill")
 691         TEXTS_FOR_ZONES("pgsteal")
 692         TEXTS_FOR_ZONES("pgscan_kswapd")
 693         TEXTS_FOR_ZONES("pgscan_direct")
 694
 695 #ifdef CONFIG_NUMA
 696         "zone_reclaim_failed",
 697 #endif
 698         "pginodesteal",
 699         "slabs_scanned",
 700         "kswapd_steal",
 701         "kswapd_inodesteal",
 702         "kswapd_low_wmark_hit_quickly",
 703         "kswapd_high_wmark_hit_quickly",
 704         "kswapd_skip_congestion_wait",
 705         "pageoutrun",
 706         "allocstall",
 707
 708         "pgrotated",
 709 #ifdef CONFIG_HUGETLB_PAGE
 710         "htlb_buddy_alloc_success",
 711         "htlb_buddy_alloc_fail",
 712 #endif
 713         "unevictable_pgs_culled",
 714         "unevictable_pgs_scanned",
 715         "unevictable_pgs_rescued",
 716         "unevictable_pgs_mlocked",
 717         "unevictable_pgs_munlocked",
 718         "unevictable_pgs_cleared",
 719         "unevictable_pgs_stranded",
 720         "unevictable_pgs_mlockfreed",
 721 #endif
 722 };
 723
 724 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
 725                                                         struct zone *zone)
 726 {
 727         int i;
 728         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
 729         seq_printf(m,
 730                    "\n  pages free     %lu"
 731                    "\n        min      %lu"
 732                    "\n        low      %lu"
 733                    "\n        high     %lu"
 734                    "\n        scanned  %lu"
 735                    "\n        spanned  %lu"
 736                    "\n        present  %lu",
 737                    zone_nr_free_pages(zone),
 738                    min_wmark_pages(zone),
 739                    low_wmark_pages(zone),
 740                    high_wmark_pages(zone),
 741                    zone->pages_scanned,
 742                    zone->spanned_pages,
 743                    zone->present_pages);
 744
 745         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 746                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
 747                                 zone_page_state(zone, i));
 748
 749         seq_printf(m,
 750                    "\n        protection: (%lu",
 751                    zone->lowmem_reserve[0]);
 752         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
 753                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
 754         seq_printf(m,
 755                    ")"
 756                    "\n  pagesets");
 757         for_each_online_cpu(i) {
 758                 struct per_cpu_pageset *pageset;
 759
 760                 pageset = per_cpu_ptr(zone->pageset, i);
 761                 seq_printf(m,
 762                            "\n    cpu: %i"
 763                            "\n              count: %i"
 764                            "\n              high:  %i"
 765                            "\n              batch: %i",
 766                            i,
 767                            pageset->pcp.count,
 768                            pageset->pcp.high,
 769                            pageset->pcp.batch);
 770 #ifdef CONFIG_SMP
 771                 seq_printf(m, "\n  vm stats threshold: %d",
 772                                 pageset->stat_threshold);
 773 #endif
 774         }
 775         seq_printf(m,
 776                    "\n  all_unreclaimable: %u"
 777                    "\n  prev_priority:     %i"
 778                    "\n  start_pfn:         %lu"
 779                    "\n  inactive_ratio:    %u",
 780                    zone->all_unreclaimable,
 781                    zone->prev_priority,
 782                    zone->zone_start_pfn,
 783                    zone->inactive_ratio);
 784         seq_putc(m, '\n');
 785 }
 786
 787 /*
 788  * Output information about zones in @pgdat.
 789  */
 790 static int zoneinfo_show(struct seq_file *m, void *arg)
 791 {
 792         pg_data_t *pgdat = (pg_data_t *)arg;
 793         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
 794         return 0;
 795 }
 796
 797 static const struct seq_operations zoneinfo_op = {
 798         .start  = frag_start, /* iterate over all zones. The same as in
 799                                * fragmentation. */
 800         .next   = frag_next,
 801         .stop   = frag_stop,
 802         .show   = zoneinfo_show,
 803 };
 804
 805 static int zoneinfo_open(struct inode *inode, struct file *file)
 806 {
 807         return seq_open(file, &zoneinfo_op);
 808 }
 809
 810 static const struct file_operations proc_zoneinfo_file_operations = {
 811         .open           = zoneinfo_open,
 812         .read           = seq_read,
 813         .llseek         = seq_lseek,
 814         .release        = seq_release,
 815 };
 816
 817 static void *vmstat_start(struct seq_file *m, loff_t *pos)
 818 {
 819         unsigned long *v;
 820 #ifdef CONFIG_VM_EVENT_COUNTERS
 821         unsigned long *e;
 822 #endif
 823         int i;
 824
 825         if (*pos >= ARRAY_SIZE(vmstat_text))
 826                 return NULL;
 827
 828 #ifdef CONFIG_VM_EVENT_COUNTERS
 829         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
 830                         + sizeof(struct vm_event_state), GFP_KERNEL);
 831 #else
 832         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
 833                         GFP_KERNEL);
 834 #endif
 835         m->private = v;
 836         if (!v)
 837                 return ERR_PTR(-ENOMEM);
 838         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 839                 v[i] = global_page_state(i);
 840 #ifdef CONFIG_VM_EVENT_COUNTERS
 841         e = v + NR_VM_ZONE_STAT_ITEMS;
 842         all_vm_events(e);
 843         e[PGPGIN] /= 2;         /* sectors -> kbytes */
 844         e[PGPGOUT] /= 2;
 845 #endif
 846         return v + *pos;
 847 }
 848
 849 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
 850 {
 851         (*pos)++;
 852         if (*pos >= ARRAY_SIZE(vmstat_text))
 853                 return NULL;
 854         return (unsigned long *)m->private + *pos;
 855 }
 856
 857 static int vmstat_show(struct seq_file *m, void *arg)
 858 {
 859         unsigned long *l = arg;
 860         unsigned long off = l - (unsigned long *)m->private;
 861
 862         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
 863         return 0;
 864 }
 865
 866 static void vmstat_stop(struct seq_file *m, void *arg)
 867 {
 868         kfree(m->private);
 869         m->private = NULL;
 870 }
 871
 872 static const struct seq_operations vmstat_op = {
 873         .start  = vmstat_start,
 874         .next   = vmstat_next,
 875         .stop   = vmstat_stop,
 876         .show   = vmstat_show,
 877 };
 878
 879 static int vmstat_open(struct inode *inode, struct file *file)
 880 {
 881         return seq_open(file, &vmstat_op);
 882 }
 883
 884 static const struct file_operations proc_vmstat_file_operations = {
 885         .open           = vmstat_open,
 886         .read           = seq_read,
 887         .llseek         = seq_lseek,
 888         .release        = seq_release,
 889 };
 890 #endif /* CONFIG_PROC_FS */
 891
 892 #ifdef CONFIG_SMP
 893 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
 894 int sysctl_stat_interval __read_mostly = HZ;
 895
 896 static void vmstat_update(struct work_struct *w)
 897 {
 898         refresh_cpu_vm_stats(smp_processor_id());
 899         schedule_delayed_work(&__get_cpu_var(vmstat_work),
 900                 round_jiffies_relative(sysctl_stat_interval));
 901 }
 902
 903 static void __cpuinit start_cpu_timer(int cpu)
 904 {
 905         struct delayed_work *work = &per_cpu(vmstat_work, cpu);
 906
 907         INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
 908         schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
 909 }
 910
 911 /*
 912  * Use the cpu notifier to insure that the thresholds are recalculated
 913  * when necessary.
 914  */
 915 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
 916                 unsigned long action,
 917                 void *hcpu)
 918 {
 919         long cpu = (long)hcpu;
 920
 921         switch (action) {
 922         case CPU_ONLINE:
 923         case CPU_ONLINE_FROZEN:
 924                 start_cpu_timer(cpu);
 925                 node_set_state(cpu_to_node(cpu), N_CPU);
 926                 break;
 927         case CPU_DOWN_PREPARE:
 928         case CPU_DOWN_PREPARE_FROZEN:
 929                 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
 930                 per_cpu(vmstat_work, cpu).work.func = NULL;
 931                 break;
 932         case CPU_DOWN_FAILED:
 933         case CPU_DOWN_FAILED_FROZEN:
 934                 start_cpu_timer(cpu);
 935                 break;
 936         case CPU_DEAD:
 937         case CPU_DEAD_FROZEN:
 938                 refresh_zone_stat_thresholds();
 939                 break;
 940         default:
 941                 break;
 942         }
 943         return NOTIFY_OK;
 944 }
 945
 946 static struct notifier_block __cpuinitdata vmstat_notifier =
 947         { &vmstat_cpuup_callback, NULL, 0 };
 948 #endif
 949
 950 static int __init setup_vmstat(void)
 951 {
 952 #ifdef CONFIG_SMP
 953         int cpu;
 954
 955         refresh_zone_stat_thresholds();
 956         register_cpu_notifier(&vmstat_notifier);
 957
 958         for_each_online_cpu(cpu)
 959                 start_cpu_timer(cpu);
 960 #endif
 961 #ifdef CONFIG_PROC_FS
 962         proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
 963         proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
 964         proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
 965         proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
 966 #endif
 967         return 0;
 968 }
 969 module_init(setup_vmstat)