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                 threshold = calculate_threshold(zone);
 140
 141                 for_each_online_cpu(cpu)
 142                         per_cpu_ptr(zone->pageset, cpu)->stat_threshold
 143                                                         = 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 = this_cpu_ptr(zone->pageset);
 154
 155         s8 *p = pcp->vm_stat_diff + item;
 156         long x;
 157
 158         x = delta + *p;
 159
 160         if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
 161                 zone_page_state_add(x, zone, item);
 162                 x = 0;
 163         }
 164         *p = x;
 165 }
 166 EXPORT_SYMBOL(__mod_zone_page_state);
 167
 168 /*
 169  * For an unknown interrupt state
 170  */
 171 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 172                                         int delta)
 173 {
 174         unsigned long flags;
 175
 176         local_irq_save(flags);
 177         __mod_zone_page_state(zone, item, delta);
 178         local_irq_restore(flags);
 179 }
 180 EXPORT_SYMBOL(mod_zone_page_state);
 181
 182 /*
 183  * Optimized increment and decrement functions.
 184  *
 185  * These are only for a single page and therefore can take a struct page *
 186  * argument instead of struct zone *. This allows the inclusion of the code
 187  * generated for page_zone(page) into the optimized functions.
 188  *
 189  * No overflow check is necessary and therefore the differential can be
 190  * incremented or decremented in place which may allow the compilers to
 191  * generate better code.
 192  * The increment or decrement is known and therefore one boundary check can
 193  * be omitted.
 194  *
 195  * NOTE: These functions are very performance sensitive. Change only
 196  * with care.
 197  *
 198  * Some processors have inc/dec instructions that are atomic vs an interrupt.
 199  * However, the code must first determine the differential location in a zone
 200  * based on the processor number and then inc/dec the counter. There is no
 201  * guarantee without disabling preemption that the processor will not change
 202  * in between and therefore the atomicity vs. interrupt cannot be exploited
 203  * in a useful way here.
 204  */
 205 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
 206 {
 207         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
 208         s8 *p = pcp->vm_stat_diff + item;
 209
 210         (*p)++;
 211
 212         if (unlikely(*p > pcp->stat_threshold)) {
 213                 int overstep = pcp->stat_threshold / 2;
 214
 215                 zone_page_state_add(*p + overstep, zone, item);
 216                 *p = -overstep;
 217         }
 218 }
 219
 220 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
 221 {
 222         __inc_zone_state(page_zone(page), item);
 223 }
 224 EXPORT_SYMBOL(__inc_zone_page_state);
 225
 226 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
 227 {
 228         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
 229         s8 *p = pcp->vm_stat_diff + item;
 230
 231         (*p)--;
 232
 233         if (unlikely(*p < - pcp->stat_threshold)) {
 234                 int overstep = pcp->stat_threshold / 2;
 235
 236                 zone_page_state_add(*p - overstep, zone, item);
 237                 *p = overstep;
 238         }
 239 }
 240
 241 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
 242 {
 243         __dec_zone_state(page_zone(page), item);
 244 }
 245 EXPORT_SYMBOL(__dec_zone_page_state);
 246
 247 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
 248 {
 249         unsigned long flags;
 250
 251         local_irq_save(flags);
 252         __inc_zone_state(zone, item);
 253         local_irq_restore(flags);
 254 }
 255
 256 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
 257 {
 258         unsigned long flags;
 259         struct zone *zone;
 260
 261         zone = page_zone(page);
 262         local_irq_save(flags);
 263         __inc_zone_state(zone, item);
 264         local_irq_restore(flags);
 265 }
 266 EXPORT_SYMBOL(inc_zone_page_state);
 267
 268 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
 269 {
 270         unsigned long flags;
 271
 272         local_irq_save(flags);
 273         __dec_zone_page_state(page, item);
 274         local_irq_restore(flags);
 275 }
 276 EXPORT_SYMBOL(dec_zone_page_state);
 277
 278 /*
 279  * Update the zone counters for one cpu.
 280  *
 281  * The cpu specified must be either the current cpu or a processor that
 282  * is not online. If it is the current cpu then the execution thread must
 283  * be pinned to the current cpu.
 284  *
 285  * Note that refresh_cpu_vm_stats strives to only access
 286  * node local memory. The per cpu pagesets on remote zones are placed
 287  * in the memory local to the processor using that pageset. So the
 288  * loop over all zones will access a series of cachelines local to
 289  * the processor.
 290  *
 291  * The call to zone_page_state_add updates the cachelines with the
 292  * statistics in the remote zone struct as well as the global cachelines
 293  * with the global counters. These could cause remote node cache line
 294  * bouncing and will have to be only done when necessary.
 295  */
 296 void refresh_cpu_vm_stats(int cpu)
 297 {
 298         struct zone *zone;
 299         int i;
 300         int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
 301
 302         for_each_populated_zone(zone) {
 303                 struct per_cpu_pageset *p;
 304
 305                 p = per_cpu_ptr(zone->pageset, cpu);
 306
 307                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 308                         if (p->vm_stat_diff[i]) {
 309                                 unsigned long flags;
 310                                 int v;
 311
 312                                 local_irq_save(flags);
 313                                 v = p->vm_stat_diff[i];
 314                                 p->vm_stat_diff[i] = 0;
 315                                 local_irq_restore(flags);
 316                                 atomic_long_add(v, &zone->vm_stat[i]);
 317                                 global_diff[i] += v;
 318 #ifdef CONFIG_NUMA
 319                                 /* 3 seconds idle till flush */
 320                                 p->expire = 3;
 321 #endif
 322                         }
 323                 cond_resched();
 324 #ifdef CONFIG_NUMA
 325                 /*
 326                  * Deal with draining the remote pageset of this
 327                  * processor
 328                  *
 329                  * Check if there are pages remaining in this pageset
 330                  * if not then there is nothing to expire.
 331                  */
 332                 if (!p->expire || !p->pcp.count)
 333                         continue;
 334
 335                 /*
 336                  * We never drain zones local to this processor.
 337                  */
 338                 if (zone_to_nid(zone) == numa_node_id()) {
 339                         p->expire = 0;
 340                         continue;
 341                 }
 342
 343                 p->expire--;
 344                 if (p->expire)
 345                         continue;
 346
 347                 if (p->pcp.count)
 348                         drain_zone_pages(zone, &p->pcp);
 349 #endif
 350         }
 351
 352         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 353                 if (global_diff[i])
 354                         atomic_long_add(global_diff[i], &vm_stat[i]);
 355 }
 356
 357 #endif
 358
 359 #ifdef CONFIG_NUMA
 360 /*
 361  * zonelist = the list of zones passed to the allocator
 362  * z        = the zone from which the allocation occurred.
 363  *
 364  * Must be called with interrupts disabled.
 365  */
 366 void zone_statistics(struct zone *preferred_zone, struct zone *z)
 367 {
 368         if (z->zone_pgdat == preferred_zone->zone_pgdat) {
 369                 __inc_zone_state(z, NUMA_HIT);
 370         } else {
 371                 __inc_zone_state(z, NUMA_MISS);
 372                 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
 373         }
 374         if (z->node == numa_node_id())
 375                 __inc_zone_state(z, NUMA_LOCAL);
 376         else
 377                 __inc_zone_state(z, NUMA_OTHER);
 378 }
 379 #endif
 380
 381 #ifdef CONFIG_PROC_FS
 382 #include <linux/proc_fs.h>
 383 #include <linux/seq_file.h>
 384
 385 static char * const migratetype_names[MIGRATE_TYPES] = {
 386         "Unmovable",
 387         "Reclaimable",
 388         "Movable",
 389         "Reserve",
 390         "Isolate",
 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
 515                 /* Watch for unexpected holes punched in the memmap */
 516                 if (!memmap_valid_within(pfn, page, zone))
 517                         continue;
 518
 519                 mtype = get_pageblock_migratetype(page);
 520
 521                 if (mtype < MIGRATE_TYPES)
 522                         count[mtype]++;
 523         }
 524
 525         /* Print counts */
 526         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 527         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 528                 seq_printf(m, "%12lu ", count[mtype]);
 529         seq_putc(m, '\n');
 530 }
 531
 532 /* Print out the free pages at each order for each migratetype */
 533 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
 534 {
 535         int mtype;
 536         pg_data_t *pgdat = (pg_data_t *)arg;
 537
 538         seq_printf(m, "\n%-23s", "Number of blocks type ");
 539         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 540                 seq_printf(m, "%12s ", migratetype_names[mtype]);
 541         seq_putc(m, '\n');
 542         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
 543
 544         return 0;
 545 }
 546
 547 /*
 548  * This prints out statistics in relation to grouping pages by mobility.
 549  * It is expensive to collect so do not constantly read the file.
 550  */
 551 static int pagetypeinfo_show(struct seq_file *m, void *arg)
 552 {
 553         pg_data_t *pgdat = (pg_data_t *)arg;
 554
 555         /* check memoryless node */
 556         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
 557                 return 0;
 558
 559         seq_printf(m, "Page block order: %d\n", pageblock_order);
 560         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
 561         seq_putc(m, '\n');
 562         pagetypeinfo_showfree(m, pgdat);
 563         pagetypeinfo_showblockcount(m, pgdat);
 564
 565         return 0;
 566 }
 567
 568 static const struct seq_operations fragmentation_op = {
 569         .start  = frag_start,
 570         .next   = frag_next,
 571         .stop   = frag_stop,
 572         .show   = frag_show,
 573 };
 574
 575 static int fragmentation_open(struct inode *inode, struct file *file)
 576 {
 577         return seq_open(file, &fragmentation_op);
 578 }
 579
 580 static const struct file_operations fragmentation_file_operations = {
 581         .open           = fragmentation_open,
 582         .read           = seq_read,
 583         .llseek         = seq_lseek,
 584         .release        = seq_release,
 585 };
 586
 587 static const struct seq_operations pagetypeinfo_op = {
 588         .start  = frag_start,
 589         .next   = frag_next,
 590         .stop   = frag_stop,
 591         .show   = pagetypeinfo_show,
 592 };
 593
 594 static int pagetypeinfo_open(struct inode *inode, struct file *file)
 595 {
 596         return seq_open(file, &pagetypeinfo_op);
 597 }
 598
 599 static const struct file_operations pagetypeinfo_file_ops = {
 600         .open           = pagetypeinfo_open,
 601         .read           = seq_read,
 602         .llseek         = seq_lseek,
 603         .release        = seq_release,
 604 };
 605
 606 #ifdef CONFIG_ZONE_DMA
 607 #define TEXT_FOR_DMA(xx) xx "_dma",
 608 #else
 609 #define TEXT_FOR_DMA(xx)
 610 #endif
 611
 612 #ifdef CONFIG_ZONE_DMA32
 613 #define TEXT_FOR_DMA32(xx) xx "_dma32",
 614 #else
 615 #define TEXT_FOR_DMA32(xx)
 616 #endif
 617
 618 #ifdef CONFIG_HIGHMEM
 619 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
 620 #else
 621 #define TEXT_FOR_HIGHMEM(xx)
 622 #endif
 623
 624 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
 625                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
 626
 627 static const char * const vmstat_text[] = {
 628         /* Zoned VM counters */
 629         "nr_free_pages",
 630         "nr_inactive_anon",
 631         "nr_active_anon",
 632         "nr_inactive_file",
 633         "nr_active_file",
 634         "nr_unevictable",
 635         "nr_mlock",
 636         "nr_anon_pages",
 637         "nr_mapped",
 638         "nr_file_pages",
 639         "nr_dirty",
 640         "nr_writeback",
 641         "nr_slab_reclaimable",
 642         "nr_slab_unreclaimable",
 643         "nr_page_table_pages",
 644         "nr_kernel_stack",
 645         "nr_unstable",
 646         "nr_bounce",
 647         "nr_vmscan_write",
 648         "nr_writeback_temp",
 649         "nr_isolated_anon",
 650         "nr_isolated_file",
 651         "nr_shmem",
 652 #ifdef CONFIG_NUMA
 653         "numa_hit",
 654         "numa_miss",
 655         "numa_foreign",
 656         "numa_interleave",
 657         "numa_local",
 658         "numa_other",
 659 #endif
 660
 661 #ifdef CONFIG_VM_EVENT_COUNTERS
 662         "pgpgin",
 663         "pgpgout",
 664         "pswpin",
 665         "pswpout",
 666
 667         TEXTS_FOR_ZONES("pgalloc")
 668
 669         "pgfree",
 670         "pgactivate",
 671         "pgdeactivate",
 672
 673         "pgfault",
 674         "pgmajfault",
 675
 676         TEXTS_FOR_ZONES("pgrefill")
 677         TEXTS_FOR_ZONES("pgsteal")
 678         TEXTS_FOR_ZONES("pgscan_kswapd")
 679         TEXTS_FOR_ZONES("pgscan_direct")
 680
 681 #ifdef CONFIG_NUMA
 682         "zone_reclaim_failed",
 683 #endif
 684         "pginodesteal",
 685         "slabs_scanned",
 686         "kswapd_steal",
 687         "kswapd_inodesteal",
 688         "kswapd_low_wmark_hit_quickly",
 689         "kswapd_high_wmark_hit_quickly",
 690         "kswapd_skip_congestion_wait",
 691         "pageoutrun",
 692         "allocstall",
 693
 694         "pgrotated",
 695 #ifdef CONFIG_HUGETLB_PAGE
 696         "htlb_buddy_alloc_success",
 697         "htlb_buddy_alloc_fail",
 698 #endif
 699         "unevictable_pgs_culled",
 700         "unevictable_pgs_scanned",
 701         "unevictable_pgs_rescued",
 702         "unevictable_pgs_mlocked",
 703         "unevictable_pgs_munlocked",
 704         "unevictable_pgs_cleared",
 705         "unevictable_pgs_stranded",
 706         "unevictable_pgs_mlockfreed",
 707 #endif
 708 };
 709
 710 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
 711                                                         struct zone *zone)
 712 {
 713         int i;
 714         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
 715         seq_printf(m,
 716                    "\n  pages free     %lu"
 717                    "\n        min      %lu"
 718                    "\n        low      %lu"
 719                    "\n        high     %lu"
 720                    "\n        scanned  %lu"
 721                    "\n        spanned  %lu"
 722                    "\n        present  %lu",
 723                    zone_page_state(zone, NR_FREE_PAGES),
 724                    min_wmark_pages(zone),
 725                    low_wmark_pages(zone),
 726                    high_wmark_pages(zone),
 727                    zone->pages_scanned,
 728                    zone->spanned_pages,
 729                    zone->present_pages);
 730
 731         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 732                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
 733                                 zone_page_state(zone, i));
 734
 735         seq_printf(m,
 736                    "\n        protection: (%lu",
 737                    zone->lowmem_reserve[0]);
 738         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
 739                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
 740         seq_printf(m,
 741                    ")"
 742                    "\n  pagesets");
 743         for_each_online_cpu(i) {
 744                 struct per_cpu_pageset *pageset;
 745
 746                 pageset = per_cpu_ptr(zone->pageset, i);
 747                 seq_printf(m,
 748                            "\n    cpu: %i"
 749                            "\n              count: %i"
 750                            "\n              high:  %i"
 751                            "\n              batch: %i",
 752                            i,
 753                            pageset->pcp.count,
 754                            pageset->pcp.high,
 755                            pageset->pcp.batch);
 756 #ifdef CONFIG_SMP
 757                 seq_printf(m, "\n  vm stats threshold: %d",
 758                                 pageset->stat_threshold);
 759 #endif
 760         }
 761         seq_printf(m,
 762                    "\n  all_unreclaimable: %u"
 763                    "\n  prev_priority:     %i"
 764                    "\n  start_pfn:         %lu"
 765                    "\n  inactive_ratio:    %u",
 766                    zone->all_unreclaimable,
 767                    zone->prev_priority,
 768                    zone->zone_start_pfn,
 769                    zone->inactive_ratio);
 770         seq_putc(m, '\n');
 771 }
 772
 773 /*
 774  * Output information about zones in @pgdat.
 775  */
 776 static int zoneinfo_show(struct seq_file *m, void *arg)
 777 {
 778         pg_data_t *pgdat = (pg_data_t *)arg;
 779         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
 780         return 0;
 781 }
 782
 783 static const struct seq_operations zoneinfo_op = {
 784         .start  = frag_start, /* iterate over all zones. The same as in
 785                                * fragmentation. */
 786         .next   = frag_next,
 787         .stop   = frag_stop,
 788         .show   = zoneinfo_show,
 789 };
 790
 791 static int zoneinfo_open(struct inode *inode, struct file *file)
 792 {
 793         return seq_open(file, &zoneinfo_op);
 794 }
 795
 796 static const struct file_operations proc_zoneinfo_file_operations = {
 797         .open           = zoneinfo_open,
 798         .read           = seq_read,
 799         .llseek         = seq_lseek,
 800         .release        = seq_release,
 801 };
 802
 803 static void *vmstat_start(struct seq_file *m, loff_t *pos)
 804 {
 805         unsigned long *v;
 806 #ifdef CONFIG_VM_EVENT_COUNTERS
 807         unsigned long *e;
 808 #endif
 809         int i;
 810
 811         if (*pos >= ARRAY_SIZE(vmstat_text))
 812                 return NULL;
 813
 814 #ifdef CONFIG_VM_EVENT_COUNTERS
 815         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
 816                         + sizeof(struct vm_event_state), GFP_KERNEL);
 817 #else
 818         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
 819                         GFP_KERNEL);
 820 #endif
 821         m->private = v;
 822         if (!v)
 823                 return ERR_PTR(-ENOMEM);
 824         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 825                 v[i] = global_page_state(i);
 826 #ifdef CONFIG_VM_EVENT_COUNTERS
 827         e = v + NR_VM_ZONE_STAT_ITEMS;
 828         all_vm_events(e);
 829         e[PGPGIN] /= 2;         /* sectors -> kbytes */
 830         e[PGPGOUT] /= 2;
 831 #endif
 832         return v + *pos;
 833 }
 834
 835 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
 836 {
 837         (*pos)++;
 838         if (*pos >= ARRAY_SIZE(vmstat_text))
 839                 return NULL;
 840         return (unsigned long *)m->private + *pos;
 841 }
 842
 843 static int vmstat_show(struct seq_file *m, void *arg)
 844 {
 845         unsigned long *l = arg;
 846         unsigned long off = l - (unsigned long *)m->private;
 847
 848         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
 849         return 0;
 850 }
 851
 852 static void vmstat_stop(struct seq_file *m, void *arg)
 853 {
 854         kfree(m->private);
 855         m->private = NULL;
 856 }
 857
 858 static const struct seq_operations vmstat_op = {
 859         .start  = vmstat_start,
 860         .next   = vmstat_next,
 861         .stop   = vmstat_stop,
 862         .show   = vmstat_show,
 863 };
 864
 865 static int vmstat_open(struct inode *inode, struct file *file)
 866 {
 867         return seq_open(file, &vmstat_op);
 868 }
 869
 870 static const struct file_operations proc_vmstat_file_operations = {
 871         .open           = vmstat_open,
 872         .read           = seq_read,
 873         .llseek         = seq_lseek,
 874         .release        = seq_release,
 875 };
 876 #endif /* CONFIG_PROC_FS */
 877
 878 #ifdef CONFIG_SMP
 879 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
 880 int sysctl_stat_interval __read_mostly = HZ;
 881
 882 static void vmstat_update(struct work_struct *w)
 883 {
 884         refresh_cpu_vm_stats(smp_processor_id());
 885         schedule_delayed_work(&__get_cpu_var(vmstat_work),
 886                 round_jiffies_relative(sysctl_stat_interval));
 887 }
 888
 889 static void __cpuinit start_cpu_timer(int cpu)
 890 {
 891         struct delayed_work *work = &per_cpu(vmstat_work, cpu);
 892
 893         INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
 894         schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
 895 }
 896
 897 /*
 898  * Use the cpu notifier to insure that the thresholds are recalculated
 899  * when necessary.
 900  */
 901 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
 902                 unsigned long action,
 903                 void *hcpu)
 904 {
 905         long cpu = (long)hcpu;
 906
 907         switch (action) {
 908         case CPU_ONLINE:
 909         case CPU_ONLINE_FROZEN:
 910                 start_cpu_timer(cpu);
 911                 node_set_state(cpu_to_node(cpu), N_CPU);
 912                 break;
 913         case CPU_DOWN_PREPARE:
 914         case CPU_DOWN_PREPARE_FROZEN:
 915                 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
 916                 per_cpu(vmstat_work, cpu).work.func = NULL;
 917                 break;
 918         case CPU_DOWN_FAILED:
 919         case CPU_DOWN_FAILED_FROZEN:
 920                 start_cpu_timer(cpu);
 921                 break;
 922         case CPU_DEAD:
 923         case CPU_DEAD_FROZEN:
 924                 refresh_zone_stat_thresholds();
 925                 break;
 926         default:
 927                 break;
 928         }
 929         return NOTIFY_OK;
 930 }
 931
 932 static struct notifier_block __cpuinitdata vmstat_notifier =
 933         { &vmstat_cpuup_callback, NULL, 0 };
 934 #endif
 935
 936 static int __init setup_vmstat(void)
 937 {
 938 #ifdef CONFIG_SMP
 939         int cpu;
 940
 941         refresh_zone_stat_thresholds();
 942         register_cpu_notifier(&vmstat_notifier);
 943
 944         for_each_online_cpu(cpu)
 945                 start_cpu_timer(cpu);
 946 #endif
 947 #ifdef CONFIG_PROC_FS
 948         proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
 949         proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
 950         proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
 951         proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
 952 #endif
 953         return 0;
 954 }
 955 module_init(setup_vmstat)