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_mask_nr(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_zone(zone) {
 139
 140                 if (!zone->present_pages)
 141                         continue;
 142
 143                 threshold = calculate_threshold(zone);
 144
 145                 for_each_online_cpu(cpu)
 146                         zone_pcp(zone, cpu)->stat_threshold = threshold;
 147         }
 148 }
 149
 150 /*
 151  * For use when we know that interrupts are disabled.
 152  */
 153 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 154                                 int delta)
 155 {
 156         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
 157         s8 *p = pcp->vm_stat_diff + item;
 158         long x;
 159
 160         x = delta + *p;
 161
 162         if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
 163                 zone_page_state_add(x, zone, item);
 164                 x = 0;
 165         }
 166         *p = x;
 167 }
 168 EXPORT_SYMBOL(__mod_zone_page_state);
 169
 170 /*
 171  * For an unknown interrupt state
 172  */
 173 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 174                                         int delta)
 175 {
 176         unsigned long flags;
 177
 178         local_irq_save(flags);
 179         __mod_zone_page_state(zone, item, delta);
 180         local_irq_restore(flags);
 181 }
 182 EXPORT_SYMBOL(mod_zone_page_state);
 183
 184 /*
 185  * Optimized increment and decrement functions.
 186  *
 187  * These are only for a single page and therefore can take a struct page *
 188  * argument instead of struct zone *. This allows the inclusion of the code
 189  * generated for page_zone(page) into the optimized functions.
 190  *
 191  * No overflow check is necessary and therefore the differential can be
 192  * incremented or decremented in place which may allow the compilers to
 193  * generate better code.
 194  * The increment or decrement is known and therefore one boundary check can
 195  * be omitted.
 196  *
 197  * NOTE: These functions are very performance sensitive. Change only
 198  * with care.
 199  *
 200  * Some processors have inc/dec instructions that are atomic vs an interrupt.
 201  * However, the code must first determine the differential location in a zone
 202  * based on the processor number and then inc/dec the counter. There is no
 203  * guarantee without disabling preemption that the processor will not change
 204  * in between and therefore the atomicity vs. interrupt cannot be exploited
 205  * in a useful way here.
 206  */
 207 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
 208 {
 209         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
 210         s8 *p = pcp->vm_stat_diff + item;
 211
 212         (*p)++;
 213
 214         if (unlikely(*p > pcp->stat_threshold)) {
 215                 int overstep = pcp->stat_threshold / 2;
 216
 217                 zone_page_state_add(*p + overstep, zone, item);
 218                 *p = -overstep;
 219         }
 220 }
 221
 222 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
 223 {
 224         __inc_zone_state(page_zone(page), item);
 225 }
 226 EXPORT_SYMBOL(__inc_zone_page_state);
 227
 228 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
 229 {
 230         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
 231         s8 *p = pcp->vm_stat_diff + item;
 232
 233         (*p)--;
 234
 235         if (unlikely(*p < - pcp->stat_threshold)) {
 236                 int overstep = pcp->stat_threshold / 2;
 237
 238                 zone_page_state_add(*p - overstep, zone, item);
 239                 *p = overstep;
 240         }
 241 }
 242
 243 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
 244 {
 245         __dec_zone_state(page_zone(page), item);
 246 }
 247 EXPORT_SYMBOL(__dec_zone_page_state);
 248
 249 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
 250 {
 251         unsigned long flags;
 252
 253         local_irq_save(flags);
 254         __inc_zone_state(zone, item);
 255         local_irq_restore(flags);
 256 }
 257
 258 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
 259 {
 260         unsigned long flags;
 261         struct zone *zone;
 262
 263         zone = page_zone(page);
 264         local_irq_save(flags);
 265         __inc_zone_state(zone, item);
 266         local_irq_restore(flags);
 267 }
 268 EXPORT_SYMBOL(inc_zone_page_state);
 269
 270 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
 271 {
 272         unsigned long flags;
 273
 274         local_irq_save(flags);
 275         __dec_zone_page_state(page, item);
 276         local_irq_restore(flags);
 277 }
 278 EXPORT_SYMBOL(dec_zone_page_state);
 279
 280 /*
 281  * Update the zone counters for one cpu.
 282  *
 283  * The cpu specified must be either the current cpu or a processor that
 284  * is not online. If it is the current cpu then the execution thread must
 285  * be pinned to the current cpu.
 286  *
 287  * Note that refresh_cpu_vm_stats strives to only access
 288  * node local memory. The per cpu pagesets on remote zones are placed
 289  * in the memory local to the processor using that pageset. So the
 290  * loop over all zones will access a series of cachelines local to
 291  * the processor.
 292  *
 293  * The call to zone_page_state_add updates the cachelines with the
 294  * statistics in the remote zone struct as well as the global cachelines
 295  * with the global counters. These could cause remote node cache line
 296  * bouncing and will have to be only done when necessary.
 297  */
 298 void refresh_cpu_vm_stats(int cpu)
 299 {
 300         struct zone *zone;
 301         int i;
 302         int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
 303
 304         for_each_zone(zone) {
 305                 struct per_cpu_pageset *p;
 306
 307                 if (!populated_zone(zone))
 308                         continue;
 309
 310                 p = zone_pcp(zone, cpu);
 311
 312                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 313                         if (p->vm_stat_diff[i]) {
 314                                 unsigned long flags;
 315                                 int v;
 316
 317                                 local_irq_save(flags);
 318                                 v = p->vm_stat_diff[i];
 319                                 p->vm_stat_diff[i] = 0;
 320                                 local_irq_restore(flags);
 321                                 atomic_long_add(v, &zone->vm_stat[i]);
 322                                 global_diff[i] += v;
 323 #ifdef CONFIG_NUMA
 324                                 /* 3 seconds idle till flush */
 325                                 p->expire = 3;
 326 #endif
 327                         }
 328                 cond_resched();
 329 #ifdef CONFIG_NUMA
 330                 /*
 331                  * Deal with draining the remote pageset of this
 332                  * processor
 333                  *
 334                  * Check if there are pages remaining in this pageset
 335                  * if not then there is nothing to expire.
 336                  */
 337                 if (!p->expire || !p->pcp.count)
 338                         continue;
 339
 340                 /*
 341                  * We never drain zones local to this processor.
 342                  */
 343                 if (zone_to_nid(zone) == numa_node_id()) {
 344                         p->expire = 0;
 345                         continue;
 346                 }
 347
 348                 p->expire--;
 349                 if (p->expire)
 350                         continue;
 351
 352                 if (p->pcp.count)
 353                         drain_zone_pages(zone, &p->pcp);
 354 #endif
 355         }
 356
 357         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 358                 if (global_diff[i])
 359                         atomic_long_add(global_diff[i], &vm_stat[i]);
 360 }
 361
 362 #endif
 363
 364 #ifdef CONFIG_NUMA
 365 /*
 366  * zonelist = the list of zones passed to the allocator
 367  * z        = the zone from which the allocation occurred.
 368  *
 369  * Must be called with interrupts disabled.
 370  */
 371 void zone_statistics(struct zone *preferred_zone, struct zone *z)
 372 {
 373         if (z->zone_pgdat == preferred_zone->zone_pgdat) {
 374                 __inc_zone_state(z, NUMA_HIT);
 375         } else {
 376                 __inc_zone_state(z, NUMA_MISS);
 377                 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
 378         }
 379         if (z->node == numa_node_id())
 380                 __inc_zone_state(z, NUMA_LOCAL);
 381         else
 382                 __inc_zone_state(z, NUMA_OTHER);
 383 }
 384 #endif
 385
 386 #ifdef CONFIG_PROC_FS
 387 #include <linux/proc_fs.h>
 388 #include <linux/seq_file.h>
 389
 390 static char * const migratetype_names[MIGRATE_TYPES] = {
 391         "Unmovable",
 392         "Reclaimable",
 393         "Movable",
 394         "Reserve",
 395         "Isolate",
 396 };
 397
 398 static void *frag_start(struct seq_file *m, loff_t *pos)
 399 {
 400         pg_data_t *pgdat;
 401         loff_t node = *pos;
 402         for (pgdat = first_online_pgdat();
 403              pgdat && node;
 404              pgdat = next_online_pgdat(pgdat))
 405                 --node;
 406
 407         return pgdat;
 408 }
 409
 410 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
 411 {
 412         pg_data_t *pgdat = (pg_data_t *)arg;
 413
 414         (*pos)++;
 415         return next_online_pgdat(pgdat);
 416 }
 417
 418 static void frag_stop(struct seq_file *m, void *arg)
 419 {
 420 }
 421
 422 /* Walk all the zones in a node and print using a callback */
 423 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
 424                 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
 425 {
 426         struct zone *zone;
 427         struct zone *node_zones = pgdat->node_zones;
 428         unsigned long flags;
 429
 430         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
 431                 if (!populated_zone(zone))
 432                         continue;
 433
 434                 spin_lock_irqsave(&zone->lock, flags);
 435                 print(m, pgdat, zone);
 436                 spin_unlock_irqrestore(&zone->lock, flags);
 437         }
 438 }
 439
 440 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
 441                                                 struct zone *zone)
 442 {
 443         int order;
 444
 445         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 446         for (order = 0; order < MAX_ORDER; ++order)
 447                 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
 448         seq_putc(m, '\n');
 449 }
 450
 451 /*
 452  * This walks the free areas for each zone.
 453  */
 454 static int frag_show(struct seq_file *m, void *arg)
 455 {
 456         pg_data_t *pgdat = (pg_data_t *)arg;
 457         walk_zones_in_node(m, pgdat, frag_show_print);
 458         return 0;
 459 }
 460
 461 static void pagetypeinfo_showfree_print(struct seq_file *m,
 462                                         pg_data_t *pgdat, struct zone *zone)
 463 {
 464         int order, mtype;
 465
 466         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
 467                 seq_printf(m, "Node %4d, zone %8s, type %12s ",
 468                                         pgdat->node_id,
 469                                         zone->name,
 470                                         migratetype_names[mtype]);
 471                 for (order = 0; order < MAX_ORDER; ++order) {
 472                         unsigned long freecount = 0;
 473                         struct free_area *area;
 474                         struct list_head *curr;
 475
 476                         area = &(zone->free_area[order]);
 477
 478                         list_for_each(curr, &area->free_list[mtype])
 479                                 freecount++;
 480                         seq_printf(m, "%6lu ", freecount);
 481                 }
 482                 seq_putc(m, '\n');
 483         }
 484 }
 485
 486 /* Print out the free pages at each order for each migatetype */
 487 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
 488 {
 489         int order;
 490         pg_data_t *pgdat = (pg_data_t *)arg;
 491
 492         /* Print header */
 493         seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
 494         for (order = 0; order < MAX_ORDER; ++order)
 495                 seq_printf(m, "%6d ", order);
 496         seq_putc(m, '\n');
 497
 498         walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
 499
 500         return 0;
 501 }
 502
 503 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
 504                                         pg_data_t *pgdat, struct zone *zone)
 505 {
 506         int mtype;
 507         unsigned long pfn;
 508         unsigned long start_pfn = zone->zone_start_pfn;
 509         unsigned long end_pfn = start_pfn + zone->spanned_pages;
 510         unsigned long count[MIGRATE_TYPES] = { 0, };
 511
 512         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
 513                 struct page *page;
 514
 515                 if (!pfn_valid(pfn))
 516                         continue;
 517
 518                 page = pfn_to_page(pfn);
 519 #ifdef CONFIG_ARCH_FLATMEM_HAS_HOLES
 520                 /*
 521                  * Ordinarily, memory holes in flatmem still have a valid
 522                  * memmap for the PFN range. However, an architecture for
 523                  * embedded systems (e.g. ARM) can free up the memmap backing
 524                  * holes to save memory on the assumption the memmap is
 525                  * never used. The page_zone linkages are then broken even
 526                  * though pfn_valid() returns true. Skip the page if the
 527                  * linkages are broken. Even if this test passed, the impact
 528                  * is that the counters for the movable type are off but
 529                  * fragmentation monitoring is likely meaningless on small
 530                  * systems.
 531                  */
 532                 if (page_zone(page) != zone)
 533                         continue;
 534 #endif
 535                 mtype = get_pageblock_migratetype(page);
 536
 537                 if (mtype < MIGRATE_TYPES)
 538                         count[mtype]++;
 539         }
 540
 541         /* Print counts */
 542         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 543         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 544                 seq_printf(m, "%12lu ", count[mtype]);
 545         seq_putc(m, '\n');
 546 }
 547
 548 /* Print out the free pages at each order for each migratetype */
 549 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
 550 {
 551         int mtype;
 552         pg_data_t *pgdat = (pg_data_t *)arg;
 553
 554         seq_printf(m, "\n%-23s", "Number of blocks type ");
 555         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 556                 seq_printf(m, "%12s ", migratetype_names[mtype]);
 557         seq_putc(m, '\n');
 558         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
 559
 560         return 0;
 561 }
 562
 563 /*
 564  * This prints out statistics in relation to grouping pages by mobility.
 565  * It is expensive to collect so do not constantly read the file.
 566  */
 567 static int pagetypeinfo_show(struct seq_file *m, void *arg)
 568 {
 569         pg_data_t *pgdat = (pg_data_t *)arg;
 570
 571         /* check memoryless node */
 572         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
 573                 return 0;
 574
 575         seq_printf(m, "Page block order: %d\n", pageblock_order);
 576         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
 577         seq_putc(m, '\n');
 578         pagetypeinfo_showfree(m, pgdat);
 579         pagetypeinfo_showblockcount(m, pgdat);
 580
 581         return 0;
 582 }
 583
 584 static const struct seq_operations fragmentation_op = {
 585         .start  = frag_start,
 586         .next   = frag_next,
 587         .stop   = frag_stop,
 588         .show   = frag_show,
 589 };
 590
 591 static int fragmentation_open(struct inode *inode, struct file *file)
 592 {
 593         return seq_open(file, &fragmentation_op);
 594 }
 595
 596 static const struct file_operations fragmentation_file_operations = {
 597         .open           = fragmentation_open,
 598         .read           = seq_read,
 599         .llseek         = seq_lseek,
 600         .release        = seq_release,
 601 };
 602
 603 static const struct seq_operations pagetypeinfo_op = {
 604         .start  = frag_start,
 605         .next   = frag_next,
 606         .stop   = frag_stop,
 607         .show   = pagetypeinfo_show,
 608 };
 609
 610 static int pagetypeinfo_open(struct inode *inode, struct file *file)
 611 {
 612         return seq_open(file, &pagetypeinfo_op);
 613 }
 614
 615 static const struct file_operations pagetypeinfo_file_ops = {
 616         .open           = pagetypeinfo_open,
 617         .read           = seq_read,
 618         .llseek         = seq_lseek,
 619         .release        = seq_release,
 620 };
 621
 622 #ifdef CONFIG_ZONE_DMA
 623 #define TEXT_FOR_DMA(xx) xx "_dma",
 624 #else
 625 #define TEXT_FOR_DMA(xx)
 626 #endif
 627
 628 #ifdef CONFIG_ZONE_DMA32
 629 #define TEXT_FOR_DMA32(xx) xx "_dma32",
 630 #else
 631 #define TEXT_FOR_DMA32(xx)
 632 #endif
 633
 634 #ifdef CONFIG_HIGHMEM
 635 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
 636 #else
 637 #define TEXT_FOR_HIGHMEM(xx)
 638 #endif
 639
 640 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
 641                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
 642
 643 static const char * const vmstat_text[] = {
 644         /* Zoned VM counters */
 645         "nr_free_pages",
 646         "nr_inactive_anon",
 647         "nr_active_anon",
 648         "nr_inactive_file",
 649         "nr_active_file",
 650 #ifdef CONFIG_UNEVICTABLE_LRU
 651         "nr_unevictable",
 652         "nr_mlock",
 653 #endif
 654         "nr_anon_pages",
 655         "nr_mapped",
 656         "nr_file_pages",
 657         "nr_dirty",
 658         "nr_writeback",
 659         "nr_slab_reclaimable",
 660         "nr_slab_unreclaimable",
 661         "nr_page_table_pages",
 662         "nr_unstable",
 663         "nr_bounce",
 664         "nr_vmscan_write",
 665         "nr_writeback_temp",
 666
 667 #ifdef CONFIG_NUMA
 668         "numa_hit",
 669         "numa_miss",
 670         "numa_foreign",
 671         "numa_interleave",
 672         "numa_local",
 673         "numa_other",
 674 #endif
 675
 676 #ifdef CONFIG_VM_EVENT_COUNTERS
 677         "pgpgin",
 678         "pgpgout",
 679         "pswpin",
 680         "pswpout",
 681
 682         TEXTS_FOR_ZONES("pgalloc")
 683
 684         "pgfree",
 685         "pgactivate",
 686         "pgdeactivate",
 687
 688         "pgfault",
 689         "pgmajfault",
 690
 691         TEXTS_FOR_ZONES("pgrefill")
 692         TEXTS_FOR_ZONES("pgsteal")
 693         TEXTS_FOR_ZONES("pgscan_kswapd")
 694         TEXTS_FOR_ZONES("pgscan_direct")
 695
 696         "pginodesteal",
 697         "slabs_scanned",
 698         "kswapd_steal",
 699         "kswapd_inodesteal",
 700         "pageoutrun",
 701         "allocstall",
 702
 703         "pgrotated",
 704 #ifdef CONFIG_HUGETLB_PAGE
 705         "htlb_buddy_alloc_success",
 706         "htlb_buddy_alloc_fail",
 707 #endif
 708 #ifdef CONFIG_UNEVICTABLE_LRU
 709         "unevictable_pgs_culled",
 710         "unevictable_pgs_scanned",
 711         "unevictable_pgs_rescued",
 712         "unevictable_pgs_mlocked",
 713         "unevictable_pgs_munlocked",
 714         "unevictable_pgs_cleared",
 715         "unevictable_pgs_stranded",
 716         "unevictable_pgs_mlockfreed",
 717 #endif
 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 (aa: %lu ia: %lu af: %lu if: %lu)"
 732                    "\n        spanned  %lu"
 733                    "\n        present  %lu",
 734                    zone_page_state(zone, NR_FREE_PAGES),
 735                    zone->pages_min,
 736                    zone->pages_low,
 737                    zone->pages_high,
 738                    zone->pages_scanned,
 739                    zone->lru[LRU_ACTIVE_ANON].nr_scan,
 740                    zone->lru[LRU_INACTIVE_ANON].nr_scan,
 741                    zone->lru[LRU_ACTIVE_FILE].nr_scan,
 742                    zone->lru[LRU_INACTIVE_FILE].nr_scan,
 743                    zone->spanned_pages,
 744                    zone->present_pages);
 745
 746         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 747                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
 748                                 zone_page_state(zone, i));
 749
 750         seq_printf(m,
 751                    "\n        protection: (%lu",
 752                    zone->lowmem_reserve[0]);
 753         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
 754                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
 755         seq_printf(m,
 756                    ")"
 757                    "\n  pagesets");
 758         for_each_online_cpu(i) {
 759                 struct per_cpu_pageset *pageset;
 760
 761                 pageset = zone_pcp(zone, i);
 762                 seq_printf(m,
 763                            "\n    cpu: %i"
 764                            "\n              count: %i"
 765                            "\n              high:  %i"
 766                            "\n              batch: %i",
 767                            i,
 768                            pageset->pcp.count,
 769                            pageset->pcp.high,
 770                            pageset->pcp.batch);
 771 #ifdef CONFIG_SMP
 772                 seq_printf(m, "\n  vm stats threshold: %d",
 773                                 pageset->stat_threshold);
 774 #endif
 775         }
 776         seq_printf(m,
 777                    "\n  all_unreclaimable: %u"
 778                    "\n  prev_priority:     %i"
 779                    "\n  start_pfn:         %lu"
 780                    "\n  inactive_ratio:    %u",
 781                            zone_is_all_unreclaimable(zone),
 782                    zone->prev_priority,
 783                    zone->zone_start_pfn,
 784                    zone->inactive_ratio);
 785         seq_putc(m, '\n');
 786 }
 787
 788 /*
 789  * Output information about zones in @pgdat.
 790  */
 791 static int zoneinfo_show(struct seq_file *m, void *arg)
 792 {
 793         pg_data_t *pgdat = (pg_data_t *)arg;
 794         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
 795         return 0;
 796 }
 797
 798 static const struct seq_operations zoneinfo_op = {
 799         .start  = frag_start, /* iterate over all zones. The same as in
 800                                * fragmentation. */
 801         .next   = frag_next,
 802         .stop   = frag_stop,
 803         .show   = zoneinfo_show,
 804 };
 805
 806 static int zoneinfo_open(struct inode *inode, struct file *file)
 807 {
 808         return seq_open(file, &zoneinfo_op);
 809 }
 810
 811 static const struct file_operations proc_zoneinfo_file_operations = {
 812         .open           = zoneinfo_open,
 813         .read           = seq_read,
 814         .llseek         = seq_lseek,
 815         .release        = seq_release,
 816 };
 817
 818 static void *vmstat_start(struct seq_file *m, loff_t *pos)
 819 {
 820         unsigned long *v;
 821 #ifdef CONFIG_VM_EVENT_COUNTERS
 822         unsigned long *e;
 823 #endif
 824         int i;
 825
 826         if (*pos >= ARRAY_SIZE(vmstat_text))
 827                 return NULL;
 828
 829 #ifdef CONFIG_VM_EVENT_COUNTERS
 830         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
 831                         + sizeof(struct vm_event_state), GFP_KERNEL);
 832 #else
 833         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
 834                         GFP_KERNEL);
 835 #endif
 836         m->private = v;
 837         if (!v)
 838                 return ERR_PTR(-ENOMEM);
 839         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 840                 v[i] = global_page_state(i);
 841 #ifdef CONFIG_VM_EVENT_COUNTERS
 842         e = v + NR_VM_ZONE_STAT_ITEMS;
 843         all_vm_events(e);
 844         e[PGPGIN] /= 2;         /* sectors -> kbytes */
 845         e[PGPGOUT] /= 2;
 846 #endif
 847         return v + *pos;
 848 }
 849
 850 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
 851 {
 852         (*pos)++;
 853         if (*pos >= ARRAY_SIZE(vmstat_text))
 854                 return NULL;
 855         return (unsigned long *)m->private + *pos;
 856 }
 857
 858 static int vmstat_show(struct seq_file *m, void *arg)
 859 {
 860         unsigned long *l = arg;
 861         unsigned long off = l - (unsigned long *)m->private;
 862
 863         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
 864         return 0;
 865 }
 866
 867 static void vmstat_stop(struct seq_file *m, void *arg)
 868 {
 869         kfree(m->private);
 870         m->private = NULL;
 871 }
 872
 873 static const struct seq_operations vmstat_op = {
 874         .start  = vmstat_start,
 875         .next   = vmstat_next,
 876         .stop   = vmstat_stop,
 877         .show   = vmstat_show,
 878 };
 879
 880 static int vmstat_open(struct inode *inode, struct file *file)
 881 {
 882         return seq_open(file, &vmstat_op);
 883 }
 884
 885 static const struct file_operations proc_vmstat_file_operations = {
 886         .open           = vmstat_open,
 887         .read           = seq_read,
 888         .llseek         = seq_lseek,
 889         .release        = seq_release,
 890 };
 891 #endif /* CONFIG_PROC_FS */
 892
 893 #ifdef CONFIG_SMP
 894 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
 895 int sysctl_stat_interval __read_mostly = HZ;
 896
 897 static void vmstat_update(struct work_struct *w)
 898 {
 899         refresh_cpu_vm_stats(smp_processor_id());
 900         schedule_delayed_work(&__get_cpu_var(vmstat_work),
 901                 sysctl_stat_interval);
 902 }
 903
 904 static void __cpuinit start_cpu_timer(int cpu)
 905 {
 906         struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
 907
 908         INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
 909         schedule_delayed_work_on(cpu, vmstat_work, HZ + cpu);
 910 }
 911
 912 /*
 913  * Use the cpu notifier to insure that the thresholds are recalculated
 914  * when necessary.
 915  */
 916 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
 917                 unsigned long action,
 918                 void *hcpu)
 919 {
 920         long cpu = (long)hcpu;
 921
 922         switch (action) {
 923         case CPU_ONLINE:
 924         case CPU_ONLINE_FROZEN:
 925                 start_cpu_timer(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)