mm/vmstat.c

   1 /*
   2  *  linux/mm/vmstat.c
   3  *
   4  *  Manages VM statistics
   5  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   6  *
   7  *  zoned VM statistics
   8  *  Copyright (C) 2006 Silicon Graphics, Inc.,
   9  *              Christoph Lameter <christoph@lameter.com>
  10  */
  11
  12 #include <linux/mm.h>
  13 #include <linux/err.h>
  14 #include <linux/module.h>
  15 #include <linux/cpu.h>
  16 #include <linux/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, cpumask_t *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_map);
  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
 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 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 const struct seq_operations pagetypeinfo_op = {
 592         .start  = frag_start,
 593         .next   = frag_next,
 594         .stop   = frag_stop,
 595         .show   = pagetypeinfo_show,
 596 };
 597
 598 #ifdef CONFIG_ZONE_DMA
 599 #define TEXT_FOR_DMA(xx) xx "_dma",
 600 #else
 601 #define TEXT_FOR_DMA(xx)
 602 #endif
 603
 604 #ifdef CONFIG_ZONE_DMA32
 605 #define TEXT_FOR_DMA32(xx) xx "_dma32",
 606 #else
 607 #define TEXT_FOR_DMA32(xx)
 608 #endif
 609
 610 #ifdef CONFIG_HIGHMEM
 611 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
 612 #else
 613 #define TEXT_FOR_HIGHMEM(xx)
 614 #endif
 615
 616 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
 617                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
 618
 619 static const char * const vmstat_text[] = {
 620         /* Zoned VM counters */
 621         "nr_free_pages",
 622         "nr_inactive_anon",
 623         "nr_active_anon",
 624         "nr_inactive_file",
 625         "nr_active_file",
 626 #ifdef CONFIG_UNEVICTABLE_LRU
 627         "nr_unevictable",
 628         "nr_mlock",
 629 #endif
 630         "nr_anon_pages",
 631         "nr_mapped",
 632         "nr_file_pages",
 633         "nr_dirty",
 634         "nr_writeback",
 635         "nr_slab_reclaimable",
 636         "nr_slab_unreclaimable",
 637         "nr_page_table_pages",
 638         "nr_unstable",
 639         "nr_bounce",
 640         "nr_vmscan_write",
 641         "nr_writeback_temp",
 642
 643 #ifdef CONFIG_NUMA
 644         "numa_hit",
 645         "numa_miss",
 646         "numa_foreign",
 647         "numa_interleave",
 648         "numa_local",
 649         "numa_other",
 650 #endif
 651
 652 #ifdef CONFIG_VM_EVENT_COUNTERS
 653         "pgpgin",
 654         "pgpgout",
 655         "pswpin",
 656         "pswpout",
 657
 658         TEXTS_FOR_ZONES("pgalloc")
 659
 660         "pgfree",
 661         "pgactivate",
 662         "pgdeactivate",
 663
 664         "pgfault",
 665         "pgmajfault",
 666
 667         TEXTS_FOR_ZONES("pgrefill")
 668         TEXTS_FOR_ZONES("pgsteal")
 669         TEXTS_FOR_ZONES("pgscan_kswapd")
 670         TEXTS_FOR_ZONES("pgscan_direct")
 671
 672         "pginodesteal",
 673         "slabs_scanned",
 674         "kswapd_steal",
 675         "kswapd_inodesteal",
 676         "pageoutrun",
 677         "allocstall",
 678
 679         "pgrotated",
 680 #ifdef CONFIG_HUGETLB_PAGE
 681         "htlb_buddy_alloc_success",
 682         "htlb_buddy_alloc_fail",
 683 #endif
 684 #ifdef CONFIG_UNEVICTABLE_LRU
 685         "unevictable_pgs_culled",
 686         "unevictable_pgs_scanned",
 687         "unevictable_pgs_rescued",
 688         "unevictable_pgs_mlocked",
 689         "unevictable_pgs_munlocked",
 690         "unevictable_pgs_cleared",
 691         "unevictable_pgs_stranded",
 692         "unevictable_pgs_mlockfreed",
 693 #endif
 694 #endif
 695 };
 696
 697 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
 698                                                         struct zone *zone)
 699 {
 700         int i;
 701         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
 702         seq_printf(m,
 703                    "\n  pages free     %lu"
 704                    "\n        min      %lu"
 705                    "\n        low      %lu"
 706                    "\n        high     %lu"
 707                    "\n        scanned  %lu (aa: %lu ia: %lu af: %lu if: %lu)"
 708                    "\n        spanned  %lu"
 709                    "\n        present  %lu",
 710                    zone_page_state(zone, NR_FREE_PAGES),
 711                    zone->pages_min,
 712                    zone->pages_low,
 713                    zone->pages_high,
 714                    zone->pages_scanned,
 715                    zone->lru[LRU_ACTIVE_ANON].nr_scan,
 716                    zone->lru[LRU_INACTIVE_ANON].nr_scan,
 717                    zone->lru[LRU_ACTIVE_FILE].nr_scan,
 718                    zone->lru[LRU_INACTIVE_FILE].nr_scan,
 719                    zone->spanned_pages,
 720                    zone->present_pages);
 721
 722         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 723                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
 724                                 zone_page_state(zone, i));
 725
 726         seq_printf(m,
 727                    "\n        protection: (%lu",
 728                    zone->lowmem_reserve[0]);
 729         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
 730                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
 731         seq_printf(m,
 732                    ")"
 733                    "\n  pagesets");
 734         for_each_online_cpu(i) {
 735                 struct per_cpu_pageset *pageset;
 736
 737                 pageset = zone_pcp(zone, i);
 738                 seq_printf(m,
 739                            "\n    cpu: %i"
 740                            "\n              count: %i"
 741                            "\n              high:  %i"
 742                            "\n              batch: %i",
 743                            i,
 744                            pageset->pcp.count,
 745                            pageset->pcp.high,
 746                            pageset->pcp.batch);
 747 #ifdef CONFIG_SMP
 748                 seq_printf(m, "\n  vm stats threshold: %d",
 749                                 pageset->stat_threshold);
 750 #endif
 751         }
 752         seq_printf(m,
 753                    "\n  all_unreclaimable: %u"
 754                    "\n  prev_priority:     %i"
 755                    "\n  start_pfn:         %lu"
 756                    "\n  inactive_ratio:    %u",
 757                            zone_is_all_unreclaimable(zone),
 758                    zone->prev_priority,
 759                    zone->zone_start_pfn,
 760                    zone->inactive_ratio);
 761         seq_putc(m, '\n');
 762 }
 763
 764 /*
 765  * Output information about zones in @pgdat.
 766  */
 767 static int zoneinfo_show(struct seq_file *m, void *arg)
 768 {
 769         pg_data_t *pgdat = (pg_data_t *)arg;
 770         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
 771         return 0;
 772 }
 773
 774 const struct seq_operations zoneinfo_op = {
 775         .start  = frag_start, /* iterate over all zones. The same as in
 776                                * fragmentation. */
 777         .next   = frag_next,
 778         .stop   = frag_stop,
 779         .show   = zoneinfo_show,
 780 };
 781
 782 static void *vmstat_start(struct seq_file *m, loff_t *pos)
 783 {
 784         unsigned long *v;
 785 #ifdef CONFIG_VM_EVENT_COUNTERS
 786         unsigned long *e;
 787 #endif
 788         int i;
 789
 790         if (*pos >= ARRAY_SIZE(vmstat_text))
 791                 return NULL;
 792
 793 #ifdef CONFIG_VM_EVENT_COUNTERS
 794         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
 795                         + sizeof(struct vm_event_state), GFP_KERNEL);
 796 #else
 797         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
 798                         GFP_KERNEL);
 799 #endif
 800         m->private = v;
 801         if (!v)
 802                 return ERR_PTR(-ENOMEM);
 803         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 804                 v[i] = global_page_state(i);
 805 #ifdef CONFIG_VM_EVENT_COUNTERS
 806         e = v + NR_VM_ZONE_STAT_ITEMS;
 807         all_vm_events(e);
 808         e[PGPGIN] /= 2;         /* sectors -> kbytes */
 809         e[PGPGOUT] /= 2;
 810 #endif
 811         return v + *pos;
 812 }
 813
 814 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
 815 {
 816         (*pos)++;
 817         if (*pos >= ARRAY_SIZE(vmstat_text))
 818                 return NULL;
 819         return (unsigned long *)m->private + *pos;
 820 }
 821
 822 static int vmstat_show(struct seq_file *m, void *arg)
 823 {
 824         unsigned long *l = arg;
 825         unsigned long off = l - (unsigned long *)m->private;
 826
 827         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
 828         return 0;
 829 }
 830
 831 static void vmstat_stop(struct seq_file *m, void *arg)
 832 {
 833         kfree(m->private);
 834         m->private = NULL;
 835 }
 836
 837 const struct seq_operations vmstat_op = {
 838         .start  = vmstat_start,
 839         .next   = vmstat_next,
 840         .stop   = vmstat_stop,
 841         .show   = vmstat_show,
 842 };
 843
 844 #endif /* CONFIG_PROC_FS */
 845
 846 #ifdef CONFIG_SMP
 847 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
 848 int sysctl_stat_interval __read_mostly = HZ;
 849
 850 static void vmstat_update(struct work_struct *w)
 851 {
 852         refresh_cpu_vm_stats(smp_processor_id());
 853         schedule_delayed_work(&__get_cpu_var(vmstat_work),
 854                 sysctl_stat_interval);
 855 }
 856
 857 static void __cpuinit start_cpu_timer(int cpu)
 858 {
 859         struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
 860
 861         INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
 862         schedule_delayed_work_on(cpu, vmstat_work, HZ + cpu);
 863 }
 864
 865 /*
 866  * Use the cpu notifier to insure that the thresholds are recalculated
 867  * when necessary.
 868  */
 869 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
 870                 unsigned long action,
 871                 void *hcpu)
 872 {
 873         long cpu = (long)hcpu;
 874
 875         switch (action) {
 876         case CPU_ONLINE:
 877         case CPU_ONLINE_FROZEN:
 878                 start_cpu_timer(cpu);
 879                 break;
 880         case CPU_DOWN_PREPARE:
 881         case CPU_DOWN_PREPARE_FROZEN:
 882                 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
 883                 per_cpu(vmstat_work, cpu).work.func = NULL;
 884                 break;
 885         case CPU_DOWN_FAILED:
 886         case CPU_DOWN_FAILED_FROZEN:
 887                 start_cpu_timer(cpu);
 888                 break;
 889         case CPU_DEAD:
 890         case CPU_DEAD_FROZEN:
 891                 refresh_zone_stat_thresholds();
 892                 break;
 893         default:
 894                 break;
 895         }
 896         return NOTIFY_OK;
 897 }
 898
 899 static struct notifier_block __cpuinitdata vmstat_notifier =
 900         { &vmstat_cpuup_callback, NULL, 0 };
 901
 902 static int __init setup_vmstat(void)
 903 {
 904         int cpu;
 905
 906         refresh_zone_stat_thresholds();
 907         register_cpu_notifier(&vmstat_notifier);
 908
 909         for_each_online_cpu(cpu)
 910                 start_cpu_timer(cpu);
 911         return 0;
 912 }
 913 module_init(setup_vmstat)
 914 #endif