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/config.h>
  13 #include <linux/mm.h>
  14 #include <linux/module.h>
  15
  16 /*
  17  * Accumulate the page_state information across all CPUs.
  18  * The result is unavoidably approximate - it can change
  19  * during and after execution of this function.
  20  */
  21 DEFINE_PER_CPU(struct page_state, page_states) = {0};
  22
  23 static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
  24 {
  25         unsigned cpu;
  26
  27         memset(ret, 0, nr * sizeof(unsigned long));
  28         cpus_and(*cpumask, *cpumask, cpu_online_map);
  29
  30         for_each_cpu_mask(cpu, *cpumask) {
  31                 unsigned long *in;
  32                 unsigned long *out;
  33                 unsigned off;
  34                 unsigned next_cpu;
  35
  36                 in = (unsigned long *)&per_cpu(page_states, cpu);
  37
  38                 next_cpu = next_cpu(cpu, *cpumask);
  39                 if (likely(next_cpu < NR_CPUS))
  40                         prefetch(&per_cpu(page_states, next_cpu));
  41
  42                 out = (unsigned long *)ret;
  43                 for (off = 0; off < nr; off++)
  44                         *out++ += *in++;
  45         }
  46 }
  47
  48 void get_full_page_state(struct page_state *ret)
  49 {
  50         cpumask_t mask = CPU_MASK_ALL;
  51
  52         __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
  53 }
  54
  55 void __mod_page_state_offset(unsigned long offset, unsigned long delta)
  56 {
  57         void *ptr;
  58
  59         ptr = &__get_cpu_var(page_states);
  60         *(unsigned long *)(ptr + offset) += delta;
  61 }
  62 EXPORT_SYMBOL(__mod_page_state_offset);
  63
  64 void mod_page_state_offset(unsigned long offset, unsigned long delta)
  65 {
  66         unsigned long flags;
  67         void *ptr;
  68
  69         local_irq_save(flags);
  70         ptr = &__get_cpu_var(page_states);
  71         *(unsigned long *)(ptr + offset) += delta;
  72         local_irq_restore(flags);
  73 }
  74 EXPORT_SYMBOL(mod_page_state_offset);
  75
  76 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
  77                         unsigned long *free, struct pglist_data *pgdat)
  78 {
  79         struct zone *zones = pgdat->node_zones;
  80         int i;
  81
  82         *active = 0;
  83         *inactive = 0;
  84         *free = 0;
  85         for (i = 0; i < MAX_NR_ZONES; i++) {
  86                 *active += zones[i].nr_active;
  87                 *inactive += zones[i].nr_inactive;
  88                 *free += zones[i].free_pages;
  89         }
  90 }
  91
  92 void get_zone_counts(unsigned long *active,
  93                 unsigned long *inactive, unsigned long *free)
  94 {
  95         struct pglist_data *pgdat;
  96
  97         *active = 0;
  98         *inactive = 0;
  99         *free = 0;
 100         for_each_online_pgdat(pgdat) {
 101                 unsigned long l, m, n;
 102                 __get_zone_counts(&l, &m, &n, pgdat);
 103                 *active += l;
 104                 *inactive += m;
 105                 *free += n;
 106         }
 107 }
 108
 109 /*
 110  * Manage combined zone based / global counters
 111  *
 112  * vm_stat contains the global counters
 113  */
 114 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
 115 EXPORT_SYMBOL(vm_stat);
 116
 117 #ifdef CONFIG_SMP
 118
 119 #define STAT_THRESHOLD 32
 120
 121 /*
 122  * Determine pointer to currently valid differential byte given a zone and
 123  * the item number.
 124  *
 125  * Preemption must be off
 126  */
 127 static inline s8 *diff_pointer(struct zone *zone, enum zone_stat_item item)
 128 {
 129         return &zone_pcp(zone, smp_processor_id())->vm_stat_diff[item];
 130 }
 131
 132 /*
 133  * For use when we know that interrupts are disabled.
 134  */
 135 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 136                                 int delta)
 137 {
 138         s8 *p;
 139         long x;
 140
 141         p = diff_pointer(zone, item);
 142         x = delta + *p;
 143
 144         if (unlikely(x > STAT_THRESHOLD || x < -STAT_THRESHOLD)) {
 145                 zone_page_state_add(x, zone, item);
 146                 x = 0;
 147         }
 148
 149         *p = x;
 150 }
 151 EXPORT_SYMBOL(__mod_zone_page_state);
 152
 153 /*
 154  * For an unknown interrupt state
 155  */
 156 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 157                                         int delta)
 158 {
 159         unsigned long flags;
 160
 161         local_irq_save(flags);
 162         __mod_zone_page_state(zone, item, delta);
 163         local_irq_restore(flags);
 164 }
 165 EXPORT_SYMBOL(mod_zone_page_state);
 166
 167 /*
 168  * Optimized increment and decrement functions.
 169  *
 170  * These are only for a single page and therefore can take a struct page *
 171  * argument instead of struct zone *. This allows the inclusion of the code
 172  * generated for page_zone(page) into the optimized functions.
 173  *
 174  * No overflow check is necessary and therefore the differential can be
 175  * incremented or decremented in place which may allow the compilers to
 176  * generate better code.
 177  *
 178  * The increment or decrement is known and therefore one boundary check can
 179  * be omitted.
 180  *
 181  * Some processors have inc/dec instructions that are atomic vs an interrupt.
 182  * However, the code must first determine the differential location in a zone
 183  * based on the processor number and then inc/dec the counter. There is no
 184  * guarantee without disabling preemption that the processor will not change
 185  * in between and therefore the atomicity vs. interrupt cannot be exploited
 186  * in a useful way here.
 187  */
 188 static void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
 189 {
 190         s8 *p = diff_pointer(zone, item);
 191
 192         (*p)++;
 193
 194         if (unlikely(*p > STAT_THRESHOLD)) {
 195                 zone_page_state_add(*p, zone, item);
 196                 *p = 0;
 197         }
 198 }
 199
 200 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
 201 {
 202         __inc_zone_state(page_zone(page), item);
 203 }
 204 EXPORT_SYMBOL(__inc_zone_page_state);
 205
 206 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
 207 {
 208         struct zone *zone = page_zone(page);
 209         s8 *p = diff_pointer(zone, item);
 210
 211         (*p)--;
 212
 213         if (unlikely(*p < -STAT_THRESHOLD)) {
 214                 zone_page_state_add(*p, zone, item);
 215                 *p = 0;
 216         }
 217 }
 218 EXPORT_SYMBOL(__dec_zone_page_state);
 219
 220 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
 221 {
 222         unsigned long flags;
 223
 224         local_irq_save(flags);
 225         __inc_zone_state(zone, item);
 226         local_irq_restore(flags);
 227 }
 228
 229 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
 230 {
 231         unsigned long flags;
 232         struct zone *zone;
 233
 234         zone = page_zone(page);
 235         local_irq_save(flags);
 236         __inc_zone_state(zone, item);
 237         local_irq_restore(flags);
 238 }
 239 EXPORT_SYMBOL(inc_zone_page_state);
 240
 241 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
 242 {
 243         unsigned long flags;
 244         struct zone *zone;
 245         s8 *p;
 246
 247         zone = page_zone(page);
 248         local_irq_save(flags);
 249         p = diff_pointer(zone, item);
 250
 251         (*p)--;
 252
 253         if (unlikely(*p < -STAT_THRESHOLD)) {
 254                 zone_page_state_add(*p, zone, item);
 255                 *p = 0;
 256         }
 257         local_irq_restore(flags);
 258 }
 259 EXPORT_SYMBOL(dec_zone_page_state);
 260
 261 /*
 262  * Update the zone counters for one cpu.
 263  */
 264 void refresh_cpu_vm_stats(int cpu)
 265 {
 266         struct zone *zone;
 267         int i;
 268         unsigned long flags;
 269
 270         for_each_zone(zone) {
 271                 struct per_cpu_pageset *pcp;
 272
 273                 pcp = zone_pcp(zone, cpu);
 274
 275                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 276                         if (pcp->vm_stat_diff[i]) {
 277                                 local_irq_save(flags);
 278                                 zone_page_state_add(pcp->vm_stat_diff[i],
 279                                         zone, i);
 280                                 pcp->vm_stat_diff[i] = 0;
 281                                 local_irq_restore(flags);
 282                         }
 283         }
 284 }
 285
 286 static void __refresh_cpu_vm_stats(void *dummy)
 287 {
 288         refresh_cpu_vm_stats(smp_processor_id());
 289 }
 290
 291 /*
 292  * Consolidate all counters.
 293  *
 294  * Note that the result is less inaccurate but still inaccurate
 295  * if concurrent processes are allowed to run.
 296  */
 297 void refresh_vm_stats(void)
 298 {
 299         on_each_cpu(__refresh_cpu_vm_stats, NULL, 0, 1);
 300 }
 301 EXPORT_SYMBOL(refresh_vm_stats);
 302
 303 #endif
 304
 305 #ifdef CONFIG_NUMA
 306 /*
 307  * zonelist = the list of zones passed to the allocator
 308  * z        = the zone from which the allocation occurred.
 309  *
 310  * Must be called with interrupts disabled.
 311  */
 312 void zone_statistics(struct zonelist *zonelist, struct zone *z)
 313 {
 314         if (z->zone_pgdat == zonelist->zones[0]->zone_pgdat) {
 315                 __inc_zone_state(z, NUMA_HIT);
 316         } else {
 317                 __inc_zone_state(z, NUMA_MISS);
 318                 __inc_zone_state(zonelist->zones[0], NUMA_FOREIGN);
 319         }
 320         if (z->zone_pgdat == NODE_DATA(numa_node_id()))
 321                 __inc_zone_state(z, NUMA_LOCAL);
 322         else
 323                 __inc_zone_state(z, NUMA_OTHER);
 324 }
 325 #endif
 326
 327 #ifdef CONFIG_PROC_FS
 328
 329 #include <linux/seq_file.h>
 330
 331 static void *frag_start(struct seq_file *m, loff_t *pos)
 332 {
 333         pg_data_t *pgdat;
 334         loff_t node = *pos;
 335         for (pgdat = first_online_pgdat();
 336              pgdat && node;
 337              pgdat = next_online_pgdat(pgdat))
 338                 --node;
 339
 340         return pgdat;
 341 }
 342
 343 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
 344 {
 345         pg_data_t *pgdat = (pg_data_t *)arg;
 346
 347         (*pos)++;
 348         return next_online_pgdat(pgdat);
 349 }
 350
 351 static void frag_stop(struct seq_file *m, void *arg)
 352 {
 353 }
 354
 355 /*
 356  * This walks the free areas for each zone.
 357  */
 358 static int frag_show(struct seq_file *m, void *arg)
 359 {
 360         pg_data_t *pgdat = (pg_data_t *)arg;
 361         struct zone *zone;
 362         struct zone *node_zones = pgdat->node_zones;
 363         unsigned long flags;
 364         int order;
 365
 366         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
 367                 if (!populated_zone(zone))
 368                         continue;
 369
 370                 spin_lock_irqsave(&zone->lock, flags);
 371                 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 372                 for (order = 0; order < MAX_ORDER; ++order)
 373                         seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
 374                 spin_unlock_irqrestore(&zone->lock, flags);
 375                 seq_putc(m, '\n');
 376         }
 377         return 0;
 378 }
 379
 380 struct seq_operations fragmentation_op = {
 381         .start  = frag_start,
 382         .next   = frag_next,
 383         .stop   = frag_stop,
 384         .show   = frag_show,
 385 };
 386
 387 static char *vmstat_text[] = {
 388         /* Zoned VM counters */
 389         "nr_anon_pages",
 390         "nr_mapped",
 391         "nr_file_pages",
 392         "nr_slab",
 393         "nr_page_table_pages",
 394         "nr_dirty",
 395         "nr_writeback",
 396         "nr_unstable",
 397         "nr_bounce",
 398
 399 #ifdef CONFIG_NUMA
 400         "numa_hit",
 401         "numa_miss",
 402         "numa_foreign",
 403         "numa_interleave",
 404         "numa_local",
 405         "numa_other",
 406 #endif
 407
 408         /* Event counters */
 409         "pgpgin",
 410         "pgpgout",
 411         "pswpin",
 412         "pswpout",
 413
 414         "pgalloc_high",
 415         "pgalloc_normal",
 416         "pgalloc_dma32",
 417         "pgalloc_dma",
 418
 419         "pgfree",
 420         "pgactivate",
 421         "pgdeactivate",
 422
 423         "pgfault",
 424         "pgmajfault",
 425
 426         "pgrefill_high",
 427         "pgrefill_normal",
 428         "pgrefill_dma32",
 429         "pgrefill_dma",
 430
 431         "pgsteal_high",
 432         "pgsteal_normal",
 433         "pgsteal_dma32",
 434         "pgsteal_dma",
 435
 436         "pgscan_kswapd_high",
 437         "pgscan_kswapd_normal",
 438         "pgscan_kswapd_dma32",
 439         "pgscan_kswapd_dma",
 440
 441         "pgscan_direct_high",
 442         "pgscan_direct_normal",
 443         "pgscan_direct_dma32",
 444         "pgscan_direct_dma",
 445
 446         "pginodesteal",
 447         "slabs_scanned",
 448         "kswapd_steal",
 449         "kswapd_inodesteal",
 450         "pageoutrun",
 451         "allocstall",
 452
 453         "pgrotated",
 454 };
 455
 456 /*
 457  * Output information about zones in @pgdat.
 458  */
 459 static int zoneinfo_show(struct seq_file *m, void *arg)
 460 {
 461         pg_data_t *pgdat = arg;
 462         struct zone *zone;
 463         struct zone *node_zones = pgdat->node_zones;
 464         unsigned long flags;
 465
 466         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
 467                 int i;
 468
 469                 if (!populated_zone(zone))
 470                         continue;
 471
 472                 spin_lock_irqsave(&zone->lock, flags);
 473                 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
 474                 seq_printf(m,
 475                            "\n  pages free     %lu"
 476                            "\n        min      %lu"
 477                            "\n        low      %lu"
 478                            "\n        high     %lu"
 479                            "\n        active   %lu"
 480                            "\n        inactive %lu"
 481                            "\n        scanned  %lu (a: %lu i: %lu)"
 482                            "\n        spanned  %lu"
 483                            "\n        present  %lu",
 484                            zone->free_pages,
 485                            zone->pages_min,
 486                            zone->pages_low,
 487                            zone->pages_high,
 488                            zone->nr_active,
 489                            zone->nr_inactive,
 490                            zone->pages_scanned,
 491                            zone->nr_scan_active, zone->nr_scan_inactive,
 492                            zone->spanned_pages,
 493                            zone->present_pages);
 494
 495                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 496                         seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
 497                                         zone_page_state(zone, i));
 498
 499                 seq_printf(m,
 500                            "\n        protection: (%lu",
 501                            zone->lowmem_reserve[0]);
 502                 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
 503                         seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
 504                 seq_printf(m,
 505                            ")"
 506                            "\n  pagesets");
 507                 for_each_online_cpu(i) {
 508                         struct per_cpu_pageset *pageset;
 509                         int j;
 510
 511                         pageset = zone_pcp(zone, i);
 512                         for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
 513                                 if (pageset->pcp[j].count)
 514                                         break;
 515                         }
 516                         if (j == ARRAY_SIZE(pageset->pcp))
 517                                 continue;
 518                         for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
 519                                 seq_printf(m,
 520                                            "\n    cpu: %i pcp: %i"
 521                                            "\n              count: %i"
 522                                            "\n              high:  %i"
 523                                            "\n              batch: %i",
 524                                            i, j,
 525                                            pageset->pcp[j].count,
 526                                            pageset->pcp[j].high,
 527                                            pageset->pcp[j].batch);
 528                         }
 529                 }
 530                 seq_printf(m,
 531                            "\n  all_unreclaimable: %u"
 532                            "\n  prev_priority:     %i"
 533                            "\n  temp_priority:     %i"
 534                            "\n  start_pfn:         %lu",
 535                            zone->all_unreclaimable,
 536                            zone->prev_priority,
 537                            zone->temp_priority,
 538                            zone->zone_start_pfn);
 539                 spin_unlock_irqrestore(&zone->lock, flags);
 540                 seq_putc(m, '\n');
 541         }
 542         return 0;
 543 }
 544
 545 struct seq_operations zoneinfo_op = {
 546         .start  = frag_start, /* iterate over all zones. The same as in
 547                                * fragmentation. */
 548         .next   = frag_next,
 549         .stop   = frag_stop,
 550         .show   = zoneinfo_show,
 551 };
 552
 553 static void *vmstat_start(struct seq_file *m, loff_t *pos)
 554 {
 555         unsigned long *v;
 556         struct page_state *ps;
 557         int i;
 558
 559         if (*pos >= ARRAY_SIZE(vmstat_text))
 560                 return NULL;
 561
 562         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
 563                         + sizeof(*ps), GFP_KERNEL);
 564         m->private = v;
 565         if (!v)
 566                 return ERR_PTR(-ENOMEM);
 567         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 568                 v[i] = global_page_state(i);
 569         ps = (struct page_state *)(v + NR_VM_ZONE_STAT_ITEMS);
 570         get_full_page_state(ps);
 571         ps->pgpgin /= 2;                /* sectors -> kbytes */
 572         ps->pgpgout /= 2;
 573         return v + *pos;
 574 }
 575
 576 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
 577 {
 578         (*pos)++;
 579         if (*pos >= ARRAY_SIZE(vmstat_text))
 580                 return NULL;
 581         return (unsigned long *)m->private + *pos;
 582 }
 583
 584 static int vmstat_show(struct seq_file *m, void *arg)
 585 {
 586         unsigned long *l = arg;
 587         unsigned long off = l - (unsigned long *)m->private;
 588
 589         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
 590         return 0;
 591 }
 592
 593 static void vmstat_stop(struct seq_file *m, void *arg)
 594 {
 595         kfree(m->private);
 596         m->private = NULL;
 597 }
 598
 599 struct seq_operations vmstat_op = {
 600         .start  = vmstat_start,
 601         .next   = vmstat_next,
 602         .stop   = vmstat_stop,
 603         .show   = vmstat_show,
 604 };
 605
 606 #endif /* CONFIG_PROC_FS */
 607