mm/vmstat.c

   1 /*
   2  *  linux/mm/vmstat.c
   3  *
   4  *  Manages VM statistics
   5  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   6  *
   7  *  zoned VM statistics
   8  *  Copyright (C) 2006 Silicon Graphics, Inc.,
   9  *              Christoph Lameter <christoph@lameter.com>
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/mm.h>
  13 #include <linux/err.h>
  14 #include <linux/module.h>
  15 #include <linux/slab.h>
  16 #include <linux/cpu.h>
  17 #include <linux/vmstat.h>
  18 #include <linux/sched.h>
  19 #include <linux/math64.h>
  20
  21 #ifdef CONFIG_VM_EVENT_COUNTERS
  22 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
  23 EXPORT_PER_CPU_SYMBOL(vm_event_states);
  24
  25 static void sum_vm_events(unsigned long *ret)
  26 {
  27         int cpu;
  28         int i;
  29
  30         memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
  31
  32         for_each_online_cpu(cpu) {
  33                 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
  34
  35                 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
  36                         ret[i] += this->event[i];
  37         }
  38 }
  39
  40 /*
  41  * Accumulate the vm event counters across all CPUs.
  42  * The result is unavoidably approximate - it can change
  43  * during and after execution of this function.
  44 */
  45 void all_vm_events(unsigned long *ret)
  46 {
  47         get_online_cpus();
  48         sum_vm_events(ret);
  49         put_online_cpus();
  50 }
  51 EXPORT_SYMBOL_GPL(all_vm_events);
  52
  53 #ifdef CONFIG_HOTPLUG
  54 /*
  55  * Fold the foreign cpu events into our own.
  56  *
  57  * This is adding to the events on one processor
  58  * but keeps the global counts constant.
  59  */
  60 void vm_events_fold_cpu(int cpu)
  61 {
  62         struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
  63         int i;
  64
  65         for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
  66                 count_vm_events(i, fold_state->event[i]);
  67                 fold_state->event[i] = 0;
  68         }
  69 }
  70 #endif /* CONFIG_HOTPLUG */
  71
  72 #endif /* CONFIG_VM_EVENT_COUNTERS */
  73
  74 /*
  75  * Manage combined zone based / global counters
  76  *
  77  * vm_stat contains the global counters
  78  */
  79 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
  80 EXPORT_SYMBOL(vm_stat);
  81
  82 #ifdef CONFIG_SMP
  83
  84 static int calculate_threshold(struct zone *zone)
  85 {
  86         int threshold;
  87         int mem;        /* memory in 128 MB units */
  88
  89         /*
  90          * The threshold scales with the number of processors and the amount
  91          * of memory per zone. More memory means that we can defer updates for
  92          * longer, more processors could lead to more contention.
  93          * fls() is used to have a cheap way of logarithmic scaling.
  94          *
  95          * Some sample thresholds:
  96          *
  97          * Threshold    Processors      (fls)   Zonesize        fls(mem+1)
  98          * ------------------------------------------------------------------
  99          * 8            1               1       0.9-1 GB        4
 100          * 16           2               2       0.9-1 GB        4
 101          * 20           2               2       1-2 GB          5
 102          * 24           2               2       2-4 GB          6
 103          * 28           2               2       4-8 GB          7
 104          * 32           2               2       8-16 GB         8
 105          * 4            2               2       <128M           1
 106          * 30           4               3       2-4 GB          5
 107          * 48           4               3       8-16 GB         8
 108          * 32           8               4       1-2 GB          4
 109          * 32           8               4       0.9-1GB         4
 110          * 10           16              5       <128M           1
 111          * 40           16              5       900M            4
 112          * 70           64              7       2-4 GB          5
 113          * 84           64              7       4-8 GB          6
 114          * 108          512             9       4-8 GB          6
 115          * 125          1024            10      8-16 GB         8
 116          * 125          1024            10      16-32 GB        9
 117          */
 118
 119         mem = zone->present_pages >> (27 - PAGE_SHIFT);
 120
 121         threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
 122
 123         /*
 124          * Maximum threshold is 125
 125          */
 126         threshold = min(125, threshold);
 127
 128         return threshold;
 129 }
 130
 131 /*
 132  * Refresh the thresholds for each zone.
 133  */
 134 static void refresh_zone_stat_thresholds(void)
 135 {
 136         struct zone *zone;
 137         int cpu;
 138         int threshold;
 139
 140         for_each_populated_zone(zone) {
 141                 unsigned long max_drift, tolerate_drift;
 142
 143                 threshold = calculate_threshold(zone);
 144
 145                 for_each_online_cpu(cpu)
 146                         per_cpu_ptr(zone->pageset, cpu)->stat_threshold
 147                                                         = threshold;
 148
 149                 /*
 150                  * Only set percpu_drift_mark if there is a danger that
 151                  * NR_FREE_PAGES reports the low watermark is ok when in fact
 152                  * the min watermark could be breached by an allocation
 153                  */
 154                 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
 155                 max_drift = num_online_cpus() * threshold;
 156                 if (max_drift > tolerate_drift)
 157                         zone->percpu_drift_mark = high_wmark_pages(zone) +
 158                                         max_drift;
 159         }
 160 }
 161
 162 /*
 163  * For use when we know that interrupts are disabled.
 164  */
 165 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 166                                 int delta)
 167 {
 168         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
 169
 170         s8 *p = pcp->vm_stat_diff + item;
 171         long x;
 172
 173         x = delta + *p;
 174
 175         if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
 176                 zone_page_state_add(x, zone, item);
 177                 x = 0;
 178         }
 179         *p = x;
 180 }
 181 EXPORT_SYMBOL(__mod_zone_page_state);
 182
 183 /*
 184  * For an unknown interrupt state
 185  */
 186 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 187                                         int delta)
 188 {
 189         unsigned long flags;
 190
 191         local_irq_save(flags);
 192         __mod_zone_page_state(zone, item, delta);
 193         local_irq_restore(flags);
 194 }
 195 EXPORT_SYMBOL(mod_zone_page_state);
 196
 197 /*
 198  * Optimized increment and decrement functions.
 199  *
 200  * These are only for a single page and therefore can take a struct page *
 201  * argument instead of struct zone *. This allows the inclusion of the code
 202  * generated for page_zone(page) into the optimized functions.
 203  *
 204  * No overflow check is necessary and therefore the differential can be
 205  * incremented or decremented in place which may allow the compilers to
 206  * generate better code.
 207  * The increment or decrement is known and therefore one boundary check can
 208  * be omitted.
 209  *
 210  * NOTE: These functions are very performance sensitive. Change only
 211  * with care.
 212  *
 213  * Some processors have inc/dec instructions that are atomic vs an interrupt.
 214  * However, the code must first determine the differential location in a zone
 215  * based on the processor number and then inc/dec the counter. There is no
 216  * guarantee without disabling preemption that the processor will not change
 217  * in between and therefore the atomicity vs. interrupt cannot be exploited
 218  * in a useful way here.
 219  */
 220 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
 221 {
 222         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
 223         s8 *p = pcp->vm_stat_diff + item;
 224
 225         (*p)++;
 226
 227         if (unlikely(*p > pcp->stat_threshold)) {
 228                 int overstep = pcp->stat_threshold / 2;
 229
 230                 zone_page_state_add(*p + overstep, zone, item);
 231                 *p = -overstep;
 232         }
 233 }
 234
 235 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
 236 {
 237         __inc_zone_state(page_zone(page), item);
 238 }
 239 EXPORT_SYMBOL(__inc_zone_page_state);
 240
 241 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
 242 {
 243         struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
 244         s8 *p = pcp->vm_stat_diff + item;
 245
 246         (*p)--;
 247
 248         if (unlikely(*p < - pcp->stat_threshold)) {
 249                 int overstep = pcp->stat_threshold / 2;
 250
 251                 zone_page_state_add(*p - overstep, zone, item);
 252                 *p = overstep;
 253         }
 254 }
 255
 256 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
 257 {
 258         __dec_zone_state(page_zone(page), item);
 259 }
 260 EXPORT_SYMBOL(__dec_zone_page_state);
 261
 262 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
 263 {
 264         unsigned long flags;
 265
 266         local_irq_save(flags);
 267         __inc_zone_state(zone, item);
 268         local_irq_restore(flags);
 269 }
 270
 271 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
 272 {
 273         unsigned long flags;
 274         struct zone *zone;
 275
 276         zone = page_zone(page);
 277         local_irq_save(flags);
 278         __inc_zone_state(zone, item);
 279         local_irq_restore(flags);
 280 }
 281 EXPORT_SYMBOL(inc_zone_page_state);
 282
 283 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
 284 {
 285         unsigned long flags;
 286
 287         local_irq_save(flags);
 288         __dec_zone_page_state(page, item);
 289         local_irq_restore(flags);
 290 }
 291 EXPORT_SYMBOL(dec_zone_page_state);
 292
 293 /*
 294  * Update the zone counters for one cpu.
 295  *
 296  * The cpu specified must be either the current cpu or a processor that
 297  * is not online. If it is the current cpu then the execution thread must
 298  * be pinned to the current cpu.
 299  *
 300  * Note that refresh_cpu_vm_stats strives to only access
 301  * node local memory. The per cpu pagesets on remote zones are placed
 302  * in the memory local to the processor using that pageset. So the
 303  * loop over all zones will access a series of cachelines local to
 304  * the processor.
 305  *
 306  * The call to zone_page_state_add updates the cachelines with the
 307  * statistics in the remote zone struct as well as the global cachelines
 308  * with the global counters. These could cause remote node cache line
 309  * bouncing and will have to be only done when necessary.
 310  */
 311 void refresh_cpu_vm_stats(int cpu)
 312 {
 313         struct zone *zone;
 314         int i;
 315         int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
 316
 317         for_each_populated_zone(zone) {
 318                 struct per_cpu_pageset *p;
 319
 320                 p = per_cpu_ptr(zone->pageset, cpu);
 321
 322                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 323                         if (p->vm_stat_diff[i]) {
 324                                 unsigned long flags;
 325                                 int v;
 326
 327                                 local_irq_save(flags);
 328                                 v = p->vm_stat_diff[i];
 329                                 p->vm_stat_diff[i] = 0;
 330                                 local_irq_restore(flags);
 331                                 atomic_long_add(v, &zone->vm_stat[i]);
 332                                 global_diff[i] += v;
 333 #ifdef CONFIG_NUMA
 334                                 /* 3 seconds idle till flush */
 335                                 p->expire = 3;
 336 #endif
 337                         }
 338                 cond_resched();
 339 #ifdef CONFIG_NUMA
 340                 /*
 341                  * Deal with draining the remote pageset of this
 342                  * processor
 343                  *
 344                  * Check if there are pages remaining in this pageset
 345                  * if not then there is nothing to expire.
 346                  */
 347                 if (!p->expire || !p->pcp.count)
 348                         continue;
 349
 350                 /*
 351                  * We never drain zones local to this processor.
 352                  */
 353                 if (zone_to_nid(zone) == numa_node_id()) {
 354                         p->expire = 0;
 355                         continue;
 356                 }
 357
 358                 p->expire--;
 359                 if (p->expire)
 360                         continue;
 361
 362                 if (p->pcp.count)
 363                         drain_zone_pages(zone, &p->pcp);
 364 #endif
 365         }
 366
 367         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 368                 if (global_diff[i])
 369                         atomic_long_add(global_diff[i], &vm_stat[i]);
 370 }
 371
 372 #endif
 373
 374 #ifdef CONFIG_NUMA
 375 /*
 376  * zonelist = the list of zones passed to the allocator
 377  * z        = the zone from which the allocation occurred.
 378  *
 379  * Must be called with interrupts disabled.
 380  */
 381 void zone_statistics(struct zone *preferred_zone, struct zone *z)
 382 {
 383         if (z->zone_pgdat == preferred_zone->zone_pgdat) {
 384                 __inc_zone_state(z, NUMA_HIT);
 385         } else {
 386                 __inc_zone_state(z, NUMA_MISS);
 387                 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
 388         }
 389         if (z->node == numa_node_id())
 390                 __inc_zone_state(z, NUMA_LOCAL);
 391         else
 392                 __inc_zone_state(z, NUMA_OTHER);
 393 }
 394 #endif
 395
 396 #ifdef CONFIG_COMPACTION
 397 struct contig_page_info {
 398         unsigned long free_pages;
 399         unsigned long free_blocks_total;
 400         unsigned long free_blocks_suitable;
 401 };
 402
 403 /*
 404  * Calculate the number of free pages in a zone, how many contiguous
 405  * pages are free and how many are large enough to satisfy an allocation of
 406  * the target size. Note that this function makes no attempt to estimate
 407  * how many suitable free blocks there *might* be if MOVABLE pages were
 408  * migrated. Calculating that is possible, but expensive and can be
 409  * figured out from userspace
 410  */
 411 static void fill_contig_page_info(struct zone *zone,
 412                                 unsigned int suitable_order,
 413                                 struct contig_page_info *info)
 414 {
 415         unsigned int order;
 416
 417         info->free_pages = 0;
 418         info->free_blocks_total = 0;
 419         info->free_blocks_suitable = 0;
 420
 421         for (order = 0; order < MAX_ORDER; order++) {
 422                 unsigned long blocks;
 423
 424                 /* Count number of free blocks */
 425                 blocks = zone->free_area[order].nr_free;
 426                 info->free_blocks_total += blocks;
 427
 428                 /* Count free base pages */
 429                 info->free_pages += blocks << order;
 430
 431                 /* Count the suitable free blocks */
 432                 if (order >= suitable_order)
 433                         info->free_blocks_suitable += blocks <<
 434                                                 (order - suitable_order);
 435         }
 436 }
 437
 438 /*
 439  * A fragmentation index only makes sense if an allocation of a requested
 440  * size would fail. If that is true, the fragmentation index indicates
 441  * whether external fragmentation or a lack of memory was the problem.
 442  * The value can be used to determine if page reclaim or compaction
 443  * should be used
 444  */
 445 static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
 446 {
 447         unsigned long requested = 1UL << order;
 448
 449         if (!info->free_blocks_total)
 450                 return 0;
 451
 452         /* Fragmentation index only makes sense when a request would fail */
 453         if (info->free_blocks_suitable)
 454                 return -1000;
 455
 456         /*
 457          * Index is between 0 and 1 so return within 3 decimal places
 458          *
 459          * 0 => allocation would fail due to lack of memory
 460          * 1 => allocation would fail due to fragmentation
 461          */
 462         return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
 463 }
 464
 465 /* Same as __fragmentation index but allocs contig_page_info on stack */
 466 int fragmentation_index(struct zone *zone, unsigned int order)
 467 {
 468         struct contig_page_info info;
 469
 470         fill_contig_page_info(zone, order, &info);
 471         return __fragmentation_index(order, &info);
 472 }
 473 #endif
 474
 475 #if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
 476 #include <linux/proc_fs.h>
 477 #include <linux/seq_file.h>
 478
 479 static char * const migratetype_names[MIGRATE_TYPES] = {
 480         "Unmovable",
 481         "Reclaimable",
 482         "Movable",
 483         "Reserve",
 484         "Isolate",
 485 };
 486
 487 static void *frag_start(struct seq_file *m, loff_t *pos)
 488 {
 489         pg_data_t *pgdat;
 490         loff_t node = *pos;
 491         for (pgdat = first_online_pgdat();
 492              pgdat && node;
 493              pgdat = next_online_pgdat(pgdat))
 494                 --node;
 495
 496         return pgdat;
 497 }
 498
 499 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
 500 {
 501         pg_data_t *pgdat = (pg_data_t *)arg;
 502
 503         (*pos)++;
 504         return next_online_pgdat(pgdat);
 505 }
 506
 507 static void frag_stop(struct seq_file *m, void *arg)
 508 {
 509 }
 510
 511 /* Walk all the zones in a node and print using a callback */
 512 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
 513                 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
 514 {
 515         struct zone *zone;
 516         struct zone *node_zones = pgdat->node_zones;
 517         unsigned long flags;
 518
 519         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
 520                 if (!populated_zone(zone))
 521                         continue;
 522
 523                 spin_lock_irqsave(&zone->lock, flags);
 524                 print(m, pgdat, zone);
 525                 spin_unlock_irqrestore(&zone->lock, flags);
 526         }
 527 }
 528 #endif
 529
 530 #ifdef CONFIG_PROC_FS
 531 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
 532                                                 struct zone *zone)
 533 {
 534         int order;
 535
 536         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 537         for (order = 0; order < MAX_ORDER; ++order)
 538                 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
 539         seq_putc(m, '\n');
 540 }
 541
 542 /*
 543  * This walks the free areas for each zone.
 544  */
 545 static int frag_show(struct seq_file *m, void *arg)
 546 {
 547         pg_data_t *pgdat = (pg_data_t *)arg;
 548         walk_zones_in_node(m, pgdat, frag_show_print);
 549         return 0;
 550 }
 551
 552 static void pagetypeinfo_showfree_print(struct seq_file *m,
 553                                         pg_data_t *pgdat, struct zone *zone)
 554 {
 555         int order, mtype;
 556
 557         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
 558                 seq_printf(m, "Node %4d, zone %8s, type %12s ",
 559                                         pgdat->node_id,
 560                                         zone->name,
 561                                         migratetype_names[mtype]);
 562                 for (order = 0; order < MAX_ORDER; ++order) {
 563                         unsigned long freecount = 0;
 564                         struct free_area *area;
 565                         struct list_head *curr;
 566
 567                         area = &(zone->free_area[order]);
 568
 569                         list_for_each(curr, &area->free_list[mtype])
 570                                 freecount++;
 571                         seq_printf(m, "%6lu ", freecount);
 572                 }
 573                 seq_putc(m, '\n');
 574         }
 575 }
 576
 577 /* Print out the free pages at each order for each migatetype */
 578 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
 579 {
 580         int order;
 581         pg_data_t *pgdat = (pg_data_t *)arg;
 582
 583         /* Print header */
 584         seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
 585         for (order = 0; order < MAX_ORDER; ++order)
 586                 seq_printf(m, "%6d ", order);
 587         seq_putc(m, '\n');
 588
 589         walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
 590
 591         return 0;
 592 }
 593
 594 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
 595                                         pg_data_t *pgdat, struct zone *zone)
 596 {
 597         int mtype;
 598         unsigned long pfn;
 599         unsigned long start_pfn = zone->zone_start_pfn;
 600         unsigned long end_pfn = start_pfn + zone->spanned_pages;
 601         unsigned long count[MIGRATE_TYPES] = { 0, };
 602
 603         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
 604                 struct page *page;
 605
 606                 if (!pfn_valid(pfn))
 607                         continue;
 608
 609                 page = pfn_to_page(pfn);
 610
 611                 /* Watch for unexpected holes punched in the memmap */
 612                 if (!memmap_valid_within(pfn, page, zone))
 613                         continue;
 614
 615                 mtype = get_pageblock_migratetype(page);
 616
 617                 if (mtype < MIGRATE_TYPES)
 618                         count[mtype]++;
 619         }
 620
 621         /* Print counts */
 622         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 623         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 624                 seq_printf(m, "%12lu ", count[mtype]);
 625         seq_putc(m, '\n');
 626 }
 627
 628 /* Print out the free pages at each order for each migratetype */
 629 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
 630 {
 631         int mtype;
 632         pg_data_t *pgdat = (pg_data_t *)arg;
 633
 634         seq_printf(m, "\n%-23s", "Number of blocks type ");
 635         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 636                 seq_printf(m, "%12s ", migratetype_names[mtype]);
 637         seq_putc(m, '\n');
 638         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
 639
 640         return 0;
 641 }
 642
 643 /*
 644  * This prints out statistics in relation to grouping pages by mobility.
 645  * It is expensive to collect so do not constantly read the file.
 646  */
 647 static int pagetypeinfo_show(struct seq_file *m, void *arg)
 648 {
 649         pg_data_t *pgdat = (pg_data_t *)arg;
 650
 651         /* check memoryless node */
 652         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
 653                 return 0;
 654
 655         seq_printf(m, "Page block order: %d\n", pageblock_order);
 656         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
 657         seq_putc(m, '\n');
 658         pagetypeinfo_showfree(m, pgdat);
 659         pagetypeinfo_showblockcount(m, pgdat);
 660
 661         return 0;
 662 }
 663
 664 static const struct seq_operations fragmentation_op = {
 665         .start  = frag_start,
 666         .next   = frag_next,
 667         .stop   = frag_stop,
 668         .show   = frag_show,
 669 };
 670
 671 static int fragmentation_open(struct inode *inode, struct file *file)
 672 {
 673         return seq_open(file, &fragmentation_op);
 674 }
 675
 676 static const struct file_operations fragmentation_file_operations = {
 677         .open           = fragmentation_open,
 678         .read           = seq_read,
 679         .llseek         = seq_lseek,
 680         .release        = seq_release,
 681 };
 682
 683 static const struct seq_operations pagetypeinfo_op = {
 684         .start  = frag_start,
 685         .next   = frag_next,
 686         .stop   = frag_stop,
 687         .show   = pagetypeinfo_show,
 688 };
 689
 690 static int pagetypeinfo_open(struct inode *inode, struct file *file)
 691 {
 692         return seq_open(file, &pagetypeinfo_op);
 693 }
 694
 695 static const struct file_operations pagetypeinfo_file_ops = {
 696         .open           = pagetypeinfo_open,
 697         .read           = seq_read,
 698         .llseek         = seq_lseek,
 699         .release        = seq_release,
 700 };
 701
 702 #ifdef CONFIG_ZONE_DMA
 703 #define TEXT_FOR_DMA(xx) xx "_dma",
 704 #else
 705 #define TEXT_FOR_DMA(xx)
 706 #endif
 707
 708 #ifdef CONFIG_ZONE_DMA32
 709 #define TEXT_FOR_DMA32(xx) xx "_dma32",
 710 #else
 711 #define TEXT_FOR_DMA32(xx)
 712 #endif
 713
 714 #ifdef CONFIG_HIGHMEM
 715 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
 716 #else
 717 #define TEXT_FOR_HIGHMEM(xx)
 718 #endif
 719
 720 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
 721                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
 722
 723 static const char * const vmstat_text[] = {
 724         /* Zoned VM counters */
 725         "nr_free_pages",
 726         "nr_inactive_anon",
 727         "nr_active_anon",
 728         "nr_inactive_file",
 729         "nr_active_file",
 730         "nr_unevictable",
 731         "nr_mlock",
 732         "nr_anon_pages",
 733         "nr_mapped",
 734         "nr_file_pages",
 735         "nr_dirty",
 736         "nr_writeback",
 737         "nr_slab_reclaimable",
 738         "nr_slab_unreclaimable",
 739         "nr_page_table_pages",
 740         "nr_kernel_stack",
 741         "nr_unstable",
 742         "nr_bounce",
 743         "nr_vmscan_write",
 744         "nr_writeback_temp",
 745         "nr_isolated_anon",
 746         "nr_isolated_file",
 747         "nr_shmem",
 748         "nr_dirtied",
 749         "nr_written",
 750
 751 #ifdef CONFIG_NUMA
 752         "numa_hit",
 753         "numa_miss",
 754         "numa_foreign",
 755         "numa_interleave",
 756         "numa_local",
 757         "numa_other",
 758 #endif
 759
 760 #ifdef CONFIG_VM_EVENT_COUNTERS
 761         "pgpgin",
 762         "pgpgout",
 763         "pswpin",
 764         "pswpout",
 765
 766         TEXTS_FOR_ZONES("pgalloc")
 767
 768         "pgfree",
 769         "pgactivate",
 770         "pgdeactivate",
 771
 772         "pgfault",
 773         "pgmajfault",
 774
 775         TEXTS_FOR_ZONES("pgrefill")
 776         TEXTS_FOR_ZONES("pgsteal")
 777         TEXTS_FOR_ZONES("pgscan_kswapd")
 778         TEXTS_FOR_ZONES("pgscan_direct")
 779
 780 #ifdef CONFIG_NUMA
 781         "zone_reclaim_failed",
 782 #endif
 783         "pginodesteal",
 784         "slabs_scanned",
 785         "kswapd_steal",
 786         "kswapd_inodesteal",
 787         "kswapd_low_wmark_hit_quickly",
 788         "kswapd_high_wmark_hit_quickly",
 789         "kswapd_skip_congestion_wait",
 790         "pageoutrun",
 791         "allocstall",
 792
 793         "pgrotated",
 794
 795 #ifdef CONFIG_COMPACTION
 796         "compact_blocks_moved",
 797         "compact_pages_moved",
 798         "compact_pagemigrate_failed",
 799         "compact_stall",
 800         "compact_fail",
 801         "compact_success",
 802 #endif
 803
 804 #ifdef CONFIG_HUGETLB_PAGE
 805         "htlb_buddy_alloc_success",
 806         "htlb_buddy_alloc_fail",
 807 #endif
 808         "unevictable_pgs_culled",
 809         "unevictable_pgs_scanned",
 810         "unevictable_pgs_rescued",
 811         "unevictable_pgs_mlocked",
 812         "unevictable_pgs_munlocked",
 813         "unevictable_pgs_cleared",
 814         "unevictable_pgs_stranded",
 815         "unevictable_pgs_mlockfreed",
 816 #endif
 817 };
 818
 819 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
 820                                                         struct zone *zone)
 821 {
 822         int i;
 823         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
 824         seq_printf(m,
 825                    "\n  pages free     %lu"
 826                    "\n        min      %lu"
 827                    "\n        low      %lu"
 828                    "\n        high     %lu"
 829                    "\n        scanned  %lu"
 830                    "\n        spanned  %lu"
 831                    "\n        present  %lu",
 832                    zone_nr_free_pages(zone),
 833                    min_wmark_pages(zone),
 834                    low_wmark_pages(zone),
 835                    high_wmark_pages(zone),
 836                    zone->pages_scanned,
 837                    zone->spanned_pages,
 838                    zone->present_pages);
 839
 840         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 841                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
 842                                 zone_page_state(zone, i));
 843
 844         seq_printf(m,
 845                    "\n        protection: (%lu",
 846                    zone->lowmem_reserve[0]);
 847         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
 848                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
 849         seq_printf(m,
 850                    ")"
 851                    "\n  pagesets");
 852         for_each_online_cpu(i) {
 853                 struct per_cpu_pageset *pageset;
 854
 855                 pageset = per_cpu_ptr(zone->pageset, i);
 856                 seq_printf(m,
 857                            "\n    cpu: %i"
 858                            "\n              count: %i"
 859                            "\n              high:  %i"
 860                            "\n              batch: %i",
 861                            i,
 862                            pageset->pcp.count,
 863                            pageset->pcp.high,
 864                            pageset->pcp.batch);
 865 #ifdef CONFIG_SMP
 866                 seq_printf(m, "\n  vm stats threshold: %d",
 867                                 pageset->stat_threshold);
 868 #endif
 869         }
 870         seq_printf(m,
 871                    "\n  all_unreclaimable: %u"
 872                    "\n  start_pfn:         %lu"
 873                    "\n  inactive_ratio:    %u",
 874                    zone->all_unreclaimable,
 875                    zone->zone_start_pfn,
 876                    zone->inactive_ratio);
 877         seq_putc(m, '\n');
 878 }
 879
 880 /*
 881  * Output information about zones in @pgdat.
 882  */
 883 static int zoneinfo_show(struct seq_file *m, void *arg)
 884 {
 885         pg_data_t *pgdat = (pg_data_t *)arg;
 886         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
 887         return 0;
 888 }
 889
 890 static const struct seq_operations zoneinfo_op = {
 891         .start  = frag_start, /* iterate over all zones. The same as in
 892                                * fragmentation. */
 893         .next   = frag_next,
 894         .stop   = frag_stop,
 895         .show   = zoneinfo_show,
 896 };
 897
 898 static int zoneinfo_open(struct inode *inode, struct file *file)
 899 {
 900         return seq_open(file, &zoneinfo_op);
 901 }
 902
 903 static const struct file_operations proc_zoneinfo_file_operations = {
 904         .open           = zoneinfo_open,
 905         .read           = seq_read,
 906         .llseek         = seq_lseek,
 907         .release        = seq_release,
 908 };
 909
 910 static void *vmstat_start(struct seq_file *m, loff_t *pos)
 911 {
 912         unsigned long *v;
 913 #ifdef CONFIG_VM_EVENT_COUNTERS
 914         unsigned long *e;
 915 #endif
 916         int i;
 917
 918         if (*pos >= ARRAY_SIZE(vmstat_text))
 919                 return NULL;
 920
 921 #ifdef CONFIG_VM_EVENT_COUNTERS
 922         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
 923                         + sizeof(struct vm_event_state), GFP_KERNEL);
 924 #else
 925         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
 926                         GFP_KERNEL);
 927 #endif
 928         m->private = v;
 929         if (!v)
 930                 return ERR_PTR(-ENOMEM);
 931         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 932                 v[i] = global_page_state(i);
 933 #ifdef CONFIG_VM_EVENT_COUNTERS
 934         e = v + NR_VM_ZONE_STAT_ITEMS;
 935         all_vm_events(e);
 936         e[PGPGIN] /= 2;         /* sectors -> kbytes */
 937         e[PGPGOUT] /= 2;
 938 #endif
 939         return v + *pos;
 940 }
 941
 942 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
 943 {
 944         (*pos)++;
 945         if (*pos >= ARRAY_SIZE(vmstat_text))
 946                 return NULL;
 947         return (unsigned long *)m->private + *pos;
 948 }
 949
 950 static int vmstat_show(struct seq_file *m, void *arg)
 951 {
 952         unsigned long *l = arg;
 953         unsigned long off = l - (unsigned long *)m->private;
 954
 955         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
 956         return 0;
 957 }
 958
 959 static void vmstat_stop(struct seq_file *m, void *arg)
 960 {
 961         kfree(m->private);
 962         m->private = NULL;
 963 }
 964
 965 static const struct seq_operations vmstat_op = {
 966         .start  = vmstat_start,
 967         .next   = vmstat_next,
 968         .stop   = vmstat_stop,
 969         .show   = vmstat_show,
 970 };
 971
 972 static int vmstat_open(struct inode *inode, struct file *file)
 973 {
 974         return seq_open(file, &vmstat_op);
 975 }
 976
 977 static const struct file_operations proc_vmstat_file_operations = {
 978         .open           = vmstat_open,
 979         .read           = seq_read,
 980         .llseek         = seq_lseek,
 981         .release        = seq_release,
 982 };
 983 #endif /* CONFIG_PROC_FS */
 984
 985 #ifdef CONFIG_SMP
 986 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
 987 int sysctl_stat_interval __read_mostly = HZ;
 988
 989 static void vmstat_update(struct work_struct *w)
 990 {
 991         refresh_cpu_vm_stats(smp_processor_id());
 992         schedule_delayed_work(&__get_cpu_var(vmstat_work),
 993                 round_jiffies_relative(sysctl_stat_interval));
 994 }
 995
 996 static void __cpuinit start_cpu_timer(int cpu)
 997 {
 998         struct delayed_work *work = &per_cpu(vmstat_work, cpu);
 999
1000         INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
1001         schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
1002 }
1003
1004 /*
1005  * Use the cpu notifier to insure that the thresholds are recalculated
1006  * when necessary.
1007  */
1008 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
1009                 unsigned long action,
1010                 void *hcpu)
1011 {
1012         long cpu = (long)hcpu;
1013
1014         switch (action) {
1015         case CPU_ONLINE:
1016         case CPU_ONLINE_FROZEN:
1017                 refresh_zone_stat_thresholds();
1018                 start_cpu_timer(cpu);
1019                 node_set_state(cpu_to_node(cpu), N_CPU);
1020                 break;
1021         case CPU_DOWN_PREPARE:
1022         case CPU_DOWN_PREPARE_FROZEN:
1023                 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
1024                 per_cpu(vmstat_work, cpu).work.func = NULL;
1025                 break;
1026         case CPU_DOWN_FAILED:
1027         case CPU_DOWN_FAILED_FROZEN:
1028                 start_cpu_timer(cpu);
1029                 break;
1030         case CPU_DEAD:
1031         case CPU_DEAD_FROZEN:
1032                 refresh_zone_stat_thresholds();
1033                 break;
1034         default:
1035                 break;
1036         }
1037         return NOTIFY_OK;
1038 }
1039
1040 static struct notifier_block __cpuinitdata vmstat_notifier =
1041         { &vmstat_cpuup_callback, NULL, 0 };
1042 #endif
1043
1044 static int __init setup_vmstat(void)
1045 {
1046 #ifdef CONFIG_SMP
1047         int cpu;
1048
1049         refresh_zone_stat_thresholds();
1050         register_cpu_notifier(&vmstat_notifier);
1051
1052         for_each_online_cpu(cpu)
1053                 start_cpu_timer(cpu);
1054 #endif
1055 #ifdef CONFIG_PROC_FS
1056         proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1057         proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1058         proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1059         proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1060 #endif
1061         return 0;
1062 }
1063 module_init(setup_vmstat)
1064
1065 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1066 #include <linux/debugfs.h>
1067
1068 static struct dentry *extfrag_debug_root;
1069
1070 /*
1071  * Return an index indicating how much of the available free memory is
1072  * unusable for an allocation of the requested size.
1073  */
1074 static int unusable_free_index(unsigned int order,
1075                                 struct contig_page_info *info)
1076 {
1077         /* No free memory is interpreted as all free memory is unusable */
1078         if (info->free_pages == 0)
1079                 return 1000;
1080
1081         /*
1082          * Index should be a value between 0 and 1. Return a value to 3
1083          * decimal places.
1084          *
1085          * 0 => no fragmentation
1086          * 1 => high fragmentation
1087          */
1088         return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1089
1090 }
1091
1092 static void unusable_show_print(struct seq_file *m,
1093                                         pg_data_t *pgdat, struct zone *zone)
1094 {
1095         unsigned int order;
1096         int index;
1097         struct contig_page_info info;
1098
1099         seq_printf(m, "Node %d, zone %8s ",
1100                                 pgdat->node_id,
1101                                 zone->name);
1102         for (order = 0; order < MAX_ORDER; ++order) {
1103                 fill_contig_page_info(zone, order, &info);
1104                 index = unusable_free_index(order, &info);
1105                 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1106         }
1107
1108         seq_putc(m, '\n');
1109 }
1110
1111 /*
1112  * Display unusable free space index
1113  *
1114  * The unusable free space index measures how much of the available free
1115  * memory cannot be used to satisfy an allocation of a given size and is a
1116  * value between 0 and 1. The higher the value, the more of free memory is
1117  * unusable and by implication, the worse the external fragmentation is. This
1118  * can be expressed as a percentage by multiplying by 100.
1119  */
1120 static int unusable_show(struct seq_file *m, void *arg)
1121 {
1122         pg_data_t *pgdat = (pg_data_t *)arg;
1123
1124         /* check memoryless node */
1125         if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
1126                 return 0;
1127
1128         walk_zones_in_node(m, pgdat, unusable_show_print);
1129
1130         return 0;
1131 }
1132
1133 static const struct seq_operations unusable_op = {
1134         .start  = frag_start,
1135         .next   = frag_next,
1136         .stop   = frag_stop,
1137         .show   = unusable_show,
1138 };
1139
1140 static int unusable_open(struct inode *inode, struct file *file)
1141 {
1142         return seq_open(file, &unusable_op);
1143 }
1144
1145 static const struct file_operations unusable_file_ops = {
1146         .open           = unusable_open,
1147         .read           = seq_read,
1148         .llseek         = seq_lseek,
1149         .release        = seq_release,
1150 };
1151
1152 static void extfrag_show_print(struct seq_file *m,
1153                                         pg_data_t *pgdat, struct zone *zone)
1154 {
1155         unsigned int order;
1156         int index;
1157
1158         /* Alloc on stack as interrupts are disabled for zone walk */
1159         struct contig_page_info info;
1160
1161         seq_printf(m, "Node %d, zone %8s ",
1162                                 pgdat->node_id,
1163                                 zone->name);
1164         for (order = 0; order < MAX_ORDER; ++order) {
1165                 fill_contig_page_info(zone, order, &info);
1166                 index = __fragmentation_index(order, &info);
1167                 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1168         }
1169
1170         seq_putc(m, '\n');
1171 }
1172
1173 /*
1174  * Display fragmentation index for orders that allocations would fail for
1175  */
1176 static int extfrag_show(struct seq_file *m, void *arg)
1177 {
1178         pg_data_t *pgdat = (pg_data_t *)arg;
1179
1180         walk_zones_in_node(m, pgdat, extfrag_show_print);
1181
1182         return 0;
1183 }
1184
1185 static const struct seq_operations extfrag_op = {
1186         .start  = frag_start,
1187         .next   = frag_next,
1188         .stop   = frag_stop,
1189         .show   = extfrag_show,
1190 };
1191
1192 static int extfrag_open(struct inode *inode, struct file *file)
1193 {
1194         return seq_open(file, &extfrag_op);
1195 }
1196
1197 static const struct file_operations extfrag_file_ops = {
1198         .open           = extfrag_open,
1199         .read           = seq_read,
1200         .llseek         = seq_lseek,
1201         .release        = seq_release,
1202 };
1203
1204 static int __init extfrag_debug_init(void)
1205 {
1206         extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1207         if (!extfrag_debug_root)
1208                 return -ENOMEM;
1209
1210         if (!debugfs_create_file("unusable_index", 0444,
1211                         extfrag_debug_root, NULL, &unusable_file_ops))
1212                 return -ENOMEM;
1213
1214         if (!debugfs_create_file("extfrag_index", 0444,
1215                         extfrag_debug_root, NULL, &extfrag_file_ops))
1216                 return -ENOMEM;
1217
1218         return 0;
1219 }
1220
1221 module_init(extfrag_debug_init);
1222 #endif