linux/mm/vmstat.c
<<
>>
Prefs
   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#include <linux/writeback.h>
  21#include <linux/compaction.h>
  22
  23#ifdef CONFIG_VM_EVENT_COUNTERS
  24DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
  25EXPORT_PER_CPU_SYMBOL(vm_event_states);
  26
  27static void sum_vm_events(unsigned long *ret)
  28{
  29        int cpu;
  30        int i;
  31
  32        memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
  33
  34        for_each_online_cpu(cpu) {
  35                struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
  36
  37                for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
  38                        ret[i] += this->event[i];
  39        }
  40}
  41
  42/*
  43 * Accumulate the vm event counters across all CPUs.
  44 * The result is unavoidably approximate - it can change
  45 * during and after execution of this function.
  46*/
  47void all_vm_events(unsigned long *ret)
  48{
  49        get_online_cpus();
  50        sum_vm_events(ret);
  51        put_online_cpus();
  52}
  53EXPORT_SYMBOL_GPL(all_vm_events);
  54
  55#ifdef CONFIG_HOTPLUG
  56/*
  57 * Fold the foreign cpu events into our own.
  58 *
  59 * This is adding to the events on one processor
  60 * but keeps the global counts constant.
  61 */
  62void vm_events_fold_cpu(int cpu)
  63{
  64        struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
  65        int i;
  66
  67        for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
  68                count_vm_events(i, fold_state->event[i]);
  69                fold_state->event[i] = 0;
  70        }
  71}
  72#endif /* CONFIG_HOTPLUG */
  73
  74#endif /* CONFIG_VM_EVENT_COUNTERS */
  75
  76/*
  77 * Manage combined zone based / global counters
  78 *
  79 * vm_stat contains the global counters
  80 */
  81atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
  82EXPORT_SYMBOL(vm_stat);
  83
  84#ifdef CONFIG_SMP
  85
  86int calculate_pressure_threshold(struct zone *zone)
  87{
  88        int threshold;
  89        int watermark_distance;
  90
  91        /*
  92         * As vmstats are not up to date, there is drift between the estimated
  93         * and real values. For high thresholds and a high number of CPUs, it
  94         * is possible for the min watermark to be breached while the estimated
  95         * value looks fine. The pressure threshold is a reduced value such
  96         * that even the maximum amount of drift will not accidentally breach
  97         * the min watermark
  98         */
  99        watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
 100        threshold = max(1, (int)(watermark_distance / num_online_cpus()));
 101
 102        /*
 103         * Maximum threshold is 125
 104         */
 105        threshold = min(125, threshold);
 106
 107        return threshold;
 108}
 109
 110int calculate_normal_threshold(struct zone *zone)
 111{
 112        int threshold;
 113        int mem;        /* memory in 128 MB units */
 114
 115        /*
 116         * The threshold scales with the number of processors and the amount
 117         * of memory per zone. More memory means that we can defer updates for
 118         * longer, more processors could lead to more contention.
 119         * fls() is used to have a cheap way of logarithmic scaling.
 120         *
 121         * Some sample thresholds:
 122         *
 123         * Threshold    Processors      (fls)   Zonesize        fls(mem+1)
 124         * ------------------------------------------------------------------
 125         * 8            1               1       0.9-1 GB        4
 126         * 16           2               2       0.9-1 GB        4
 127         * 20           2               2       1-2 GB          5
 128         * 24           2               2       2-4 GB          6
 129         * 28           2               2       4-8 GB          7
 130         * 32           2               2       8-16 GB         8
 131         * 4            2               2       <128M           1
 132         * 30           4               3       2-4 GB          5
 133         * 48           4               3       8-16 GB         8
 134         * 32           8               4       1-2 GB          4
 135         * 32           8               4       0.9-1GB         4
 136         * 10           16              5       <128M           1
 137         * 40           16              5       900M            4
 138         * 70           64              7       2-4 GB          5
 139         * 84           64              7       4-8 GB          6
 140         * 108          512             9       4-8 GB          6
 141         * 125          1024            10      8-16 GB         8
 142         * 125          1024            10      16-32 GB        9
 143         */
 144
 145        mem = zone->present_pages >> (27 - PAGE_SHIFT);
 146
 147        threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
 148
 149        /*
 150         * Maximum threshold is 125
 151         */
 152        threshold = min(125, threshold);
 153
 154        return threshold;
 155}
 156
 157/*
 158 * Refresh the thresholds for each zone.
 159 */
 160void refresh_zone_stat_thresholds(void)
 161{
 162        struct zone *zone;
 163        int cpu;
 164        int threshold;
 165
 166        for_each_populated_zone(zone) {
 167                unsigned long max_drift, tolerate_drift;
 168
 169                threshold = calculate_normal_threshold(zone);
 170
 171                for_each_online_cpu(cpu)
 172                        per_cpu_ptr(zone->pageset, cpu)->stat_threshold
 173                                                        = threshold;
 174
 175                /*
 176                 * Only set percpu_drift_mark if there is a danger that
 177                 * NR_FREE_PAGES reports the low watermark is ok when in fact
 178                 * the min watermark could be breached by an allocation
 179                 */
 180                tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
 181                max_drift = num_online_cpus() * threshold;
 182                if (max_drift > tolerate_drift)
 183                        zone->percpu_drift_mark = high_wmark_pages(zone) +
 184                                        max_drift;
 185        }
 186}
 187
 188void set_pgdat_percpu_threshold(pg_data_t *pgdat,
 189                                int (*calculate_pressure)(struct zone *))
 190{
 191        struct zone *zone;
 192        int cpu;
 193        int threshold;
 194        int i;
 195
 196        for (i = 0; i < pgdat->nr_zones; i++) {
 197                zone = &pgdat->node_zones[i];
 198                if (!zone->percpu_drift_mark)
 199                        continue;
 200
 201                threshold = (*calculate_pressure)(zone);
 202                for_each_possible_cpu(cpu)
 203                        per_cpu_ptr(zone->pageset, cpu)->stat_threshold
 204                                                        = threshold;
 205        }
 206}
 207
 208/*
 209 * For use when we know that interrupts are disabled.
 210 */
 211void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 212                                int delta)
 213{
 214        struct per_cpu_pageset __percpu *pcp = zone->pageset;
 215        s8 __percpu *p = pcp->vm_stat_diff + item;
 216        long x;
 217        long t;
 218
 219        x = delta + __this_cpu_read(*p);
 220
 221        t = __this_cpu_read(pcp->stat_threshold);
 222
 223        if (unlikely(x > t || x < -t)) {
 224                zone_page_state_add(x, zone, item);
 225                x = 0;
 226        }
 227        __this_cpu_write(*p, x);
 228}
 229EXPORT_SYMBOL(__mod_zone_page_state);
 230
 231/*
 232 * Optimized increment and decrement functions.
 233 *
 234 * These are only for a single page and therefore can take a struct page *
 235 * argument instead of struct zone *. This allows the inclusion of the code
 236 * generated for page_zone(page) into the optimized functions.
 237 *
 238 * No overflow check is necessary and therefore the differential can be
 239 * incremented or decremented in place which may allow the compilers to
 240 * generate better code.
 241 * The increment or decrement is known and therefore one boundary check can
 242 * be omitted.
 243 *
 244 * NOTE: These functions are very performance sensitive. Change only
 245 * with care.
 246 *
 247 * Some processors have inc/dec instructions that are atomic vs an interrupt.
 248 * However, the code must first determine the differential location in a zone
 249 * based on the processor number and then inc/dec the counter. There is no
 250 * guarantee without disabling preemption that the processor will not change
 251 * in between and therefore the atomicity vs. interrupt cannot be exploited
 252 * in a useful way here.
 253 */
 254void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
 255{
 256        struct per_cpu_pageset __percpu *pcp = zone->pageset;
 257        s8 __percpu *p = pcp->vm_stat_diff + item;
 258        s8 v, t;
 259
 260        v = __this_cpu_inc_return(*p);
 261        t = __this_cpu_read(pcp->stat_threshold);
 262        if (unlikely(v > t)) {
 263                s8 overstep = t >> 1;
 264
 265                zone_page_state_add(v + overstep, zone, item);
 266                __this_cpu_write(*p, -overstep);
 267        }
 268}
 269
 270void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
 271{
 272        __inc_zone_state(page_zone(page), item);
 273}
 274EXPORT_SYMBOL(__inc_zone_page_state);
 275
 276void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
 277{
 278        struct per_cpu_pageset __percpu *pcp = zone->pageset;
 279        s8 __percpu *p = pcp->vm_stat_diff + item;
 280        s8 v, t;
 281
 282        v = __this_cpu_dec_return(*p);
 283        t = __this_cpu_read(pcp->stat_threshold);
 284        if (unlikely(v < - t)) {
 285                s8 overstep = t >> 1;
 286
 287                zone_page_state_add(v - overstep, zone, item);
 288                __this_cpu_write(*p, overstep);
 289        }
 290}
 291
 292void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
 293{
 294        __dec_zone_state(page_zone(page), item);
 295}
 296EXPORT_SYMBOL(__dec_zone_page_state);
 297
 298#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
 299/*
 300 * If we have cmpxchg_local support then we do not need to incur the overhead
 301 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
 302 *
 303 * mod_state() modifies the zone counter state through atomic per cpu
 304 * operations.
 305 *
 306 * Overstep mode specifies how overstep should handled:
 307 *     0       No overstepping
 308 *     1       Overstepping half of threshold
 309 *     -1      Overstepping minus half of threshold
 310*/
 311static inline void mod_state(struct zone *zone,
 312       enum zone_stat_item item, int delta, int overstep_mode)
 313{
 314        struct per_cpu_pageset __percpu *pcp = zone->pageset;
 315        s8 __percpu *p = pcp->vm_stat_diff + item;
 316        long o, n, t, z;
 317
 318        do {
 319                z = 0;  /* overflow to zone counters */
 320
 321                /*
 322                 * The fetching of the stat_threshold is racy. We may apply
 323                 * a counter threshold to the wrong the cpu if we get
 324                 * rescheduled while executing here. However, the next
 325                 * counter update will apply the threshold again and
 326                 * therefore bring the counter under the threshold again.
 327                 *
 328                 * Most of the time the thresholds are the same anyways
 329                 * for all cpus in a zone.
 330                 */
 331                t = this_cpu_read(pcp->stat_threshold);
 332
 333                o = this_cpu_read(*p);
 334                n = delta + o;
 335
 336                if (n > t || n < -t) {
 337                        int os = overstep_mode * (t >> 1) ;
 338
 339                        /* Overflow must be added to zone counters */
 340                        z = n + os;
 341                        n = -os;
 342                }
 343        } while (this_cpu_cmpxchg(*p, o, n) != o);
 344
 345        if (z)
 346                zone_page_state_add(z, zone, item);
 347}
 348
 349void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 350                                        int delta)
 351{
 352        mod_state(zone, item, delta, 0);
 353}
 354EXPORT_SYMBOL(mod_zone_page_state);
 355
 356void inc_zone_state(struct zone *zone, enum zone_stat_item item)
 357{
 358        mod_state(zone, item, 1, 1);
 359}
 360
 361void inc_zone_page_state(struct page *page, enum zone_stat_item item)
 362{
 363        mod_state(page_zone(page), item, 1, 1);
 364}
 365EXPORT_SYMBOL(inc_zone_page_state);
 366
 367void dec_zone_page_state(struct page *page, enum zone_stat_item item)
 368{
 369        mod_state(page_zone(page), item, -1, -1);
 370}
 371EXPORT_SYMBOL(dec_zone_page_state);
 372#else
 373/*
 374 * Use interrupt disable to serialize counter updates
 375 */
 376void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
 377                                        int delta)
 378{
 379        unsigned long flags;
 380
 381        local_irq_save(flags);
 382        __mod_zone_page_state(zone, item, delta);
 383        local_irq_restore(flags);
 384}
 385EXPORT_SYMBOL(mod_zone_page_state);
 386
 387void inc_zone_state(struct zone *zone, enum zone_stat_item item)
 388{
 389        unsigned long flags;
 390
 391        local_irq_save(flags);
 392        __inc_zone_state(zone, item);
 393        local_irq_restore(flags);
 394}
 395
 396void inc_zone_page_state(struct page *page, enum zone_stat_item item)
 397{
 398        unsigned long flags;
 399        struct zone *zone;
 400
 401        zone = page_zone(page);
 402        local_irq_save(flags);
 403        __inc_zone_state(zone, item);
 404        local_irq_restore(flags);
 405}
 406EXPORT_SYMBOL(inc_zone_page_state);
 407
 408void dec_zone_page_state(struct page *page, enum zone_stat_item item)
 409{
 410        unsigned long flags;
 411
 412        local_irq_save(flags);
 413        __dec_zone_page_state(page, item);
 414        local_irq_restore(flags);
 415}
 416EXPORT_SYMBOL(dec_zone_page_state);
 417#endif
 418
 419/*
 420 * Update the zone counters for one cpu.
 421 *
 422 * The cpu specified must be either the current cpu or a processor that
 423 * is not online. If it is the current cpu then the execution thread must
 424 * be pinned to the current cpu.
 425 *
 426 * Note that refresh_cpu_vm_stats strives to only access
 427 * node local memory. The per cpu pagesets on remote zones are placed
 428 * in the memory local to the processor using that pageset. So the
 429 * loop over all zones will access a series of cachelines local to
 430 * the processor.
 431 *
 432 * The call to zone_page_state_add updates the cachelines with the
 433 * statistics in the remote zone struct as well as the global cachelines
 434 * with the global counters. These could cause remote node cache line
 435 * bouncing and will have to be only done when necessary.
 436 */
 437void refresh_cpu_vm_stats(int cpu)
 438{
 439        struct zone *zone;
 440        int i;
 441        int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
 442
 443        for_each_populated_zone(zone) {
 444                struct per_cpu_pageset *p;
 445
 446                p = per_cpu_ptr(zone->pageset, cpu);
 447
 448                for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 449                        if (p->vm_stat_diff[i]) {
 450                                unsigned long flags;
 451                                int v;
 452
 453                                local_irq_save(flags);
 454                                v = p->vm_stat_diff[i];
 455                                p->vm_stat_diff[i] = 0;
 456                                local_irq_restore(flags);
 457                                atomic_long_add(v, &zone->vm_stat[i]);
 458                                global_diff[i] += v;
 459#ifdef CONFIG_NUMA
 460                                /* 3 seconds idle till flush */
 461                                p->expire = 3;
 462#endif
 463                        }
 464                cond_resched();
 465#ifdef CONFIG_NUMA
 466                /*
 467                 * Deal with draining the remote pageset of this
 468                 * processor
 469                 *
 470                 * Check if there are pages remaining in this pageset
 471                 * if not then there is nothing to expire.
 472                 */
 473                if (!p->expire || !p->pcp.count)
 474                        continue;
 475
 476                /*
 477                 * We never drain zones local to this processor.
 478                 */
 479                if (zone_to_nid(zone) == numa_node_id()) {
 480                        p->expire = 0;
 481                        continue;
 482                }
 483
 484                p->expire--;
 485                if (p->expire)
 486                        continue;
 487
 488                if (p->pcp.count)
 489                        drain_zone_pages(zone, &p->pcp);
 490#endif
 491        }
 492
 493        for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 494                if (global_diff[i])
 495                        atomic_long_add(global_diff[i], &vm_stat[i]);
 496}
 497
 498#endif
 499
 500#ifdef CONFIG_NUMA
 501/*
 502 * zonelist = the list of zones passed to the allocator
 503 * z        = the zone from which the allocation occurred.
 504 *
 505 * Must be called with interrupts disabled.
 506 *
 507 * When __GFP_OTHER_NODE is set assume the node of the preferred
 508 * zone is the local node. This is useful for daemons who allocate
 509 * memory on behalf of other processes.
 510 */
 511void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags)
 512{
 513        if (z->zone_pgdat == preferred_zone->zone_pgdat) {
 514                __inc_zone_state(z, NUMA_HIT);
 515        } else {
 516                __inc_zone_state(z, NUMA_MISS);
 517                __inc_zone_state(preferred_zone, NUMA_FOREIGN);
 518        }
 519        if (z->node == ((flags & __GFP_OTHER_NODE) ?
 520                        preferred_zone->node : numa_node_id()))
 521                __inc_zone_state(z, NUMA_LOCAL);
 522        else
 523                __inc_zone_state(z, NUMA_OTHER);
 524}
 525#endif
 526
 527#ifdef CONFIG_COMPACTION
 528
 529struct contig_page_info {
 530        unsigned long free_pages;
 531        unsigned long free_blocks_total;
 532        unsigned long free_blocks_suitable;
 533};
 534
 535/*
 536 * Calculate the number of free pages in a zone, how many contiguous
 537 * pages are free and how many are large enough to satisfy an allocation of
 538 * the target size. Note that this function makes no attempt to estimate
 539 * how many suitable free blocks there *might* be if MOVABLE pages were
 540 * migrated. Calculating that is possible, but expensive and can be
 541 * figured out from userspace
 542 */
 543static void fill_contig_page_info(struct zone *zone,
 544                                unsigned int suitable_order,
 545                                struct contig_page_info *info)
 546{
 547        unsigned int order;
 548
 549        info->free_pages = 0;
 550        info->free_blocks_total = 0;
 551        info->free_blocks_suitable = 0;
 552
 553        for (order = 0; order < MAX_ORDER; order++) {
 554                unsigned long blocks;
 555
 556                /* Count number of free blocks */
 557                blocks = zone->free_area[order].nr_free;
 558                info->free_blocks_total += blocks;
 559
 560                /* Count free base pages */
 561                info->free_pages += blocks << order;
 562
 563                /* Count the suitable free blocks */
 564                if (order >= suitable_order)
 565                        info->free_blocks_suitable += blocks <<
 566                                                (order - suitable_order);
 567        }
 568}
 569
 570/*
 571 * A fragmentation index only makes sense if an allocation of a requested
 572 * size would fail. If that is true, the fragmentation index indicates
 573 * whether external fragmentation or a lack of memory was the problem.
 574 * The value can be used to determine if page reclaim or compaction
 575 * should be used
 576 */
 577static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
 578{
 579        unsigned long requested = 1UL << order;
 580
 581        if (!info->free_blocks_total)
 582                return 0;
 583
 584        /* Fragmentation index only makes sense when a request would fail */
 585        if (info->free_blocks_suitable)
 586                return -1000;
 587
 588        /*
 589         * Index is between 0 and 1 so return within 3 decimal places
 590         *
 591         * 0 => allocation would fail due to lack of memory
 592         * 1 => allocation would fail due to fragmentation
 593         */
 594        return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
 595}
 596
 597/* Same as __fragmentation index but allocs contig_page_info on stack */
 598int fragmentation_index(struct zone *zone, unsigned int order)
 599{
 600        struct contig_page_info info;
 601
 602        fill_contig_page_info(zone, order, &info);
 603        return __fragmentation_index(order, &info);
 604}
 605#endif
 606
 607#if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
 608#include <linux/proc_fs.h>
 609#include <linux/seq_file.h>
 610
 611static char * const migratetype_names[MIGRATE_TYPES] = {
 612        "Unmovable",
 613        "Reclaimable",
 614        "Movable",
 615        "Reserve",
 616#ifdef CONFIG_CMA
 617        "CMA",
 618#endif
 619        "Isolate",
 620};
 621
 622static void *frag_start(struct seq_file *m, loff_t *pos)
 623{
 624        pg_data_t *pgdat;
 625        loff_t node = *pos;
 626        for (pgdat = first_online_pgdat();
 627             pgdat && node;
 628             pgdat = next_online_pgdat(pgdat))
 629                --node;
 630
 631        return pgdat;
 632}
 633
 634static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
 635{
 636        pg_data_t *pgdat = (pg_data_t *)arg;
 637
 638        (*pos)++;
 639        return next_online_pgdat(pgdat);
 640}
 641
 642static void frag_stop(struct seq_file *m, void *arg)
 643{
 644}
 645
 646/* Walk all the zones in a node and print using a callback */
 647static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
 648                void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
 649{
 650        struct zone *zone;
 651        struct zone *node_zones = pgdat->node_zones;
 652        unsigned long flags;
 653
 654        for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
 655                if (!populated_zone(zone))
 656                        continue;
 657
 658                spin_lock_irqsave(&zone->lock, flags);
 659                print(m, pgdat, zone);
 660                spin_unlock_irqrestore(&zone->lock, flags);
 661        }
 662}
 663#endif
 664
 665#if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
 666#ifdef CONFIG_ZONE_DMA
 667#define TEXT_FOR_DMA(xx) xx "_dma",
 668#else
 669#define TEXT_FOR_DMA(xx)
 670#endif
 671
 672#ifdef CONFIG_ZONE_DMA32
 673#define TEXT_FOR_DMA32(xx) xx "_dma32",
 674#else
 675#define TEXT_FOR_DMA32(xx)
 676#endif
 677
 678#ifdef CONFIG_HIGHMEM
 679#define TEXT_FOR_HIGHMEM(xx) xx "_high",
 680#else
 681#define TEXT_FOR_HIGHMEM(xx)
 682#endif
 683
 684#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
 685                                        TEXT_FOR_HIGHMEM(xx) xx "_movable",
 686
 687const char * const vmstat_text[] = {
 688        /* Zoned VM counters */
 689        "nr_free_pages",
 690        "nr_inactive_anon",
 691        "nr_active_anon",
 692        "nr_inactive_file",
 693        "nr_active_file",
 694        "nr_unevictable",
 695        "nr_mlock",
 696        "nr_anon_pages",
 697        "nr_mapped",
 698        "nr_file_pages",
 699        "nr_dirty",
 700        "nr_writeback",
 701        "nr_slab_reclaimable",
 702        "nr_slab_unreclaimable",
 703        "nr_page_table_pages",
 704        "nr_kernel_stack",
 705        "nr_unstable",
 706        "nr_bounce",
 707        "nr_vmscan_write",
 708        "nr_vmscan_immediate_reclaim",
 709        "nr_writeback_temp",
 710        "nr_isolated_anon",
 711        "nr_isolated_file",
 712        "nr_shmem",
 713        "nr_dirtied",
 714        "nr_written",
 715
 716#ifdef CONFIG_NUMA
 717        "numa_hit",
 718        "numa_miss",
 719        "numa_foreign",
 720        "numa_interleave",
 721        "numa_local",
 722        "numa_other",
 723#endif
 724        "nr_anon_transparent_hugepages",
 725        "nr_dirty_threshold",
 726        "nr_dirty_background_threshold",
 727
 728#ifdef CONFIG_VM_EVENT_COUNTERS
 729        "pgpgin",
 730        "pgpgout",
 731        "pswpin",
 732        "pswpout",
 733
 734        TEXTS_FOR_ZONES("pgalloc")
 735
 736        "pgfree",
 737        "pgactivate",
 738        "pgdeactivate",
 739
 740        "pgfault",
 741        "pgmajfault",
 742
 743        TEXTS_FOR_ZONES("pgrefill")
 744        TEXTS_FOR_ZONES("pgsteal_kswapd")
 745        TEXTS_FOR_ZONES("pgsteal_direct")
 746        TEXTS_FOR_ZONES("pgscan_kswapd")
 747        TEXTS_FOR_ZONES("pgscan_direct")
 748        "pgscan_direct_throttle",
 749
 750#ifdef CONFIG_NUMA
 751        "zone_reclaim_failed",
 752#endif
 753        "pginodesteal",
 754        "slabs_scanned",
 755        "kswapd_inodesteal",
 756        "kswapd_low_wmark_hit_quickly",
 757        "kswapd_high_wmark_hit_quickly",
 758        "kswapd_skip_congestion_wait",
 759        "pageoutrun",
 760        "allocstall",
 761
 762        "pgrotated",
 763
 764#ifdef CONFIG_COMPACTION
 765        "compact_blocks_moved",
 766        "compact_pages_moved",
 767        "compact_pagemigrate_failed",
 768        "compact_stall",
 769        "compact_fail",
 770        "compact_success",
 771#endif
 772
 773#ifdef CONFIG_HUGETLB_PAGE
 774        "htlb_buddy_alloc_success",
 775        "htlb_buddy_alloc_fail",
 776#endif
 777        "unevictable_pgs_culled",
 778        "unevictable_pgs_scanned",
 779        "unevictable_pgs_rescued",
 780        "unevictable_pgs_mlocked",
 781        "unevictable_pgs_munlocked",
 782        "unevictable_pgs_cleared",
 783        "unevictable_pgs_stranded",
 784        "unevictable_pgs_mlockfreed",
 785
 786#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 787        "thp_fault_alloc",
 788        "thp_fault_fallback",
 789        "thp_collapse_alloc",
 790        "thp_collapse_alloc_failed",
 791        "thp_split",
 792#endif
 793
 794#endif /* CONFIG_VM_EVENTS_COUNTERS */
 795};
 796#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
 797
 798
 799#ifdef CONFIG_PROC_FS
 800static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
 801                                                struct zone *zone)
 802{
 803        int order;
 804
 805        seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 806        for (order = 0; order < MAX_ORDER; ++order)
 807                seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
 808        seq_putc(m, '\n');
 809}
 810
 811/*
 812 * This walks the free areas for each zone.
 813 */
 814static int frag_show(struct seq_file *m, void *arg)
 815{
 816        pg_data_t *pgdat = (pg_data_t *)arg;
 817        walk_zones_in_node(m, pgdat, frag_show_print);
 818        return 0;
 819}
 820
 821static void pagetypeinfo_showfree_print(struct seq_file *m,
 822                                        pg_data_t *pgdat, struct zone *zone)
 823{
 824        int order, mtype;
 825
 826        for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
 827                seq_printf(m, "Node %4d, zone %8s, type %12s ",
 828                                        pgdat->node_id,
 829                                        zone->name,
 830                                        migratetype_names[mtype]);
 831                for (order = 0; order < MAX_ORDER; ++order) {
 832                        unsigned long freecount = 0;
 833                        struct free_area *area;
 834                        struct list_head *curr;
 835
 836                        area = &(zone->free_area[order]);
 837
 838                        list_for_each(curr, &area->free_list[mtype])
 839                                freecount++;
 840                        seq_printf(m, "%6lu ", freecount);
 841                }
 842                seq_putc(m, '\n');
 843        }
 844}
 845
 846/* Print out the free pages at each order for each migatetype */
 847static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
 848{
 849        int order;
 850        pg_data_t *pgdat = (pg_data_t *)arg;
 851
 852        /* Print header */
 853        seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
 854        for (order = 0; order < MAX_ORDER; ++order)
 855                seq_printf(m, "%6d ", order);
 856        seq_putc(m, '\n');
 857
 858        walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
 859
 860        return 0;
 861}
 862
 863static void pagetypeinfo_showblockcount_print(struct seq_file *m,
 864                                        pg_data_t *pgdat, struct zone *zone)
 865{
 866        int mtype;
 867        unsigned long pfn;
 868        unsigned long start_pfn = zone->zone_start_pfn;
 869        unsigned long end_pfn = start_pfn + zone->spanned_pages;
 870        unsigned long count[MIGRATE_TYPES] = { 0, };
 871
 872        for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
 873                struct page *page;
 874
 875                if (!pfn_valid(pfn))
 876                        continue;
 877
 878                page = pfn_to_page(pfn);
 879
 880                /* Watch for unexpected holes punched in the memmap */
 881                if (!memmap_valid_within(pfn, page, zone))
 882                        continue;
 883
 884                mtype = get_pageblock_migratetype(page);
 885
 886                if (mtype < MIGRATE_TYPES)
 887                        count[mtype]++;
 888        }
 889
 890        /* Print counts */
 891        seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 892        for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 893                seq_printf(m, "%12lu ", count[mtype]);
 894        seq_putc(m, '\n');
 895}
 896
 897/* Print out the free pages at each order for each migratetype */
 898static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
 899{
 900        int mtype;
 901        pg_data_t *pgdat = (pg_data_t *)arg;
 902
 903        seq_printf(m, "\n%-23s", "Number of blocks type ");
 904        for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
 905                seq_printf(m, "%12s ", migratetype_names[mtype]);
 906        seq_putc(m, '\n');
 907        walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
 908
 909        return 0;
 910}
 911
 912/*
 913 * This prints out statistics in relation to grouping pages by mobility.
 914 * It is expensive to collect so do not constantly read the file.
 915 */
 916static int pagetypeinfo_show(struct seq_file *m, void *arg)
 917{
 918        pg_data_t *pgdat = (pg_data_t *)arg;
 919
 920        /* check memoryless node */
 921        if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
 922                return 0;
 923
 924        seq_printf(m, "Page block order: %d\n", pageblock_order);
 925        seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
 926        seq_putc(m, '\n');
 927        pagetypeinfo_showfree(m, pgdat);
 928        pagetypeinfo_showblockcount(m, pgdat);
 929
 930        return 0;
 931}
 932
 933static const struct seq_operations fragmentation_op = {
 934        .start  = frag_start,
 935        .next   = frag_next,
 936        .stop   = frag_stop,
 937        .show   = frag_show,
 938};
 939
 940static int fragmentation_open(struct inode *inode, struct file *file)
 941{
 942        return seq_open(file, &fragmentation_op);
 943}
 944
 945static const struct file_operations fragmentation_file_operations = {
 946        .open           = fragmentation_open,
 947        .read           = seq_read,
 948        .llseek         = seq_lseek,
 949        .release        = seq_release,
 950};
 951
 952static const struct seq_operations pagetypeinfo_op = {
 953        .start  = frag_start,
 954        .next   = frag_next,
 955        .stop   = frag_stop,
 956        .show   = pagetypeinfo_show,
 957};
 958
 959static int pagetypeinfo_open(struct inode *inode, struct file *file)
 960{
 961        return seq_open(file, &pagetypeinfo_op);
 962}
 963
 964static const struct file_operations pagetypeinfo_file_ops = {
 965        .open           = pagetypeinfo_open,
 966        .read           = seq_read,
 967        .llseek         = seq_lseek,
 968        .release        = seq_release,
 969};
 970
 971static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
 972                                                        struct zone *zone)
 973{
 974        int i;
 975        seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
 976        seq_printf(m,
 977                   "\n  pages free     %lu"
 978                   "\n        min      %lu"
 979                   "\n        low      %lu"
 980                   "\n        high     %lu"
 981                   "\n        scanned  %lu"
 982                   "\n        spanned  %lu"
 983                   "\n        present  %lu",
 984                   zone_page_state(zone, NR_FREE_PAGES),
 985                   min_wmark_pages(zone),
 986                   low_wmark_pages(zone),
 987                   high_wmark_pages(zone),
 988                   zone->pages_scanned,
 989                   zone->spanned_pages,
 990                   zone->present_pages);
 991
 992        for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 993                seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
 994                                zone_page_state(zone, i));
 995
 996        seq_printf(m,
 997                   "\n        protection: (%lu",
 998                   zone->lowmem_reserve[0]);
 999        for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1000                seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
1001        seq_printf(m,
1002                   ")"
1003                   "\n  pagesets");
1004        for_each_online_cpu(i) {
1005                struct per_cpu_pageset *pageset;
1006
1007                pageset = per_cpu_ptr(zone->pageset, i);
1008                seq_printf(m,
1009                           "\n    cpu: %i"
1010                           "\n              count: %i"
1011                           "\n              high:  %i"
1012                           "\n              batch: %i",
1013                           i,
1014                           pageset->pcp.count,
1015                           pageset->pcp.high,
1016                           pageset->pcp.batch);
1017#ifdef CONFIG_SMP
1018                seq_printf(m, "\n  vm stats threshold: %d",
1019                                pageset->stat_threshold);
1020#endif
1021        }
1022        seq_printf(m,
1023                   "\n  all_unreclaimable: %u"
1024                   "\n  start_pfn:         %lu"
1025                   "\n  inactive_ratio:    %u",
1026                   zone->all_unreclaimable,
1027                   zone->zone_start_pfn,
1028                   zone->inactive_ratio);
1029        seq_putc(m, '\n');
1030}
1031
1032/*
1033 * Output information about zones in @pgdat.
1034 */
1035static int zoneinfo_show(struct seq_file *m, void *arg)
1036{
1037        pg_data_t *pgdat = (pg_data_t *)arg;
1038        walk_zones_in_node(m, pgdat, zoneinfo_show_print);
1039        return 0;
1040}
1041
1042static const struct seq_operations zoneinfo_op = {
1043        .start  = frag_start, /* iterate over all zones. The same as in
1044                               * fragmentation. */
1045        .next   = frag_next,
1046        .stop   = frag_stop,
1047        .show   = zoneinfo_show,
1048};
1049
1050static int zoneinfo_open(struct inode *inode, struct file *file)
1051{
1052        return seq_open(file, &zoneinfo_op);
1053}
1054
1055static const struct file_operations proc_zoneinfo_file_operations = {
1056        .open           = zoneinfo_open,
1057        .read           = seq_read,
1058        .llseek         = seq_lseek,
1059        .release        = seq_release,
1060};
1061
1062enum writeback_stat_item {
1063        NR_DIRTY_THRESHOLD,
1064        NR_DIRTY_BG_THRESHOLD,
1065        NR_VM_WRITEBACK_STAT_ITEMS,
1066};
1067
1068static void *vmstat_start(struct seq_file *m, loff_t *pos)
1069{
1070        unsigned long *v;
1071        int i, stat_items_size;
1072
1073        if (*pos >= ARRAY_SIZE(vmstat_text))
1074                return NULL;
1075        stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1076                          NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
1077
1078#ifdef CONFIG_VM_EVENT_COUNTERS
1079        stat_items_size += sizeof(struct vm_event_state);
1080#endif
1081
1082        v = kmalloc(stat_items_size, GFP_KERNEL);
1083        m->private = v;
1084        if (!v)
1085                return ERR_PTR(-ENOMEM);
1086        for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1087                v[i] = global_page_state(i);
1088        v += NR_VM_ZONE_STAT_ITEMS;
1089
1090        global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1091                            v + NR_DIRTY_THRESHOLD);
1092        v += NR_VM_WRITEBACK_STAT_ITEMS;
1093
1094#ifdef CONFIG_VM_EVENT_COUNTERS
1095        all_vm_events(v);
1096        v[PGPGIN] /= 2;         /* sectors -> kbytes */
1097        v[PGPGOUT] /= 2;
1098#endif
1099        return (unsigned long *)m->private + *pos;
1100}
1101
1102static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1103{
1104        (*pos)++;
1105        if (*pos >= ARRAY_SIZE(vmstat_text))
1106                return NULL;
1107        return (unsigned long *)m->private + *pos;
1108}
1109
1110static int vmstat_show(struct seq_file *m, void *arg)
1111{
1112        unsigned long *l = arg;
1113        unsigned long off = l - (unsigned long *)m->private;
1114
1115        seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1116        return 0;
1117}
1118
1119static void vmstat_stop(struct seq_file *m, void *arg)
1120{
1121        kfree(m->private);
1122        m->private = NULL;
1123}
1124
1125static const struct seq_operations vmstat_op = {
1126        .start  = vmstat_start,
1127        .next   = vmstat_next,
1128        .stop   = vmstat_stop,
1129        .show   = vmstat_show,
1130};
1131
1132static int vmstat_open(struct inode *inode, struct file *file)
1133{
1134        return seq_open(file, &vmstat_op);
1135}
1136
1137static const struct file_operations proc_vmstat_file_operations = {
1138        .open           = vmstat_open,
1139        .read           = seq_read,
1140        .llseek         = seq_lseek,
1141        .release        = seq_release,
1142};
1143#endif /* CONFIG_PROC_FS */
1144
1145#ifdef CONFIG_SMP
1146static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
1147int sysctl_stat_interval __read_mostly = HZ;
1148
1149static void vmstat_update(struct work_struct *w)
1150{
1151        refresh_cpu_vm_stats(smp_processor_id());
1152        schedule_delayed_work(&__get_cpu_var(vmstat_work),
1153                round_jiffies_relative(sysctl_stat_interval));
1154}
1155
1156static void __cpuinit start_cpu_timer(int cpu)
1157{
1158        struct delayed_work *work = &per_cpu(vmstat_work, cpu);
1159
1160        INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
1161        schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
1162}
1163
1164/*
1165 * Use the cpu notifier to insure that the thresholds are recalculated
1166 * when necessary.
1167 */
1168static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
1169                unsigned long action,
1170                void *hcpu)
1171{
1172        long cpu = (long)hcpu;
1173
1174        switch (action) {
1175        case CPU_ONLINE:
1176        case CPU_ONLINE_FROZEN:
1177                refresh_zone_stat_thresholds();
1178                start_cpu_timer(cpu);
1179                node_set_state(cpu_to_node(cpu), N_CPU);
1180                break;
1181        case CPU_DOWN_PREPARE:
1182        case CPU_DOWN_PREPARE_FROZEN:
1183                cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1184                per_cpu(vmstat_work, cpu).work.func = NULL;
1185                break;
1186        case CPU_DOWN_FAILED:
1187        case CPU_DOWN_FAILED_FROZEN:
1188                start_cpu_timer(cpu);
1189                break;
1190        case CPU_DEAD:
1191        case CPU_DEAD_FROZEN:
1192                refresh_zone_stat_thresholds();
1193                break;
1194        default:
1195                break;
1196        }
1197        return NOTIFY_OK;
1198}
1199
1200static struct notifier_block __cpuinitdata vmstat_notifier =
1201        { &vmstat_cpuup_callback, NULL, 0 };
1202#endif
1203
1204static int __init setup_vmstat(void)
1205{
1206#ifdef CONFIG_SMP
1207        int cpu;
1208
1209        register_cpu_notifier(&vmstat_notifier);
1210
1211        for_each_online_cpu(cpu)
1212                start_cpu_timer(cpu);
1213#endif
1214#ifdef CONFIG_PROC_FS
1215        proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
1216        proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
1217        proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
1218        proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
1219#endif
1220        return 0;
1221}
1222module_init(setup_vmstat)
1223
1224#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1225#include <linux/debugfs.h>
1226
1227
1228/*
1229 * Return an index indicating how much of the available free memory is
1230 * unusable for an allocation of the requested size.
1231 */
1232static int unusable_free_index(unsigned int order,
1233                                struct contig_page_info *info)
1234{
1235        /* No free memory is interpreted as all free memory is unusable */
1236        if (info->free_pages == 0)
1237                return 1000;
1238
1239        /*
1240         * Index should be a value between 0 and 1. Return a value to 3
1241         * decimal places.
1242         *
1243         * 0 => no fragmentation
1244         * 1 => high fragmentation
1245         */
1246        return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1247
1248}
1249
1250static void unusable_show_print(struct seq_file *m,
1251                                        pg_data_t *pgdat, struct zone *zone)
1252{
1253        unsigned int order;
1254        int index;
1255        struct contig_page_info info;
1256
1257        seq_printf(m, "Node %d, zone %8s ",
1258                                pgdat->node_id,
1259                                zone->name);
1260        for (order = 0; order < MAX_ORDER; ++order) {
1261                fill_contig_page_info(zone, order, &info);
1262                index = unusable_free_index(order, &info);
1263                seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1264        }
1265
1266        seq_putc(m, '\n');
1267}
1268
1269/*
1270 * Display unusable free space index
1271 *
1272 * The unusable free space index measures how much of the available free
1273 * memory cannot be used to satisfy an allocation of a given size and is a
1274 * value between 0 and 1. The higher the value, the more of free memory is
1275 * unusable and by implication, the worse the external fragmentation is. This
1276 * can be expressed as a percentage by multiplying by 100.
1277 */
1278static int unusable_show(struct seq_file *m, void *arg)
1279{
1280        pg_data_t *pgdat = (pg_data_t *)arg;
1281
1282        /* check memoryless node */
1283        if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
1284                return 0;
1285
1286        walk_zones_in_node(m, pgdat, unusable_show_print);
1287
1288        return 0;
1289}
1290
1291static const struct seq_operations unusable_op = {
1292        .start  = frag_start,
1293        .next   = frag_next,
1294        .stop   = frag_stop,
1295        .show   = unusable_show,
1296};
1297
1298static int unusable_open(struct inode *inode, struct file *file)
1299{
1300        return seq_open(file, &unusable_op);
1301}
1302
1303static const struct file_operations unusable_file_ops = {
1304        .open           = unusable_open,
1305        .read           = seq_read,
1306        .llseek         = seq_lseek,
1307        .release        = seq_release,
1308};
1309
1310static void extfrag_show_print(struct seq_file *m,
1311                                        pg_data_t *pgdat, struct zone *zone)
1312{
1313        unsigned int order;
1314        int index;
1315
1316        /* Alloc on stack as interrupts are disabled for zone walk */
1317        struct contig_page_info info;
1318
1319        seq_printf(m, "Node %d, zone %8s ",
1320                                pgdat->node_id,
1321                                zone->name);
1322        for (order = 0; order < MAX_ORDER; ++order) {
1323                fill_contig_page_info(zone, order, &info);
1324                index = __fragmentation_index(order, &info);
1325                seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1326        }
1327
1328        seq_putc(m, '\n');
1329}
1330
1331/*
1332 * Display fragmentation index for orders that allocations would fail for
1333 */
1334static int extfrag_show(struct seq_file *m, void *arg)
1335{
1336        pg_data_t *pgdat = (pg_data_t *)arg;
1337
1338        walk_zones_in_node(m, pgdat, extfrag_show_print);
1339
1340        return 0;
1341}
1342
1343static const struct seq_operations extfrag_op = {
1344        .start  = frag_start,
1345        .next   = frag_next,
1346        .stop   = frag_stop,
1347        .show   = extfrag_show,
1348};
1349
1350static int extfrag_open(struct inode *inode, struct file *file)
1351{
1352        return seq_open(file, &extfrag_op);
1353}
1354
1355static const struct file_operations extfrag_file_ops = {
1356        .open           = extfrag_open,
1357        .read           = seq_read,
1358        .llseek         = seq_lseek,
1359        .release        = seq_release,
1360};
1361
1362static int __init extfrag_debug_init(void)
1363{
1364        struct dentry *extfrag_debug_root;
1365
1366        extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1367        if (!extfrag_debug_root)
1368                return -ENOMEM;
1369
1370        if (!debugfs_create_file("unusable_index", 0444,
1371                        extfrag_debug_root, NULL, &unusable_file_ops))
1372                goto fail;
1373
1374        if (!debugfs_create_file("extfrag_index", 0444,
1375                        extfrag_debug_root, NULL, &extfrag_file_ops))
1376                goto fail;
1377
1378        return 0;
1379fail:
1380        debugfs_remove_recursive(extfrag_debug_root);
1381        return -ENOMEM;
1382}
1383
1384module_init(extfrag_debug_init);
1385#endif
1386
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.