1f6ac2354SChristoph Lameter /* 2f6ac2354SChristoph Lameter * linux/mm/vmstat.c 3f6ac2354SChristoph Lameter * 4f6ac2354SChristoph Lameter * Manages VM statistics 5f6ac2354SChristoph Lameter * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 62244b95aSChristoph Lameter * 72244b95aSChristoph Lameter * zoned VM statistics 82244b95aSChristoph Lameter * Copyright (C) 2006 Silicon Graphics, Inc., 92244b95aSChristoph Lameter * Christoph Lameter <christoph@lameter.com> 107cc36bbdSChristoph Lameter * Copyright (C) 2008-2014 Christoph Lameter 11f6ac2354SChristoph Lameter */ 128f32f7e5SAlexey Dobriyan #include <linux/fs.h> 13f6ac2354SChristoph Lameter #include <linux/mm.h> 144e950f6fSAlexey Dobriyan #include <linux/err.h> 152244b95aSChristoph Lameter #include <linux/module.h> 165a0e3ad6STejun Heo #include <linux/slab.h> 17df9ecabaSChristoph Lameter #include <linux/cpu.h> 187cc36bbdSChristoph Lameter #include <linux/cpumask.h> 19c748e134SAdrian Bunk #include <linux/vmstat.h> 20*3c486871SAndrew Morton #include <linux/proc_fs.h> 21*3c486871SAndrew Morton #include <linux/seq_file.h> 22*3c486871SAndrew Morton #include <linux/debugfs.h> 23e8edc6e0SAlexey Dobriyan #include <linux/sched.h> 24f1a5ab12SMel Gorman #include <linux/math64.h> 2579da826aSMichael Rubin #include <linux/writeback.h> 2636deb0beSNamhyung Kim #include <linux/compaction.h> 276e543d57SLisa Du #include <linux/mm_inline.h> 2848c96a36SJoonsoo Kim #include <linux/page_ext.h> 2948c96a36SJoonsoo Kim #include <linux/page_owner.h> 306e543d57SLisa Du 316e543d57SLisa Du #include "internal.h" 32f6ac2354SChristoph Lameter 33f8891e5eSChristoph Lameter #ifdef CONFIG_VM_EVENT_COUNTERS 34f8891e5eSChristoph Lameter DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}}; 35f8891e5eSChristoph Lameter EXPORT_PER_CPU_SYMBOL(vm_event_states); 36f8891e5eSChristoph Lameter 3731f961a8SMinchan Kim static void sum_vm_events(unsigned long *ret) 38f8891e5eSChristoph Lameter { 399eccf2a8SChristoph Lameter int cpu; 40f8891e5eSChristoph Lameter int i; 41f8891e5eSChristoph Lameter 42f8891e5eSChristoph Lameter memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long)); 43f8891e5eSChristoph Lameter 4431f961a8SMinchan Kim for_each_online_cpu(cpu) { 45f8891e5eSChristoph Lameter struct vm_event_state *this = &per_cpu(vm_event_states, cpu); 46f8891e5eSChristoph Lameter 47f8891e5eSChristoph Lameter for (i = 0; i < NR_VM_EVENT_ITEMS; i++) 48f8891e5eSChristoph Lameter ret[i] += this->event[i]; 49f8891e5eSChristoph Lameter } 50f8891e5eSChristoph Lameter } 51f8891e5eSChristoph Lameter 52f8891e5eSChristoph Lameter /* 53f8891e5eSChristoph Lameter * Accumulate the vm event counters across all CPUs. 54f8891e5eSChristoph Lameter * The result is unavoidably approximate - it can change 55f8891e5eSChristoph Lameter * during and after execution of this function. 56f8891e5eSChristoph Lameter */ 57f8891e5eSChristoph Lameter void all_vm_events(unsigned long *ret) 58f8891e5eSChristoph Lameter { 59b5be1132SKOSAKI Motohiro get_online_cpus(); 6031f961a8SMinchan Kim sum_vm_events(ret); 61b5be1132SKOSAKI Motohiro put_online_cpus(); 62f8891e5eSChristoph Lameter } 6332dd66fcSHeiko Carstens EXPORT_SYMBOL_GPL(all_vm_events); 64f8891e5eSChristoph Lameter 65f8891e5eSChristoph Lameter /* 66f8891e5eSChristoph Lameter * Fold the foreign cpu events into our own. 67f8891e5eSChristoph Lameter * 68f8891e5eSChristoph Lameter * This is adding to the events on one processor 69f8891e5eSChristoph Lameter * but keeps the global counts constant. 70f8891e5eSChristoph Lameter */ 71f8891e5eSChristoph Lameter void vm_events_fold_cpu(int cpu) 72f8891e5eSChristoph Lameter { 73f8891e5eSChristoph Lameter struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu); 74f8891e5eSChristoph Lameter int i; 75f8891e5eSChristoph Lameter 76f8891e5eSChristoph Lameter for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { 77f8891e5eSChristoph Lameter count_vm_events(i, fold_state->event[i]); 78f8891e5eSChristoph Lameter fold_state->event[i] = 0; 79f8891e5eSChristoph Lameter } 80f8891e5eSChristoph Lameter } 81f8891e5eSChristoph Lameter 82f8891e5eSChristoph Lameter #endif /* CONFIG_VM_EVENT_COUNTERS */ 83f8891e5eSChristoph Lameter 842244b95aSChristoph Lameter /* 852244b95aSChristoph Lameter * Manage combined zone based / global counters 862244b95aSChristoph Lameter * 872244b95aSChristoph Lameter * vm_stat contains the global counters 882244b95aSChristoph Lameter */ 89a1cb2c60SDimitri Sivanich atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp; 902244b95aSChristoph Lameter EXPORT_SYMBOL(vm_stat); 912244b95aSChristoph Lameter 922244b95aSChristoph Lameter #ifdef CONFIG_SMP 932244b95aSChristoph Lameter 94b44129b3SMel Gorman int calculate_pressure_threshold(struct zone *zone) 9588f5acf8SMel Gorman { 9688f5acf8SMel Gorman int threshold; 9788f5acf8SMel Gorman int watermark_distance; 9888f5acf8SMel Gorman 9988f5acf8SMel Gorman /* 10088f5acf8SMel Gorman * As vmstats are not up to date, there is drift between the estimated 10188f5acf8SMel Gorman * and real values. For high thresholds and a high number of CPUs, it 10288f5acf8SMel Gorman * is possible for the min watermark to be breached while the estimated 10388f5acf8SMel Gorman * value looks fine. The pressure threshold is a reduced value such 10488f5acf8SMel Gorman * that even the maximum amount of drift will not accidentally breach 10588f5acf8SMel Gorman * the min watermark 10688f5acf8SMel Gorman */ 10788f5acf8SMel Gorman watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone); 10888f5acf8SMel Gorman threshold = max(1, (int)(watermark_distance / num_online_cpus())); 10988f5acf8SMel Gorman 11088f5acf8SMel Gorman /* 11188f5acf8SMel Gorman * Maximum threshold is 125 11288f5acf8SMel Gorman */ 11388f5acf8SMel Gorman threshold = min(125, threshold); 11488f5acf8SMel Gorman 11588f5acf8SMel Gorman return threshold; 11688f5acf8SMel Gorman } 11788f5acf8SMel Gorman 118b44129b3SMel Gorman int calculate_normal_threshold(struct zone *zone) 119df9ecabaSChristoph Lameter { 120df9ecabaSChristoph Lameter int threshold; 121df9ecabaSChristoph Lameter int mem; /* memory in 128 MB units */ 1222244b95aSChristoph Lameter 1232244b95aSChristoph Lameter /* 124df9ecabaSChristoph Lameter * The threshold scales with the number of processors and the amount 125df9ecabaSChristoph Lameter * of memory per zone. More memory means that we can defer updates for 126df9ecabaSChristoph Lameter * longer, more processors could lead to more contention. 127df9ecabaSChristoph Lameter * fls() is used to have a cheap way of logarithmic scaling. 1282244b95aSChristoph Lameter * 129df9ecabaSChristoph Lameter * Some sample thresholds: 130df9ecabaSChristoph Lameter * 131df9ecabaSChristoph Lameter * Threshold Processors (fls) Zonesize fls(mem+1) 132df9ecabaSChristoph Lameter * ------------------------------------------------------------------ 133df9ecabaSChristoph Lameter * 8 1 1 0.9-1 GB 4 134df9ecabaSChristoph Lameter * 16 2 2 0.9-1 GB 4 135df9ecabaSChristoph Lameter * 20 2 2 1-2 GB 5 136df9ecabaSChristoph Lameter * 24 2 2 2-4 GB 6 137df9ecabaSChristoph Lameter * 28 2 2 4-8 GB 7 138df9ecabaSChristoph Lameter * 32 2 2 8-16 GB 8 139df9ecabaSChristoph Lameter * 4 2 2 <128M 1 140df9ecabaSChristoph Lameter * 30 4 3 2-4 GB 5 141df9ecabaSChristoph Lameter * 48 4 3 8-16 GB 8 142df9ecabaSChristoph Lameter * 32 8 4 1-2 GB 4 143df9ecabaSChristoph Lameter * 32 8 4 0.9-1GB 4 144df9ecabaSChristoph Lameter * 10 16 5 <128M 1 145df9ecabaSChristoph Lameter * 40 16 5 900M 4 146df9ecabaSChristoph Lameter * 70 64 7 2-4 GB 5 147df9ecabaSChristoph Lameter * 84 64 7 4-8 GB 6 148df9ecabaSChristoph Lameter * 108 512 9 4-8 GB 6 149df9ecabaSChristoph Lameter * 125 1024 10 8-16 GB 8 150df9ecabaSChristoph Lameter * 125 1024 10 16-32 GB 9 1512244b95aSChristoph Lameter */ 152df9ecabaSChristoph Lameter 153b40da049SJiang Liu mem = zone->managed_pages >> (27 - PAGE_SHIFT); 154df9ecabaSChristoph Lameter 155df9ecabaSChristoph Lameter threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem)); 156df9ecabaSChristoph Lameter 157df9ecabaSChristoph Lameter /* 158df9ecabaSChristoph Lameter * Maximum threshold is 125 159df9ecabaSChristoph Lameter */ 160df9ecabaSChristoph Lameter threshold = min(125, threshold); 161df9ecabaSChristoph Lameter 162df9ecabaSChristoph Lameter return threshold; 163df9ecabaSChristoph Lameter } 164df9ecabaSChristoph Lameter 165df9ecabaSChristoph Lameter /* 166df9ecabaSChristoph Lameter * Refresh the thresholds for each zone. 167df9ecabaSChristoph Lameter */ 168a6cccdc3SKOSAKI Motohiro void refresh_zone_stat_thresholds(void) 1692244b95aSChristoph Lameter { 170df9ecabaSChristoph Lameter struct zone *zone; 171df9ecabaSChristoph Lameter int cpu; 172df9ecabaSChristoph Lameter int threshold; 173df9ecabaSChristoph Lameter 174ee99c71cSKOSAKI Motohiro for_each_populated_zone(zone) { 175aa454840SChristoph Lameter unsigned long max_drift, tolerate_drift; 176aa454840SChristoph Lameter 177b44129b3SMel Gorman threshold = calculate_normal_threshold(zone); 178df9ecabaSChristoph Lameter 179df9ecabaSChristoph Lameter for_each_online_cpu(cpu) 18099dcc3e5SChristoph Lameter per_cpu_ptr(zone->pageset, cpu)->stat_threshold 18199dcc3e5SChristoph Lameter = threshold; 182aa454840SChristoph Lameter 183aa454840SChristoph Lameter /* 184aa454840SChristoph Lameter * Only set percpu_drift_mark if there is a danger that 185aa454840SChristoph Lameter * NR_FREE_PAGES reports the low watermark is ok when in fact 186aa454840SChristoph Lameter * the min watermark could be breached by an allocation 187aa454840SChristoph Lameter */ 188aa454840SChristoph Lameter tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone); 189aa454840SChristoph Lameter max_drift = num_online_cpus() * threshold; 190aa454840SChristoph Lameter if (max_drift > tolerate_drift) 191aa454840SChristoph Lameter zone->percpu_drift_mark = high_wmark_pages(zone) + 192aa454840SChristoph Lameter max_drift; 193df9ecabaSChristoph Lameter } 1942244b95aSChristoph Lameter } 1952244b95aSChristoph Lameter 196b44129b3SMel Gorman void set_pgdat_percpu_threshold(pg_data_t *pgdat, 197b44129b3SMel Gorman int (*calculate_pressure)(struct zone *)) 19888f5acf8SMel Gorman { 19988f5acf8SMel Gorman struct zone *zone; 20088f5acf8SMel Gorman int cpu; 20188f5acf8SMel Gorman int threshold; 20288f5acf8SMel Gorman int i; 20388f5acf8SMel Gorman 20488f5acf8SMel Gorman for (i = 0; i < pgdat->nr_zones; i++) { 20588f5acf8SMel Gorman zone = &pgdat->node_zones[i]; 20688f5acf8SMel Gorman if (!zone->percpu_drift_mark) 20788f5acf8SMel Gorman continue; 20888f5acf8SMel Gorman 209b44129b3SMel Gorman threshold = (*calculate_pressure)(zone); 210bb0b6dffSMel Gorman for_each_online_cpu(cpu) 21188f5acf8SMel Gorman per_cpu_ptr(zone->pageset, cpu)->stat_threshold 21288f5acf8SMel Gorman = threshold; 21388f5acf8SMel Gorman } 21488f5acf8SMel Gorman } 21588f5acf8SMel Gorman 2162244b95aSChristoph Lameter /* 217bea04b07SJianyu Zhan * For use when we know that interrupts are disabled, 218bea04b07SJianyu Zhan * or when we know that preemption is disabled and that 219bea04b07SJianyu Zhan * particular counter cannot be updated from interrupt context. 2202244b95aSChristoph Lameter */ 2212244b95aSChristoph Lameter void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item, 2222244b95aSChristoph Lameter int delta) 2232244b95aSChristoph Lameter { 22412938a92SChristoph Lameter struct per_cpu_pageset __percpu *pcp = zone->pageset; 22512938a92SChristoph Lameter s8 __percpu *p = pcp->vm_stat_diff + item; 2262244b95aSChristoph Lameter long x; 22712938a92SChristoph Lameter long t; 2282244b95aSChristoph Lameter 22912938a92SChristoph Lameter x = delta + __this_cpu_read(*p); 2302244b95aSChristoph Lameter 23112938a92SChristoph Lameter t = __this_cpu_read(pcp->stat_threshold); 23212938a92SChristoph Lameter 23312938a92SChristoph Lameter if (unlikely(x > t || x < -t)) { 2342244b95aSChristoph Lameter zone_page_state_add(x, zone, item); 2352244b95aSChristoph Lameter x = 0; 2362244b95aSChristoph Lameter } 23712938a92SChristoph Lameter __this_cpu_write(*p, x); 2382244b95aSChristoph Lameter } 2392244b95aSChristoph Lameter EXPORT_SYMBOL(__mod_zone_page_state); 2402244b95aSChristoph Lameter 2412244b95aSChristoph Lameter /* 2422244b95aSChristoph Lameter * Optimized increment and decrement functions. 2432244b95aSChristoph Lameter * 2442244b95aSChristoph Lameter * These are only for a single page and therefore can take a struct page * 2452244b95aSChristoph Lameter * argument instead of struct zone *. This allows the inclusion of the code 2462244b95aSChristoph Lameter * generated for page_zone(page) into the optimized functions. 2472244b95aSChristoph Lameter * 2482244b95aSChristoph Lameter * No overflow check is necessary and therefore the differential can be 2492244b95aSChristoph Lameter * incremented or decremented in place which may allow the compilers to 2502244b95aSChristoph Lameter * generate better code. 2512244b95aSChristoph Lameter * The increment or decrement is known and therefore one boundary check can 2522244b95aSChristoph Lameter * be omitted. 2532244b95aSChristoph Lameter * 254df9ecabaSChristoph Lameter * NOTE: These functions are very performance sensitive. Change only 255df9ecabaSChristoph Lameter * with care. 256df9ecabaSChristoph Lameter * 2572244b95aSChristoph Lameter * Some processors have inc/dec instructions that are atomic vs an interrupt. 2582244b95aSChristoph Lameter * However, the code must first determine the differential location in a zone 2592244b95aSChristoph Lameter * based on the processor number and then inc/dec the counter. There is no 2602244b95aSChristoph Lameter * guarantee without disabling preemption that the processor will not change 2612244b95aSChristoph Lameter * in between and therefore the atomicity vs. interrupt cannot be exploited 2622244b95aSChristoph Lameter * in a useful way here. 2632244b95aSChristoph Lameter */ 264c8785385SChristoph Lameter void __inc_zone_state(struct zone *zone, enum zone_stat_item item) 2652244b95aSChristoph Lameter { 26612938a92SChristoph Lameter struct per_cpu_pageset __percpu *pcp = zone->pageset; 26712938a92SChristoph Lameter s8 __percpu *p = pcp->vm_stat_diff + item; 26812938a92SChristoph Lameter s8 v, t; 2692244b95aSChristoph Lameter 270908ee0f1SChristoph Lameter v = __this_cpu_inc_return(*p); 27112938a92SChristoph Lameter t = __this_cpu_read(pcp->stat_threshold); 27212938a92SChristoph Lameter if (unlikely(v > t)) { 27312938a92SChristoph Lameter s8 overstep = t >> 1; 2742244b95aSChristoph Lameter 27512938a92SChristoph Lameter zone_page_state_add(v + overstep, zone, item); 27612938a92SChristoph Lameter __this_cpu_write(*p, -overstep); 2772244b95aSChristoph Lameter } 2782244b95aSChristoph Lameter } 279ca889e6cSChristoph Lameter 280ca889e6cSChristoph Lameter void __inc_zone_page_state(struct page *page, enum zone_stat_item item) 281ca889e6cSChristoph Lameter { 282ca889e6cSChristoph Lameter __inc_zone_state(page_zone(page), item); 283ca889e6cSChristoph Lameter } 2842244b95aSChristoph Lameter EXPORT_SYMBOL(__inc_zone_page_state); 2852244b95aSChristoph Lameter 286c8785385SChristoph Lameter void __dec_zone_state(struct zone *zone, enum zone_stat_item item) 2872244b95aSChristoph Lameter { 28812938a92SChristoph Lameter struct per_cpu_pageset __percpu *pcp = zone->pageset; 28912938a92SChristoph Lameter s8 __percpu *p = pcp->vm_stat_diff + item; 29012938a92SChristoph Lameter s8 v, t; 2912244b95aSChristoph Lameter 292908ee0f1SChristoph Lameter v = __this_cpu_dec_return(*p); 29312938a92SChristoph Lameter t = __this_cpu_read(pcp->stat_threshold); 29412938a92SChristoph Lameter if (unlikely(v < - t)) { 29512938a92SChristoph Lameter s8 overstep = t >> 1; 2962244b95aSChristoph Lameter 29712938a92SChristoph Lameter zone_page_state_add(v - overstep, zone, item); 29812938a92SChristoph Lameter __this_cpu_write(*p, overstep); 2992244b95aSChristoph Lameter } 3002244b95aSChristoph Lameter } 301c8785385SChristoph Lameter 302c8785385SChristoph Lameter void __dec_zone_page_state(struct page *page, enum zone_stat_item item) 303c8785385SChristoph Lameter { 304c8785385SChristoph Lameter __dec_zone_state(page_zone(page), item); 305c8785385SChristoph Lameter } 3062244b95aSChristoph Lameter EXPORT_SYMBOL(__dec_zone_page_state); 3072244b95aSChristoph Lameter 3084156153cSHeiko Carstens #ifdef CONFIG_HAVE_CMPXCHG_LOCAL 3097c839120SChristoph Lameter /* 3107c839120SChristoph Lameter * If we have cmpxchg_local support then we do not need to incur the overhead 3117c839120SChristoph Lameter * that comes with local_irq_save/restore if we use this_cpu_cmpxchg. 3127c839120SChristoph Lameter * 3137c839120SChristoph Lameter * mod_state() modifies the zone counter state through atomic per cpu 3147c839120SChristoph Lameter * operations. 3157c839120SChristoph Lameter * 3167c839120SChristoph Lameter * Overstep mode specifies how overstep should handled: 3177c839120SChristoph Lameter * 0 No overstepping 3187c839120SChristoph Lameter * 1 Overstepping half of threshold 3197c839120SChristoph Lameter * -1 Overstepping minus half of threshold 3207c839120SChristoph Lameter */ 3217c839120SChristoph Lameter static inline void mod_state(struct zone *zone, 3227c839120SChristoph Lameter enum zone_stat_item item, int delta, int overstep_mode) 3237c839120SChristoph Lameter { 3247c839120SChristoph Lameter struct per_cpu_pageset __percpu *pcp = zone->pageset; 3257c839120SChristoph Lameter s8 __percpu *p = pcp->vm_stat_diff + item; 3267c839120SChristoph Lameter long o, n, t, z; 3277c839120SChristoph Lameter 3287c839120SChristoph Lameter do { 3297c839120SChristoph Lameter z = 0; /* overflow to zone counters */ 3307c839120SChristoph Lameter 3317c839120SChristoph Lameter /* 3327c839120SChristoph Lameter * The fetching of the stat_threshold is racy. We may apply 3337c839120SChristoph Lameter * a counter threshold to the wrong the cpu if we get 334d3bc2367SChristoph Lameter * rescheduled while executing here. However, the next 335d3bc2367SChristoph Lameter * counter update will apply the threshold again and 336d3bc2367SChristoph Lameter * therefore bring the counter under the threshold again. 337d3bc2367SChristoph Lameter * 338d3bc2367SChristoph Lameter * Most of the time the thresholds are the same anyways 339d3bc2367SChristoph Lameter * for all cpus in a zone. 3407c839120SChristoph Lameter */ 3417c839120SChristoph Lameter t = this_cpu_read(pcp->stat_threshold); 3427c839120SChristoph Lameter 3437c839120SChristoph Lameter o = this_cpu_read(*p); 3447c839120SChristoph Lameter n = delta + o; 3457c839120SChristoph Lameter 3467c839120SChristoph Lameter if (n > t || n < -t) { 3477c839120SChristoph Lameter int os = overstep_mode * (t >> 1) ; 3487c839120SChristoph Lameter 3497c839120SChristoph Lameter /* Overflow must be added to zone counters */ 3507c839120SChristoph Lameter z = n + os; 3517c839120SChristoph Lameter n = -os; 3527c839120SChristoph Lameter } 3537c839120SChristoph Lameter } while (this_cpu_cmpxchg(*p, o, n) != o); 3547c839120SChristoph Lameter 3557c839120SChristoph Lameter if (z) 3567c839120SChristoph Lameter zone_page_state_add(z, zone, item); 3577c839120SChristoph Lameter } 3587c839120SChristoph Lameter 3597c839120SChristoph Lameter void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, 3607c839120SChristoph Lameter int delta) 3617c839120SChristoph Lameter { 3627c839120SChristoph Lameter mod_state(zone, item, delta, 0); 3637c839120SChristoph Lameter } 3647c839120SChristoph Lameter EXPORT_SYMBOL(mod_zone_page_state); 3657c839120SChristoph Lameter 3667c839120SChristoph Lameter void inc_zone_state(struct zone *zone, enum zone_stat_item item) 3677c839120SChristoph Lameter { 3687c839120SChristoph Lameter mod_state(zone, item, 1, 1); 3697c839120SChristoph Lameter } 3707c839120SChristoph Lameter 3717c839120SChristoph Lameter void inc_zone_page_state(struct page *page, enum zone_stat_item item) 3727c839120SChristoph Lameter { 3737c839120SChristoph Lameter mod_state(page_zone(page), item, 1, 1); 3747c839120SChristoph Lameter } 3757c839120SChristoph Lameter EXPORT_SYMBOL(inc_zone_page_state); 3767c839120SChristoph Lameter 3777c839120SChristoph Lameter void dec_zone_page_state(struct page *page, enum zone_stat_item item) 3787c839120SChristoph Lameter { 3797c839120SChristoph Lameter mod_state(page_zone(page), item, -1, -1); 3807c839120SChristoph Lameter } 3817c839120SChristoph Lameter EXPORT_SYMBOL(dec_zone_page_state); 3827c839120SChristoph Lameter #else 3837c839120SChristoph Lameter /* 3847c839120SChristoph Lameter * Use interrupt disable to serialize counter updates 3857c839120SChristoph Lameter */ 3867c839120SChristoph Lameter void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, 3877c839120SChristoph Lameter int delta) 3887c839120SChristoph Lameter { 3897c839120SChristoph Lameter unsigned long flags; 3907c839120SChristoph Lameter 3917c839120SChristoph Lameter local_irq_save(flags); 3927c839120SChristoph Lameter __mod_zone_page_state(zone, item, delta); 3937c839120SChristoph Lameter local_irq_restore(flags); 3947c839120SChristoph Lameter } 3957c839120SChristoph Lameter EXPORT_SYMBOL(mod_zone_page_state); 3967c839120SChristoph Lameter 397ca889e6cSChristoph Lameter void inc_zone_state(struct zone *zone, enum zone_stat_item item) 398ca889e6cSChristoph Lameter { 399ca889e6cSChristoph Lameter unsigned long flags; 400ca889e6cSChristoph Lameter 401ca889e6cSChristoph Lameter local_irq_save(flags); 402ca889e6cSChristoph Lameter __inc_zone_state(zone, item); 403ca889e6cSChristoph Lameter local_irq_restore(flags); 404ca889e6cSChristoph Lameter } 405ca889e6cSChristoph Lameter 4062244b95aSChristoph Lameter void inc_zone_page_state(struct page *page, enum zone_stat_item item) 4072244b95aSChristoph Lameter { 4082244b95aSChristoph Lameter unsigned long flags; 4092244b95aSChristoph Lameter struct zone *zone; 4102244b95aSChristoph Lameter 4112244b95aSChristoph Lameter zone = page_zone(page); 4122244b95aSChristoph Lameter local_irq_save(flags); 413ca889e6cSChristoph Lameter __inc_zone_state(zone, item); 4142244b95aSChristoph Lameter local_irq_restore(flags); 4152244b95aSChristoph Lameter } 4162244b95aSChristoph Lameter EXPORT_SYMBOL(inc_zone_page_state); 4172244b95aSChristoph Lameter 4182244b95aSChristoph Lameter void dec_zone_page_state(struct page *page, enum zone_stat_item item) 4192244b95aSChristoph Lameter { 4202244b95aSChristoph Lameter unsigned long flags; 4212244b95aSChristoph Lameter 4222244b95aSChristoph Lameter local_irq_save(flags); 423a302eb4eSChristoph Lameter __dec_zone_page_state(page, item); 4242244b95aSChristoph Lameter local_irq_restore(flags); 4252244b95aSChristoph Lameter } 4262244b95aSChristoph Lameter EXPORT_SYMBOL(dec_zone_page_state); 4277c839120SChristoph Lameter #endif 4282244b95aSChristoph Lameter 4297cc36bbdSChristoph Lameter 4307cc36bbdSChristoph Lameter /* 4317cc36bbdSChristoph Lameter * Fold a differential into the global counters. 4327cc36bbdSChristoph Lameter * Returns the number of counters updated. 4337cc36bbdSChristoph Lameter */ 4347cc36bbdSChristoph Lameter static int fold_diff(int *diff) 4354edb0748SChristoph Lameter { 4364edb0748SChristoph Lameter int i; 4377cc36bbdSChristoph Lameter int changes = 0; 4384edb0748SChristoph Lameter 4394edb0748SChristoph Lameter for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) 4407cc36bbdSChristoph Lameter if (diff[i]) { 4414edb0748SChristoph Lameter atomic_long_add(diff[i], &vm_stat[i]); 4427cc36bbdSChristoph Lameter changes++; 4437cc36bbdSChristoph Lameter } 4447cc36bbdSChristoph Lameter return changes; 4454edb0748SChristoph Lameter } 4464edb0748SChristoph Lameter 4472244b95aSChristoph Lameter /* 4482bb921e5SChristoph Lameter * Update the zone counters for the current cpu. 449a7f75e25SChristoph Lameter * 4504037d452SChristoph Lameter * Note that refresh_cpu_vm_stats strives to only access 4514037d452SChristoph Lameter * node local memory. The per cpu pagesets on remote zones are placed 4524037d452SChristoph Lameter * in the memory local to the processor using that pageset. So the 4534037d452SChristoph Lameter * loop over all zones will access a series of cachelines local to 4544037d452SChristoph Lameter * the processor. 4554037d452SChristoph Lameter * 4564037d452SChristoph Lameter * The call to zone_page_state_add updates the cachelines with the 4574037d452SChristoph Lameter * statistics in the remote zone struct as well as the global cachelines 4584037d452SChristoph Lameter * with the global counters. These could cause remote node cache line 4594037d452SChristoph Lameter * bouncing and will have to be only done when necessary. 4607cc36bbdSChristoph Lameter * 4617cc36bbdSChristoph Lameter * The function returns the number of global counters updated. 4622244b95aSChristoph Lameter */ 4637cc36bbdSChristoph Lameter static int refresh_cpu_vm_stats(void) 4642244b95aSChristoph Lameter { 4652244b95aSChristoph Lameter struct zone *zone; 4662244b95aSChristoph Lameter int i; 467a7f75e25SChristoph Lameter int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, }; 4687cc36bbdSChristoph Lameter int changes = 0; 4692244b95aSChristoph Lameter 470ee99c71cSKOSAKI Motohiro for_each_populated_zone(zone) { 471fbc2edb0SChristoph Lameter struct per_cpu_pageset __percpu *p = zone->pageset; 4722244b95aSChristoph Lameter 473fbc2edb0SChristoph Lameter for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) { 474a7f75e25SChristoph Lameter int v; 475a7f75e25SChristoph Lameter 476fbc2edb0SChristoph Lameter v = this_cpu_xchg(p->vm_stat_diff[i], 0); 477fbc2edb0SChristoph Lameter if (v) { 478fbc2edb0SChristoph Lameter 479a7f75e25SChristoph Lameter atomic_long_add(v, &zone->vm_stat[i]); 480a7f75e25SChristoph Lameter global_diff[i] += v; 4814037d452SChristoph Lameter #ifdef CONFIG_NUMA 4824037d452SChristoph Lameter /* 3 seconds idle till flush */ 483fbc2edb0SChristoph Lameter __this_cpu_write(p->expire, 3); 4844037d452SChristoph Lameter #endif 4852244b95aSChristoph Lameter } 486fbc2edb0SChristoph Lameter } 487468fd62eSDimitri Sivanich cond_resched(); 4884037d452SChristoph Lameter #ifdef CONFIG_NUMA 4894037d452SChristoph Lameter /* 4904037d452SChristoph Lameter * Deal with draining the remote pageset of this 4914037d452SChristoph Lameter * processor 4924037d452SChristoph Lameter * 4934037d452SChristoph Lameter * Check if there are pages remaining in this pageset 4944037d452SChristoph Lameter * if not then there is nothing to expire. 4954037d452SChristoph Lameter */ 496fbc2edb0SChristoph Lameter if (!__this_cpu_read(p->expire) || 497fbc2edb0SChristoph Lameter !__this_cpu_read(p->pcp.count)) 4984037d452SChristoph Lameter continue; 4994037d452SChristoph Lameter 5004037d452SChristoph Lameter /* 5014037d452SChristoph Lameter * We never drain zones local to this processor. 5024037d452SChristoph Lameter */ 5034037d452SChristoph Lameter if (zone_to_nid(zone) == numa_node_id()) { 504fbc2edb0SChristoph Lameter __this_cpu_write(p->expire, 0); 5054037d452SChristoph Lameter continue; 5064037d452SChristoph Lameter } 5074037d452SChristoph Lameter 508fbc2edb0SChristoph Lameter if (__this_cpu_dec_return(p->expire)) 5094037d452SChristoph Lameter continue; 5104037d452SChristoph Lameter 5117cc36bbdSChristoph Lameter if (__this_cpu_read(p->pcp.count)) { 5127c8e0181SChristoph Lameter drain_zone_pages(zone, this_cpu_ptr(&p->pcp)); 5137cc36bbdSChristoph Lameter changes++; 5147cc36bbdSChristoph Lameter } 5154037d452SChristoph Lameter #endif 5162244b95aSChristoph Lameter } 5177cc36bbdSChristoph Lameter changes += fold_diff(global_diff); 5187cc36bbdSChristoph Lameter return changes; 5192244b95aSChristoph Lameter } 5202244b95aSChristoph Lameter 52140f4b1eaSCody P Schafer /* 5222bb921e5SChristoph Lameter * Fold the data for an offline cpu into the global array. 5232bb921e5SChristoph Lameter * There cannot be any access by the offline cpu and therefore 5242bb921e5SChristoph Lameter * synchronization is simplified. 5252bb921e5SChristoph Lameter */ 5262bb921e5SChristoph Lameter void cpu_vm_stats_fold(int cpu) 5272bb921e5SChristoph Lameter { 5282bb921e5SChristoph Lameter struct zone *zone; 5292bb921e5SChristoph Lameter int i; 5302bb921e5SChristoph Lameter int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, }; 5312bb921e5SChristoph Lameter 5322bb921e5SChristoph Lameter for_each_populated_zone(zone) { 5332bb921e5SChristoph Lameter struct per_cpu_pageset *p; 5342bb921e5SChristoph Lameter 5352bb921e5SChristoph Lameter p = per_cpu_ptr(zone->pageset, cpu); 5362bb921e5SChristoph Lameter 5372bb921e5SChristoph Lameter for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) 5382bb921e5SChristoph Lameter if (p->vm_stat_diff[i]) { 5392bb921e5SChristoph Lameter int v; 5402bb921e5SChristoph Lameter 5412bb921e5SChristoph Lameter v = p->vm_stat_diff[i]; 5422bb921e5SChristoph Lameter p->vm_stat_diff[i] = 0; 5432bb921e5SChristoph Lameter atomic_long_add(v, &zone->vm_stat[i]); 5442bb921e5SChristoph Lameter global_diff[i] += v; 5452bb921e5SChristoph Lameter } 5462bb921e5SChristoph Lameter } 5472bb921e5SChristoph Lameter 5484edb0748SChristoph Lameter fold_diff(global_diff); 5492bb921e5SChristoph Lameter } 5502bb921e5SChristoph Lameter 5512bb921e5SChristoph Lameter /* 55240f4b1eaSCody P Schafer * this is only called if !populated_zone(zone), which implies no other users of 55340f4b1eaSCody P Schafer * pset->vm_stat_diff[] exsist. 55440f4b1eaSCody P Schafer */ 5555a883813SMinchan Kim void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset) 5565a883813SMinchan Kim { 5575a883813SMinchan Kim int i; 5585a883813SMinchan Kim 5595a883813SMinchan Kim for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) 5605a883813SMinchan Kim if (pset->vm_stat_diff[i]) { 5615a883813SMinchan Kim int v = pset->vm_stat_diff[i]; 5625a883813SMinchan Kim pset->vm_stat_diff[i] = 0; 5635a883813SMinchan Kim atomic_long_add(v, &zone->vm_stat[i]); 5645a883813SMinchan Kim atomic_long_add(v, &vm_stat[i]); 5655a883813SMinchan Kim } 5665a883813SMinchan Kim } 5672244b95aSChristoph Lameter #endif 5682244b95aSChristoph Lameter 569ca889e6cSChristoph Lameter #ifdef CONFIG_NUMA 570ca889e6cSChristoph Lameter /* 571ca889e6cSChristoph Lameter * zonelist = the list of zones passed to the allocator 572ca889e6cSChristoph Lameter * z = the zone from which the allocation occurred. 573ca889e6cSChristoph Lameter * 574ca889e6cSChristoph Lameter * Must be called with interrupts disabled. 57578afd561SAndi Kleen * 57678afd561SAndi Kleen * When __GFP_OTHER_NODE is set assume the node of the preferred 57778afd561SAndi Kleen * zone is the local node. This is useful for daemons who allocate 57878afd561SAndi Kleen * memory on behalf of other processes. 579ca889e6cSChristoph Lameter */ 58078afd561SAndi Kleen void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags) 581ca889e6cSChristoph Lameter { 58218ea7e71SMel Gorman if (z->zone_pgdat == preferred_zone->zone_pgdat) { 583ca889e6cSChristoph Lameter __inc_zone_state(z, NUMA_HIT); 584ca889e6cSChristoph Lameter } else { 585ca889e6cSChristoph Lameter __inc_zone_state(z, NUMA_MISS); 58618ea7e71SMel Gorman __inc_zone_state(preferred_zone, NUMA_FOREIGN); 587ca889e6cSChristoph Lameter } 58878afd561SAndi Kleen if (z->node == ((flags & __GFP_OTHER_NODE) ? 58978afd561SAndi Kleen preferred_zone->node : numa_node_id())) 590ca889e6cSChristoph Lameter __inc_zone_state(z, NUMA_LOCAL); 591ca889e6cSChristoph Lameter else 592ca889e6cSChristoph Lameter __inc_zone_state(z, NUMA_OTHER); 593ca889e6cSChristoph Lameter } 594ca889e6cSChristoph Lameter #endif 595ca889e6cSChristoph Lameter 596d7a5752cSMel Gorman #ifdef CONFIG_COMPACTION 59736deb0beSNamhyung Kim 598d7a5752cSMel Gorman struct contig_page_info { 599d7a5752cSMel Gorman unsigned long free_pages; 600d7a5752cSMel Gorman unsigned long free_blocks_total; 601d7a5752cSMel Gorman unsigned long free_blocks_suitable; 602d7a5752cSMel Gorman }; 603d7a5752cSMel Gorman 604d7a5752cSMel Gorman /* 605d7a5752cSMel Gorman * Calculate the number of free pages in a zone, how many contiguous 606d7a5752cSMel Gorman * pages are free and how many are large enough to satisfy an allocation of 607d7a5752cSMel Gorman * the target size. Note that this function makes no attempt to estimate 608d7a5752cSMel Gorman * how many suitable free blocks there *might* be if MOVABLE pages were 609d7a5752cSMel Gorman * migrated. Calculating that is possible, but expensive and can be 610d7a5752cSMel Gorman * figured out from userspace 611d7a5752cSMel Gorman */ 612d7a5752cSMel Gorman static void fill_contig_page_info(struct zone *zone, 613d7a5752cSMel Gorman unsigned int suitable_order, 614d7a5752cSMel Gorman struct contig_page_info *info) 615d7a5752cSMel Gorman { 616d7a5752cSMel Gorman unsigned int order; 617d7a5752cSMel Gorman 618d7a5752cSMel Gorman info->free_pages = 0; 619d7a5752cSMel Gorman info->free_blocks_total = 0; 620d7a5752cSMel Gorman info->free_blocks_suitable = 0; 621d7a5752cSMel Gorman 622d7a5752cSMel Gorman for (order = 0; order < MAX_ORDER; order++) { 623d7a5752cSMel Gorman unsigned long blocks; 624d7a5752cSMel Gorman 625d7a5752cSMel Gorman /* Count number of free blocks */ 626d7a5752cSMel Gorman blocks = zone->free_area[order].nr_free; 627d7a5752cSMel Gorman info->free_blocks_total += blocks; 628d7a5752cSMel Gorman 629d7a5752cSMel Gorman /* Count free base pages */ 630d7a5752cSMel Gorman info->free_pages += blocks << order; 631d7a5752cSMel Gorman 632d7a5752cSMel Gorman /* Count the suitable free blocks */ 633d7a5752cSMel Gorman if (order >= suitable_order) 634d7a5752cSMel Gorman info->free_blocks_suitable += blocks << 635d7a5752cSMel Gorman (order - suitable_order); 636d7a5752cSMel Gorman } 637d7a5752cSMel Gorman } 638f1a5ab12SMel Gorman 639f1a5ab12SMel Gorman /* 640f1a5ab12SMel Gorman * A fragmentation index only makes sense if an allocation of a requested 641f1a5ab12SMel Gorman * size would fail. If that is true, the fragmentation index indicates 642f1a5ab12SMel Gorman * whether external fragmentation or a lack of memory was the problem. 643f1a5ab12SMel Gorman * The value can be used to determine if page reclaim or compaction 644f1a5ab12SMel Gorman * should be used 645f1a5ab12SMel Gorman */ 64656de7263SMel Gorman static int __fragmentation_index(unsigned int order, struct contig_page_info *info) 647f1a5ab12SMel Gorman { 648f1a5ab12SMel Gorman unsigned long requested = 1UL << order; 649f1a5ab12SMel Gorman 650f1a5ab12SMel Gorman if (!info->free_blocks_total) 651f1a5ab12SMel Gorman return 0; 652f1a5ab12SMel Gorman 653f1a5ab12SMel Gorman /* Fragmentation index only makes sense when a request would fail */ 654f1a5ab12SMel Gorman if (info->free_blocks_suitable) 655f1a5ab12SMel Gorman return -1000; 656f1a5ab12SMel Gorman 657f1a5ab12SMel Gorman /* 658f1a5ab12SMel Gorman * Index is between 0 and 1 so return within 3 decimal places 659f1a5ab12SMel Gorman * 660f1a5ab12SMel Gorman * 0 => allocation would fail due to lack of memory 661f1a5ab12SMel Gorman * 1 => allocation would fail due to fragmentation 662f1a5ab12SMel Gorman */ 663f1a5ab12SMel Gorman return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total); 664f1a5ab12SMel Gorman } 66556de7263SMel Gorman 66656de7263SMel Gorman /* Same as __fragmentation index but allocs contig_page_info on stack */ 66756de7263SMel Gorman int fragmentation_index(struct zone *zone, unsigned int order) 66856de7263SMel Gorman { 66956de7263SMel Gorman struct contig_page_info info; 67056de7263SMel Gorman 67156de7263SMel Gorman fill_contig_page_info(zone, order, &info); 67256de7263SMel Gorman return __fragmentation_index(order, &info); 67356de7263SMel Gorman } 674d7a5752cSMel Gorman #endif 675d7a5752cSMel Gorman 6760d6617c7SDavid Rientjes #if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA) 677fa25c503SKOSAKI Motohiro #ifdef CONFIG_ZONE_DMA 678fa25c503SKOSAKI Motohiro #define TEXT_FOR_DMA(xx) xx "_dma", 679fa25c503SKOSAKI Motohiro #else 680fa25c503SKOSAKI Motohiro #define TEXT_FOR_DMA(xx) 681fa25c503SKOSAKI Motohiro #endif 682fa25c503SKOSAKI Motohiro 683fa25c503SKOSAKI Motohiro #ifdef CONFIG_ZONE_DMA32 684fa25c503SKOSAKI Motohiro #define TEXT_FOR_DMA32(xx) xx "_dma32", 685fa25c503SKOSAKI Motohiro #else 686fa25c503SKOSAKI Motohiro #define TEXT_FOR_DMA32(xx) 687fa25c503SKOSAKI Motohiro #endif 688fa25c503SKOSAKI Motohiro 689fa25c503SKOSAKI Motohiro #ifdef CONFIG_HIGHMEM 690fa25c503SKOSAKI Motohiro #define TEXT_FOR_HIGHMEM(xx) xx "_high", 691fa25c503SKOSAKI Motohiro #else 692fa25c503SKOSAKI Motohiro #define TEXT_FOR_HIGHMEM(xx) 693fa25c503SKOSAKI Motohiro #endif 694fa25c503SKOSAKI Motohiro 695fa25c503SKOSAKI Motohiro #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \ 696fa25c503SKOSAKI Motohiro TEXT_FOR_HIGHMEM(xx) xx "_movable", 697fa25c503SKOSAKI Motohiro 698fa25c503SKOSAKI Motohiro const char * const vmstat_text[] = { 69909316c09SKonstantin Khlebnikov /* enum zone_stat_item countes */ 700fa25c503SKOSAKI Motohiro "nr_free_pages", 70181c0a2bbSJohannes Weiner "nr_alloc_batch", 702fa25c503SKOSAKI Motohiro "nr_inactive_anon", 703fa25c503SKOSAKI Motohiro "nr_active_anon", 704fa25c503SKOSAKI Motohiro "nr_inactive_file", 705fa25c503SKOSAKI Motohiro "nr_active_file", 706fa25c503SKOSAKI Motohiro "nr_unevictable", 707fa25c503SKOSAKI Motohiro "nr_mlock", 708fa25c503SKOSAKI Motohiro "nr_anon_pages", 709fa25c503SKOSAKI Motohiro "nr_mapped", 710fa25c503SKOSAKI Motohiro "nr_file_pages", 711fa25c503SKOSAKI Motohiro "nr_dirty", 712fa25c503SKOSAKI Motohiro "nr_writeback", 713fa25c503SKOSAKI Motohiro "nr_slab_reclaimable", 714fa25c503SKOSAKI Motohiro "nr_slab_unreclaimable", 715fa25c503SKOSAKI Motohiro "nr_page_table_pages", 716fa25c503SKOSAKI Motohiro "nr_kernel_stack", 717fa25c503SKOSAKI Motohiro "nr_unstable", 718fa25c503SKOSAKI Motohiro "nr_bounce", 719fa25c503SKOSAKI Motohiro "nr_vmscan_write", 72049ea7eb6SMel Gorman "nr_vmscan_immediate_reclaim", 721fa25c503SKOSAKI Motohiro "nr_writeback_temp", 722fa25c503SKOSAKI Motohiro "nr_isolated_anon", 723fa25c503SKOSAKI Motohiro "nr_isolated_file", 724fa25c503SKOSAKI Motohiro "nr_shmem", 725fa25c503SKOSAKI Motohiro "nr_dirtied", 726fa25c503SKOSAKI Motohiro "nr_written", 7270d5d823aSMel Gorman "nr_pages_scanned", 728fa25c503SKOSAKI Motohiro 729fa25c503SKOSAKI Motohiro #ifdef CONFIG_NUMA 730fa25c503SKOSAKI Motohiro "numa_hit", 731fa25c503SKOSAKI Motohiro "numa_miss", 732fa25c503SKOSAKI Motohiro "numa_foreign", 733fa25c503SKOSAKI Motohiro "numa_interleave", 734fa25c503SKOSAKI Motohiro "numa_local", 735fa25c503SKOSAKI Motohiro "numa_other", 736fa25c503SKOSAKI Motohiro #endif 737a528910eSJohannes Weiner "workingset_refault", 738a528910eSJohannes Weiner "workingset_activate", 739449dd698SJohannes Weiner "workingset_nodereclaim", 740fa25c503SKOSAKI Motohiro "nr_anon_transparent_hugepages", 741d1ce749aSBartlomiej Zolnierkiewicz "nr_free_cma", 74209316c09SKonstantin Khlebnikov 74309316c09SKonstantin Khlebnikov /* enum writeback_stat_item counters */ 744fa25c503SKOSAKI Motohiro "nr_dirty_threshold", 745fa25c503SKOSAKI Motohiro "nr_dirty_background_threshold", 746fa25c503SKOSAKI Motohiro 747fa25c503SKOSAKI Motohiro #ifdef CONFIG_VM_EVENT_COUNTERS 74809316c09SKonstantin Khlebnikov /* enum vm_event_item counters */ 749fa25c503SKOSAKI Motohiro "pgpgin", 750fa25c503SKOSAKI Motohiro "pgpgout", 751fa25c503SKOSAKI Motohiro "pswpin", 752fa25c503SKOSAKI Motohiro "pswpout", 753fa25c503SKOSAKI Motohiro 754fa25c503SKOSAKI Motohiro TEXTS_FOR_ZONES("pgalloc") 755fa25c503SKOSAKI Motohiro 756fa25c503SKOSAKI Motohiro "pgfree", 757fa25c503SKOSAKI Motohiro "pgactivate", 758fa25c503SKOSAKI Motohiro "pgdeactivate", 759fa25c503SKOSAKI Motohiro 760fa25c503SKOSAKI Motohiro "pgfault", 761fa25c503SKOSAKI Motohiro "pgmajfault", 762fa25c503SKOSAKI Motohiro 763fa25c503SKOSAKI Motohiro TEXTS_FOR_ZONES("pgrefill") 764904249aaSYing Han TEXTS_FOR_ZONES("pgsteal_kswapd") 765904249aaSYing Han TEXTS_FOR_ZONES("pgsteal_direct") 766fa25c503SKOSAKI Motohiro TEXTS_FOR_ZONES("pgscan_kswapd") 767fa25c503SKOSAKI Motohiro TEXTS_FOR_ZONES("pgscan_direct") 76868243e76SMel Gorman "pgscan_direct_throttle", 769fa25c503SKOSAKI Motohiro 770fa25c503SKOSAKI Motohiro #ifdef CONFIG_NUMA 771fa25c503SKOSAKI Motohiro "zone_reclaim_failed", 772fa25c503SKOSAKI Motohiro #endif 773fa25c503SKOSAKI Motohiro "pginodesteal", 774fa25c503SKOSAKI Motohiro "slabs_scanned", 775fa25c503SKOSAKI Motohiro "kswapd_inodesteal", 776fa25c503SKOSAKI Motohiro "kswapd_low_wmark_hit_quickly", 777fa25c503SKOSAKI Motohiro "kswapd_high_wmark_hit_quickly", 778fa25c503SKOSAKI Motohiro "pageoutrun", 779fa25c503SKOSAKI Motohiro "allocstall", 780fa25c503SKOSAKI Motohiro 781fa25c503SKOSAKI Motohiro "pgrotated", 782fa25c503SKOSAKI Motohiro 7835509a5d2SDave Hansen "drop_pagecache", 7845509a5d2SDave Hansen "drop_slab", 7855509a5d2SDave Hansen 78603c5a6e1SMel Gorman #ifdef CONFIG_NUMA_BALANCING 78703c5a6e1SMel Gorman "numa_pte_updates", 78872403b4aSMel Gorman "numa_huge_pte_updates", 78903c5a6e1SMel Gorman "numa_hint_faults", 79003c5a6e1SMel Gorman "numa_hint_faults_local", 79103c5a6e1SMel Gorman "numa_pages_migrated", 79203c5a6e1SMel Gorman #endif 7935647bc29SMel Gorman #ifdef CONFIG_MIGRATION 7945647bc29SMel Gorman "pgmigrate_success", 7955647bc29SMel Gorman "pgmigrate_fail", 7965647bc29SMel Gorman #endif 797fa25c503SKOSAKI Motohiro #ifdef CONFIG_COMPACTION 798397487dbSMel Gorman "compact_migrate_scanned", 799397487dbSMel Gorman "compact_free_scanned", 800397487dbSMel Gorman "compact_isolated", 801fa25c503SKOSAKI Motohiro "compact_stall", 802fa25c503SKOSAKI Motohiro "compact_fail", 803fa25c503SKOSAKI Motohiro "compact_success", 804fa25c503SKOSAKI Motohiro #endif 805fa25c503SKOSAKI Motohiro 806fa25c503SKOSAKI Motohiro #ifdef CONFIG_HUGETLB_PAGE 807fa25c503SKOSAKI Motohiro "htlb_buddy_alloc_success", 808fa25c503SKOSAKI Motohiro "htlb_buddy_alloc_fail", 809fa25c503SKOSAKI Motohiro #endif 810fa25c503SKOSAKI Motohiro "unevictable_pgs_culled", 811fa25c503SKOSAKI Motohiro "unevictable_pgs_scanned", 812fa25c503SKOSAKI Motohiro "unevictable_pgs_rescued", 813fa25c503SKOSAKI Motohiro "unevictable_pgs_mlocked", 814fa25c503SKOSAKI Motohiro "unevictable_pgs_munlocked", 815fa25c503SKOSAKI Motohiro "unevictable_pgs_cleared", 816fa25c503SKOSAKI Motohiro "unevictable_pgs_stranded", 817fa25c503SKOSAKI Motohiro 818fa25c503SKOSAKI Motohiro #ifdef CONFIG_TRANSPARENT_HUGEPAGE 819fa25c503SKOSAKI Motohiro "thp_fault_alloc", 820fa25c503SKOSAKI Motohiro "thp_fault_fallback", 821fa25c503SKOSAKI Motohiro "thp_collapse_alloc", 822fa25c503SKOSAKI Motohiro "thp_collapse_alloc_failed", 823fa25c503SKOSAKI Motohiro "thp_split", 824d8a8e1f0SKirill A. Shutemov "thp_zero_page_alloc", 825d8a8e1f0SKirill A. Shutemov "thp_zero_page_alloc_failed", 826fa25c503SKOSAKI Motohiro #endif 82709316c09SKonstantin Khlebnikov #ifdef CONFIG_MEMORY_BALLOON 82809316c09SKonstantin Khlebnikov "balloon_inflate", 82909316c09SKonstantin Khlebnikov "balloon_deflate", 83009316c09SKonstantin Khlebnikov #ifdef CONFIG_BALLOON_COMPACTION 83109316c09SKonstantin Khlebnikov "balloon_migrate", 83209316c09SKonstantin Khlebnikov #endif 83309316c09SKonstantin Khlebnikov #endif /* CONFIG_MEMORY_BALLOON */ 834ec659934SMel Gorman #ifdef CONFIG_DEBUG_TLBFLUSH 8356df46865SDave Hansen #ifdef CONFIG_SMP 8369824cf97SDave Hansen "nr_tlb_remote_flush", 8379824cf97SDave Hansen "nr_tlb_remote_flush_received", 838ec659934SMel Gorman #endif /* CONFIG_SMP */ 8399824cf97SDave Hansen "nr_tlb_local_flush_all", 8409824cf97SDave Hansen "nr_tlb_local_flush_one", 841ec659934SMel Gorman #endif /* CONFIG_DEBUG_TLBFLUSH */ 842fa25c503SKOSAKI Motohiro 8434f115147SDavidlohr Bueso #ifdef CONFIG_DEBUG_VM_VMACACHE 8444f115147SDavidlohr Bueso "vmacache_find_calls", 8454f115147SDavidlohr Bueso "vmacache_find_hits", 846f5f302e2SDavidlohr Bueso "vmacache_full_flushes", 8474f115147SDavidlohr Bueso #endif 848fa25c503SKOSAKI Motohiro #endif /* CONFIG_VM_EVENTS_COUNTERS */ 849fa25c503SKOSAKI Motohiro }; 8500d6617c7SDavid Rientjes #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */ 851fa25c503SKOSAKI Motohiro 852fa25c503SKOSAKI Motohiro 853*3c486871SAndrew Morton #if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \ 854*3c486871SAndrew Morton defined(CONFIG_PROC_FS) 855*3c486871SAndrew Morton static void *frag_start(struct seq_file *m, loff_t *pos) 856*3c486871SAndrew Morton { 857*3c486871SAndrew Morton pg_data_t *pgdat; 858*3c486871SAndrew Morton loff_t node = *pos; 859*3c486871SAndrew Morton 860*3c486871SAndrew Morton for (pgdat = first_online_pgdat(); 861*3c486871SAndrew Morton pgdat && node; 862*3c486871SAndrew Morton pgdat = next_online_pgdat(pgdat)) 863*3c486871SAndrew Morton --node; 864*3c486871SAndrew Morton 865*3c486871SAndrew Morton return pgdat; 866*3c486871SAndrew Morton } 867*3c486871SAndrew Morton 868*3c486871SAndrew Morton static void *frag_next(struct seq_file *m, void *arg, loff_t *pos) 869*3c486871SAndrew Morton { 870*3c486871SAndrew Morton pg_data_t *pgdat = (pg_data_t *)arg; 871*3c486871SAndrew Morton 872*3c486871SAndrew Morton (*pos)++; 873*3c486871SAndrew Morton return next_online_pgdat(pgdat); 874*3c486871SAndrew Morton } 875*3c486871SAndrew Morton 876*3c486871SAndrew Morton static void frag_stop(struct seq_file *m, void *arg) 877*3c486871SAndrew Morton { 878*3c486871SAndrew Morton } 879*3c486871SAndrew Morton 880*3c486871SAndrew Morton /* Walk all the zones in a node and print using a callback */ 881*3c486871SAndrew Morton static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat, 882*3c486871SAndrew Morton void (*print)(struct seq_file *m, pg_data_t *, struct zone *)) 883*3c486871SAndrew Morton { 884*3c486871SAndrew Morton struct zone *zone; 885*3c486871SAndrew Morton struct zone *node_zones = pgdat->node_zones; 886*3c486871SAndrew Morton unsigned long flags; 887*3c486871SAndrew Morton 888*3c486871SAndrew Morton for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { 889*3c486871SAndrew Morton if (!populated_zone(zone)) 890*3c486871SAndrew Morton continue; 891*3c486871SAndrew Morton 892*3c486871SAndrew Morton spin_lock_irqsave(&zone->lock, flags); 893*3c486871SAndrew Morton print(m, pgdat, zone); 894*3c486871SAndrew Morton spin_unlock_irqrestore(&zone->lock, flags); 895*3c486871SAndrew Morton } 896*3c486871SAndrew Morton } 897*3c486871SAndrew Morton #endif 898*3c486871SAndrew Morton 899d7a5752cSMel Gorman #ifdef CONFIG_PROC_FS 900*3c486871SAndrew Morton static char * const migratetype_names[MIGRATE_TYPES] = { 901*3c486871SAndrew Morton "Unmovable", 902*3c486871SAndrew Morton "Reclaimable", 903*3c486871SAndrew Morton "Movable", 904*3c486871SAndrew Morton "Reserve", 905*3c486871SAndrew Morton #ifdef CONFIG_CMA 906*3c486871SAndrew Morton "CMA", 907*3c486871SAndrew Morton #endif 908*3c486871SAndrew Morton #ifdef CONFIG_MEMORY_ISOLATION 909*3c486871SAndrew Morton "Isolate", 910*3c486871SAndrew Morton #endif 911*3c486871SAndrew Morton }; 912*3c486871SAndrew Morton 913467c996cSMel Gorman static void frag_show_print(struct seq_file *m, pg_data_t *pgdat, 914467c996cSMel Gorman struct zone *zone) 915467c996cSMel Gorman { 916467c996cSMel Gorman int order; 917467c996cSMel Gorman 918f6ac2354SChristoph Lameter seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); 919f6ac2354SChristoph Lameter for (order = 0; order < MAX_ORDER; ++order) 920f6ac2354SChristoph Lameter seq_printf(m, "%6lu ", zone->free_area[order].nr_free); 921f6ac2354SChristoph Lameter seq_putc(m, '\n'); 922f6ac2354SChristoph Lameter } 923467c996cSMel Gorman 924467c996cSMel Gorman /* 925467c996cSMel Gorman * This walks the free areas for each zone. 926467c996cSMel Gorman */ 927467c996cSMel Gorman static int frag_show(struct seq_file *m, void *arg) 928467c996cSMel Gorman { 929467c996cSMel Gorman pg_data_t *pgdat = (pg_data_t *)arg; 930467c996cSMel Gorman walk_zones_in_node(m, pgdat, frag_show_print); 931467c996cSMel Gorman return 0; 932467c996cSMel Gorman } 933467c996cSMel Gorman 934467c996cSMel Gorman static void pagetypeinfo_showfree_print(struct seq_file *m, 935467c996cSMel Gorman pg_data_t *pgdat, struct zone *zone) 936467c996cSMel Gorman { 937467c996cSMel Gorman int order, mtype; 938467c996cSMel Gorman 939467c996cSMel Gorman for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) { 940467c996cSMel Gorman seq_printf(m, "Node %4d, zone %8s, type %12s ", 941467c996cSMel Gorman pgdat->node_id, 942467c996cSMel Gorman zone->name, 943467c996cSMel Gorman migratetype_names[mtype]); 944467c996cSMel Gorman for (order = 0; order < MAX_ORDER; ++order) { 945467c996cSMel Gorman unsigned long freecount = 0; 946467c996cSMel Gorman struct free_area *area; 947467c996cSMel Gorman struct list_head *curr; 948467c996cSMel Gorman 949467c996cSMel Gorman area = &(zone->free_area[order]); 950467c996cSMel Gorman 951467c996cSMel Gorman list_for_each(curr, &area->free_list[mtype]) 952467c996cSMel Gorman freecount++; 953467c996cSMel Gorman seq_printf(m, "%6lu ", freecount); 954467c996cSMel Gorman } 955467c996cSMel Gorman seq_putc(m, '\n'); 956467c996cSMel Gorman } 957467c996cSMel Gorman } 958467c996cSMel Gorman 959467c996cSMel Gorman /* Print out the free pages at each order for each migatetype */ 960467c996cSMel Gorman static int pagetypeinfo_showfree(struct seq_file *m, void *arg) 961467c996cSMel Gorman { 962467c996cSMel Gorman int order; 963467c996cSMel Gorman pg_data_t *pgdat = (pg_data_t *)arg; 964467c996cSMel Gorman 965467c996cSMel Gorman /* Print header */ 966467c996cSMel Gorman seq_printf(m, "%-43s ", "Free pages count per migrate type at order"); 967467c996cSMel Gorman for (order = 0; order < MAX_ORDER; ++order) 968467c996cSMel Gorman seq_printf(m, "%6d ", order); 969467c996cSMel Gorman seq_putc(m, '\n'); 970467c996cSMel Gorman 971467c996cSMel Gorman walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print); 972467c996cSMel Gorman 973467c996cSMel Gorman return 0; 974467c996cSMel Gorman } 975467c996cSMel Gorman 976467c996cSMel Gorman static void pagetypeinfo_showblockcount_print(struct seq_file *m, 977467c996cSMel Gorman pg_data_t *pgdat, struct zone *zone) 978467c996cSMel Gorman { 979467c996cSMel Gorman int mtype; 980467c996cSMel Gorman unsigned long pfn; 981467c996cSMel Gorman unsigned long start_pfn = zone->zone_start_pfn; 982108bcc96SCody P Schafer unsigned long end_pfn = zone_end_pfn(zone); 983467c996cSMel Gorman unsigned long count[MIGRATE_TYPES] = { 0, }; 984467c996cSMel Gorman 985467c996cSMel Gorman for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { 986467c996cSMel Gorman struct page *page; 987467c996cSMel Gorman 988467c996cSMel Gorman if (!pfn_valid(pfn)) 989467c996cSMel Gorman continue; 990467c996cSMel Gorman 991467c996cSMel Gorman page = pfn_to_page(pfn); 992eb33575cSMel Gorman 993eb33575cSMel Gorman /* Watch for unexpected holes punched in the memmap */ 994eb33575cSMel Gorman if (!memmap_valid_within(pfn, page, zone)) 995e80d6a24SMel Gorman continue; 996eb33575cSMel Gorman 997467c996cSMel Gorman mtype = get_pageblock_migratetype(page); 998467c996cSMel Gorman 999e80d6a24SMel Gorman if (mtype < MIGRATE_TYPES) 1000467c996cSMel Gorman count[mtype]++; 1001467c996cSMel Gorman } 1002467c996cSMel Gorman 1003467c996cSMel Gorman /* Print counts */ 1004467c996cSMel Gorman seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); 1005467c996cSMel Gorman for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) 1006467c996cSMel Gorman seq_printf(m, "%12lu ", count[mtype]); 1007467c996cSMel Gorman seq_putc(m, '\n'); 1008467c996cSMel Gorman } 1009467c996cSMel Gorman 1010467c996cSMel Gorman /* Print out the free pages at each order for each migratetype */ 1011467c996cSMel Gorman static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg) 1012467c996cSMel Gorman { 1013467c996cSMel Gorman int mtype; 1014467c996cSMel Gorman pg_data_t *pgdat = (pg_data_t *)arg; 1015467c996cSMel Gorman 1016467c996cSMel Gorman seq_printf(m, "\n%-23s", "Number of blocks type "); 1017467c996cSMel Gorman for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) 1018467c996cSMel Gorman seq_printf(m, "%12s ", migratetype_names[mtype]); 1019467c996cSMel Gorman seq_putc(m, '\n'); 1020467c996cSMel Gorman walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print); 1021467c996cSMel Gorman 1022467c996cSMel Gorman return 0; 1023467c996cSMel Gorman } 1024467c996cSMel Gorman 102548c96a36SJoonsoo Kim #ifdef CONFIG_PAGE_OWNER 102648c96a36SJoonsoo Kim static void pagetypeinfo_showmixedcount_print(struct seq_file *m, 102748c96a36SJoonsoo Kim pg_data_t *pgdat, 102848c96a36SJoonsoo Kim struct zone *zone) 102948c96a36SJoonsoo Kim { 103048c96a36SJoonsoo Kim struct page *page; 103148c96a36SJoonsoo Kim struct page_ext *page_ext; 103248c96a36SJoonsoo Kim unsigned long pfn = zone->zone_start_pfn, block_end_pfn; 103348c96a36SJoonsoo Kim unsigned long end_pfn = pfn + zone->spanned_pages; 103448c96a36SJoonsoo Kim unsigned long count[MIGRATE_TYPES] = { 0, }; 103548c96a36SJoonsoo Kim int pageblock_mt, page_mt; 103648c96a36SJoonsoo Kim int i; 103748c96a36SJoonsoo Kim 103848c96a36SJoonsoo Kim /* Scan block by block. First and last block may be incomplete */ 103948c96a36SJoonsoo Kim pfn = zone->zone_start_pfn; 104048c96a36SJoonsoo Kim 104148c96a36SJoonsoo Kim /* 104248c96a36SJoonsoo Kim * Walk the zone in pageblock_nr_pages steps. If a page block spans 104348c96a36SJoonsoo Kim * a zone boundary, it will be double counted between zones. This does 104448c96a36SJoonsoo Kim * not matter as the mixed block count will still be correct 104548c96a36SJoonsoo Kim */ 104648c96a36SJoonsoo Kim for (; pfn < end_pfn; ) { 104748c96a36SJoonsoo Kim if (!pfn_valid(pfn)) { 104848c96a36SJoonsoo Kim pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES); 104948c96a36SJoonsoo Kim continue; 105048c96a36SJoonsoo Kim } 105148c96a36SJoonsoo Kim 105248c96a36SJoonsoo Kim block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); 105348c96a36SJoonsoo Kim block_end_pfn = min(block_end_pfn, end_pfn); 105448c96a36SJoonsoo Kim 105548c96a36SJoonsoo Kim page = pfn_to_page(pfn); 105648c96a36SJoonsoo Kim pageblock_mt = get_pfnblock_migratetype(page, pfn); 105748c96a36SJoonsoo Kim 105848c96a36SJoonsoo Kim for (; pfn < block_end_pfn; pfn++) { 105948c96a36SJoonsoo Kim if (!pfn_valid_within(pfn)) 106048c96a36SJoonsoo Kim continue; 106148c96a36SJoonsoo Kim 106248c96a36SJoonsoo Kim page = pfn_to_page(pfn); 106348c96a36SJoonsoo Kim if (PageBuddy(page)) { 106448c96a36SJoonsoo Kim pfn += (1UL << page_order(page)) - 1; 106548c96a36SJoonsoo Kim continue; 106648c96a36SJoonsoo Kim } 106748c96a36SJoonsoo Kim 106848c96a36SJoonsoo Kim if (PageReserved(page)) 106948c96a36SJoonsoo Kim continue; 107048c96a36SJoonsoo Kim 107148c96a36SJoonsoo Kim page_ext = lookup_page_ext(page); 107248c96a36SJoonsoo Kim 107348c96a36SJoonsoo Kim if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) 107448c96a36SJoonsoo Kim continue; 107548c96a36SJoonsoo Kim 107648c96a36SJoonsoo Kim page_mt = gfpflags_to_migratetype(page_ext->gfp_mask); 107748c96a36SJoonsoo Kim if (pageblock_mt != page_mt) { 107848c96a36SJoonsoo Kim if (is_migrate_cma(pageblock_mt)) 107948c96a36SJoonsoo Kim count[MIGRATE_MOVABLE]++; 108048c96a36SJoonsoo Kim else 108148c96a36SJoonsoo Kim count[pageblock_mt]++; 108248c96a36SJoonsoo Kim 108348c96a36SJoonsoo Kim pfn = block_end_pfn; 108448c96a36SJoonsoo Kim break; 108548c96a36SJoonsoo Kim } 108648c96a36SJoonsoo Kim pfn += (1UL << page_ext->order) - 1; 108748c96a36SJoonsoo Kim } 108848c96a36SJoonsoo Kim } 108948c96a36SJoonsoo Kim 109048c96a36SJoonsoo Kim /* Print counts */ 109148c96a36SJoonsoo Kim seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); 109248c96a36SJoonsoo Kim for (i = 0; i < MIGRATE_TYPES; i++) 109348c96a36SJoonsoo Kim seq_printf(m, "%12lu ", count[i]); 109448c96a36SJoonsoo Kim seq_putc(m, '\n'); 109548c96a36SJoonsoo Kim } 109648c96a36SJoonsoo Kim #endif /* CONFIG_PAGE_OWNER */ 109748c96a36SJoonsoo Kim 109848c96a36SJoonsoo Kim /* 109948c96a36SJoonsoo Kim * Print out the number of pageblocks for each migratetype that contain pages 110048c96a36SJoonsoo Kim * of other types. This gives an indication of how well fallbacks are being 110148c96a36SJoonsoo Kim * contained by rmqueue_fallback(). It requires information from PAGE_OWNER 110248c96a36SJoonsoo Kim * to determine what is going on 110348c96a36SJoonsoo Kim */ 110448c96a36SJoonsoo Kim static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat) 110548c96a36SJoonsoo Kim { 110648c96a36SJoonsoo Kim #ifdef CONFIG_PAGE_OWNER 110748c96a36SJoonsoo Kim int mtype; 110848c96a36SJoonsoo Kim 110948c96a36SJoonsoo Kim if (!page_owner_inited) 111048c96a36SJoonsoo Kim return; 111148c96a36SJoonsoo Kim 111248c96a36SJoonsoo Kim drain_all_pages(NULL); 111348c96a36SJoonsoo Kim 111448c96a36SJoonsoo Kim seq_printf(m, "\n%-23s", "Number of mixed blocks "); 111548c96a36SJoonsoo Kim for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) 111648c96a36SJoonsoo Kim seq_printf(m, "%12s ", migratetype_names[mtype]); 111748c96a36SJoonsoo Kim seq_putc(m, '\n'); 111848c96a36SJoonsoo Kim 111948c96a36SJoonsoo Kim walk_zones_in_node(m, pgdat, pagetypeinfo_showmixedcount_print); 112048c96a36SJoonsoo Kim #endif /* CONFIG_PAGE_OWNER */ 112148c96a36SJoonsoo Kim } 112248c96a36SJoonsoo Kim 1123467c996cSMel Gorman /* 1124467c996cSMel Gorman * This prints out statistics in relation to grouping pages by mobility. 1125467c996cSMel Gorman * It is expensive to collect so do not constantly read the file. 1126467c996cSMel Gorman */ 1127467c996cSMel Gorman static int pagetypeinfo_show(struct seq_file *m, void *arg) 1128467c996cSMel Gorman { 1129467c996cSMel Gorman pg_data_t *pgdat = (pg_data_t *)arg; 1130467c996cSMel Gorman 113141b25a37SKOSAKI Motohiro /* check memoryless node */ 1132a47b53c5SLai Jiangshan if (!node_state(pgdat->node_id, N_MEMORY)) 113341b25a37SKOSAKI Motohiro return 0; 113441b25a37SKOSAKI Motohiro 1135467c996cSMel Gorman seq_printf(m, "Page block order: %d\n", pageblock_order); 1136467c996cSMel Gorman seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages); 1137467c996cSMel Gorman seq_putc(m, '\n'); 1138467c996cSMel Gorman pagetypeinfo_showfree(m, pgdat); 1139467c996cSMel Gorman pagetypeinfo_showblockcount(m, pgdat); 114048c96a36SJoonsoo Kim pagetypeinfo_showmixedcount(m, pgdat); 1141467c996cSMel Gorman 1142f6ac2354SChristoph Lameter return 0; 1143f6ac2354SChristoph Lameter } 1144f6ac2354SChristoph Lameter 11458f32f7e5SAlexey Dobriyan static const struct seq_operations fragmentation_op = { 1146f6ac2354SChristoph Lameter .start = frag_start, 1147f6ac2354SChristoph Lameter .next = frag_next, 1148f6ac2354SChristoph Lameter .stop = frag_stop, 1149f6ac2354SChristoph Lameter .show = frag_show, 1150f6ac2354SChristoph Lameter }; 1151f6ac2354SChristoph Lameter 11528f32f7e5SAlexey Dobriyan static int fragmentation_open(struct inode *inode, struct file *file) 11538f32f7e5SAlexey Dobriyan { 11548f32f7e5SAlexey Dobriyan return seq_open(file, &fragmentation_op); 11558f32f7e5SAlexey Dobriyan } 11568f32f7e5SAlexey Dobriyan 11578f32f7e5SAlexey Dobriyan static const struct file_operations fragmentation_file_operations = { 11588f32f7e5SAlexey Dobriyan .open = fragmentation_open, 11598f32f7e5SAlexey Dobriyan .read = seq_read, 11608f32f7e5SAlexey Dobriyan .llseek = seq_lseek, 11618f32f7e5SAlexey Dobriyan .release = seq_release, 11628f32f7e5SAlexey Dobriyan }; 11638f32f7e5SAlexey Dobriyan 116474e2e8e8SAlexey Dobriyan static const struct seq_operations pagetypeinfo_op = { 1165467c996cSMel Gorman .start = frag_start, 1166467c996cSMel Gorman .next = frag_next, 1167467c996cSMel Gorman .stop = frag_stop, 1168467c996cSMel Gorman .show = pagetypeinfo_show, 1169467c996cSMel Gorman }; 1170467c996cSMel Gorman 117174e2e8e8SAlexey Dobriyan static int pagetypeinfo_open(struct inode *inode, struct file *file) 117274e2e8e8SAlexey Dobriyan { 117374e2e8e8SAlexey Dobriyan return seq_open(file, &pagetypeinfo_op); 117474e2e8e8SAlexey Dobriyan } 117574e2e8e8SAlexey Dobriyan 117674e2e8e8SAlexey Dobriyan static const struct file_operations pagetypeinfo_file_ops = { 117774e2e8e8SAlexey Dobriyan .open = pagetypeinfo_open, 117874e2e8e8SAlexey Dobriyan .read = seq_read, 117974e2e8e8SAlexey Dobriyan .llseek = seq_lseek, 118074e2e8e8SAlexey Dobriyan .release = seq_release, 118174e2e8e8SAlexey Dobriyan }; 118274e2e8e8SAlexey Dobriyan 1183467c996cSMel Gorman static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat, 1184467c996cSMel Gorman struct zone *zone) 1185f6ac2354SChristoph Lameter { 1186f6ac2354SChristoph Lameter int i; 1187f6ac2354SChristoph Lameter seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name); 1188f6ac2354SChristoph Lameter seq_printf(m, 1189f6ac2354SChristoph Lameter "\n pages free %lu" 1190f6ac2354SChristoph Lameter "\n min %lu" 1191f6ac2354SChristoph Lameter "\n low %lu" 1192f6ac2354SChristoph Lameter "\n high %lu" 119308d9ae7cSWu Fengguang "\n scanned %lu" 1194f6ac2354SChristoph Lameter "\n spanned %lu" 11959feedc9dSJiang Liu "\n present %lu" 11969feedc9dSJiang Liu "\n managed %lu", 119788f5acf8SMel Gorman zone_page_state(zone, NR_FREE_PAGES), 119841858966SMel Gorman min_wmark_pages(zone), 119941858966SMel Gorman low_wmark_pages(zone), 120041858966SMel Gorman high_wmark_pages(zone), 12010d5d823aSMel Gorman zone_page_state(zone, NR_PAGES_SCANNED), 1202f6ac2354SChristoph Lameter zone->spanned_pages, 12039feedc9dSJiang Liu zone->present_pages, 12049feedc9dSJiang Liu zone->managed_pages); 12052244b95aSChristoph Lameter 12062244b95aSChristoph Lameter for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) 12072244b95aSChristoph Lameter seq_printf(m, "\n %-12s %lu", vmstat_text[i], 12082244b95aSChristoph Lameter zone_page_state(zone, i)); 12092244b95aSChristoph Lameter 1210f6ac2354SChristoph Lameter seq_printf(m, 12113484b2deSMel Gorman "\n protection: (%ld", 1212f6ac2354SChristoph Lameter zone->lowmem_reserve[0]); 1213f6ac2354SChristoph Lameter for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++) 12143484b2deSMel Gorman seq_printf(m, ", %ld", zone->lowmem_reserve[i]); 1215f6ac2354SChristoph Lameter seq_printf(m, 1216f6ac2354SChristoph Lameter ")" 1217f6ac2354SChristoph Lameter "\n pagesets"); 1218f6ac2354SChristoph Lameter for_each_online_cpu(i) { 1219f6ac2354SChristoph Lameter struct per_cpu_pageset *pageset; 1220f6ac2354SChristoph Lameter 122199dcc3e5SChristoph Lameter pageset = per_cpu_ptr(zone->pageset, i); 1222f6ac2354SChristoph Lameter seq_printf(m, 12233dfa5721SChristoph Lameter "\n cpu: %i" 1224f6ac2354SChristoph Lameter "\n count: %i" 1225f6ac2354SChristoph Lameter "\n high: %i" 1226f6ac2354SChristoph Lameter "\n batch: %i", 12273dfa5721SChristoph Lameter i, 12283dfa5721SChristoph Lameter pageset->pcp.count, 12293dfa5721SChristoph Lameter pageset->pcp.high, 12303dfa5721SChristoph Lameter pageset->pcp.batch); 1231df9ecabaSChristoph Lameter #ifdef CONFIG_SMP 1232df9ecabaSChristoph Lameter seq_printf(m, "\n vm stats threshold: %d", 1233df9ecabaSChristoph Lameter pageset->stat_threshold); 1234df9ecabaSChristoph Lameter #endif 1235f6ac2354SChristoph Lameter } 1236f6ac2354SChristoph Lameter seq_printf(m, 1237f6ac2354SChristoph Lameter "\n all_unreclaimable: %u" 1238556adecbSRik van Riel "\n start_pfn: %lu" 1239556adecbSRik van Riel "\n inactive_ratio: %u", 12406e543d57SLisa Du !zone_reclaimable(zone), 1241556adecbSRik van Riel zone->zone_start_pfn, 1242556adecbSRik van Riel zone->inactive_ratio); 1243f6ac2354SChristoph Lameter seq_putc(m, '\n'); 1244f6ac2354SChristoph Lameter } 1245467c996cSMel Gorman 1246467c996cSMel Gorman /* 1247467c996cSMel Gorman * Output information about zones in @pgdat. 1248467c996cSMel Gorman */ 1249467c996cSMel Gorman static int zoneinfo_show(struct seq_file *m, void *arg) 1250467c996cSMel Gorman { 1251467c996cSMel Gorman pg_data_t *pgdat = (pg_data_t *)arg; 1252467c996cSMel Gorman walk_zones_in_node(m, pgdat, zoneinfo_show_print); 1253f6ac2354SChristoph Lameter return 0; 1254f6ac2354SChristoph Lameter } 1255f6ac2354SChristoph Lameter 12565c9fe628SAlexey Dobriyan static const struct seq_operations zoneinfo_op = { 1257f6ac2354SChristoph Lameter .start = frag_start, /* iterate over all zones. The same as in 1258f6ac2354SChristoph Lameter * fragmentation. */ 1259f6ac2354SChristoph Lameter .next = frag_next, 1260f6ac2354SChristoph Lameter .stop = frag_stop, 1261f6ac2354SChristoph Lameter .show = zoneinfo_show, 1262f6ac2354SChristoph Lameter }; 1263f6ac2354SChristoph Lameter 12645c9fe628SAlexey Dobriyan static int zoneinfo_open(struct inode *inode, struct file *file) 12655c9fe628SAlexey Dobriyan { 12665c9fe628SAlexey Dobriyan return seq_open(file, &zoneinfo_op); 12675c9fe628SAlexey Dobriyan } 12685c9fe628SAlexey Dobriyan 12695c9fe628SAlexey Dobriyan static const struct file_operations proc_zoneinfo_file_operations = { 12705c9fe628SAlexey Dobriyan .open = zoneinfo_open, 12715c9fe628SAlexey Dobriyan .read = seq_read, 12725c9fe628SAlexey Dobriyan .llseek = seq_lseek, 12735c9fe628SAlexey Dobriyan .release = seq_release, 12745c9fe628SAlexey Dobriyan }; 12755c9fe628SAlexey Dobriyan 127679da826aSMichael Rubin enum writeback_stat_item { 127779da826aSMichael Rubin NR_DIRTY_THRESHOLD, 127879da826aSMichael Rubin NR_DIRTY_BG_THRESHOLD, 127979da826aSMichael Rubin NR_VM_WRITEBACK_STAT_ITEMS, 128079da826aSMichael Rubin }; 128179da826aSMichael Rubin 1282f6ac2354SChristoph Lameter static void *vmstat_start(struct seq_file *m, loff_t *pos) 1283f6ac2354SChristoph Lameter { 12842244b95aSChristoph Lameter unsigned long *v; 128579da826aSMichael Rubin int i, stat_items_size; 1286f6ac2354SChristoph Lameter 1287f6ac2354SChristoph Lameter if (*pos >= ARRAY_SIZE(vmstat_text)) 1288f6ac2354SChristoph Lameter return NULL; 128979da826aSMichael Rubin stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) + 129079da826aSMichael Rubin NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long); 1291f6ac2354SChristoph Lameter 1292f8891e5eSChristoph Lameter #ifdef CONFIG_VM_EVENT_COUNTERS 129379da826aSMichael Rubin stat_items_size += sizeof(struct vm_event_state); 1294f8891e5eSChristoph Lameter #endif 129579da826aSMichael Rubin 129679da826aSMichael Rubin v = kmalloc(stat_items_size, GFP_KERNEL); 12972244b95aSChristoph Lameter m->private = v; 12982244b95aSChristoph Lameter if (!v) 1299f6ac2354SChristoph Lameter return ERR_PTR(-ENOMEM); 13002244b95aSChristoph Lameter for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) 13012244b95aSChristoph Lameter v[i] = global_page_state(i); 130279da826aSMichael Rubin v += NR_VM_ZONE_STAT_ITEMS; 130379da826aSMichael Rubin 130479da826aSMichael Rubin global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD, 130579da826aSMichael Rubin v + NR_DIRTY_THRESHOLD); 130679da826aSMichael Rubin v += NR_VM_WRITEBACK_STAT_ITEMS; 130779da826aSMichael Rubin 1308f8891e5eSChristoph Lameter #ifdef CONFIG_VM_EVENT_COUNTERS 130979da826aSMichael Rubin all_vm_events(v); 131079da826aSMichael Rubin v[PGPGIN] /= 2; /* sectors -> kbytes */ 131179da826aSMichael Rubin v[PGPGOUT] /= 2; 1312f8891e5eSChristoph Lameter #endif 1313ff8b16d7SWu Fengguang return (unsigned long *)m->private + *pos; 1314f6ac2354SChristoph Lameter } 1315f6ac2354SChristoph Lameter 1316f6ac2354SChristoph Lameter static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos) 1317f6ac2354SChristoph Lameter { 1318f6ac2354SChristoph Lameter (*pos)++; 1319f6ac2354SChristoph Lameter if (*pos >= ARRAY_SIZE(vmstat_text)) 1320f6ac2354SChristoph Lameter return NULL; 1321f6ac2354SChristoph Lameter return (unsigned long *)m->private + *pos; 1322f6ac2354SChristoph Lameter } 1323f6ac2354SChristoph Lameter 1324f6ac2354SChristoph Lameter static int vmstat_show(struct seq_file *m, void *arg) 1325f6ac2354SChristoph Lameter { 1326f6ac2354SChristoph Lameter unsigned long *l = arg; 1327f6ac2354SChristoph Lameter unsigned long off = l - (unsigned long *)m->private; 1328f6ac2354SChristoph Lameter 1329f6ac2354SChristoph Lameter seq_printf(m, "%s %lu\n", vmstat_text[off], *l); 1330f6ac2354SChristoph Lameter return 0; 1331f6ac2354SChristoph Lameter } 1332f6ac2354SChristoph Lameter 1333f6ac2354SChristoph Lameter static void vmstat_stop(struct seq_file *m, void *arg) 1334f6ac2354SChristoph Lameter { 1335f6ac2354SChristoph Lameter kfree(m->private); 1336f6ac2354SChristoph Lameter m->private = NULL; 1337f6ac2354SChristoph Lameter } 1338f6ac2354SChristoph Lameter 1339b6aa44abSAlexey Dobriyan static const struct seq_operations vmstat_op = { 1340f6ac2354SChristoph Lameter .start = vmstat_start, 1341f6ac2354SChristoph Lameter .next = vmstat_next, 1342f6ac2354SChristoph Lameter .stop = vmstat_stop, 1343f6ac2354SChristoph Lameter .show = vmstat_show, 1344f6ac2354SChristoph Lameter }; 1345f6ac2354SChristoph Lameter 1346b6aa44abSAlexey Dobriyan static int vmstat_open(struct inode *inode, struct file *file) 1347b6aa44abSAlexey Dobriyan { 1348b6aa44abSAlexey Dobriyan return seq_open(file, &vmstat_op); 1349b6aa44abSAlexey Dobriyan } 1350b6aa44abSAlexey Dobriyan 1351b6aa44abSAlexey Dobriyan static const struct file_operations proc_vmstat_file_operations = { 1352b6aa44abSAlexey Dobriyan .open = vmstat_open, 1353b6aa44abSAlexey Dobriyan .read = seq_read, 1354b6aa44abSAlexey Dobriyan .llseek = seq_lseek, 1355b6aa44abSAlexey Dobriyan .release = seq_release, 1356b6aa44abSAlexey Dobriyan }; 1357f6ac2354SChristoph Lameter #endif /* CONFIG_PROC_FS */ 1358f6ac2354SChristoph Lameter 1359df9ecabaSChristoph Lameter #ifdef CONFIG_SMP 1360d1187ed2SChristoph Lameter static DEFINE_PER_CPU(struct delayed_work, vmstat_work); 136177461ab3SChristoph Lameter int sysctl_stat_interval __read_mostly = HZ; 13627cc36bbdSChristoph Lameter static cpumask_var_t cpu_stat_off; 1363d1187ed2SChristoph Lameter 1364d1187ed2SChristoph Lameter static void vmstat_update(struct work_struct *w) 1365d1187ed2SChristoph Lameter { 13667cc36bbdSChristoph Lameter if (refresh_cpu_vm_stats()) 13677cc36bbdSChristoph Lameter /* 13687cc36bbdSChristoph Lameter * Counters were updated so we expect more updates 13697cc36bbdSChristoph Lameter * to occur in the future. Keep on running the 13707cc36bbdSChristoph Lameter * update worker thread. 13717cc36bbdSChristoph Lameter */ 13727c8e0181SChristoph Lameter schedule_delayed_work(this_cpu_ptr(&vmstat_work), 137398f4ebb2SAnton Blanchard round_jiffies_relative(sysctl_stat_interval)); 13747cc36bbdSChristoph Lameter else { 13757cc36bbdSChristoph Lameter /* 13767cc36bbdSChristoph Lameter * We did not update any counters so the app may be in 13777cc36bbdSChristoph Lameter * a mode where it does not cause counter updates. 13787cc36bbdSChristoph Lameter * We may be uselessly running vmstat_update. 13797cc36bbdSChristoph Lameter * Defer the checking for differentials to the 13807cc36bbdSChristoph Lameter * shepherd thread on a different processor. 13817cc36bbdSChristoph Lameter */ 13827cc36bbdSChristoph Lameter int r; 13837cc36bbdSChristoph Lameter /* 13847cc36bbdSChristoph Lameter * Shepherd work thread does not race since it never 13857cc36bbdSChristoph Lameter * changes the bit if its zero but the cpu 13867cc36bbdSChristoph Lameter * online / off line code may race if 13877cc36bbdSChristoph Lameter * worker threads are still allowed during 13887cc36bbdSChristoph Lameter * shutdown / startup. 13897cc36bbdSChristoph Lameter */ 13907cc36bbdSChristoph Lameter r = cpumask_test_and_set_cpu(smp_processor_id(), 13917cc36bbdSChristoph Lameter cpu_stat_off); 13927cc36bbdSChristoph Lameter VM_BUG_ON(r); 13937cc36bbdSChristoph Lameter } 1394d1187ed2SChristoph Lameter } 1395d1187ed2SChristoph Lameter 13967cc36bbdSChristoph Lameter /* 13977cc36bbdSChristoph Lameter * Check if the diffs for a certain cpu indicate that 13987cc36bbdSChristoph Lameter * an update is needed. 13997cc36bbdSChristoph Lameter */ 14007cc36bbdSChristoph Lameter static bool need_update(int cpu) 1401d1187ed2SChristoph Lameter { 14027cc36bbdSChristoph Lameter struct zone *zone; 1403d1187ed2SChristoph Lameter 14047cc36bbdSChristoph Lameter for_each_populated_zone(zone) { 14057cc36bbdSChristoph Lameter struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu); 14067cc36bbdSChristoph Lameter 14077cc36bbdSChristoph Lameter BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1); 14087cc36bbdSChristoph Lameter /* 14097cc36bbdSChristoph Lameter * The fast way of checking if there are any vmstat diffs. 14107cc36bbdSChristoph Lameter * This works because the diffs are byte sized items. 14117cc36bbdSChristoph Lameter */ 14127cc36bbdSChristoph Lameter if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS)) 14137cc36bbdSChristoph Lameter return true; 14147cc36bbdSChristoph Lameter 14157cc36bbdSChristoph Lameter } 14167cc36bbdSChristoph Lameter return false; 14177cc36bbdSChristoph Lameter } 14187cc36bbdSChristoph Lameter 14197cc36bbdSChristoph Lameter 14207cc36bbdSChristoph Lameter /* 14217cc36bbdSChristoph Lameter * Shepherd worker thread that checks the 14227cc36bbdSChristoph Lameter * differentials of processors that have their worker 14237cc36bbdSChristoph Lameter * threads for vm statistics updates disabled because of 14247cc36bbdSChristoph Lameter * inactivity. 14257cc36bbdSChristoph Lameter */ 14267cc36bbdSChristoph Lameter static void vmstat_shepherd(struct work_struct *w); 14277cc36bbdSChristoph Lameter 14287cc36bbdSChristoph Lameter static DECLARE_DELAYED_WORK(shepherd, vmstat_shepherd); 14297cc36bbdSChristoph Lameter 14307cc36bbdSChristoph Lameter static void vmstat_shepherd(struct work_struct *w) 14317cc36bbdSChristoph Lameter { 14327cc36bbdSChristoph Lameter int cpu; 14337cc36bbdSChristoph Lameter 14347cc36bbdSChristoph Lameter get_online_cpus(); 14357cc36bbdSChristoph Lameter /* Check processors whose vmstat worker threads have been disabled */ 14367cc36bbdSChristoph Lameter for_each_cpu(cpu, cpu_stat_off) 14377cc36bbdSChristoph Lameter if (need_update(cpu) && 14387cc36bbdSChristoph Lameter cpumask_test_and_clear_cpu(cpu, cpu_stat_off)) 14397cc36bbdSChristoph Lameter 14407cc36bbdSChristoph Lameter schedule_delayed_work_on(cpu, &per_cpu(vmstat_work, cpu), 14417cc36bbdSChristoph Lameter __round_jiffies_relative(sysctl_stat_interval, cpu)); 14427cc36bbdSChristoph Lameter 14437cc36bbdSChristoph Lameter put_online_cpus(); 14447cc36bbdSChristoph Lameter 14457cc36bbdSChristoph Lameter schedule_delayed_work(&shepherd, 14467cc36bbdSChristoph Lameter round_jiffies_relative(sysctl_stat_interval)); 14477cc36bbdSChristoph Lameter 14487cc36bbdSChristoph Lameter } 14497cc36bbdSChristoph Lameter 14507cc36bbdSChristoph Lameter static void __init start_shepherd_timer(void) 14517cc36bbdSChristoph Lameter { 14527cc36bbdSChristoph Lameter int cpu; 14537cc36bbdSChristoph Lameter 14547cc36bbdSChristoph Lameter for_each_possible_cpu(cpu) 14557cc36bbdSChristoph Lameter INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu), 14567cc36bbdSChristoph Lameter vmstat_update); 14577cc36bbdSChristoph Lameter 14587cc36bbdSChristoph Lameter if (!alloc_cpumask_var(&cpu_stat_off, GFP_KERNEL)) 14597cc36bbdSChristoph Lameter BUG(); 14607cc36bbdSChristoph Lameter cpumask_copy(cpu_stat_off, cpu_online_mask); 14617cc36bbdSChristoph Lameter 14627cc36bbdSChristoph Lameter schedule_delayed_work(&shepherd, 14637cc36bbdSChristoph Lameter round_jiffies_relative(sysctl_stat_interval)); 1464d1187ed2SChristoph Lameter } 1465d1187ed2SChristoph Lameter 1466807a1bd2SToshi Kani static void vmstat_cpu_dead(int node) 1467807a1bd2SToshi Kani { 1468807a1bd2SToshi Kani int cpu; 1469807a1bd2SToshi Kani 1470807a1bd2SToshi Kani get_online_cpus(); 1471807a1bd2SToshi Kani for_each_online_cpu(cpu) 1472807a1bd2SToshi Kani if (cpu_to_node(cpu) == node) 1473807a1bd2SToshi Kani goto end; 1474807a1bd2SToshi Kani 1475807a1bd2SToshi Kani node_clear_state(node, N_CPU); 1476807a1bd2SToshi Kani end: 1477807a1bd2SToshi Kani put_online_cpus(); 1478807a1bd2SToshi Kani } 1479807a1bd2SToshi Kani 1480df9ecabaSChristoph Lameter /* 1481df9ecabaSChristoph Lameter * Use the cpu notifier to insure that the thresholds are recalculated 1482df9ecabaSChristoph Lameter * when necessary. 1483df9ecabaSChristoph Lameter */ 14840db0628dSPaul Gortmaker static int vmstat_cpuup_callback(struct notifier_block *nfb, 1485df9ecabaSChristoph Lameter unsigned long action, 1486df9ecabaSChristoph Lameter void *hcpu) 1487df9ecabaSChristoph Lameter { 1488d1187ed2SChristoph Lameter long cpu = (long)hcpu; 1489d1187ed2SChristoph Lameter 1490df9ecabaSChristoph Lameter switch (action) { 1491d1187ed2SChristoph Lameter case CPU_ONLINE: 1492d1187ed2SChristoph Lameter case CPU_ONLINE_FROZEN: 14935ee28a44SKAMEZAWA Hiroyuki refresh_zone_stat_thresholds(); 1494ad596925SChristoph Lameter node_set_state(cpu_to_node(cpu), N_CPU); 14957cc36bbdSChristoph Lameter cpumask_set_cpu(cpu, cpu_stat_off); 1496d1187ed2SChristoph Lameter break; 1497d1187ed2SChristoph Lameter case CPU_DOWN_PREPARE: 1498d1187ed2SChristoph Lameter case CPU_DOWN_PREPARE_FROZEN: 1499afe2c511STejun Heo cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu)); 15007cc36bbdSChristoph Lameter cpumask_clear_cpu(cpu, cpu_stat_off); 1501d1187ed2SChristoph Lameter break; 1502d1187ed2SChristoph Lameter case CPU_DOWN_FAILED: 1503d1187ed2SChristoph Lameter case CPU_DOWN_FAILED_FROZEN: 15047cc36bbdSChristoph Lameter cpumask_set_cpu(cpu, cpu_stat_off); 1505d1187ed2SChristoph Lameter break; 1506df9ecabaSChristoph Lameter case CPU_DEAD: 15078bb78442SRafael J. Wysocki case CPU_DEAD_FROZEN: 1508df9ecabaSChristoph Lameter refresh_zone_stat_thresholds(); 1509807a1bd2SToshi Kani vmstat_cpu_dead(cpu_to_node(cpu)); 1510df9ecabaSChristoph Lameter break; 1511df9ecabaSChristoph Lameter default: 1512df9ecabaSChristoph Lameter break; 1513df9ecabaSChristoph Lameter } 1514df9ecabaSChristoph Lameter return NOTIFY_OK; 1515df9ecabaSChristoph Lameter } 1516df9ecabaSChristoph Lameter 15170db0628dSPaul Gortmaker static struct notifier_block vmstat_notifier = 1518df9ecabaSChristoph Lameter { &vmstat_cpuup_callback, NULL, 0 }; 15198f32f7e5SAlexey Dobriyan #endif 1520df9ecabaSChristoph Lameter 1521e2fc88d0SAdrian Bunk static int __init setup_vmstat(void) 1522df9ecabaSChristoph Lameter { 15238f32f7e5SAlexey Dobriyan #ifdef CONFIG_SMP 15240be94badSSrivatsa S. Bhat cpu_notifier_register_begin(); 15250be94badSSrivatsa S. Bhat __register_cpu_notifier(&vmstat_notifier); 1526d1187ed2SChristoph Lameter 15277cc36bbdSChristoph Lameter start_shepherd_timer(); 15280be94badSSrivatsa S. Bhat cpu_notifier_register_done(); 15298f32f7e5SAlexey Dobriyan #endif 15308f32f7e5SAlexey Dobriyan #ifdef CONFIG_PROC_FS 15318f32f7e5SAlexey Dobriyan proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations); 153274e2e8e8SAlexey Dobriyan proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops); 1533b6aa44abSAlexey Dobriyan proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations); 15345c9fe628SAlexey Dobriyan proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations); 15358f32f7e5SAlexey Dobriyan #endif 1536df9ecabaSChristoph Lameter return 0; 1537df9ecabaSChristoph Lameter } 1538df9ecabaSChristoph Lameter module_init(setup_vmstat) 1539d7a5752cSMel Gorman 1540d7a5752cSMel Gorman #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION) 1541d7a5752cSMel Gorman 1542d7a5752cSMel Gorman /* 1543d7a5752cSMel Gorman * Return an index indicating how much of the available free memory is 1544d7a5752cSMel Gorman * unusable for an allocation of the requested size. 1545d7a5752cSMel Gorman */ 1546d7a5752cSMel Gorman static int unusable_free_index(unsigned int order, 1547d7a5752cSMel Gorman struct contig_page_info *info) 1548d7a5752cSMel Gorman { 1549d7a5752cSMel Gorman /* No free memory is interpreted as all free memory is unusable */ 1550d7a5752cSMel Gorman if (info->free_pages == 0) 1551d7a5752cSMel Gorman return 1000; 1552d7a5752cSMel Gorman 1553d7a5752cSMel Gorman /* 1554d7a5752cSMel Gorman * Index should be a value between 0 and 1. Return a value to 3 1555d7a5752cSMel Gorman * decimal places. 1556d7a5752cSMel Gorman * 1557d7a5752cSMel Gorman * 0 => no fragmentation 1558d7a5752cSMel Gorman * 1 => high fragmentation 1559d7a5752cSMel Gorman */ 1560d7a5752cSMel Gorman return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages); 1561d7a5752cSMel Gorman 1562d7a5752cSMel Gorman } 1563d7a5752cSMel Gorman 1564d7a5752cSMel Gorman static void unusable_show_print(struct seq_file *m, 1565d7a5752cSMel Gorman pg_data_t *pgdat, struct zone *zone) 1566d7a5752cSMel Gorman { 1567d7a5752cSMel Gorman unsigned int order; 1568d7a5752cSMel Gorman int index; 1569d7a5752cSMel Gorman struct contig_page_info info; 1570d7a5752cSMel Gorman 1571d7a5752cSMel Gorman seq_printf(m, "Node %d, zone %8s ", 1572d7a5752cSMel Gorman pgdat->node_id, 1573d7a5752cSMel Gorman zone->name); 1574d7a5752cSMel Gorman for (order = 0; order < MAX_ORDER; ++order) { 1575d7a5752cSMel Gorman fill_contig_page_info(zone, order, &info); 1576d7a5752cSMel Gorman index = unusable_free_index(order, &info); 1577d7a5752cSMel Gorman seq_printf(m, "%d.%03d ", index / 1000, index % 1000); 1578d7a5752cSMel Gorman } 1579d7a5752cSMel Gorman 1580d7a5752cSMel Gorman seq_putc(m, '\n'); 1581d7a5752cSMel Gorman } 1582d7a5752cSMel Gorman 1583d7a5752cSMel Gorman /* 1584d7a5752cSMel Gorman * Display unusable free space index 1585d7a5752cSMel Gorman * 1586d7a5752cSMel Gorman * The unusable free space index measures how much of the available free 1587d7a5752cSMel Gorman * memory cannot be used to satisfy an allocation of a given size and is a 1588d7a5752cSMel Gorman * value between 0 and 1. The higher the value, the more of free memory is 1589d7a5752cSMel Gorman * unusable and by implication, the worse the external fragmentation is. This 1590d7a5752cSMel Gorman * can be expressed as a percentage by multiplying by 100. 1591d7a5752cSMel Gorman */ 1592d7a5752cSMel Gorman static int unusable_show(struct seq_file *m, void *arg) 1593d7a5752cSMel Gorman { 1594d7a5752cSMel Gorman pg_data_t *pgdat = (pg_data_t *)arg; 1595d7a5752cSMel Gorman 1596d7a5752cSMel Gorman /* check memoryless node */ 1597a47b53c5SLai Jiangshan if (!node_state(pgdat->node_id, N_MEMORY)) 1598d7a5752cSMel Gorman return 0; 1599d7a5752cSMel Gorman 1600d7a5752cSMel Gorman walk_zones_in_node(m, pgdat, unusable_show_print); 1601d7a5752cSMel Gorman 1602d7a5752cSMel Gorman return 0; 1603d7a5752cSMel Gorman } 1604d7a5752cSMel Gorman 1605d7a5752cSMel Gorman static const struct seq_operations unusable_op = { 1606d7a5752cSMel Gorman .start = frag_start, 1607d7a5752cSMel Gorman .next = frag_next, 1608d7a5752cSMel Gorman .stop = frag_stop, 1609d7a5752cSMel Gorman .show = unusable_show, 1610d7a5752cSMel Gorman }; 1611d7a5752cSMel Gorman 1612d7a5752cSMel Gorman static int unusable_open(struct inode *inode, struct file *file) 1613d7a5752cSMel Gorman { 1614d7a5752cSMel Gorman return seq_open(file, &unusable_op); 1615d7a5752cSMel Gorman } 1616d7a5752cSMel Gorman 1617d7a5752cSMel Gorman static const struct file_operations unusable_file_ops = { 1618d7a5752cSMel Gorman .open = unusable_open, 1619d7a5752cSMel Gorman .read = seq_read, 1620d7a5752cSMel Gorman .llseek = seq_lseek, 1621d7a5752cSMel Gorman .release = seq_release, 1622d7a5752cSMel Gorman }; 1623d7a5752cSMel Gorman 1624f1a5ab12SMel Gorman static void extfrag_show_print(struct seq_file *m, 1625f1a5ab12SMel Gorman pg_data_t *pgdat, struct zone *zone) 1626f1a5ab12SMel Gorman { 1627f1a5ab12SMel Gorman unsigned int order; 1628f1a5ab12SMel Gorman int index; 1629f1a5ab12SMel Gorman 1630f1a5ab12SMel Gorman /* Alloc on stack as interrupts are disabled for zone walk */ 1631f1a5ab12SMel Gorman struct contig_page_info info; 1632f1a5ab12SMel Gorman 1633f1a5ab12SMel Gorman seq_printf(m, "Node %d, zone %8s ", 1634f1a5ab12SMel Gorman pgdat->node_id, 1635f1a5ab12SMel Gorman zone->name); 1636f1a5ab12SMel Gorman for (order = 0; order < MAX_ORDER; ++order) { 1637f1a5ab12SMel Gorman fill_contig_page_info(zone, order, &info); 163856de7263SMel Gorman index = __fragmentation_index(order, &info); 1639f1a5ab12SMel Gorman seq_printf(m, "%d.%03d ", index / 1000, index % 1000); 1640f1a5ab12SMel Gorman } 1641f1a5ab12SMel Gorman 1642f1a5ab12SMel Gorman seq_putc(m, '\n'); 1643f1a5ab12SMel Gorman } 1644f1a5ab12SMel Gorman 1645f1a5ab12SMel Gorman /* 1646f1a5ab12SMel Gorman * Display fragmentation index for orders that allocations would fail for 1647f1a5ab12SMel Gorman */ 1648f1a5ab12SMel Gorman static int extfrag_show(struct seq_file *m, void *arg) 1649f1a5ab12SMel Gorman { 1650f1a5ab12SMel Gorman pg_data_t *pgdat = (pg_data_t *)arg; 1651f1a5ab12SMel Gorman 1652f1a5ab12SMel Gorman walk_zones_in_node(m, pgdat, extfrag_show_print); 1653f1a5ab12SMel Gorman 1654f1a5ab12SMel Gorman return 0; 1655f1a5ab12SMel Gorman } 1656f1a5ab12SMel Gorman 1657f1a5ab12SMel Gorman static const struct seq_operations extfrag_op = { 1658f1a5ab12SMel Gorman .start = frag_start, 1659f1a5ab12SMel Gorman .next = frag_next, 1660f1a5ab12SMel Gorman .stop = frag_stop, 1661f1a5ab12SMel Gorman .show = extfrag_show, 1662f1a5ab12SMel Gorman }; 1663f1a5ab12SMel Gorman 1664f1a5ab12SMel Gorman static int extfrag_open(struct inode *inode, struct file *file) 1665f1a5ab12SMel Gorman { 1666f1a5ab12SMel Gorman return seq_open(file, &extfrag_op); 1667f1a5ab12SMel Gorman } 1668f1a5ab12SMel Gorman 1669f1a5ab12SMel Gorman static const struct file_operations extfrag_file_ops = { 1670f1a5ab12SMel Gorman .open = extfrag_open, 1671f1a5ab12SMel Gorman .read = seq_read, 1672f1a5ab12SMel Gorman .llseek = seq_lseek, 1673f1a5ab12SMel Gorman .release = seq_release, 1674f1a5ab12SMel Gorman }; 1675f1a5ab12SMel Gorman 1676d7a5752cSMel Gorman static int __init extfrag_debug_init(void) 1677d7a5752cSMel Gorman { 1678bde8bd8aSSasikantha babu struct dentry *extfrag_debug_root; 1679bde8bd8aSSasikantha babu 1680d7a5752cSMel Gorman extfrag_debug_root = debugfs_create_dir("extfrag", NULL); 1681d7a5752cSMel Gorman if (!extfrag_debug_root) 1682d7a5752cSMel Gorman return -ENOMEM; 1683d7a5752cSMel Gorman 1684d7a5752cSMel Gorman if (!debugfs_create_file("unusable_index", 0444, 1685d7a5752cSMel Gorman extfrag_debug_root, NULL, &unusable_file_ops)) 1686bde8bd8aSSasikantha babu goto fail; 1687d7a5752cSMel Gorman 1688f1a5ab12SMel Gorman if (!debugfs_create_file("extfrag_index", 0444, 1689f1a5ab12SMel Gorman extfrag_debug_root, NULL, &extfrag_file_ops)) 1690bde8bd8aSSasikantha babu goto fail; 1691f1a5ab12SMel Gorman 1692d7a5752cSMel Gorman return 0; 1693bde8bd8aSSasikantha babu fail: 1694bde8bd8aSSasikantha babu debugfs_remove_recursive(extfrag_debug_root); 1695bde8bd8aSSasikantha babu return -ENOMEM; 1696d7a5752cSMel Gorman } 1697d7a5752cSMel Gorman 1698d7a5752cSMel Gorman module_init(extfrag_debug_init); 1699d7a5752cSMel Gorman #endif 1700