include/linux/vmstat.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _LINUX_VMSTAT_H
   3 #define _LINUX_VMSTAT_H
   4
   5 #include <linux/types.h>
   6 #include <linux/percpu.h>
   7 #include <linux/mmzone.h>
   8 #include <linux/vm_event_item.h>
   9 #include <linux/atomic.h>
  10 #include <linux/static_key.h>
  11
  12 extern int sysctl_stat_interval;
  13
  14 #ifdef CONFIG_NUMA
  15 #define ENABLE_NUMA_STAT   1
  16 #define DISABLE_NUMA_STAT   0
  17 extern int sysctl_vm_numa_stat;
  18 DECLARE_STATIC_KEY_TRUE(vm_numa_stat_key);
  19 extern int sysctl_vm_numa_stat_handler(struct ctl_table *table,
  20                 int write, void __user *buffer, size_t *length, loff_t *ppos);
  21 #endif
  22
  23 struct reclaim_stat {
  24         unsigned nr_dirty;
  25         unsigned nr_unqueued_dirty;
  26         unsigned nr_congested;
  27         unsigned nr_writeback;
  28         unsigned nr_immediate;
  29         unsigned nr_activate;
  30         unsigned nr_ref_keep;
  31         unsigned nr_unmap_fail;
  32 };
  33
  34 #ifdef CONFIG_VM_EVENT_COUNTERS
  35 /*
  36  * Light weight per cpu counter implementation.
  37  *
  38  * Counters should only be incremented and no critical kernel component
  39  * should rely on the counter values.
  40  *
  41  * Counters are handled completely inline. On many platforms the code
  42  * generated will simply be the increment of a global address.
  43  */
  44
  45 struct vm_event_state {
  46         unsigned long event[NR_VM_EVENT_ITEMS];
  47 };
  48
  49 DECLARE_PER_CPU(struct vm_event_state, vm_event_states);
  50
  51 /*
  52  * vm counters are allowed to be racy. Use raw_cpu_ops to avoid the
  53  * local_irq_disable overhead.
  54  */
  55 static inline void __count_vm_event(enum vm_event_item item)
  56 {
  57         raw_cpu_inc(vm_event_states.event[item]);
  58 }
  59
  60 static inline void count_vm_event(enum vm_event_item item)
  61 {
  62         this_cpu_inc(vm_event_states.event[item]);
  63 }
  64
  65 static inline void __count_vm_events(enum vm_event_item item, long delta)
  66 {
  67         raw_cpu_add(vm_event_states.event[item], delta);
  68 }
  69
  70 static inline void count_vm_events(enum vm_event_item item, long delta)
  71 {
  72         this_cpu_add(vm_event_states.event[item], delta);
  73 }
  74
  75 extern void all_vm_events(unsigned long *);
  76
  77 extern void vm_events_fold_cpu(int cpu);
  78
  79 #else
  80
  81 /* Disable counters */
  82 static inline void count_vm_event(enum vm_event_item item)
  83 {
  84 }
  85 static inline void count_vm_events(enum vm_event_item item, long delta)
  86 {
  87 }
  88 static inline void __count_vm_event(enum vm_event_item item)
  89 {
  90 }
  91 static inline void __count_vm_events(enum vm_event_item item, long delta)
  92 {
  93 }
  94 static inline void all_vm_events(unsigned long *ret)
  95 {
  96 }
  97 static inline void vm_events_fold_cpu(int cpu)
  98 {
  99 }
 100
 101 #endif /* CONFIG_VM_EVENT_COUNTERS */
 102
 103 #ifdef CONFIG_NUMA_BALANCING
 104 #define count_vm_numa_event(x)     count_vm_event(x)
 105 #define count_vm_numa_events(x, y) count_vm_events(x, y)
 106 #else
 107 #define count_vm_numa_event(x) do {} while (0)
 108 #define count_vm_numa_events(x, y) do { (void)(y); } while (0)
 109 #endif /* CONFIG_NUMA_BALANCING */
 110
 111 #ifdef CONFIG_DEBUG_TLBFLUSH
 112 #define count_vm_tlb_event(x)      count_vm_event(x)
 113 #define count_vm_tlb_events(x, y)  count_vm_events(x, y)
 114 #else
 115 #define count_vm_tlb_event(x)     do {} while (0)
 116 #define count_vm_tlb_events(x, y) do { (void)(y); } while (0)
 117 #endif
 118
 119 #ifdef CONFIG_DEBUG_VM_VMACACHE
 120 #define count_vm_vmacache_event(x) count_vm_event(x)
 121 #else
 122 #define count_vm_vmacache_event(x) do {} while (0)
 123 #endif
 124
 125 #define __count_zid_vm_events(item, zid, delta) \
 126         __count_vm_events(item##_NORMAL - ZONE_NORMAL + zid, delta)
 127
 128 /*
 129  * Zone and node-based page accounting with per cpu differentials.
 130  */
 131 extern atomic_long_t vm_zone_stat[NR_VM_ZONE_STAT_ITEMS];
 132 extern atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
 133 extern atomic_long_t vm_node_stat[NR_VM_NODE_STAT_ITEMS];
 134
 135 #ifdef CONFIG_NUMA
 136 static inline void zone_numa_state_add(long x, struct zone *zone,
 137                                  enum numa_stat_item item)
 138 {
 139         atomic_long_add(x, &zone->vm_numa_stat[item]);
 140         atomic_long_add(x, &vm_numa_stat[item]);
 141 }
 142
 143 static inline unsigned long global_numa_state(enum numa_stat_item item)
 144 {
 145         long x = atomic_long_read(&vm_numa_stat[item]);
 146
 147         return x;
 148 }
 149
 150 static inline unsigned long zone_numa_state_snapshot(struct zone *zone,
 151                                         enum numa_stat_item item)
 152 {
 153         long x = atomic_long_read(&zone->vm_numa_stat[item]);
 154         int cpu;
 155
 156         for_each_online_cpu(cpu)
 157                 x += per_cpu_ptr(zone->pageset, cpu)->vm_numa_stat_diff[item];
 158
 159         return x;
 160 }
 161 #endif /* CONFIG_NUMA */
 162
 163 static inline void zone_page_state_add(long x, struct zone *zone,
 164                                  enum zone_stat_item item)
 165 {
 166         atomic_long_add(x, &zone->vm_stat[item]);
 167         atomic_long_add(x, &vm_zone_stat[item]);
 168 }
 169
 170 static inline void node_page_state_add(long x, struct pglist_data *pgdat,
 171                                  enum node_stat_item item)
 172 {
 173         atomic_long_add(x, &pgdat->vm_stat[item]);
 174         atomic_long_add(x, &vm_node_stat[item]);
 175 }
 176
 177 static inline unsigned long global_zone_page_state(enum zone_stat_item item)
 178 {
 179         long x = atomic_long_read(&vm_zone_stat[item]);
 180 #ifdef CONFIG_SMP
 181         if (x < 0)
 182                 x = 0;
 183 #endif
 184         return x;
 185 }
 186
 187 static inline unsigned long global_node_page_state(enum node_stat_item item)
 188 {
 189         long x = atomic_long_read(&vm_node_stat[item]);
 190 #ifdef CONFIG_SMP
 191         if (x < 0)
 192                 x = 0;
 193 #endif
 194         return x;
 195 }
 196
 197 static inline unsigned long zone_page_state(struct zone *zone,
 198                                         enum zone_stat_item item)
 199 {
 200         long x = atomic_long_read(&zone->vm_stat[item]);
 201 #ifdef CONFIG_SMP
 202         if (x < 0)
 203                 x = 0;
 204 #endif
 205         return x;
 206 }
 207
 208 /*
 209  * More accurate version that also considers the currently pending
 210  * deltas. For that we need to loop over all cpus to find the current
 211  * deltas. There is no synchronization so the result cannot be
 212  * exactly accurate either.
 213  */
 214 static inline unsigned long zone_page_state_snapshot(struct zone *zone,
 215                                         enum zone_stat_item item)
 216 {
 217         long x = atomic_long_read(&zone->vm_stat[item]);
 218
 219 #ifdef CONFIG_SMP
 220         int cpu;
 221         for_each_online_cpu(cpu)
 222                 x += per_cpu_ptr(zone->pageset, cpu)->vm_stat_diff[item];
 223
 224         if (x < 0)
 225                 x = 0;
 226 #endif
 227         return x;
 228 }
 229
 230 #ifdef CONFIG_NUMA
 231 extern void __inc_numa_state(struct zone *zone, enum numa_stat_item item);
 232 extern unsigned long sum_zone_node_page_state(int node,
 233                                               enum zone_stat_item item);
 234 extern unsigned long sum_zone_numa_state(int node, enum numa_stat_item item);
 235 extern unsigned long node_page_state(struct pglist_data *pgdat,
 236                                                 enum node_stat_item item);
 237 #else
 238 #define sum_zone_node_page_state(node, item) global_zone_page_state(item)
 239 #define node_page_state(node, item) global_node_page_state(item)
 240 #endif /* CONFIG_NUMA */
 241
 242 #ifdef CONFIG_SMP
 243 void __mod_zone_page_state(struct zone *, enum zone_stat_item item, long);
 244 void __inc_zone_page_state(struct page *, enum zone_stat_item);
 245 void __dec_zone_page_state(struct page *, enum zone_stat_item);
 246
 247 void __mod_node_page_state(struct pglist_data *, enum node_stat_item item, long);
 248 void __inc_node_page_state(struct page *, enum node_stat_item);
 249 void __dec_node_page_state(struct page *, enum node_stat_item);
 250
 251 void mod_zone_page_state(struct zone *, enum zone_stat_item, long);
 252 void inc_zone_page_state(struct page *, enum zone_stat_item);
 253 void dec_zone_page_state(struct page *, enum zone_stat_item);
 254
 255 void mod_node_page_state(struct pglist_data *, enum node_stat_item, long);
 256 void inc_node_page_state(struct page *, enum node_stat_item);
 257 void dec_node_page_state(struct page *, enum node_stat_item);
 258
 259 extern void inc_node_state(struct pglist_data *, enum node_stat_item);
 260 extern void __inc_zone_state(struct zone *, enum zone_stat_item);
 261 extern void __inc_node_state(struct pglist_data *, enum node_stat_item);
 262 extern void dec_zone_state(struct zone *, enum zone_stat_item);
 263 extern void __dec_zone_state(struct zone *, enum zone_stat_item);
 264 extern void __dec_node_state(struct pglist_data *, enum node_stat_item);
 265
 266 void quiet_vmstat(void);
 267 void cpu_vm_stats_fold(int cpu);
 268 void refresh_zone_stat_thresholds(void);
 269
 270 struct ctl_table;
 271 int vmstat_refresh(struct ctl_table *, int write,
 272                    void __user *buffer, size_t *lenp, loff_t *ppos);
 273
 274 void drain_zonestat(struct zone *zone, struct per_cpu_pageset *);
 275
 276 int calculate_pressure_threshold(struct zone *zone);
 277 int calculate_normal_threshold(struct zone *zone);
 278 void set_pgdat_percpu_threshold(pg_data_t *pgdat,
 279                                 int (*calculate_pressure)(struct zone *));
 280 #else /* CONFIG_SMP */
 281
 282 /*
 283  * We do not maintain differentials in a single processor configuration.
 284  * The functions directly modify the zone and global counters.
 285  */
 286 static inline void __mod_zone_page_state(struct zone *zone,
 287                         enum zone_stat_item item, long delta)
 288 {
 289         zone_page_state_add(delta, zone, item);
 290 }
 291
 292 static inline void __mod_node_page_state(struct pglist_data *pgdat,
 293                         enum node_stat_item item, int delta)
 294 {
 295         node_page_state_add(delta, pgdat, item);
 296 }
 297
 298 static inline void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
 299 {
 300         atomic_long_inc(&zone->vm_stat[item]);
 301         atomic_long_inc(&vm_zone_stat[item]);
 302 }
 303
 304 static inline void __inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
 305 {
 306         atomic_long_inc(&pgdat->vm_stat[item]);
 307         atomic_long_inc(&vm_node_stat[item]);
 308 }
 309
 310 static inline void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
 311 {
 312         atomic_long_dec(&zone->vm_stat[item]);
 313         atomic_long_dec(&vm_zone_stat[item]);
 314 }
 315
 316 static inline void __dec_node_state(struct pglist_data *pgdat, enum node_stat_item item)
 317 {
 318         atomic_long_dec(&pgdat->vm_stat[item]);
 319         atomic_long_dec(&vm_node_stat[item]);
 320 }
 321
 322 static inline void __inc_zone_page_state(struct page *page,
 323                         enum zone_stat_item item)
 324 {
 325         __inc_zone_state(page_zone(page), item);
 326 }
 327
 328 static inline void __inc_node_page_state(struct page *page,
 329                         enum node_stat_item item)
 330 {
 331         __inc_node_state(page_pgdat(page), item);
 332 }
 333
 334
 335 static inline void __dec_zone_page_state(struct page *page,
 336                         enum zone_stat_item item)
 337 {
 338         __dec_zone_state(page_zone(page), item);
 339 }
 340
 341 static inline void __dec_node_page_state(struct page *page,
 342                         enum node_stat_item item)
 343 {
 344         __dec_node_state(page_pgdat(page), item);
 345 }
 346
 347
 348 /*
 349  * We only use atomic operations to update counters. So there is no need to
 350  * disable interrupts.
 351  */
 352 #define inc_zone_page_state __inc_zone_page_state
 353 #define dec_zone_page_state __dec_zone_page_state
 354 #define mod_zone_page_state __mod_zone_page_state
 355
 356 #define inc_node_page_state __inc_node_page_state
 357 #define dec_node_page_state __dec_node_page_state
 358 #define mod_node_page_state __mod_node_page_state
 359
 360 #define inc_zone_state __inc_zone_state
 361 #define inc_node_state __inc_node_state
 362 #define dec_zone_state __dec_zone_state
 363
 364 #define set_pgdat_percpu_threshold(pgdat, callback) { }
 365
 366 static inline void refresh_zone_stat_thresholds(void) { }
 367 static inline void cpu_vm_stats_fold(int cpu) { }
 368 static inline void quiet_vmstat(void) { }
 369
 370 static inline void drain_zonestat(struct zone *zone,
 371                         struct per_cpu_pageset *pset) { }
 372 #endif          /* CONFIG_SMP */
 373
 374 static inline void __mod_zone_freepage_state(struct zone *zone, int nr_pages,
 375                                              int migratetype)
 376 {
 377         __mod_zone_page_state(zone, NR_FREE_PAGES, nr_pages);
 378         if (is_migrate_cma(migratetype))
 379                 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, nr_pages);
 380 }
 381
 382 extern const char * const vmstat_text[];
 383
 384 #endif /* _LINUX_VMSTAT_H */