1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_VMSTAT_H
3 #define _LINUX_VMSTAT_H
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 #include <linux/mmdebug.h>
13 extern int sysctl_stat_interval;
16 #define ENABLE_NUMA_STAT 1
17 #define DISABLE_NUMA_STAT 0
18 extern int sysctl_vm_numa_stat;
19 DECLARE_STATIC_KEY_TRUE(vm_numa_stat_key);
20 int sysctl_vm_numa_stat_handler(struct ctl_table *table, int write,
21 void *buffer, size_t *length, loff_t *ppos);
26 unsigned nr_unqueued_dirty;
27 unsigned nr_congested;
28 unsigned nr_writeback;
29 unsigned nr_immediate;
31 unsigned nr_activate[ANON_AND_FILE];
33 unsigned nr_unmap_fail;
34 unsigned nr_lazyfree_fail;
37 enum writeback_stat_item {
39 NR_DIRTY_BG_THRESHOLD,
40 NR_VM_WRITEBACK_STAT_ITEMS,
43 #ifdef CONFIG_VM_EVENT_COUNTERS
45 * Light weight per cpu counter implementation.
47 * Counters should only be incremented and no critical kernel component
48 * should rely on the counter values.
50 * Counters are handled completely inline. On many platforms the code
51 * generated will simply be the increment of a global address.
54 struct vm_event_state {
55 unsigned long event[NR_VM_EVENT_ITEMS];
58 DECLARE_PER_CPU(struct vm_event_state, vm_event_states);
61 * vm counters are allowed to be racy. Use raw_cpu_ops to avoid the
62 * local_irq_disable overhead.
64 static inline void __count_vm_event(enum vm_event_item item)
66 raw_cpu_inc(vm_event_states.event[item]);
69 static inline void count_vm_event(enum vm_event_item item)
71 this_cpu_inc(vm_event_states.event[item]);
74 static inline void __count_vm_events(enum vm_event_item item, long delta)
76 raw_cpu_add(vm_event_states.event[item], delta);
79 static inline void count_vm_events(enum vm_event_item item, long delta)
81 this_cpu_add(vm_event_states.event[item], delta);
84 extern void all_vm_events(unsigned long *);
86 extern void vm_events_fold_cpu(int cpu);
90 /* Disable counters */
91 static inline void count_vm_event(enum vm_event_item item)
94 static inline void count_vm_events(enum vm_event_item item, long delta)
97 static inline void __count_vm_event(enum vm_event_item item)
100 static inline void __count_vm_events(enum vm_event_item item, long delta)
103 static inline void all_vm_events(unsigned long *ret)
106 static inline void vm_events_fold_cpu(int cpu)
110 #endif /* CONFIG_VM_EVENT_COUNTERS */
112 #ifdef CONFIG_NUMA_BALANCING
113 #define count_vm_numa_event(x) count_vm_event(x)
114 #define count_vm_numa_events(x, y) count_vm_events(x, y)
116 #define count_vm_numa_event(x) do {} while (0)
117 #define count_vm_numa_events(x, y) do { (void)(y); } while (0)
118 #endif /* CONFIG_NUMA_BALANCING */
120 #ifdef CONFIG_DEBUG_TLBFLUSH
121 #define count_vm_tlb_event(x) count_vm_event(x)
122 #define count_vm_tlb_events(x, y) count_vm_events(x, y)
124 #define count_vm_tlb_event(x) do {} while (0)
125 #define count_vm_tlb_events(x, y) do { (void)(y); } while (0)
128 #ifdef CONFIG_DEBUG_VM_VMACACHE
129 #define count_vm_vmacache_event(x) count_vm_event(x)
131 #define count_vm_vmacache_event(x) do {} while (0)
134 #define __count_zid_vm_events(item, zid, delta) \
135 __count_vm_events(item##_NORMAL - ZONE_NORMAL + zid, delta)
138 * Zone and node-based page accounting with per cpu differentials.
140 extern atomic_long_t vm_zone_stat[NR_VM_ZONE_STAT_ITEMS];
141 extern atomic_long_t vm_node_stat[NR_VM_NODE_STAT_ITEMS];
142 extern atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
145 static inline void zone_numa_event_add(long x, struct zone *zone,
146 enum numa_stat_item item)
148 atomic_long_add(x, &zone->vm_numa_event[item]);
149 atomic_long_add(x, &vm_numa_event[item]);
152 static inline unsigned long zone_numa_event_state(struct zone *zone,
153 enum numa_stat_item item)
155 return atomic_long_read(&zone->vm_numa_event[item]);
158 static inline unsigned long
159 global_numa_event_state(enum numa_stat_item item)
161 return atomic_long_read(&vm_numa_event[item]);
163 #endif /* CONFIG_NUMA */
165 static inline void zone_page_state_add(long x, struct zone *zone,
166 enum zone_stat_item item)
168 atomic_long_add(x, &zone->vm_stat[item]);
169 atomic_long_add(x, &vm_zone_stat[item]);
172 static inline void node_page_state_add(long x, struct pglist_data *pgdat,
173 enum node_stat_item item)
175 atomic_long_add(x, &pgdat->vm_stat[item]);
176 atomic_long_add(x, &vm_node_stat[item]);
179 static inline unsigned long global_zone_page_state(enum zone_stat_item item)
181 long x = atomic_long_read(&vm_zone_stat[item]);
190 unsigned long global_node_page_state_pages(enum node_stat_item item)
192 long x = atomic_long_read(&vm_node_stat[item]);
200 static inline unsigned long global_node_page_state(enum node_stat_item item)
202 VM_WARN_ON_ONCE(vmstat_item_in_bytes(item));
204 return global_node_page_state_pages(item);
207 static inline unsigned long zone_page_state(struct zone *zone,
208 enum zone_stat_item item)
210 long x = atomic_long_read(&zone->vm_stat[item]);
219 * More accurate version that also considers the currently pending
220 * deltas. For that we need to loop over all cpus to find the current
221 * deltas. There is no synchronization so the result cannot be
222 * exactly accurate either.
224 static inline unsigned long zone_page_state_snapshot(struct zone *zone,
225 enum zone_stat_item item)
227 long x = atomic_long_read(&zone->vm_stat[item]);
231 for_each_online_cpu(cpu)
232 x += per_cpu_ptr(zone->per_cpu_zonestats, cpu)->vm_stat_diff[item];
241 /* See __count_vm_event comment on why raw_cpu_inc is used. */
243 __count_numa_event(struct zone *zone, enum numa_stat_item item)
245 struct per_cpu_zonestat __percpu *pzstats = zone->per_cpu_zonestats;
247 raw_cpu_inc(pzstats->vm_numa_event[item]);
251 __count_numa_events(struct zone *zone, enum numa_stat_item item, long delta)
253 struct per_cpu_zonestat __percpu *pzstats = zone->per_cpu_zonestats;
255 raw_cpu_add(pzstats->vm_numa_event[item], delta);
258 extern unsigned long sum_zone_node_page_state(int node,
259 enum zone_stat_item item);
260 extern unsigned long sum_zone_numa_event_state(int node, enum numa_stat_item item);
261 extern unsigned long node_page_state(struct pglist_data *pgdat,
262 enum node_stat_item item);
263 extern unsigned long node_page_state_pages(struct pglist_data *pgdat,
264 enum node_stat_item item);
265 extern void fold_vm_numa_events(void);
267 #define sum_zone_node_page_state(node, item) global_zone_page_state(item)
268 #define node_page_state(node, item) global_node_page_state(item)
269 #define node_page_state_pages(node, item) global_node_page_state_pages(item)
270 static inline void fold_vm_numa_events(void)
273 #endif /* CONFIG_NUMA */
276 void __mod_zone_page_state(struct zone *, enum zone_stat_item item, long);
277 void __inc_zone_page_state(struct page *, enum zone_stat_item);
278 void __dec_zone_page_state(struct page *, enum zone_stat_item);
280 void __mod_node_page_state(struct pglist_data *, enum node_stat_item item, long);
281 void __inc_node_page_state(struct page *, enum node_stat_item);
282 void __dec_node_page_state(struct page *, enum node_stat_item);
284 void mod_zone_page_state(struct zone *, enum zone_stat_item, long);
285 void inc_zone_page_state(struct page *, enum zone_stat_item);
286 void dec_zone_page_state(struct page *, enum zone_stat_item);
288 void mod_node_page_state(struct pglist_data *, enum node_stat_item, long);
289 void inc_node_page_state(struct page *, enum node_stat_item);
290 void dec_node_page_state(struct page *, enum node_stat_item);
292 extern void inc_node_state(struct pglist_data *, enum node_stat_item);
293 extern void __inc_zone_state(struct zone *, enum zone_stat_item);
294 extern void __inc_node_state(struct pglist_data *, enum node_stat_item);
295 extern void dec_zone_state(struct zone *, enum zone_stat_item);
296 extern void __dec_zone_state(struct zone *, enum zone_stat_item);
297 extern void __dec_node_state(struct pglist_data *, enum node_stat_item);
299 void quiet_vmstat(void);
300 void cpu_vm_stats_fold(int cpu);
301 void refresh_zone_stat_thresholds(void);
304 int vmstat_refresh(struct ctl_table *, int write, void *buffer, size_t *lenp,
307 void drain_zonestat(struct zone *zone, struct per_cpu_zonestat *);
309 int calculate_pressure_threshold(struct zone *zone);
310 int calculate_normal_threshold(struct zone *zone);
311 void set_pgdat_percpu_threshold(pg_data_t *pgdat,
312 int (*calculate_pressure)(struct zone *));
313 #else /* CONFIG_SMP */
316 * We do not maintain differentials in a single processor configuration.
317 * The functions directly modify the zone and global counters.
319 static inline void __mod_zone_page_state(struct zone *zone,
320 enum zone_stat_item item, long delta)
322 zone_page_state_add(delta, zone, item);
325 static inline void __mod_node_page_state(struct pglist_data *pgdat,
326 enum node_stat_item item, int delta)
328 if (vmstat_item_in_bytes(item)) {
330 * Only cgroups use subpage accounting right now; at
331 * the global level, these items still change in
332 * multiples of whole pages. Store them as pages
333 * internally to keep the per-cpu counters compact.
335 VM_WARN_ON_ONCE(delta & (PAGE_SIZE - 1));
336 delta >>= PAGE_SHIFT;
339 node_page_state_add(delta, pgdat, item);
342 static inline void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
344 atomic_long_inc(&zone->vm_stat[item]);
345 atomic_long_inc(&vm_zone_stat[item]);
348 static inline void __inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
350 atomic_long_inc(&pgdat->vm_stat[item]);
351 atomic_long_inc(&vm_node_stat[item]);
354 static inline void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
356 atomic_long_dec(&zone->vm_stat[item]);
357 atomic_long_dec(&vm_zone_stat[item]);
360 static inline void __dec_node_state(struct pglist_data *pgdat, enum node_stat_item item)
362 atomic_long_dec(&pgdat->vm_stat[item]);
363 atomic_long_dec(&vm_node_stat[item]);
366 static inline void __inc_zone_page_state(struct page *page,
367 enum zone_stat_item item)
369 __inc_zone_state(page_zone(page), item);
372 static inline void __inc_node_page_state(struct page *page,
373 enum node_stat_item item)
375 __inc_node_state(page_pgdat(page), item);
379 static inline void __dec_zone_page_state(struct page *page,
380 enum zone_stat_item item)
382 __dec_zone_state(page_zone(page), item);
385 static inline void __dec_node_page_state(struct page *page,
386 enum node_stat_item item)
388 __dec_node_state(page_pgdat(page), item);
393 * We only use atomic operations to update counters. So there is no need to
394 * disable interrupts.
396 #define inc_zone_page_state __inc_zone_page_state
397 #define dec_zone_page_state __dec_zone_page_state
398 #define mod_zone_page_state __mod_zone_page_state
400 #define inc_node_page_state __inc_node_page_state
401 #define dec_node_page_state __dec_node_page_state
402 #define mod_node_page_state __mod_node_page_state
404 #define inc_zone_state __inc_zone_state
405 #define inc_node_state __inc_node_state
406 #define dec_zone_state __dec_zone_state
408 #define set_pgdat_percpu_threshold(pgdat, callback) { }
410 static inline void refresh_zone_stat_thresholds(void) { }
411 static inline void cpu_vm_stats_fold(int cpu) { }
412 static inline void quiet_vmstat(void) { }
414 static inline void drain_zonestat(struct zone *zone,
415 struct per_cpu_zonestat *pzstats) { }
416 #endif /* CONFIG_SMP */
418 static inline void __mod_zone_freepage_state(struct zone *zone, int nr_pages,
421 __mod_zone_page_state(zone, NR_FREE_PAGES, nr_pages);
422 if (is_migrate_cma(migratetype))
423 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, nr_pages);
426 extern const char * const vmstat_text[];
428 static inline const char *zone_stat_name(enum zone_stat_item item)
430 return vmstat_text[item];
434 static inline const char *numa_stat_name(enum numa_stat_item item)
436 return vmstat_text[NR_VM_ZONE_STAT_ITEMS +
439 #endif /* CONFIG_NUMA */
441 static inline const char *node_stat_name(enum node_stat_item item)
443 return vmstat_text[NR_VM_ZONE_STAT_ITEMS +
444 NR_VM_NUMA_EVENT_ITEMS +
448 static inline const char *lru_list_name(enum lru_list lru)
450 return node_stat_name(NR_LRU_BASE + lru) + 3; // skip "nr_"
453 static inline const char *writeback_stat_name(enum writeback_stat_item item)
455 return vmstat_text[NR_VM_ZONE_STAT_ITEMS +
456 NR_VM_NUMA_EVENT_ITEMS +
457 NR_VM_NODE_STAT_ITEMS +
461 #if defined(CONFIG_VM_EVENT_COUNTERS) || defined(CONFIG_MEMCG)
462 static inline const char *vm_event_name(enum vm_event_item item)
464 return vmstat_text[NR_VM_ZONE_STAT_ITEMS +
465 NR_VM_NUMA_EVENT_ITEMS +
466 NR_VM_NODE_STAT_ITEMS +
467 NR_VM_WRITEBACK_STAT_ITEMS +
470 #endif /* CONFIG_VM_EVENT_COUNTERS || CONFIG_MEMCG */
474 void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
477 static inline void mod_lruvec_state(struct lruvec *lruvec,
478 enum node_stat_item idx, int val)
482 local_irq_save(flags);
483 __mod_lruvec_state(lruvec, idx, val);
484 local_irq_restore(flags);
487 void __mod_lruvec_page_state(struct page *page,
488 enum node_stat_item idx, int val);
490 static inline void mod_lruvec_page_state(struct page *page,
491 enum node_stat_item idx, int val)
495 local_irq_save(flags);
496 __mod_lruvec_page_state(page, idx, val);
497 local_irq_restore(flags);
502 static inline void __mod_lruvec_state(struct lruvec *lruvec,
503 enum node_stat_item idx, int val)
505 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
508 static inline void mod_lruvec_state(struct lruvec *lruvec,
509 enum node_stat_item idx, int val)
511 mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
514 static inline void __mod_lruvec_page_state(struct page *page,
515 enum node_stat_item idx, int val)
517 __mod_node_page_state(page_pgdat(page), idx, val);
520 static inline void mod_lruvec_page_state(struct page *page,
521 enum node_stat_item idx, int val)
523 mod_node_page_state(page_pgdat(page), idx, val);
526 #endif /* CONFIG_MEMCG */
528 static inline void inc_lruvec_state(struct lruvec *lruvec,
529 enum node_stat_item idx)
531 mod_lruvec_state(lruvec, idx, 1);
534 static inline void __inc_lruvec_page_state(struct page *page,
535 enum node_stat_item idx)
537 __mod_lruvec_page_state(page, idx, 1);
540 static inline void __dec_lruvec_page_state(struct page *page,
541 enum node_stat_item idx)
543 __mod_lruvec_page_state(page, idx, -1);
546 static inline void inc_lruvec_page_state(struct page *page,
547 enum node_stat_item idx)
549 mod_lruvec_page_state(page, idx, 1);
552 static inline void dec_lruvec_page_state(struct page *page,
553 enum node_stat_item idx)
555 mod_lruvec_page_state(page, idx, -1);
558 #endif /* _LINUX_VMSTAT_H */