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 #include <linux/mmdebug.h> 12 13 extern int sysctl_stat_interval; 14 15 #ifdef CONFIG_NUMA 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); 22 #endif 23 24 struct reclaim_stat { 25 unsigned nr_dirty; 26 unsigned nr_unqueued_dirty; 27 unsigned nr_congested; 28 unsigned nr_writeback; 29 unsigned nr_immediate; 30 unsigned nr_pageout; 31 unsigned nr_activate[ANON_AND_FILE]; 32 unsigned nr_ref_keep; 33 unsigned nr_unmap_fail; 34 unsigned nr_lazyfree_fail; 35 }; 36 37 enum writeback_stat_item { 38 NR_DIRTY_THRESHOLD, 39 NR_DIRTY_BG_THRESHOLD, 40 NR_VM_WRITEBACK_STAT_ITEMS, 41 }; 42 43 #ifdef CONFIG_VM_EVENT_COUNTERS 44 /* 45 * Light weight per cpu counter implementation. 46 * 47 * Counters should only be incremented and no critical kernel component 48 * should rely on the counter values. 49 * 50 * Counters are handled completely inline. On many platforms the code 51 * generated will simply be the increment of a global address. 52 */ 53 54 struct vm_event_state { 55 unsigned long event[NR_VM_EVENT_ITEMS]; 56 }; 57 58 DECLARE_PER_CPU(struct vm_event_state, vm_event_states); 59 60 /* 61 * vm counters are allowed to be racy. Use raw_cpu_ops to avoid the 62 * local_irq_disable overhead. 63 */ 64 static inline void __count_vm_event(enum vm_event_item item) 65 { 66 raw_cpu_inc(vm_event_states.event[item]); 67 } 68 69 static inline void count_vm_event(enum vm_event_item item) 70 { 71 this_cpu_inc(vm_event_states.event[item]); 72 } 73 74 static inline void __count_vm_events(enum vm_event_item item, long delta) 75 { 76 raw_cpu_add(vm_event_states.event[item], delta); 77 } 78 79 static inline void count_vm_events(enum vm_event_item item, long delta) 80 { 81 this_cpu_add(vm_event_states.event[item], delta); 82 } 83 84 extern void all_vm_events(unsigned long *); 85 86 extern void vm_events_fold_cpu(int cpu); 87 88 #else 89 90 /* Disable counters */ 91 static inline void count_vm_event(enum vm_event_item item) 92 { 93 } 94 static inline void count_vm_events(enum vm_event_item item, long delta) 95 { 96 } 97 static inline void __count_vm_event(enum vm_event_item item) 98 { 99 } 100 static inline void __count_vm_events(enum vm_event_item item, long delta) 101 { 102 } 103 static inline void all_vm_events(unsigned long *ret) 104 { 105 } 106 static inline void vm_events_fold_cpu(int cpu) 107 { 108 } 109 110 #endif /* CONFIG_VM_EVENT_COUNTERS */ 111 112 #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) 115 #else 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 */ 119 120 #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) 123 #else 124 #define count_vm_tlb_event(x) do {} while (0) 125 #define count_vm_tlb_events(x, y) do { (void)(y); } while (0) 126 #endif 127 128 #ifdef CONFIG_DEBUG_VM_VMACACHE 129 #define count_vm_vmacache_event(x) count_vm_event(x) 130 #else 131 #define count_vm_vmacache_event(x) do {} while (0) 132 #endif 133 134 #define __count_zid_vm_events(item, zid, delta) \ 135 __count_vm_events(item##_NORMAL - ZONE_NORMAL + zid, delta) 136 137 /* 138 * Zone and node-based page accounting with per cpu differentials. 139 */ 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]; 143 144 #ifdef CONFIG_NUMA 145 static inline void zone_numa_event_add(long x, struct zone *zone, 146 enum numa_stat_item item) 147 { 148 atomic_long_add(x, &zone->vm_numa_event[item]); 149 atomic_long_add(x, &vm_numa_event[item]); 150 } 151 152 static inline unsigned long zone_numa_event_state(struct zone *zone, 153 enum numa_stat_item item) 154 { 155 return atomic_long_read(&zone->vm_numa_event[item]); 156 } 157 158 static inline unsigned long 159 global_numa_event_state(enum numa_stat_item item) 160 { 161 return atomic_long_read(&vm_numa_event[item]); 162 } 163 #endif /* CONFIG_NUMA */ 164 165 static inline void zone_page_state_add(long x, struct zone *zone, 166 enum zone_stat_item item) 167 { 168 atomic_long_add(x, &zone->vm_stat[item]); 169 atomic_long_add(x, &vm_zone_stat[item]); 170 } 171 172 static inline void node_page_state_add(long x, struct pglist_data *pgdat, 173 enum node_stat_item item) 174 { 175 atomic_long_add(x, &pgdat->vm_stat[item]); 176 atomic_long_add(x, &vm_node_stat[item]); 177 } 178 179 static inline unsigned long global_zone_page_state(enum zone_stat_item item) 180 { 181 long x = atomic_long_read(&vm_zone_stat[item]); 182 #ifdef CONFIG_SMP 183 if (x < 0) 184 x = 0; 185 #endif 186 return x; 187 } 188 189 static inline 190 unsigned long global_node_page_state_pages(enum node_stat_item item) 191 { 192 long x = atomic_long_read(&vm_node_stat[item]); 193 #ifdef CONFIG_SMP 194 if (x < 0) 195 x = 0; 196 #endif 197 return x; 198 } 199 200 static inline unsigned long global_node_page_state(enum node_stat_item item) 201 { 202 VM_WARN_ON_ONCE(vmstat_item_in_bytes(item)); 203 204 return global_node_page_state_pages(item); 205 } 206 207 static inline unsigned long zone_page_state(struct zone *zone, 208 enum zone_stat_item item) 209 { 210 long x = atomic_long_read(&zone->vm_stat[item]); 211 #ifdef CONFIG_SMP 212 if (x < 0) 213 x = 0; 214 #endif 215 return x; 216 } 217 218 /* 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. 223 */ 224 static inline unsigned long zone_page_state_snapshot(struct zone *zone, 225 enum zone_stat_item item) 226 { 227 long x = atomic_long_read(&zone->vm_stat[item]); 228 229 #ifdef CONFIG_SMP 230 int cpu; 231 for_each_online_cpu(cpu) 232 x += per_cpu_ptr(zone->per_cpu_zonestats, cpu)->vm_stat_diff[item]; 233 234 if (x < 0) 235 x = 0; 236 #endif 237 return x; 238 } 239 240 #ifdef CONFIG_NUMA 241 /* See __count_vm_event comment on why raw_cpu_inc is used. */ 242 static inline void 243 __count_numa_event(struct zone *zone, enum numa_stat_item item) 244 { 245 struct per_cpu_zonestat __percpu *pzstats = zone->per_cpu_zonestats; 246 247 raw_cpu_inc(pzstats->vm_numa_event[item]); 248 } 249 250 static inline void 251 __count_numa_events(struct zone *zone, enum numa_stat_item item, long delta) 252 { 253 struct per_cpu_zonestat __percpu *pzstats = zone->per_cpu_zonestats; 254 255 raw_cpu_add(pzstats->vm_numa_event[item], delta); 256 } 257 258 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); 266 #else 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) 271 { 272 } 273 #endif /* CONFIG_NUMA */ 274 275 #ifdef CONFIG_SMP 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); 279 280 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); 283 284 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); 287 288 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); 291 292 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); 298 299 void quiet_vmstat(void); 300 void cpu_vm_stats_fold(int cpu); 301 void refresh_zone_stat_thresholds(void); 302 303 struct ctl_table; 304 int vmstat_refresh(struct ctl_table *, int write, void *buffer, size_t *lenp, 305 loff_t *ppos); 306 307 void drain_zonestat(struct zone *zone, struct per_cpu_zonestat *); 308 309 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 */ 314 315 /* 316 * We do not maintain differentials in a single processor configuration. 317 * The functions directly modify the zone and global counters. 318 */ 319 static inline void __mod_zone_page_state(struct zone *zone, 320 enum zone_stat_item item, long delta) 321 { 322 zone_page_state_add(delta, zone, item); 323 } 324 325 static inline void __mod_node_page_state(struct pglist_data *pgdat, 326 enum node_stat_item item, int delta) 327 { 328 if (vmstat_item_in_bytes(item)) { 329 /* 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. 334 */ 335 VM_WARN_ON_ONCE(delta & (PAGE_SIZE - 1)); 336 delta >>= PAGE_SHIFT; 337 } 338 339 node_page_state_add(delta, pgdat, item); 340 } 341 342 static inline void __inc_zone_state(struct zone *zone, enum zone_stat_item item) 343 { 344 atomic_long_inc(&zone->vm_stat[item]); 345 atomic_long_inc(&vm_zone_stat[item]); 346 } 347 348 static inline void __inc_node_state(struct pglist_data *pgdat, enum node_stat_item item) 349 { 350 atomic_long_inc(&pgdat->vm_stat[item]); 351 atomic_long_inc(&vm_node_stat[item]); 352 } 353 354 static inline void __dec_zone_state(struct zone *zone, enum zone_stat_item item) 355 { 356 atomic_long_dec(&zone->vm_stat[item]); 357 atomic_long_dec(&vm_zone_stat[item]); 358 } 359 360 static inline void __dec_node_state(struct pglist_data *pgdat, enum node_stat_item item) 361 { 362 atomic_long_dec(&pgdat->vm_stat[item]); 363 atomic_long_dec(&vm_node_stat[item]); 364 } 365 366 static inline void __inc_zone_page_state(struct page *page, 367 enum zone_stat_item item) 368 { 369 __inc_zone_state(page_zone(page), item); 370 } 371 372 static inline void __inc_node_page_state(struct page *page, 373 enum node_stat_item item) 374 { 375 __inc_node_state(page_pgdat(page), item); 376 } 377 378 379 static inline void __dec_zone_page_state(struct page *page, 380 enum zone_stat_item item) 381 { 382 __dec_zone_state(page_zone(page), item); 383 } 384 385 static inline void __dec_node_page_state(struct page *page, 386 enum node_stat_item item) 387 { 388 __dec_node_state(page_pgdat(page), item); 389 } 390 391 392 /* 393 * We only use atomic operations to update counters. So there is no need to 394 * disable interrupts. 395 */ 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 399 400 #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 403 404 #define inc_zone_state __inc_zone_state 405 #define inc_node_state __inc_node_state 406 #define dec_zone_state __dec_zone_state 407 408 #define set_pgdat_percpu_threshold(pgdat, callback) { } 409 410 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) { } 413 414 static inline void drain_zonestat(struct zone *zone, 415 struct per_cpu_zonestat *pzstats) { } 416 #endif /* CONFIG_SMP */ 417 418 static inline void __zone_stat_mod_folio(struct folio *folio, 419 enum zone_stat_item item, long nr) 420 { 421 __mod_zone_page_state(folio_zone(folio), item, nr); 422 } 423 424 static inline void __zone_stat_add_folio(struct folio *folio, 425 enum zone_stat_item item) 426 { 427 __mod_zone_page_state(folio_zone(folio), item, folio_nr_pages(folio)); 428 } 429 430 static inline void __zone_stat_sub_folio(struct folio *folio, 431 enum zone_stat_item item) 432 { 433 __mod_zone_page_state(folio_zone(folio), item, -folio_nr_pages(folio)); 434 } 435 436 static inline void zone_stat_mod_folio(struct folio *folio, 437 enum zone_stat_item item, long nr) 438 { 439 mod_zone_page_state(folio_zone(folio), item, nr); 440 } 441 442 static inline void zone_stat_add_folio(struct folio *folio, 443 enum zone_stat_item item) 444 { 445 mod_zone_page_state(folio_zone(folio), item, folio_nr_pages(folio)); 446 } 447 448 static inline void zone_stat_sub_folio(struct folio *folio, 449 enum zone_stat_item item) 450 { 451 mod_zone_page_state(folio_zone(folio), item, -folio_nr_pages(folio)); 452 } 453 454 static inline void __node_stat_mod_folio(struct folio *folio, 455 enum node_stat_item item, long nr) 456 { 457 __mod_node_page_state(folio_pgdat(folio), item, nr); 458 } 459 460 static inline void __node_stat_add_folio(struct folio *folio, 461 enum node_stat_item item) 462 { 463 __mod_node_page_state(folio_pgdat(folio), item, folio_nr_pages(folio)); 464 } 465 466 static inline void __node_stat_sub_folio(struct folio *folio, 467 enum node_stat_item item) 468 { 469 __mod_node_page_state(folio_pgdat(folio), item, -folio_nr_pages(folio)); 470 } 471 472 static inline void node_stat_mod_folio(struct folio *folio, 473 enum node_stat_item item, long nr) 474 { 475 mod_node_page_state(folio_pgdat(folio), item, nr); 476 } 477 478 static inline void node_stat_add_folio(struct folio *folio, 479 enum node_stat_item item) 480 { 481 mod_node_page_state(folio_pgdat(folio), item, folio_nr_pages(folio)); 482 } 483 484 static inline void node_stat_sub_folio(struct folio *folio, 485 enum node_stat_item item) 486 { 487 mod_node_page_state(folio_pgdat(folio), item, -folio_nr_pages(folio)); 488 } 489 490 static inline void __mod_zone_freepage_state(struct zone *zone, int nr_pages, 491 int migratetype) 492 { 493 __mod_zone_page_state(zone, NR_FREE_PAGES, nr_pages); 494 if (is_migrate_cma(migratetype)) 495 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, nr_pages); 496 } 497 498 extern const char * const vmstat_text[]; 499 500 static inline const char *zone_stat_name(enum zone_stat_item item) 501 { 502 return vmstat_text[item]; 503 } 504 505 #ifdef CONFIG_NUMA 506 static inline const char *numa_stat_name(enum numa_stat_item item) 507 { 508 return vmstat_text[NR_VM_ZONE_STAT_ITEMS + 509 item]; 510 } 511 #endif /* CONFIG_NUMA */ 512 513 static inline const char *node_stat_name(enum node_stat_item item) 514 { 515 return vmstat_text[NR_VM_ZONE_STAT_ITEMS + 516 NR_VM_NUMA_EVENT_ITEMS + 517 item]; 518 } 519 520 static inline const char *lru_list_name(enum lru_list lru) 521 { 522 return node_stat_name(NR_LRU_BASE + lru) + 3; // skip "nr_" 523 } 524 525 static inline const char *writeback_stat_name(enum writeback_stat_item item) 526 { 527 return vmstat_text[NR_VM_ZONE_STAT_ITEMS + 528 NR_VM_NUMA_EVENT_ITEMS + 529 NR_VM_NODE_STAT_ITEMS + 530 item]; 531 } 532 533 #if defined(CONFIG_VM_EVENT_COUNTERS) || defined(CONFIG_MEMCG) 534 static inline const char *vm_event_name(enum vm_event_item item) 535 { 536 return vmstat_text[NR_VM_ZONE_STAT_ITEMS + 537 NR_VM_NUMA_EVENT_ITEMS + 538 NR_VM_NODE_STAT_ITEMS + 539 NR_VM_WRITEBACK_STAT_ITEMS + 540 item]; 541 } 542 #endif /* CONFIG_VM_EVENT_COUNTERS || CONFIG_MEMCG */ 543 544 #ifdef CONFIG_MEMCG 545 546 void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, 547 int val); 548 549 static inline void mod_lruvec_state(struct lruvec *lruvec, 550 enum node_stat_item idx, int val) 551 { 552 unsigned long flags; 553 554 local_irq_save(flags); 555 __mod_lruvec_state(lruvec, idx, val); 556 local_irq_restore(flags); 557 } 558 559 void __mod_lruvec_page_state(struct page *page, 560 enum node_stat_item idx, int val); 561 562 static inline void mod_lruvec_page_state(struct page *page, 563 enum node_stat_item idx, int val) 564 { 565 unsigned long flags; 566 567 local_irq_save(flags); 568 __mod_lruvec_page_state(page, idx, val); 569 local_irq_restore(flags); 570 } 571 572 #else 573 574 static inline void __mod_lruvec_state(struct lruvec *lruvec, 575 enum node_stat_item idx, int val) 576 { 577 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); 578 } 579 580 static inline void mod_lruvec_state(struct lruvec *lruvec, 581 enum node_stat_item idx, int val) 582 { 583 mod_node_page_state(lruvec_pgdat(lruvec), idx, val); 584 } 585 586 static inline void __mod_lruvec_page_state(struct page *page, 587 enum node_stat_item idx, int val) 588 { 589 __mod_node_page_state(page_pgdat(page), idx, val); 590 } 591 592 static inline void mod_lruvec_page_state(struct page *page, 593 enum node_stat_item idx, int val) 594 { 595 mod_node_page_state(page_pgdat(page), idx, val); 596 } 597 598 #endif /* CONFIG_MEMCG */ 599 600 static inline void __inc_lruvec_page_state(struct page *page, 601 enum node_stat_item idx) 602 { 603 __mod_lruvec_page_state(page, idx, 1); 604 } 605 606 static inline void __dec_lruvec_page_state(struct page *page, 607 enum node_stat_item idx) 608 { 609 __mod_lruvec_page_state(page, idx, -1); 610 } 611 612 static inline void __lruvec_stat_mod_folio(struct folio *folio, 613 enum node_stat_item idx, int val) 614 { 615 __mod_lruvec_page_state(&folio->page, idx, val); 616 } 617 618 static inline void __lruvec_stat_add_folio(struct folio *folio, 619 enum node_stat_item idx) 620 { 621 __lruvec_stat_mod_folio(folio, idx, folio_nr_pages(folio)); 622 } 623 624 static inline void __lruvec_stat_sub_folio(struct folio *folio, 625 enum node_stat_item idx) 626 { 627 __lruvec_stat_mod_folio(folio, idx, -folio_nr_pages(folio)); 628 } 629 630 static inline void inc_lruvec_page_state(struct page *page, 631 enum node_stat_item idx) 632 { 633 mod_lruvec_page_state(page, idx, 1); 634 } 635 636 static inline void dec_lruvec_page_state(struct page *page, 637 enum node_stat_item idx) 638 { 639 mod_lruvec_page_state(page, idx, -1); 640 } 641 642 static inline void lruvec_stat_mod_folio(struct folio *folio, 643 enum node_stat_item idx, int val) 644 { 645 mod_lruvec_page_state(&folio->page, idx, val); 646 } 647 648 static inline void lruvec_stat_add_folio(struct folio *folio, 649 enum node_stat_item idx) 650 { 651 lruvec_stat_mod_folio(folio, idx, folio_nr_pages(folio)); 652 } 653 654 static inline void lruvec_stat_sub_folio(struct folio *folio, 655 enum node_stat_item idx) 656 { 657 lruvec_stat_mod_folio(folio, idx, -folio_nr_pages(folio)); 658 } 659 #endif /* _LINUX_VMSTAT_H */ 660