1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_MMZONE_H 3 #define _LINUX_MMZONE_H 4 5 #ifndef __ASSEMBLY__ 6 #ifndef __GENERATING_BOUNDS_H 7 8 #include <linux/spinlock.h> 9 #include <linux/list.h> 10 #include <linux/list_nulls.h> 11 #include <linux/wait.h> 12 #include <linux/bitops.h> 13 #include <linux/cache.h> 14 #include <linux/threads.h> 15 #include <linux/numa.h> 16 #include <linux/init.h> 17 #include <linux/seqlock.h> 18 #include <linux/nodemask.h> 19 #include <linux/pageblock-flags.h> 20 #include <linux/page-flags-layout.h> 21 #include <linux/atomic.h> 22 #include <linux/mm_types.h> 23 #include <linux/page-flags.h> 24 #include <linux/local_lock.h> 25 #include <asm/page.h> 26 27 /* Free memory management - zoned buddy allocator. */ 28 #ifndef CONFIG_ARCH_FORCE_MAX_ORDER 29 #define MAX_ORDER 10 30 #else 31 #define MAX_ORDER CONFIG_ARCH_FORCE_MAX_ORDER 32 #endif 33 #define MAX_ORDER_NR_PAGES (1 << MAX_ORDER) 34 35 #define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES) 36 37 /* 38 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed 39 * costly to service. That is between allocation orders which should 40 * coalesce naturally under reasonable reclaim pressure and those which 41 * will not. 42 */ 43 #define PAGE_ALLOC_COSTLY_ORDER 3 44 45 enum migratetype { 46 MIGRATE_UNMOVABLE, 47 MIGRATE_MOVABLE, 48 MIGRATE_RECLAIMABLE, 49 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */ 50 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES, 51 #ifdef CONFIG_CMA 52 /* 53 * MIGRATE_CMA migration type is designed to mimic the way 54 * ZONE_MOVABLE works. Only movable pages can be allocated 55 * from MIGRATE_CMA pageblocks and page allocator never 56 * implicitly change migration type of MIGRATE_CMA pageblock. 57 * 58 * The way to use it is to change migratetype of a range of 59 * pageblocks to MIGRATE_CMA which can be done by 60 * __free_pageblock_cma() function. 61 */ 62 MIGRATE_CMA, 63 #endif 64 #ifdef CONFIG_MEMORY_ISOLATION 65 MIGRATE_ISOLATE, /* can't allocate from here */ 66 #endif 67 MIGRATE_TYPES 68 }; 69 70 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */ 71 extern const char * const migratetype_names[MIGRATE_TYPES]; 72 73 #ifdef CONFIG_CMA 74 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) 75 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA) 76 #else 77 # define is_migrate_cma(migratetype) false 78 # define is_migrate_cma_page(_page) false 79 #endif 80 81 static inline bool is_migrate_movable(int mt) 82 { 83 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE; 84 } 85 86 /* 87 * Check whether a migratetype can be merged with another migratetype. 88 * 89 * It is only mergeable when it can fall back to other migratetypes for 90 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c. 91 */ 92 static inline bool migratetype_is_mergeable(int mt) 93 { 94 return mt < MIGRATE_PCPTYPES; 95 } 96 97 #define for_each_migratetype_order(order, type) \ 98 for (order = 0; order <= MAX_ORDER; order++) \ 99 for (type = 0; type < MIGRATE_TYPES; type++) 100 101 extern int page_group_by_mobility_disabled; 102 103 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1) 104 105 #define get_pageblock_migratetype(page) \ 106 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK) 107 108 #define folio_migratetype(folio) \ 109 get_pfnblock_flags_mask(&folio->page, folio_pfn(folio), \ 110 MIGRATETYPE_MASK) 111 struct free_area { 112 struct list_head free_list[MIGRATE_TYPES]; 113 unsigned long nr_free; 114 }; 115 116 struct pglist_data; 117 118 #ifdef CONFIG_NUMA 119 enum numa_stat_item { 120 NUMA_HIT, /* allocated in intended node */ 121 NUMA_MISS, /* allocated in non intended node */ 122 NUMA_FOREIGN, /* was intended here, hit elsewhere */ 123 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ 124 NUMA_LOCAL, /* allocation from local node */ 125 NUMA_OTHER, /* allocation from other node */ 126 NR_VM_NUMA_EVENT_ITEMS 127 }; 128 #else 129 #define NR_VM_NUMA_EVENT_ITEMS 0 130 #endif 131 132 enum zone_stat_item { 133 /* First 128 byte cacheline (assuming 64 bit words) */ 134 NR_FREE_PAGES, 135 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */ 136 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE, 137 NR_ZONE_ACTIVE_ANON, 138 NR_ZONE_INACTIVE_FILE, 139 NR_ZONE_ACTIVE_FILE, 140 NR_ZONE_UNEVICTABLE, 141 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */ 142 NR_MLOCK, /* mlock()ed pages found and moved off LRU */ 143 /* Second 128 byte cacheline */ 144 NR_BOUNCE, 145 #if IS_ENABLED(CONFIG_ZSMALLOC) 146 NR_ZSPAGES, /* allocated in zsmalloc */ 147 #endif 148 NR_FREE_CMA_PAGES, 149 #ifdef CONFIG_UNACCEPTED_MEMORY 150 NR_UNACCEPTED, 151 #endif 152 NR_VM_ZONE_STAT_ITEMS }; 153 154 enum node_stat_item { 155 NR_LRU_BASE, 156 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ 157 NR_ACTIVE_ANON, /* " " " " " */ 158 NR_INACTIVE_FILE, /* " " " " " */ 159 NR_ACTIVE_FILE, /* " " " " " */ 160 NR_UNEVICTABLE, /* " " " " " */ 161 NR_SLAB_RECLAIMABLE_B, 162 NR_SLAB_UNRECLAIMABLE_B, 163 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ 164 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ 165 WORKINGSET_NODES, 166 WORKINGSET_REFAULT_BASE, 167 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE, 168 WORKINGSET_REFAULT_FILE, 169 WORKINGSET_ACTIVATE_BASE, 170 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE, 171 WORKINGSET_ACTIVATE_FILE, 172 WORKINGSET_RESTORE_BASE, 173 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE, 174 WORKINGSET_RESTORE_FILE, 175 WORKINGSET_NODERECLAIM, 176 NR_ANON_MAPPED, /* Mapped anonymous pages */ 177 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. 178 only modified from process context */ 179 NR_FILE_PAGES, 180 NR_FILE_DIRTY, 181 NR_WRITEBACK, 182 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ 183 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ 184 NR_SHMEM_THPS, 185 NR_SHMEM_PMDMAPPED, 186 NR_FILE_THPS, 187 NR_FILE_PMDMAPPED, 188 NR_ANON_THPS, 189 NR_VMSCAN_WRITE, 190 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ 191 NR_DIRTIED, /* page dirtyings since bootup */ 192 NR_WRITTEN, /* page writings since bootup */ 193 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */ 194 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */ 195 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */ 196 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */ 197 NR_KERNEL_STACK_KB, /* measured in KiB */ 198 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK) 199 NR_KERNEL_SCS_KB, /* measured in KiB */ 200 #endif 201 NR_PAGETABLE, /* used for pagetables */ 202 NR_SECONDARY_PAGETABLE, /* secondary pagetables, e.g. KVM pagetables */ 203 #ifdef CONFIG_SWAP 204 NR_SWAPCACHE, 205 #endif 206 #ifdef CONFIG_NUMA_BALANCING 207 PGPROMOTE_SUCCESS, /* promote successfully */ 208 PGPROMOTE_CANDIDATE, /* candidate pages to promote */ 209 #endif 210 NR_VM_NODE_STAT_ITEMS 211 }; 212 213 /* 214 * Returns true if the item should be printed in THPs (/proc/vmstat 215 * currently prints number of anon, file and shmem THPs. But the item 216 * is charged in pages). 217 */ 218 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item) 219 { 220 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) 221 return false; 222 223 return item == NR_ANON_THPS || 224 item == NR_FILE_THPS || 225 item == NR_SHMEM_THPS || 226 item == NR_SHMEM_PMDMAPPED || 227 item == NR_FILE_PMDMAPPED; 228 } 229 230 /* 231 * Returns true if the value is measured in bytes (most vmstat values are 232 * measured in pages). This defines the API part, the internal representation 233 * might be different. 234 */ 235 static __always_inline bool vmstat_item_in_bytes(int idx) 236 { 237 /* 238 * Global and per-node slab counters track slab pages. 239 * It's expected that changes are multiples of PAGE_SIZE. 240 * Internally values are stored in pages. 241 * 242 * Per-memcg and per-lruvec counters track memory, consumed 243 * by individual slab objects. These counters are actually 244 * byte-precise. 245 */ 246 return (idx == NR_SLAB_RECLAIMABLE_B || 247 idx == NR_SLAB_UNRECLAIMABLE_B); 248 } 249 250 /* 251 * We do arithmetic on the LRU lists in various places in the code, 252 * so it is important to keep the active lists LRU_ACTIVE higher in 253 * the array than the corresponding inactive lists, and to keep 254 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. 255 * 256 * This has to be kept in sync with the statistics in zone_stat_item 257 * above and the descriptions in vmstat_text in mm/vmstat.c 258 */ 259 #define LRU_BASE 0 260 #define LRU_ACTIVE 1 261 #define LRU_FILE 2 262 263 enum lru_list { 264 LRU_INACTIVE_ANON = LRU_BASE, 265 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, 266 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, 267 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, 268 LRU_UNEVICTABLE, 269 NR_LRU_LISTS 270 }; 271 272 enum vmscan_throttle_state { 273 VMSCAN_THROTTLE_WRITEBACK, 274 VMSCAN_THROTTLE_ISOLATED, 275 VMSCAN_THROTTLE_NOPROGRESS, 276 VMSCAN_THROTTLE_CONGESTED, 277 NR_VMSCAN_THROTTLE, 278 }; 279 280 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) 281 282 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) 283 284 static inline bool is_file_lru(enum lru_list lru) 285 { 286 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); 287 } 288 289 static inline bool is_active_lru(enum lru_list lru) 290 { 291 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); 292 } 293 294 #define WORKINGSET_ANON 0 295 #define WORKINGSET_FILE 1 296 #define ANON_AND_FILE 2 297 298 enum lruvec_flags { 299 /* 300 * An lruvec has many dirty pages backed by a congested BDI: 301 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim. 302 * It can be cleared by cgroup reclaim or kswapd. 303 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim. 304 * It can only be cleared by kswapd. 305 * 306 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup 307 * reclaim, but not vice versa. This only applies to the root cgroup. 308 * The goal is to prevent cgroup reclaim on the root cgroup (e.g. 309 * memory.reclaim) to unthrottle an unbalanced node (that was throttled 310 * by kswapd). 311 */ 312 LRUVEC_CGROUP_CONGESTED, 313 LRUVEC_NODE_CONGESTED, 314 }; 315 316 #endif /* !__GENERATING_BOUNDS_H */ 317 318 /* 319 * Evictable pages are divided into multiple generations. The youngest and the 320 * oldest generation numbers, max_seq and min_seq, are monotonically increasing. 321 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An 322 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the 323 * corresponding generation. The gen counter in folio->flags stores gen+1 while 324 * a page is on one of lrugen->folios[]. Otherwise it stores 0. 325 * 326 * A page is added to the youngest generation on faulting. The aging needs to 327 * check the accessed bit at least twice before handing this page over to the 328 * eviction. The first check takes care of the accessed bit set on the initial 329 * fault; the second check makes sure this page hasn't been used since then. 330 * This process, AKA second chance, requires a minimum of two generations, 331 * hence MIN_NR_GENS. And to maintain ABI compatibility with the active/inactive 332 * LRU, e.g., /proc/vmstat, these two generations are considered active; the 333 * rest of generations, if they exist, are considered inactive. See 334 * lru_gen_is_active(). 335 * 336 * PG_active is always cleared while a page is on one of lrugen->folios[] so 337 * that the aging needs not to worry about it. And it's set again when a page 338 * considered active is isolated for non-reclaiming purposes, e.g., migration. 339 * See lru_gen_add_folio() and lru_gen_del_folio(). 340 * 341 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the 342 * number of categories of the active/inactive LRU when keeping track of 343 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits 344 * in folio->flags. 345 */ 346 #define MIN_NR_GENS 2U 347 #define MAX_NR_GENS 4U 348 349 /* 350 * Each generation is divided into multiple tiers. A page accessed N times 351 * through file descriptors is in tier order_base_2(N). A page in the first tier 352 * (N=0,1) is marked by PG_referenced unless it was faulted in through page 353 * tables or read ahead. A page in any other tier (N>1) is marked by 354 * PG_referenced and PG_workingset. This implies a minimum of two tiers is 355 * supported without using additional bits in folio->flags. 356 * 357 * In contrast to moving across generations which requires the LRU lock, moving 358 * across tiers only involves atomic operations on folio->flags and therefore 359 * has a negligible cost in the buffered access path. In the eviction path, 360 * comparisons of refaulted/(evicted+protected) from the first tier and the 361 * rest infer whether pages accessed multiple times through file descriptors 362 * are statistically hot and thus worth protecting. 363 * 364 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the 365 * number of categories of the active/inactive LRU when keeping track of 366 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in 367 * folio->flags. 368 */ 369 #define MAX_NR_TIERS 4U 370 371 #ifndef __GENERATING_BOUNDS_H 372 373 struct lruvec; 374 struct page_vma_mapped_walk; 375 376 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF) 377 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF) 378 379 #ifdef CONFIG_LRU_GEN 380 381 enum { 382 LRU_GEN_ANON, 383 LRU_GEN_FILE, 384 }; 385 386 enum { 387 LRU_GEN_CORE, 388 LRU_GEN_MM_WALK, 389 LRU_GEN_NONLEAF_YOUNG, 390 NR_LRU_GEN_CAPS 391 }; 392 393 #define MIN_LRU_BATCH BITS_PER_LONG 394 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64) 395 396 /* whether to keep historical stats from evicted generations */ 397 #ifdef CONFIG_LRU_GEN_STATS 398 #define NR_HIST_GENS MAX_NR_GENS 399 #else 400 #define NR_HIST_GENS 1U 401 #endif 402 403 /* 404 * The youngest generation number is stored in max_seq for both anon and file 405 * types as they are aged on an equal footing. The oldest generation numbers are 406 * stored in min_seq[] separately for anon and file types as clean file pages 407 * can be evicted regardless of swap constraints. 408 * 409 * Normally anon and file min_seq are in sync. But if swapping is constrained, 410 * e.g., out of swap space, file min_seq is allowed to advance and leave anon 411 * min_seq behind. 412 * 413 * The number of pages in each generation is eventually consistent and therefore 414 * can be transiently negative when reset_batch_size() is pending. 415 */ 416 struct lru_gen_folio { 417 /* the aging increments the youngest generation number */ 418 unsigned long max_seq; 419 /* the eviction increments the oldest generation numbers */ 420 unsigned long min_seq[ANON_AND_FILE]; 421 /* the birth time of each generation in jiffies */ 422 unsigned long timestamps[MAX_NR_GENS]; 423 /* the multi-gen LRU lists, lazily sorted on eviction */ 424 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 425 /* the multi-gen LRU sizes, eventually consistent */ 426 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 427 /* the exponential moving average of refaulted */ 428 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS]; 429 /* the exponential moving average of evicted+protected */ 430 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS]; 431 /* the first tier doesn't need protection, hence the minus one */ 432 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS - 1]; 433 /* can be modified without holding the LRU lock */ 434 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS]; 435 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS]; 436 /* whether the multi-gen LRU is enabled */ 437 bool enabled; 438 #ifdef CONFIG_MEMCG 439 /* the memcg generation this lru_gen_folio belongs to */ 440 u8 gen; 441 /* the list segment this lru_gen_folio belongs to */ 442 u8 seg; 443 /* per-node lru_gen_folio list for global reclaim */ 444 struct hlist_nulls_node list; 445 #endif 446 }; 447 448 enum { 449 MM_LEAF_TOTAL, /* total leaf entries */ 450 MM_LEAF_OLD, /* old leaf entries */ 451 MM_LEAF_YOUNG, /* young leaf entries */ 452 MM_NONLEAF_TOTAL, /* total non-leaf entries */ 453 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */ 454 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */ 455 NR_MM_STATS 456 }; 457 458 /* double-buffering Bloom filters */ 459 #define NR_BLOOM_FILTERS 2 460 461 struct lru_gen_mm_state { 462 /* set to max_seq after each iteration */ 463 unsigned long seq; 464 /* where the current iteration continues after */ 465 struct list_head *head; 466 /* where the last iteration ended before */ 467 struct list_head *tail; 468 /* Bloom filters flip after each iteration */ 469 unsigned long *filters[NR_BLOOM_FILTERS]; 470 /* the mm stats for debugging */ 471 unsigned long stats[NR_HIST_GENS][NR_MM_STATS]; 472 }; 473 474 struct lru_gen_mm_walk { 475 /* the lruvec under reclaim */ 476 struct lruvec *lruvec; 477 /* unstable max_seq from lru_gen_folio */ 478 unsigned long max_seq; 479 /* the next address within an mm to scan */ 480 unsigned long next_addr; 481 /* to batch promoted pages */ 482 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 483 /* to batch the mm stats */ 484 int mm_stats[NR_MM_STATS]; 485 /* total batched items */ 486 int batched; 487 bool can_swap; 488 bool force_scan; 489 }; 490 491 void lru_gen_init_lruvec(struct lruvec *lruvec); 492 void lru_gen_look_around(struct page_vma_mapped_walk *pvmw); 493 494 #ifdef CONFIG_MEMCG 495 496 /* 497 * For each node, memcgs are divided into two generations: the old and the 498 * young. For each generation, memcgs are randomly sharded into multiple bins 499 * to improve scalability. For each bin, the hlist_nulls is virtually divided 500 * into three segments: the head, the tail and the default. 501 * 502 * An onlining memcg is added to the tail of a random bin in the old generation. 503 * The eviction starts at the head of a random bin in the old generation. The 504 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes 505 * the old generation, is incremented when all its bins become empty. 506 * 507 * There are four operations: 508 * 1. MEMCG_LRU_HEAD, which moves an memcg to the head of a random bin in its 509 * current generation (old or young) and updates its "seg" to "head"; 510 * 2. MEMCG_LRU_TAIL, which moves an memcg to the tail of a random bin in its 511 * current generation (old or young) and updates its "seg" to "tail"; 512 * 3. MEMCG_LRU_OLD, which moves an memcg to the head of a random bin in the old 513 * generation, updates its "gen" to "old" and resets its "seg" to "default"; 514 * 4. MEMCG_LRU_YOUNG, which moves an memcg to the tail of a random bin in the 515 * young generation, updates its "gen" to "young" and resets its "seg" to 516 * "default". 517 * 518 * The events that trigger the above operations are: 519 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD; 520 * 2. The first attempt to reclaim an memcg below low, which triggers 521 * MEMCG_LRU_TAIL; 522 * 3. The first attempt to reclaim an memcg below reclaimable size threshold, 523 * which triggers MEMCG_LRU_TAIL; 524 * 4. The second attempt to reclaim an memcg below reclaimable size threshold, 525 * which triggers MEMCG_LRU_YOUNG; 526 * 5. Attempting to reclaim an memcg below min, which triggers MEMCG_LRU_YOUNG; 527 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG; 528 * 7. Offlining an memcg, which triggers MEMCG_LRU_OLD. 529 * 530 * Note that memcg LRU only applies to global reclaim, and the round-robin 531 * incrementing of their max_seq counters ensures the eventual fairness to all 532 * eligible memcgs. For memcg reclaim, it still relies on mem_cgroup_iter(). 533 */ 534 #define MEMCG_NR_GENS 2 535 #define MEMCG_NR_BINS 8 536 537 struct lru_gen_memcg { 538 /* the per-node memcg generation counter */ 539 unsigned long seq; 540 /* each memcg has one lru_gen_folio per node */ 541 unsigned long nr_memcgs[MEMCG_NR_GENS]; 542 /* per-node lru_gen_folio list for global reclaim */ 543 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS]; 544 /* protects the above */ 545 spinlock_t lock; 546 }; 547 548 void lru_gen_init_pgdat(struct pglist_data *pgdat); 549 550 void lru_gen_init_memcg(struct mem_cgroup *memcg); 551 void lru_gen_exit_memcg(struct mem_cgroup *memcg); 552 void lru_gen_online_memcg(struct mem_cgroup *memcg); 553 void lru_gen_offline_memcg(struct mem_cgroup *memcg); 554 void lru_gen_release_memcg(struct mem_cgroup *memcg); 555 void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid); 556 557 #else /* !CONFIG_MEMCG */ 558 559 #define MEMCG_NR_GENS 1 560 561 struct lru_gen_memcg { 562 }; 563 564 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat) 565 { 566 } 567 568 #endif /* CONFIG_MEMCG */ 569 570 #else /* !CONFIG_LRU_GEN */ 571 572 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat) 573 { 574 } 575 576 static inline void lru_gen_init_lruvec(struct lruvec *lruvec) 577 { 578 } 579 580 static inline void lru_gen_look_around(struct page_vma_mapped_walk *pvmw) 581 { 582 } 583 584 #ifdef CONFIG_MEMCG 585 586 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg) 587 { 588 } 589 590 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg) 591 { 592 } 593 594 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg) 595 { 596 } 597 598 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg) 599 { 600 } 601 602 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg) 603 { 604 } 605 606 static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid) 607 { 608 } 609 610 #endif /* CONFIG_MEMCG */ 611 612 #endif /* CONFIG_LRU_GEN */ 613 614 struct lruvec { 615 struct list_head lists[NR_LRU_LISTS]; 616 /* per lruvec lru_lock for memcg */ 617 spinlock_t lru_lock; 618 /* 619 * These track the cost of reclaiming one LRU - file or anon - 620 * over the other. As the observed cost of reclaiming one LRU 621 * increases, the reclaim scan balance tips toward the other. 622 */ 623 unsigned long anon_cost; 624 unsigned long file_cost; 625 /* Non-resident age, driven by LRU movement */ 626 atomic_long_t nonresident_age; 627 /* Refaults at the time of last reclaim cycle */ 628 unsigned long refaults[ANON_AND_FILE]; 629 /* Various lruvec state flags (enum lruvec_flags) */ 630 unsigned long flags; 631 #ifdef CONFIG_LRU_GEN 632 /* evictable pages divided into generations */ 633 struct lru_gen_folio lrugen; 634 /* to concurrently iterate lru_gen_mm_list */ 635 struct lru_gen_mm_state mm_state; 636 #endif 637 #ifdef CONFIG_MEMCG 638 struct pglist_data *pgdat; 639 #endif 640 }; 641 642 /* Isolate unmapped pages */ 643 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2) 644 /* Isolate for asynchronous migration */ 645 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) 646 /* Isolate unevictable pages */ 647 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) 648 649 /* LRU Isolation modes. */ 650 typedef unsigned __bitwise isolate_mode_t; 651 652 enum zone_watermarks { 653 WMARK_MIN, 654 WMARK_LOW, 655 WMARK_HIGH, 656 WMARK_PROMO, 657 NR_WMARK 658 }; 659 660 /* 661 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. One additional list 662 * for THP which will usually be GFP_MOVABLE. Even if it is another type, 663 * it should not contribute to serious fragmentation causing THP allocation 664 * failures. 665 */ 666 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 667 #define NR_PCP_THP 1 668 #else 669 #define NR_PCP_THP 0 670 #endif 671 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1)) 672 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP) 673 674 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost) 675 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost) 676 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost) 677 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost) 678 679 /* Fields and list protected by pagesets local_lock in page_alloc.c */ 680 struct per_cpu_pages { 681 spinlock_t lock; /* Protects lists field */ 682 int count; /* number of pages in the list */ 683 int high; /* high watermark, emptying needed */ 684 int batch; /* chunk size for buddy add/remove */ 685 short free_factor; /* batch scaling factor during free */ 686 #ifdef CONFIG_NUMA 687 short expire; /* When 0, remote pagesets are drained */ 688 #endif 689 690 /* Lists of pages, one per migrate type stored on the pcp-lists */ 691 struct list_head lists[NR_PCP_LISTS]; 692 } ____cacheline_aligned_in_smp; 693 694 struct per_cpu_zonestat { 695 #ifdef CONFIG_SMP 696 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; 697 s8 stat_threshold; 698 #endif 699 #ifdef CONFIG_NUMA 700 /* 701 * Low priority inaccurate counters that are only folded 702 * on demand. Use a large type to avoid the overhead of 703 * folding during refresh_cpu_vm_stats. 704 */ 705 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS]; 706 #endif 707 }; 708 709 struct per_cpu_nodestat { 710 s8 stat_threshold; 711 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS]; 712 }; 713 714 #endif /* !__GENERATING_BOUNDS.H */ 715 716 enum zone_type { 717 /* 718 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able 719 * to DMA to all of the addressable memory (ZONE_NORMAL). 720 * On architectures where this area covers the whole 32 bit address 721 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller 722 * DMA addressing constraints. This distinction is important as a 32bit 723 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit 724 * platforms may need both zones as they support peripherals with 725 * different DMA addressing limitations. 726 */ 727 #ifdef CONFIG_ZONE_DMA 728 ZONE_DMA, 729 #endif 730 #ifdef CONFIG_ZONE_DMA32 731 ZONE_DMA32, 732 #endif 733 /* 734 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be 735 * performed on pages in ZONE_NORMAL if the DMA devices support 736 * transfers to all addressable memory. 737 */ 738 ZONE_NORMAL, 739 #ifdef CONFIG_HIGHMEM 740 /* 741 * A memory area that is only addressable by the kernel through 742 * mapping portions into its own address space. This is for example 743 * used by i386 to allow the kernel to address the memory beyond 744 * 900MB. The kernel will set up special mappings (page 745 * table entries on i386) for each page that the kernel needs to 746 * access. 747 */ 748 ZONE_HIGHMEM, 749 #endif 750 /* 751 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains 752 * movable pages with few exceptional cases described below. Main use 753 * cases for ZONE_MOVABLE are to make memory offlining/unplug more 754 * likely to succeed, and to locally limit unmovable allocations - e.g., 755 * to increase the number of THP/huge pages. Notable special cases are: 756 * 757 * 1. Pinned pages: (long-term) pinning of movable pages might 758 * essentially turn such pages unmovable. Therefore, we do not allow 759 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and 760 * faulted, they come from the right zone right away. However, it is 761 * still possible that address space already has pages in 762 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has 763 * touches that memory before pinning). In such case we migrate them 764 * to a different zone. When migration fails - pinning fails. 765 * 2. memblock allocations: kernelcore/movablecore setups might create 766 * situations where ZONE_MOVABLE contains unmovable allocations 767 * after boot. Memory offlining and allocations fail early. 768 * 3. Memory holes: kernelcore/movablecore setups might create very rare 769 * situations where ZONE_MOVABLE contains memory holes after boot, 770 * for example, if we have sections that are only partially 771 * populated. Memory offlining and allocations fail early. 772 * 4. PG_hwpoison pages: while poisoned pages can be skipped during 773 * memory offlining, such pages cannot be allocated. 774 * 5. Unmovable PG_offline pages: in paravirtualized environments, 775 * hotplugged memory blocks might only partially be managed by the 776 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The 777 * parts not manged by the buddy are unmovable PG_offline pages. In 778 * some cases (virtio-mem), such pages can be skipped during 779 * memory offlining, however, cannot be moved/allocated. These 780 * techniques might use alloc_contig_range() to hide previously 781 * exposed pages from the buddy again (e.g., to implement some sort 782 * of memory unplug in virtio-mem). 783 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create 784 * situations where ZERO_PAGE(0) which is allocated differently 785 * on different platforms may end up in a movable zone. ZERO_PAGE(0) 786 * cannot be migrated. 787 * 7. Memory-hotplug: when using memmap_on_memory and onlining the 788 * memory to the MOVABLE zone, the vmemmap pages are also placed in 789 * such zone. Such pages cannot be really moved around as they are 790 * self-stored in the range, but they are treated as movable when 791 * the range they describe is about to be offlined. 792 * 793 * In general, no unmovable allocations that degrade memory offlining 794 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range()) 795 * have to expect that migrating pages in ZONE_MOVABLE can fail (even 796 * if has_unmovable_pages() states that there are no unmovable pages, 797 * there can be false negatives). 798 */ 799 ZONE_MOVABLE, 800 #ifdef CONFIG_ZONE_DEVICE 801 ZONE_DEVICE, 802 #endif 803 __MAX_NR_ZONES 804 805 }; 806 807 #ifndef __GENERATING_BOUNDS_H 808 809 #define ASYNC_AND_SYNC 2 810 811 struct zone { 812 /* Read-mostly fields */ 813 814 /* zone watermarks, access with *_wmark_pages(zone) macros */ 815 unsigned long _watermark[NR_WMARK]; 816 unsigned long watermark_boost; 817 818 unsigned long nr_reserved_highatomic; 819 820 /* 821 * We don't know if the memory that we're going to allocate will be 822 * freeable or/and it will be released eventually, so to avoid totally 823 * wasting several GB of ram we must reserve some of the lower zone 824 * memory (otherwise we risk to run OOM on the lower zones despite 825 * there being tons of freeable ram on the higher zones). This array is 826 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl 827 * changes. 828 */ 829 long lowmem_reserve[MAX_NR_ZONES]; 830 831 #ifdef CONFIG_NUMA 832 int node; 833 #endif 834 struct pglist_data *zone_pgdat; 835 struct per_cpu_pages __percpu *per_cpu_pageset; 836 struct per_cpu_zonestat __percpu *per_cpu_zonestats; 837 /* 838 * the high and batch values are copied to individual pagesets for 839 * faster access 840 */ 841 int pageset_high; 842 int pageset_batch; 843 844 #ifndef CONFIG_SPARSEMEM 845 /* 846 * Flags for a pageblock_nr_pages block. See pageblock-flags.h. 847 * In SPARSEMEM, this map is stored in struct mem_section 848 */ 849 unsigned long *pageblock_flags; 850 #endif /* CONFIG_SPARSEMEM */ 851 852 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ 853 unsigned long zone_start_pfn; 854 855 /* 856 * spanned_pages is the total pages spanned by the zone, including 857 * holes, which is calculated as: 858 * spanned_pages = zone_end_pfn - zone_start_pfn; 859 * 860 * present_pages is physical pages existing within the zone, which 861 * is calculated as: 862 * present_pages = spanned_pages - absent_pages(pages in holes); 863 * 864 * present_early_pages is present pages existing within the zone 865 * located on memory available since early boot, excluding hotplugged 866 * memory. 867 * 868 * managed_pages is present pages managed by the buddy system, which 869 * is calculated as (reserved_pages includes pages allocated by the 870 * bootmem allocator): 871 * managed_pages = present_pages - reserved_pages; 872 * 873 * cma pages is present pages that are assigned for CMA use 874 * (MIGRATE_CMA). 875 * 876 * So present_pages may be used by memory hotplug or memory power 877 * management logic to figure out unmanaged pages by checking 878 * (present_pages - managed_pages). And managed_pages should be used 879 * by page allocator and vm scanner to calculate all kinds of watermarks 880 * and thresholds. 881 * 882 * Locking rules: 883 * 884 * zone_start_pfn and spanned_pages are protected by span_seqlock. 885 * It is a seqlock because it has to be read outside of zone->lock, 886 * and it is done in the main allocator path. But, it is written 887 * quite infrequently. 888 * 889 * The span_seq lock is declared along with zone->lock because it is 890 * frequently read in proximity to zone->lock. It's good to 891 * give them a chance of being in the same cacheline. 892 * 893 * Write access to present_pages at runtime should be protected by 894 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of 895 * present_pages should use get_online_mems() to get a stable value. 896 */ 897 atomic_long_t managed_pages; 898 unsigned long spanned_pages; 899 unsigned long present_pages; 900 #if defined(CONFIG_MEMORY_HOTPLUG) 901 unsigned long present_early_pages; 902 #endif 903 #ifdef CONFIG_CMA 904 unsigned long cma_pages; 905 #endif 906 907 const char *name; 908 909 #ifdef CONFIG_MEMORY_ISOLATION 910 /* 911 * Number of isolated pageblock. It is used to solve incorrect 912 * freepage counting problem due to racy retrieving migratetype 913 * of pageblock. Protected by zone->lock. 914 */ 915 unsigned long nr_isolate_pageblock; 916 #endif 917 918 #ifdef CONFIG_MEMORY_HOTPLUG 919 /* see spanned/present_pages for more description */ 920 seqlock_t span_seqlock; 921 #endif 922 923 int initialized; 924 925 /* Write-intensive fields used from the page allocator */ 926 CACHELINE_PADDING(_pad1_); 927 928 /* free areas of different sizes */ 929 struct free_area free_area[MAX_ORDER + 1]; 930 931 #ifdef CONFIG_UNACCEPTED_MEMORY 932 /* Pages to be accepted. All pages on the list are MAX_ORDER */ 933 struct list_head unaccepted_pages; 934 #endif 935 936 /* zone flags, see below */ 937 unsigned long flags; 938 939 /* Primarily protects free_area */ 940 spinlock_t lock; 941 942 /* Write-intensive fields used by compaction and vmstats. */ 943 CACHELINE_PADDING(_pad2_); 944 945 /* 946 * When free pages are below this point, additional steps are taken 947 * when reading the number of free pages to avoid per-cpu counter 948 * drift allowing watermarks to be breached 949 */ 950 unsigned long percpu_drift_mark; 951 952 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 953 /* pfn where compaction free scanner should start */ 954 unsigned long compact_cached_free_pfn; 955 /* pfn where compaction migration scanner should start */ 956 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC]; 957 unsigned long compact_init_migrate_pfn; 958 unsigned long compact_init_free_pfn; 959 #endif 960 961 #ifdef CONFIG_COMPACTION 962 /* 963 * On compaction failure, 1<<compact_defer_shift compactions 964 * are skipped before trying again. The number attempted since 965 * last failure is tracked with compact_considered. 966 * compact_order_failed is the minimum compaction failed order. 967 */ 968 unsigned int compact_considered; 969 unsigned int compact_defer_shift; 970 int compact_order_failed; 971 #endif 972 973 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 974 /* Set to true when the PG_migrate_skip bits should be cleared */ 975 bool compact_blockskip_flush; 976 #endif 977 978 bool contiguous; 979 980 CACHELINE_PADDING(_pad3_); 981 /* Zone statistics */ 982 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; 983 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS]; 984 } ____cacheline_internodealigned_in_smp; 985 986 enum pgdat_flags { 987 PGDAT_DIRTY, /* reclaim scanning has recently found 988 * many dirty file pages at the tail 989 * of the LRU. 990 */ 991 PGDAT_WRITEBACK, /* reclaim scanning has recently found 992 * many pages under writeback 993 */ 994 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */ 995 }; 996 997 enum zone_flags { 998 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks. 999 * Cleared when kswapd is woken. 1000 */ 1001 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */ 1002 }; 1003 1004 static inline unsigned long zone_managed_pages(struct zone *zone) 1005 { 1006 return (unsigned long)atomic_long_read(&zone->managed_pages); 1007 } 1008 1009 static inline unsigned long zone_cma_pages(struct zone *zone) 1010 { 1011 #ifdef CONFIG_CMA 1012 return zone->cma_pages; 1013 #else 1014 return 0; 1015 #endif 1016 } 1017 1018 static inline unsigned long zone_end_pfn(const struct zone *zone) 1019 { 1020 return zone->zone_start_pfn + zone->spanned_pages; 1021 } 1022 1023 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn) 1024 { 1025 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone); 1026 } 1027 1028 static inline bool zone_is_initialized(struct zone *zone) 1029 { 1030 return zone->initialized; 1031 } 1032 1033 static inline bool zone_is_empty(struct zone *zone) 1034 { 1035 return zone->spanned_pages == 0; 1036 } 1037 1038 #ifndef BUILD_VDSO32_64 1039 /* 1040 * The zone field is never updated after free_area_init_core() 1041 * sets it, so none of the operations on it need to be atomic. 1042 */ 1043 1044 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */ 1045 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH) 1046 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH) 1047 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH) 1048 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH) 1049 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH) 1050 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH) 1051 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH) 1052 1053 /* 1054 * Define the bit shifts to access each section. For non-existent 1055 * sections we define the shift as 0; that plus a 0 mask ensures 1056 * the compiler will optimise away reference to them. 1057 */ 1058 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0)) 1059 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0)) 1060 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0)) 1061 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0)) 1062 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0)) 1063 1064 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */ 1065 #ifdef NODE_NOT_IN_PAGE_FLAGS 1066 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT) 1067 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \ 1068 SECTIONS_PGOFF : ZONES_PGOFF) 1069 #else 1070 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT) 1071 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \ 1072 NODES_PGOFF : ZONES_PGOFF) 1073 #endif 1074 1075 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0)) 1076 1077 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1) 1078 #define NODES_MASK ((1UL << NODES_WIDTH) - 1) 1079 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1) 1080 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1) 1081 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1) 1082 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1) 1083 1084 static inline enum zone_type page_zonenum(const struct page *page) 1085 { 1086 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT); 1087 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK; 1088 } 1089 1090 static inline enum zone_type folio_zonenum(const struct folio *folio) 1091 { 1092 return page_zonenum(&folio->page); 1093 } 1094 1095 #ifdef CONFIG_ZONE_DEVICE 1096 static inline bool is_zone_device_page(const struct page *page) 1097 { 1098 return page_zonenum(page) == ZONE_DEVICE; 1099 } 1100 1101 /* 1102 * Consecutive zone device pages should not be merged into the same sgl 1103 * or bvec segment with other types of pages or if they belong to different 1104 * pgmaps. Otherwise getting the pgmap of a given segment is not possible 1105 * without scanning the entire segment. This helper returns true either if 1106 * both pages are not zone device pages or both pages are zone device pages 1107 * with the same pgmap. 1108 */ 1109 static inline bool zone_device_pages_have_same_pgmap(const struct page *a, 1110 const struct page *b) 1111 { 1112 if (is_zone_device_page(a) != is_zone_device_page(b)) 1113 return false; 1114 if (!is_zone_device_page(a)) 1115 return true; 1116 return a->pgmap == b->pgmap; 1117 } 1118 1119 extern void memmap_init_zone_device(struct zone *, unsigned long, 1120 unsigned long, struct dev_pagemap *); 1121 #else 1122 static inline bool is_zone_device_page(const struct page *page) 1123 { 1124 return false; 1125 } 1126 static inline bool zone_device_pages_have_same_pgmap(const struct page *a, 1127 const struct page *b) 1128 { 1129 return true; 1130 } 1131 #endif 1132 1133 static inline bool folio_is_zone_device(const struct folio *folio) 1134 { 1135 return is_zone_device_page(&folio->page); 1136 } 1137 1138 static inline bool is_zone_movable_page(const struct page *page) 1139 { 1140 return page_zonenum(page) == ZONE_MOVABLE; 1141 } 1142 1143 static inline bool folio_is_zone_movable(const struct folio *folio) 1144 { 1145 return folio_zonenum(folio) == ZONE_MOVABLE; 1146 } 1147 #endif 1148 1149 /* 1150 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty 1151 * intersection with the given zone 1152 */ 1153 static inline bool zone_intersects(struct zone *zone, 1154 unsigned long start_pfn, unsigned long nr_pages) 1155 { 1156 if (zone_is_empty(zone)) 1157 return false; 1158 if (start_pfn >= zone_end_pfn(zone) || 1159 start_pfn + nr_pages <= zone->zone_start_pfn) 1160 return false; 1161 1162 return true; 1163 } 1164 1165 /* 1166 * The "priority" of VM scanning is how much of the queues we will scan in one 1167 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the 1168 * queues ("queue_length >> 12") during an aging round. 1169 */ 1170 #define DEF_PRIORITY 12 1171 1172 /* Maximum number of zones on a zonelist */ 1173 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) 1174 1175 enum { 1176 ZONELIST_FALLBACK, /* zonelist with fallback */ 1177 #ifdef CONFIG_NUMA 1178 /* 1179 * The NUMA zonelists are doubled because we need zonelists that 1180 * restrict the allocations to a single node for __GFP_THISNODE. 1181 */ 1182 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */ 1183 #endif 1184 MAX_ZONELISTS 1185 }; 1186 1187 /* 1188 * This struct contains information about a zone in a zonelist. It is stored 1189 * here to avoid dereferences into large structures and lookups of tables 1190 */ 1191 struct zoneref { 1192 struct zone *zone; /* Pointer to actual zone */ 1193 int zone_idx; /* zone_idx(zoneref->zone) */ 1194 }; 1195 1196 /* 1197 * One allocation request operates on a zonelist. A zonelist 1198 * is a list of zones, the first one is the 'goal' of the 1199 * allocation, the other zones are fallback zones, in decreasing 1200 * priority. 1201 * 1202 * To speed the reading of the zonelist, the zonerefs contain the zone index 1203 * of the entry being read. Helper functions to access information given 1204 * a struct zoneref are 1205 * 1206 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs 1207 * zonelist_zone_idx() - Return the index of the zone for an entry 1208 * zonelist_node_idx() - Return the index of the node for an entry 1209 */ 1210 struct zonelist { 1211 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; 1212 }; 1213 1214 /* 1215 * The array of struct pages for flatmem. 1216 * It must be declared for SPARSEMEM as well because there are configurations 1217 * that rely on that. 1218 */ 1219 extern struct page *mem_map; 1220 1221 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1222 struct deferred_split { 1223 spinlock_t split_queue_lock; 1224 struct list_head split_queue; 1225 unsigned long split_queue_len; 1226 }; 1227 #endif 1228 1229 #ifdef CONFIG_MEMORY_FAILURE 1230 /* 1231 * Per NUMA node memory failure handling statistics. 1232 */ 1233 struct memory_failure_stats { 1234 /* 1235 * Number of raw pages poisoned. 1236 * Cases not accounted: memory outside kernel control, offline page, 1237 * arch-specific memory_failure (SGX), hwpoison_filter() filtered 1238 * error events, and unpoison actions from hwpoison_unpoison. 1239 */ 1240 unsigned long total; 1241 /* 1242 * Recovery results of poisoned raw pages handled by memory_failure, 1243 * in sync with mf_result. 1244 * total = ignored + failed + delayed + recovered. 1245 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted. 1246 */ 1247 unsigned long ignored; 1248 unsigned long failed; 1249 unsigned long delayed; 1250 unsigned long recovered; 1251 }; 1252 #endif 1253 1254 /* 1255 * On NUMA machines, each NUMA node would have a pg_data_t to describe 1256 * it's memory layout. On UMA machines there is a single pglist_data which 1257 * describes the whole memory. 1258 * 1259 * Memory statistics and page replacement data structures are maintained on a 1260 * per-zone basis. 1261 */ 1262 typedef struct pglist_data { 1263 /* 1264 * node_zones contains just the zones for THIS node. Not all of the 1265 * zones may be populated, but it is the full list. It is referenced by 1266 * this node's node_zonelists as well as other node's node_zonelists. 1267 */ 1268 struct zone node_zones[MAX_NR_ZONES]; 1269 1270 /* 1271 * node_zonelists contains references to all zones in all nodes. 1272 * Generally the first zones will be references to this node's 1273 * node_zones. 1274 */ 1275 struct zonelist node_zonelists[MAX_ZONELISTS]; 1276 1277 int nr_zones; /* number of populated zones in this node */ 1278 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */ 1279 struct page *node_mem_map; 1280 #ifdef CONFIG_PAGE_EXTENSION 1281 struct page_ext *node_page_ext; 1282 #endif 1283 #endif 1284 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT) 1285 /* 1286 * Must be held any time you expect node_start_pfn, 1287 * node_present_pages, node_spanned_pages or nr_zones to stay constant. 1288 * Also synchronizes pgdat->first_deferred_pfn during deferred page 1289 * init. 1290 * 1291 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to 1292 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG 1293 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT. 1294 * 1295 * Nests above zone->lock and zone->span_seqlock 1296 */ 1297 spinlock_t node_size_lock; 1298 #endif 1299 unsigned long node_start_pfn; 1300 unsigned long node_present_pages; /* total number of physical pages */ 1301 unsigned long node_spanned_pages; /* total size of physical page 1302 range, including holes */ 1303 int node_id; 1304 wait_queue_head_t kswapd_wait; 1305 wait_queue_head_t pfmemalloc_wait; 1306 1307 /* workqueues for throttling reclaim for different reasons. */ 1308 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE]; 1309 1310 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */ 1311 unsigned long nr_reclaim_start; /* nr pages written while throttled 1312 * when throttling started. */ 1313 #ifdef CONFIG_MEMORY_HOTPLUG 1314 struct mutex kswapd_lock; 1315 #endif 1316 struct task_struct *kswapd; /* Protected by kswapd_lock */ 1317 int kswapd_order; 1318 enum zone_type kswapd_highest_zoneidx; 1319 1320 int kswapd_failures; /* Number of 'reclaimed == 0' runs */ 1321 1322 #ifdef CONFIG_COMPACTION 1323 int kcompactd_max_order; 1324 enum zone_type kcompactd_highest_zoneidx; 1325 wait_queue_head_t kcompactd_wait; 1326 struct task_struct *kcompactd; 1327 bool proactive_compact_trigger; 1328 #endif 1329 /* 1330 * This is a per-node reserve of pages that are not available 1331 * to userspace allocations. 1332 */ 1333 unsigned long totalreserve_pages; 1334 1335 #ifdef CONFIG_NUMA 1336 /* 1337 * node reclaim becomes active if more unmapped pages exist. 1338 */ 1339 unsigned long min_unmapped_pages; 1340 unsigned long min_slab_pages; 1341 #endif /* CONFIG_NUMA */ 1342 1343 /* Write-intensive fields used by page reclaim */ 1344 CACHELINE_PADDING(_pad1_); 1345 1346 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1347 /* 1348 * If memory initialisation on large machines is deferred then this 1349 * is the first PFN that needs to be initialised. 1350 */ 1351 unsigned long first_deferred_pfn; 1352 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1353 1354 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1355 struct deferred_split deferred_split_queue; 1356 #endif 1357 1358 #ifdef CONFIG_NUMA_BALANCING 1359 /* start time in ms of current promote rate limit period */ 1360 unsigned int nbp_rl_start; 1361 /* number of promote candidate pages at start time of current rate limit period */ 1362 unsigned long nbp_rl_nr_cand; 1363 /* promote threshold in ms */ 1364 unsigned int nbp_threshold; 1365 /* start time in ms of current promote threshold adjustment period */ 1366 unsigned int nbp_th_start; 1367 /* 1368 * number of promote candidate pages at start time of current promote 1369 * threshold adjustment period 1370 */ 1371 unsigned long nbp_th_nr_cand; 1372 #endif 1373 /* Fields commonly accessed by the page reclaim scanner */ 1374 1375 /* 1376 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED. 1377 * 1378 * Use mem_cgroup_lruvec() to look up lruvecs. 1379 */ 1380 struct lruvec __lruvec; 1381 1382 unsigned long flags; 1383 1384 #ifdef CONFIG_LRU_GEN 1385 /* kswap mm walk data */ 1386 struct lru_gen_mm_walk mm_walk; 1387 /* lru_gen_folio list */ 1388 struct lru_gen_memcg memcg_lru; 1389 #endif 1390 1391 CACHELINE_PADDING(_pad2_); 1392 1393 /* Per-node vmstats */ 1394 struct per_cpu_nodestat __percpu *per_cpu_nodestats; 1395 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS]; 1396 #ifdef CONFIG_NUMA 1397 struct memory_tier __rcu *memtier; 1398 #endif 1399 #ifdef CONFIG_MEMORY_FAILURE 1400 struct memory_failure_stats mf_stats; 1401 #endif 1402 } pg_data_t; 1403 1404 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) 1405 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) 1406 1407 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) 1408 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid)) 1409 1410 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat) 1411 { 1412 return pgdat->node_start_pfn + pgdat->node_spanned_pages; 1413 } 1414 1415 #include <linux/memory_hotplug.h> 1416 1417 void build_all_zonelists(pg_data_t *pgdat); 1418 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order, 1419 enum zone_type highest_zoneidx); 1420 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, 1421 int highest_zoneidx, unsigned int alloc_flags, 1422 long free_pages); 1423 bool zone_watermark_ok(struct zone *z, unsigned int order, 1424 unsigned long mark, int highest_zoneidx, 1425 unsigned int alloc_flags); 1426 bool zone_watermark_ok_safe(struct zone *z, unsigned int order, 1427 unsigned long mark, int highest_zoneidx); 1428 /* 1429 * Memory initialization context, use to differentiate memory added by 1430 * the platform statically or via memory hotplug interface. 1431 */ 1432 enum meminit_context { 1433 MEMINIT_EARLY, 1434 MEMINIT_HOTPLUG, 1435 }; 1436 1437 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, 1438 unsigned long size); 1439 1440 extern void lruvec_init(struct lruvec *lruvec); 1441 1442 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec) 1443 { 1444 #ifdef CONFIG_MEMCG 1445 return lruvec->pgdat; 1446 #else 1447 return container_of(lruvec, struct pglist_data, __lruvec); 1448 #endif 1449 } 1450 1451 #ifdef CONFIG_HAVE_MEMORYLESS_NODES 1452 int local_memory_node(int node_id); 1453 #else 1454 static inline int local_memory_node(int node_id) { return node_id; }; 1455 #endif 1456 1457 /* 1458 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. 1459 */ 1460 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) 1461 1462 #ifdef CONFIG_ZONE_DEVICE 1463 static inline bool zone_is_zone_device(struct zone *zone) 1464 { 1465 return zone_idx(zone) == ZONE_DEVICE; 1466 } 1467 #else 1468 static inline bool zone_is_zone_device(struct zone *zone) 1469 { 1470 return false; 1471 } 1472 #endif 1473 1474 /* 1475 * Returns true if a zone has pages managed by the buddy allocator. 1476 * All the reclaim decisions have to use this function rather than 1477 * populated_zone(). If the whole zone is reserved then we can easily 1478 * end up with populated_zone() && !managed_zone(). 1479 */ 1480 static inline bool managed_zone(struct zone *zone) 1481 { 1482 return zone_managed_pages(zone); 1483 } 1484 1485 /* Returns true if a zone has memory */ 1486 static inline bool populated_zone(struct zone *zone) 1487 { 1488 return zone->present_pages; 1489 } 1490 1491 #ifdef CONFIG_NUMA 1492 static inline int zone_to_nid(struct zone *zone) 1493 { 1494 return zone->node; 1495 } 1496 1497 static inline void zone_set_nid(struct zone *zone, int nid) 1498 { 1499 zone->node = nid; 1500 } 1501 #else 1502 static inline int zone_to_nid(struct zone *zone) 1503 { 1504 return 0; 1505 } 1506 1507 static inline void zone_set_nid(struct zone *zone, int nid) {} 1508 #endif 1509 1510 extern int movable_zone; 1511 1512 static inline int is_highmem_idx(enum zone_type idx) 1513 { 1514 #ifdef CONFIG_HIGHMEM 1515 return (idx == ZONE_HIGHMEM || 1516 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM)); 1517 #else 1518 return 0; 1519 #endif 1520 } 1521 1522 /** 1523 * is_highmem - helper function to quickly check if a struct zone is a 1524 * highmem zone or not. This is an attempt to keep references 1525 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. 1526 * @zone: pointer to struct zone variable 1527 * Return: 1 for a highmem zone, 0 otherwise 1528 */ 1529 static inline int is_highmem(struct zone *zone) 1530 { 1531 return is_highmem_idx(zone_idx(zone)); 1532 } 1533 1534 #ifdef CONFIG_ZONE_DMA 1535 bool has_managed_dma(void); 1536 #else 1537 static inline bool has_managed_dma(void) 1538 { 1539 return false; 1540 } 1541 #endif 1542 1543 1544 #ifndef CONFIG_NUMA 1545 1546 extern struct pglist_data contig_page_data; 1547 static inline struct pglist_data *NODE_DATA(int nid) 1548 { 1549 return &contig_page_data; 1550 } 1551 1552 #else /* CONFIG_NUMA */ 1553 1554 #include <asm/mmzone.h> 1555 1556 #endif /* !CONFIG_NUMA */ 1557 1558 extern struct pglist_data *first_online_pgdat(void); 1559 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); 1560 extern struct zone *next_zone(struct zone *zone); 1561 1562 /** 1563 * for_each_online_pgdat - helper macro to iterate over all online nodes 1564 * @pgdat: pointer to a pg_data_t variable 1565 */ 1566 #define for_each_online_pgdat(pgdat) \ 1567 for (pgdat = first_online_pgdat(); \ 1568 pgdat; \ 1569 pgdat = next_online_pgdat(pgdat)) 1570 /** 1571 * for_each_zone - helper macro to iterate over all memory zones 1572 * @zone: pointer to struct zone variable 1573 * 1574 * The user only needs to declare the zone variable, for_each_zone 1575 * fills it in. 1576 */ 1577 #define for_each_zone(zone) \ 1578 for (zone = (first_online_pgdat())->node_zones; \ 1579 zone; \ 1580 zone = next_zone(zone)) 1581 1582 #define for_each_populated_zone(zone) \ 1583 for (zone = (first_online_pgdat())->node_zones; \ 1584 zone; \ 1585 zone = next_zone(zone)) \ 1586 if (!populated_zone(zone)) \ 1587 ; /* do nothing */ \ 1588 else 1589 1590 static inline struct zone *zonelist_zone(struct zoneref *zoneref) 1591 { 1592 return zoneref->zone; 1593 } 1594 1595 static inline int zonelist_zone_idx(struct zoneref *zoneref) 1596 { 1597 return zoneref->zone_idx; 1598 } 1599 1600 static inline int zonelist_node_idx(struct zoneref *zoneref) 1601 { 1602 return zone_to_nid(zoneref->zone); 1603 } 1604 1605 struct zoneref *__next_zones_zonelist(struct zoneref *z, 1606 enum zone_type highest_zoneidx, 1607 nodemask_t *nodes); 1608 1609 /** 1610 * next_zones_zonelist - Returns the next zone at or below highest_zoneidx within the allowed nodemask using a cursor within a zonelist as a starting point 1611 * @z: The cursor used as a starting point for the search 1612 * @highest_zoneidx: The zone index of the highest zone to return 1613 * @nodes: An optional nodemask to filter the zonelist with 1614 * 1615 * This function returns the next zone at or below a given zone index that is 1616 * within the allowed nodemask using a cursor as the starting point for the 1617 * search. The zoneref returned is a cursor that represents the current zone 1618 * being examined. It should be advanced by one before calling 1619 * next_zones_zonelist again. 1620 * 1621 * Return: the next zone at or below highest_zoneidx within the allowed 1622 * nodemask using a cursor within a zonelist as a starting point 1623 */ 1624 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z, 1625 enum zone_type highest_zoneidx, 1626 nodemask_t *nodes) 1627 { 1628 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx)) 1629 return z; 1630 return __next_zones_zonelist(z, highest_zoneidx, nodes); 1631 } 1632 1633 /** 1634 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist 1635 * @zonelist: The zonelist to search for a suitable zone 1636 * @highest_zoneidx: The zone index of the highest zone to return 1637 * @nodes: An optional nodemask to filter the zonelist with 1638 * 1639 * This function returns the first zone at or below a given zone index that is 1640 * within the allowed nodemask. The zoneref returned is a cursor that can be 1641 * used to iterate the zonelist with next_zones_zonelist by advancing it by 1642 * one before calling. 1643 * 1644 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is 1645 * never NULL). This may happen either genuinely, or due to concurrent nodemask 1646 * update due to cpuset modification. 1647 * 1648 * Return: Zoneref pointer for the first suitable zone found 1649 */ 1650 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, 1651 enum zone_type highest_zoneidx, 1652 nodemask_t *nodes) 1653 { 1654 return next_zones_zonelist(zonelist->_zonerefs, 1655 highest_zoneidx, nodes); 1656 } 1657 1658 /** 1659 * for_each_zone_zonelist_nodemask - helper macro to iterate over valid zones in a zonelist at or below a given zone index and within a nodemask 1660 * @zone: The current zone in the iterator 1661 * @z: The current pointer within zonelist->_zonerefs being iterated 1662 * @zlist: The zonelist being iterated 1663 * @highidx: The zone index of the highest zone to return 1664 * @nodemask: Nodemask allowed by the allocator 1665 * 1666 * This iterator iterates though all zones at or below a given zone index and 1667 * within a given nodemask 1668 */ 1669 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ 1670 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \ 1671 zone; \ 1672 z = next_zones_zonelist(++z, highidx, nodemask), \ 1673 zone = zonelist_zone(z)) 1674 1675 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \ 1676 for (zone = z->zone; \ 1677 zone; \ 1678 z = next_zones_zonelist(++z, highidx, nodemask), \ 1679 zone = zonelist_zone(z)) 1680 1681 1682 /** 1683 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index 1684 * @zone: The current zone in the iterator 1685 * @z: The current pointer within zonelist->zones being iterated 1686 * @zlist: The zonelist being iterated 1687 * @highidx: The zone index of the highest zone to return 1688 * 1689 * This iterator iterates though all zones at or below a given zone index. 1690 */ 1691 #define for_each_zone_zonelist(zone, z, zlist, highidx) \ 1692 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) 1693 1694 /* Whether the 'nodes' are all movable nodes */ 1695 static inline bool movable_only_nodes(nodemask_t *nodes) 1696 { 1697 struct zonelist *zonelist; 1698 struct zoneref *z; 1699 int nid; 1700 1701 if (nodes_empty(*nodes)) 1702 return false; 1703 1704 /* 1705 * We can chose arbitrary node from the nodemask to get a 1706 * zonelist as they are interlinked. We just need to find 1707 * at least one zone that can satisfy kernel allocations. 1708 */ 1709 nid = first_node(*nodes); 1710 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK]; 1711 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes); 1712 return (!z->zone) ? true : false; 1713 } 1714 1715 1716 #ifdef CONFIG_SPARSEMEM 1717 #include <asm/sparsemem.h> 1718 #endif 1719 1720 #ifdef CONFIG_FLATMEM 1721 #define pfn_to_nid(pfn) (0) 1722 #endif 1723 1724 #ifdef CONFIG_SPARSEMEM 1725 1726 /* 1727 * PA_SECTION_SHIFT physical address to/from section number 1728 * PFN_SECTION_SHIFT pfn to/from section number 1729 */ 1730 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) 1731 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) 1732 1733 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) 1734 1735 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) 1736 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) 1737 1738 #define SECTION_BLOCKFLAGS_BITS \ 1739 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) 1740 1741 #if (MAX_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS 1742 #error Allocator MAX_ORDER exceeds SECTION_SIZE 1743 #endif 1744 1745 static inline unsigned long pfn_to_section_nr(unsigned long pfn) 1746 { 1747 return pfn >> PFN_SECTION_SHIFT; 1748 } 1749 static inline unsigned long section_nr_to_pfn(unsigned long sec) 1750 { 1751 return sec << PFN_SECTION_SHIFT; 1752 } 1753 1754 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) 1755 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) 1756 1757 #define SUBSECTION_SHIFT 21 1758 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT) 1759 1760 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT) 1761 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT) 1762 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1)) 1763 1764 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS 1765 #error Subsection size exceeds section size 1766 #else 1767 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT)) 1768 #endif 1769 1770 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION) 1771 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK) 1772 1773 struct mem_section_usage { 1774 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1775 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION); 1776 #endif 1777 /* See declaration of similar field in struct zone */ 1778 unsigned long pageblock_flags[0]; 1779 }; 1780 1781 void subsection_map_init(unsigned long pfn, unsigned long nr_pages); 1782 1783 struct page; 1784 struct page_ext; 1785 struct mem_section { 1786 /* 1787 * This is, logically, a pointer to an array of struct 1788 * pages. However, it is stored with some other magic. 1789 * (see sparse.c::sparse_init_one_section()) 1790 * 1791 * Additionally during early boot we encode node id of 1792 * the location of the section here to guide allocation. 1793 * (see sparse.c::memory_present()) 1794 * 1795 * Making it a UL at least makes someone do a cast 1796 * before using it wrong. 1797 */ 1798 unsigned long section_mem_map; 1799 1800 struct mem_section_usage *usage; 1801 #ifdef CONFIG_PAGE_EXTENSION 1802 /* 1803 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use 1804 * section. (see page_ext.h about this.) 1805 */ 1806 struct page_ext *page_ext; 1807 unsigned long pad; 1808 #endif 1809 /* 1810 * WARNING: mem_section must be a power-of-2 in size for the 1811 * calculation and use of SECTION_ROOT_MASK to make sense. 1812 */ 1813 }; 1814 1815 #ifdef CONFIG_SPARSEMEM_EXTREME 1816 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) 1817 #else 1818 #define SECTIONS_PER_ROOT 1 1819 #endif 1820 1821 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) 1822 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) 1823 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) 1824 1825 #ifdef CONFIG_SPARSEMEM_EXTREME 1826 extern struct mem_section **mem_section; 1827 #else 1828 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; 1829 #endif 1830 1831 static inline unsigned long *section_to_usemap(struct mem_section *ms) 1832 { 1833 return ms->usage->pageblock_flags; 1834 } 1835 1836 static inline struct mem_section *__nr_to_section(unsigned long nr) 1837 { 1838 unsigned long root = SECTION_NR_TO_ROOT(nr); 1839 1840 if (unlikely(root >= NR_SECTION_ROOTS)) 1841 return NULL; 1842 1843 #ifdef CONFIG_SPARSEMEM_EXTREME 1844 if (!mem_section || !mem_section[root]) 1845 return NULL; 1846 #endif 1847 return &mem_section[root][nr & SECTION_ROOT_MASK]; 1848 } 1849 extern size_t mem_section_usage_size(void); 1850 1851 /* 1852 * We use the lower bits of the mem_map pointer to store 1853 * a little bit of information. The pointer is calculated 1854 * as mem_map - section_nr_to_pfn(pnum). The result is 1855 * aligned to the minimum alignment of the two values: 1856 * 1. All mem_map arrays are page-aligned. 1857 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT 1858 * lowest bits. PFN_SECTION_SHIFT is arch-specific 1859 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the 1860 * worst combination is powerpc with 256k pages, 1861 * which results in PFN_SECTION_SHIFT equal 6. 1862 * To sum it up, at least 6 bits are available on all architectures. 1863 * However, we can exceed 6 bits on some other architectures except 1864 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available 1865 * with the worst case of 64K pages on arm64) if we make sure the 1866 * exceeded bit is not applicable to powerpc. 1867 */ 1868 enum { 1869 SECTION_MARKED_PRESENT_BIT, 1870 SECTION_HAS_MEM_MAP_BIT, 1871 SECTION_IS_ONLINE_BIT, 1872 SECTION_IS_EARLY_BIT, 1873 #ifdef CONFIG_ZONE_DEVICE 1874 SECTION_TAINT_ZONE_DEVICE_BIT, 1875 #endif 1876 SECTION_MAP_LAST_BIT, 1877 }; 1878 1879 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT) 1880 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT) 1881 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT) 1882 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT) 1883 #ifdef CONFIG_ZONE_DEVICE 1884 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT) 1885 #endif 1886 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1)) 1887 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT 1888 1889 static inline struct page *__section_mem_map_addr(struct mem_section *section) 1890 { 1891 unsigned long map = section->section_mem_map; 1892 map &= SECTION_MAP_MASK; 1893 return (struct page *)map; 1894 } 1895 1896 static inline int present_section(struct mem_section *section) 1897 { 1898 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); 1899 } 1900 1901 static inline int present_section_nr(unsigned long nr) 1902 { 1903 return present_section(__nr_to_section(nr)); 1904 } 1905 1906 static inline int valid_section(struct mem_section *section) 1907 { 1908 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); 1909 } 1910 1911 static inline int early_section(struct mem_section *section) 1912 { 1913 return (section && (section->section_mem_map & SECTION_IS_EARLY)); 1914 } 1915 1916 static inline int valid_section_nr(unsigned long nr) 1917 { 1918 return valid_section(__nr_to_section(nr)); 1919 } 1920 1921 static inline int online_section(struct mem_section *section) 1922 { 1923 return (section && (section->section_mem_map & SECTION_IS_ONLINE)); 1924 } 1925 1926 #ifdef CONFIG_ZONE_DEVICE 1927 static inline int online_device_section(struct mem_section *section) 1928 { 1929 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE; 1930 1931 return section && ((section->section_mem_map & flags) == flags); 1932 } 1933 #else 1934 static inline int online_device_section(struct mem_section *section) 1935 { 1936 return 0; 1937 } 1938 #endif 1939 1940 static inline int online_section_nr(unsigned long nr) 1941 { 1942 return online_section(__nr_to_section(nr)); 1943 } 1944 1945 #ifdef CONFIG_MEMORY_HOTPLUG 1946 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 1947 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 1948 #endif 1949 1950 static inline struct mem_section *__pfn_to_section(unsigned long pfn) 1951 { 1952 return __nr_to_section(pfn_to_section_nr(pfn)); 1953 } 1954 1955 extern unsigned long __highest_present_section_nr; 1956 1957 static inline int subsection_map_index(unsigned long pfn) 1958 { 1959 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION; 1960 } 1961 1962 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1963 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 1964 { 1965 int idx = subsection_map_index(pfn); 1966 1967 return test_bit(idx, ms->usage->subsection_map); 1968 } 1969 #else 1970 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 1971 { 1972 return 1; 1973 } 1974 #endif 1975 1976 #ifndef CONFIG_HAVE_ARCH_PFN_VALID 1977 /** 1978 * pfn_valid - check if there is a valid memory map entry for a PFN 1979 * @pfn: the page frame number to check 1980 * 1981 * Check if there is a valid memory map entry aka struct page for the @pfn. 1982 * Note, that availability of the memory map entry does not imply that 1983 * there is actual usable memory at that @pfn. The struct page may 1984 * represent a hole or an unusable page frame. 1985 * 1986 * Return: 1 for PFNs that have memory map entries and 0 otherwise 1987 */ 1988 static inline int pfn_valid(unsigned long pfn) 1989 { 1990 struct mem_section *ms; 1991 1992 /* 1993 * Ensure the upper PAGE_SHIFT bits are clear in the 1994 * pfn. Else it might lead to false positives when 1995 * some of the upper bits are set, but the lower bits 1996 * match a valid pfn. 1997 */ 1998 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn) 1999 return 0; 2000 2001 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 2002 return 0; 2003 ms = __pfn_to_section(pfn); 2004 if (!valid_section(ms)) 2005 return 0; 2006 /* 2007 * Traditionally early sections always returned pfn_valid() for 2008 * the entire section-sized span. 2009 */ 2010 return early_section(ms) || pfn_section_valid(ms, pfn); 2011 } 2012 #endif 2013 2014 static inline int pfn_in_present_section(unsigned long pfn) 2015 { 2016 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 2017 return 0; 2018 return present_section(__pfn_to_section(pfn)); 2019 } 2020 2021 static inline unsigned long next_present_section_nr(unsigned long section_nr) 2022 { 2023 while (++section_nr <= __highest_present_section_nr) { 2024 if (present_section_nr(section_nr)) 2025 return section_nr; 2026 } 2027 2028 return -1; 2029 } 2030 2031 /* 2032 * These are _only_ used during initialisation, therefore they 2033 * can use __initdata ... They could have names to indicate 2034 * this restriction. 2035 */ 2036 #ifdef CONFIG_NUMA 2037 #define pfn_to_nid(pfn) \ 2038 ({ \ 2039 unsigned long __pfn_to_nid_pfn = (pfn); \ 2040 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ 2041 }) 2042 #else 2043 #define pfn_to_nid(pfn) (0) 2044 #endif 2045 2046 void sparse_init(void); 2047 #else 2048 #define sparse_init() do {} while (0) 2049 #define sparse_index_init(_sec, _nid) do {} while (0) 2050 #define pfn_in_present_section pfn_valid 2051 #define subsection_map_init(_pfn, _nr_pages) do {} while (0) 2052 #endif /* CONFIG_SPARSEMEM */ 2053 2054 #endif /* !__GENERATING_BOUNDS.H */ 2055 #endif /* !__ASSEMBLY__ */ 2056 #endif /* _LINUX_MMZONE_H */ 2057