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 <linux/zswap.h> 26 #include <asm/page.h> 27 28 /* Free memory management - zoned buddy allocator. */ 29 #ifndef CONFIG_ARCH_FORCE_MAX_ORDER 30 #define MAX_PAGE_ORDER 10 31 #else 32 #define MAX_PAGE_ORDER CONFIG_ARCH_FORCE_MAX_ORDER 33 #endif 34 #define MAX_ORDER_NR_PAGES (1 << MAX_PAGE_ORDER) 35 36 #define IS_MAX_ORDER_ALIGNED(pfn) IS_ALIGNED(pfn, MAX_ORDER_NR_PAGES) 37 38 #define NR_PAGE_ORDERS (MAX_PAGE_ORDER + 1) 39 40 /* Defines the order for the number of pages that have a migrate type. */ 41 #ifndef CONFIG_PAGE_BLOCK_MAX_ORDER 42 #define PAGE_BLOCK_MAX_ORDER MAX_PAGE_ORDER 43 #else 44 #define PAGE_BLOCK_MAX_ORDER CONFIG_PAGE_BLOCK_MAX_ORDER 45 #endif /* CONFIG_PAGE_BLOCK_MAX_ORDER */ 46 47 /* 48 * The MAX_PAGE_ORDER, which defines the max order of pages to be allocated 49 * by the buddy allocator, has to be larger or equal to the PAGE_BLOCK_MAX_ORDER, 50 * which defines the order for the number of pages that can have a migrate type 51 */ 52 #if (PAGE_BLOCK_MAX_ORDER > MAX_PAGE_ORDER) 53 #error MAX_PAGE_ORDER must be >= PAGE_BLOCK_MAX_ORDER 54 #endif 55 56 /* 57 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed 58 * costly to service. That is between allocation orders which should 59 * coalesce naturally under reasonable reclaim pressure and those which 60 * will not. 61 */ 62 #define PAGE_ALLOC_COSTLY_ORDER 3 63 64 enum migratetype { 65 MIGRATE_UNMOVABLE, 66 MIGRATE_MOVABLE, 67 MIGRATE_RECLAIMABLE, 68 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */ 69 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES, 70 #ifdef CONFIG_CMA 71 /* 72 * MIGRATE_CMA migration type is designed to mimic the way 73 * ZONE_MOVABLE works. Only movable pages can be allocated 74 * from MIGRATE_CMA pageblocks and page allocator never 75 * implicitly change migration type of MIGRATE_CMA pageblock. 76 * 77 * The way to use it is to change migratetype of a range of 78 * pageblocks to MIGRATE_CMA which can be done by 79 * __free_pageblock_cma() function. 80 */ 81 MIGRATE_CMA, 82 __MIGRATE_TYPE_END = MIGRATE_CMA, 83 #else 84 __MIGRATE_TYPE_END = MIGRATE_HIGHATOMIC, 85 #endif 86 #ifdef CONFIG_MEMORY_ISOLATION 87 MIGRATE_ISOLATE, /* can't allocate from here */ 88 #endif 89 MIGRATE_TYPES 90 }; 91 92 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */ 93 extern const char * const migratetype_names[MIGRATE_TYPES]; 94 95 #ifdef CONFIG_CMA 96 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) 97 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA) 98 /* 99 * __dump_folio() in mm/debug.c passes a folio pointer to on-stack struct folio, 100 * so folio_pfn() cannot be used and pfn is needed. 101 */ 102 # define is_migrate_cma_folio(folio, pfn) \ 103 (get_pfnblock_migratetype(&folio->page, pfn) == MIGRATE_CMA) 104 #else 105 # define is_migrate_cma(migratetype) false 106 # define is_migrate_cma_page(_page) false 107 # define is_migrate_cma_folio(folio, pfn) false 108 #endif 109 110 static inline bool is_migrate_movable(int mt) 111 { 112 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE; 113 } 114 115 /* 116 * Check whether a migratetype can be merged with another migratetype. 117 * 118 * It is only mergeable when it can fall back to other migratetypes for 119 * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c. 120 */ 121 static inline bool migratetype_is_mergeable(int mt) 122 { 123 return mt < MIGRATE_PCPTYPES; 124 } 125 126 #define for_each_migratetype_order(order, type) \ 127 for (order = 0; order < NR_PAGE_ORDERS; order++) \ 128 for (type = 0; type < MIGRATE_TYPES; type++) 129 130 extern int page_group_by_mobility_disabled; 131 132 #define get_pageblock_migratetype(page) \ 133 get_pfnblock_migratetype(page, page_to_pfn(page)) 134 135 #define folio_migratetype(folio) \ 136 get_pageblock_migratetype(&folio->page) 137 138 struct free_area { 139 struct list_head free_list[MIGRATE_TYPES]; 140 unsigned long nr_free; 141 }; 142 143 struct pglist_data; 144 145 #ifdef CONFIG_NUMA 146 enum numa_stat_item { 147 NUMA_HIT, /* allocated in intended node */ 148 NUMA_MISS, /* allocated in non intended node */ 149 NUMA_FOREIGN, /* was intended here, hit elsewhere */ 150 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ 151 NUMA_LOCAL, /* allocation from local node */ 152 NUMA_OTHER, /* allocation from other node */ 153 NR_VM_NUMA_EVENT_ITEMS 154 }; 155 #else 156 #define NR_VM_NUMA_EVENT_ITEMS 0 157 #endif 158 159 enum zone_stat_item { 160 /* First 128 byte cacheline (assuming 64 bit words) */ 161 NR_FREE_PAGES, 162 NR_FREE_PAGES_BLOCKS, 163 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */ 164 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE, 165 NR_ZONE_ACTIVE_ANON, 166 NR_ZONE_INACTIVE_FILE, 167 NR_ZONE_ACTIVE_FILE, 168 NR_ZONE_UNEVICTABLE, 169 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */ 170 NR_MLOCK, /* mlock()ed pages found and moved off LRU */ 171 /* Second 128 byte cacheline */ 172 #if IS_ENABLED(CONFIG_ZSMALLOC) 173 NR_ZSPAGES, /* allocated in zsmalloc */ 174 #endif 175 NR_FREE_CMA_PAGES, 176 #ifdef CONFIG_UNACCEPTED_MEMORY 177 NR_UNACCEPTED, 178 #endif 179 NR_VM_ZONE_STAT_ITEMS }; 180 181 enum node_stat_item { 182 NR_LRU_BASE, 183 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ 184 NR_ACTIVE_ANON, /* " " " " " */ 185 NR_INACTIVE_FILE, /* " " " " " */ 186 NR_ACTIVE_FILE, /* " " " " " */ 187 NR_UNEVICTABLE, /* " " " " " */ 188 NR_SLAB_RECLAIMABLE_B, 189 NR_SLAB_UNRECLAIMABLE_B, 190 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ 191 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ 192 WORKINGSET_NODES, 193 WORKINGSET_REFAULT_BASE, 194 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE, 195 WORKINGSET_REFAULT_FILE, 196 WORKINGSET_ACTIVATE_BASE, 197 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE, 198 WORKINGSET_ACTIVATE_FILE, 199 WORKINGSET_RESTORE_BASE, 200 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE, 201 WORKINGSET_RESTORE_FILE, 202 WORKINGSET_NODERECLAIM, 203 NR_ANON_MAPPED, /* Mapped anonymous pages */ 204 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. 205 only modified from process context */ 206 NR_FILE_PAGES, 207 NR_FILE_DIRTY, 208 NR_WRITEBACK, 209 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ 210 NR_SHMEM_THPS, 211 NR_SHMEM_PMDMAPPED, 212 NR_FILE_THPS, 213 NR_FILE_PMDMAPPED, 214 NR_ANON_THPS, 215 NR_VMSCAN_WRITE, 216 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ 217 NR_DIRTIED, /* page dirtyings since bootup */ 218 NR_WRITTEN, /* page writings since bootup */ 219 NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */ 220 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */ 221 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */ 222 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */ 223 NR_KERNEL_STACK_KB, /* measured in KiB */ 224 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK) 225 NR_KERNEL_SCS_KB, /* measured in KiB */ 226 #endif 227 NR_PAGETABLE, /* used for pagetables */ 228 NR_SECONDARY_PAGETABLE, /* secondary pagetables, KVM & IOMMU */ 229 #ifdef CONFIG_IOMMU_SUPPORT 230 NR_IOMMU_PAGES, /* # of pages allocated by IOMMU */ 231 #endif 232 #ifdef CONFIG_SWAP 233 NR_SWAPCACHE, 234 #endif 235 #ifdef CONFIG_NUMA_BALANCING 236 PGPROMOTE_SUCCESS, /* promote successfully */ 237 PGPROMOTE_CANDIDATE, /* candidate pages to promote */ 238 #endif 239 /* PGDEMOTE_*: pages demoted */ 240 PGDEMOTE_KSWAPD, 241 PGDEMOTE_DIRECT, 242 PGDEMOTE_KHUGEPAGED, 243 PGDEMOTE_PROACTIVE, 244 #ifdef CONFIG_HUGETLB_PAGE 245 NR_HUGETLB, 246 #endif 247 NR_BALLOON_PAGES, 248 NR_VM_NODE_STAT_ITEMS 249 }; 250 251 /* 252 * Returns true if the item should be printed in THPs (/proc/vmstat 253 * currently prints number of anon, file and shmem THPs. But the item 254 * is charged in pages). 255 */ 256 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item) 257 { 258 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) 259 return false; 260 261 return item == NR_ANON_THPS || 262 item == NR_FILE_THPS || 263 item == NR_SHMEM_THPS || 264 item == NR_SHMEM_PMDMAPPED || 265 item == NR_FILE_PMDMAPPED; 266 } 267 268 /* 269 * Returns true if the value is measured in bytes (most vmstat values are 270 * measured in pages). This defines the API part, the internal representation 271 * might be different. 272 */ 273 static __always_inline bool vmstat_item_in_bytes(int idx) 274 { 275 /* 276 * Global and per-node slab counters track slab pages. 277 * It's expected that changes are multiples of PAGE_SIZE. 278 * Internally values are stored in pages. 279 * 280 * Per-memcg and per-lruvec counters track memory, consumed 281 * by individual slab objects. These counters are actually 282 * byte-precise. 283 */ 284 return (idx == NR_SLAB_RECLAIMABLE_B || 285 idx == NR_SLAB_UNRECLAIMABLE_B); 286 } 287 288 /* 289 * We do arithmetic on the LRU lists in various places in the code, 290 * so it is important to keep the active lists LRU_ACTIVE higher in 291 * the array than the corresponding inactive lists, and to keep 292 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. 293 * 294 * This has to be kept in sync with the statistics in zone_stat_item 295 * above and the descriptions in vmstat_text in mm/vmstat.c 296 */ 297 #define LRU_BASE 0 298 #define LRU_ACTIVE 1 299 #define LRU_FILE 2 300 301 enum lru_list { 302 LRU_INACTIVE_ANON = LRU_BASE, 303 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, 304 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, 305 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, 306 LRU_UNEVICTABLE, 307 NR_LRU_LISTS 308 }; 309 310 enum vmscan_throttle_state { 311 VMSCAN_THROTTLE_WRITEBACK, 312 VMSCAN_THROTTLE_ISOLATED, 313 VMSCAN_THROTTLE_NOPROGRESS, 314 VMSCAN_THROTTLE_CONGESTED, 315 NR_VMSCAN_THROTTLE, 316 }; 317 318 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) 319 320 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) 321 322 static inline bool is_file_lru(enum lru_list lru) 323 { 324 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); 325 } 326 327 static inline bool is_active_lru(enum lru_list lru) 328 { 329 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); 330 } 331 332 #define WORKINGSET_ANON 0 333 #define WORKINGSET_FILE 1 334 #define ANON_AND_FILE 2 335 336 enum lruvec_flags { 337 /* 338 * An lruvec has many dirty pages backed by a congested BDI: 339 * 1. LRUVEC_CGROUP_CONGESTED is set by cgroup-level reclaim. 340 * It can be cleared by cgroup reclaim or kswapd. 341 * 2. LRUVEC_NODE_CONGESTED is set by kswapd node-level reclaim. 342 * It can only be cleared by kswapd. 343 * 344 * Essentially, kswapd can unthrottle an lruvec throttled by cgroup 345 * reclaim, but not vice versa. This only applies to the root cgroup. 346 * The goal is to prevent cgroup reclaim on the root cgroup (e.g. 347 * memory.reclaim) to unthrottle an unbalanced node (that was throttled 348 * by kswapd). 349 */ 350 LRUVEC_CGROUP_CONGESTED, 351 LRUVEC_NODE_CONGESTED, 352 }; 353 354 #endif /* !__GENERATING_BOUNDS_H */ 355 356 /* 357 * Evictable folios are divided into multiple generations. The youngest and the 358 * oldest generation numbers, max_seq and min_seq, are monotonically increasing. 359 * They form a sliding window of a variable size [MIN_NR_GENS, MAX_NR_GENS]. An 360 * offset within MAX_NR_GENS, i.e., gen, indexes the LRU list of the 361 * corresponding generation. The gen counter in folio->flags stores gen+1 while 362 * a folio is on one of lrugen->folios[]. Otherwise it stores 0. 363 * 364 * After a folio is faulted in, the aging needs to check the accessed bit at 365 * least twice before handing this folio over to the eviction. The first check 366 * clears the accessed bit from the initial fault; the second check makes sure 367 * this folio hasn't been used since then. This process, AKA second chance, 368 * requires a minimum of two generations, hence MIN_NR_GENS. And to maintain ABI 369 * compatibility with the active/inactive LRU, e.g., /proc/vmstat, these two 370 * generations are considered active; the rest of generations, if they exist, 371 * are considered inactive. See lru_gen_is_active(). 372 * 373 * PG_active is always cleared while a folio is on one of lrugen->folios[] so 374 * that the sliding window needs not to worry about it. And it's set again when 375 * a folio considered active is isolated for non-reclaiming purposes, e.g., 376 * migration. See lru_gen_add_folio() and lru_gen_del_folio(). 377 * 378 * MAX_NR_GENS is set to 4 so that the multi-gen LRU can support twice the 379 * number of categories of the active/inactive LRU when keeping track of 380 * accesses through page tables. This requires order_base_2(MAX_NR_GENS+1) bits 381 * in folio->flags, masked by LRU_GEN_MASK. 382 */ 383 #define MIN_NR_GENS 2U 384 #define MAX_NR_GENS 4U 385 386 /* 387 * Each generation is divided into multiple tiers. A folio accessed N times 388 * through file descriptors is in tier order_base_2(N). A folio in the first 389 * tier (N=0,1) is marked by PG_referenced unless it was faulted in through page 390 * tables or read ahead. A folio in the last tier (MAX_NR_TIERS-1) is marked by 391 * PG_workingset. A folio in any other tier (1<N<5) between the first and last 392 * is marked by additional bits of LRU_REFS_WIDTH in folio->flags. 393 * 394 * In contrast to moving across generations which requires the LRU lock, moving 395 * across tiers only involves atomic operations on folio->flags and therefore 396 * has a negligible cost in the buffered access path. In the eviction path, 397 * comparisons of refaulted/(evicted+protected) from the first tier and the rest 398 * infer whether folios accessed multiple times through file descriptors are 399 * statistically hot and thus worth protecting. 400 * 401 * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice the 402 * number of categories of the active/inactive LRU when keeping track of 403 * accesses through file descriptors. This uses MAX_NR_TIERS-2 spare bits in 404 * folio->flags, masked by LRU_REFS_MASK. 405 */ 406 #define MAX_NR_TIERS 4U 407 408 #ifndef __GENERATING_BOUNDS_H 409 410 #define LRU_GEN_MASK ((BIT(LRU_GEN_WIDTH) - 1) << LRU_GEN_PGOFF) 411 #define LRU_REFS_MASK ((BIT(LRU_REFS_WIDTH) - 1) << LRU_REFS_PGOFF) 412 413 /* 414 * For folios accessed multiple times through file descriptors, 415 * lru_gen_inc_refs() sets additional bits of LRU_REFS_WIDTH in folio->flags 416 * after PG_referenced, then PG_workingset after LRU_REFS_WIDTH. After all its 417 * bits are set, i.e., LRU_REFS_FLAGS|BIT(PG_workingset), a folio is lazily 418 * promoted into the second oldest generation in the eviction path. And when 419 * folio_inc_gen() does that, it clears LRU_REFS_FLAGS so that 420 * lru_gen_inc_refs() can start over. Note that for this case, LRU_REFS_MASK is 421 * only valid when PG_referenced is set. 422 * 423 * For folios accessed multiple times through page tables, folio_update_gen() 424 * from a page table walk or lru_gen_set_refs() from a rmap walk sets 425 * PG_referenced after the accessed bit is cleared for the first time. 426 * Thereafter, those two paths set PG_workingset and promote folios to the 427 * youngest generation. Like folio_inc_gen(), folio_update_gen() also clears 428 * PG_referenced. Note that for this case, LRU_REFS_MASK is not used. 429 * 430 * For both cases above, after PG_workingset is set on a folio, it remains until 431 * this folio is either reclaimed, or "deactivated" by lru_gen_clear_refs(). It 432 * can be set again if lru_gen_test_recent() returns true upon a refault. 433 */ 434 #define LRU_REFS_FLAGS (LRU_REFS_MASK | BIT(PG_referenced)) 435 436 struct lruvec; 437 struct page_vma_mapped_walk; 438 439 #ifdef CONFIG_LRU_GEN 440 441 enum { 442 LRU_GEN_ANON, 443 LRU_GEN_FILE, 444 }; 445 446 enum { 447 LRU_GEN_CORE, 448 LRU_GEN_MM_WALK, 449 LRU_GEN_NONLEAF_YOUNG, 450 NR_LRU_GEN_CAPS 451 }; 452 453 #define MIN_LRU_BATCH BITS_PER_LONG 454 #define MAX_LRU_BATCH (MIN_LRU_BATCH * 64) 455 456 /* whether to keep historical stats from evicted generations */ 457 #ifdef CONFIG_LRU_GEN_STATS 458 #define NR_HIST_GENS MAX_NR_GENS 459 #else 460 #define NR_HIST_GENS 1U 461 #endif 462 463 /* 464 * The youngest generation number is stored in max_seq for both anon and file 465 * types as they are aged on an equal footing. The oldest generation numbers are 466 * stored in min_seq[] separately for anon and file types so that they can be 467 * incremented independently. Ideally min_seq[] are kept in sync when both anon 468 * and file types are evictable. However, to adapt to situations like extreme 469 * swappiness, they are allowed to be out of sync by at most 470 * MAX_NR_GENS-MIN_NR_GENS-1. 471 * 472 * The number of pages in each generation is eventually consistent and therefore 473 * can be transiently negative when reset_batch_size() is pending. 474 */ 475 struct lru_gen_folio { 476 /* the aging increments the youngest generation number */ 477 unsigned long max_seq; 478 /* the eviction increments the oldest generation numbers */ 479 unsigned long min_seq[ANON_AND_FILE]; 480 /* the birth time of each generation in jiffies */ 481 unsigned long timestamps[MAX_NR_GENS]; 482 /* the multi-gen LRU lists, lazily sorted on eviction */ 483 struct list_head folios[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 484 /* the multi-gen LRU sizes, eventually consistent */ 485 long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 486 /* the exponential moving average of refaulted */ 487 unsigned long avg_refaulted[ANON_AND_FILE][MAX_NR_TIERS]; 488 /* the exponential moving average of evicted+protected */ 489 unsigned long avg_total[ANON_AND_FILE][MAX_NR_TIERS]; 490 /* can only be modified under the LRU lock */ 491 unsigned long protected[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS]; 492 /* can be modified without holding the LRU lock */ 493 atomic_long_t evicted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS]; 494 atomic_long_t refaulted[NR_HIST_GENS][ANON_AND_FILE][MAX_NR_TIERS]; 495 /* whether the multi-gen LRU is enabled */ 496 bool enabled; 497 /* the memcg generation this lru_gen_folio belongs to */ 498 u8 gen; 499 /* the list segment this lru_gen_folio belongs to */ 500 u8 seg; 501 /* per-node lru_gen_folio list for global reclaim */ 502 struct hlist_nulls_node list; 503 }; 504 505 enum { 506 MM_LEAF_TOTAL, /* total leaf entries */ 507 MM_LEAF_YOUNG, /* young leaf entries */ 508 MM_NONLEAF_FOUND, /* non-leaf entries found in Bloom filters */ 509 MM_NONLEAF_ADDED, /* non-leaf entries added to Bloom filters */ 510 NR_MM_STATS 511 }; 512 513 /* double-buffering Bloom filters */ 514 #define NR_BLOOM_FILTERS 2 515 516 struct lru_gen_mm_state { 517 /* synced with max_seq after each iteration */ 518 unsigned long seq; 519 /* where the current iteration continues after */ 520 struct list_head *head; 521 /* where the last iteration ended before */ 522 struct list_head *tail; 523 /* Bloom filters flip after each iteration */ 524 unsigned long *filters[NR_BLOOM_FILTERS]; 525 /* the mm stats for debugging */ 526 unsigned long stats[NR_HIST_GENS][NR_MM_STATS]; 527 }; 528 529 struct lru_gen_mm_walk { 530 /* the lruvec under reclaim */ 531 struct lruvec *lruvec; 532 /* max_seq from lru_gen_folio: can be out of date */ 533 unsigned long seq; 534 /* the next address within an mm to scan */ 535 unsigned long next_addr; 536 /* to batch promoted pages */ 537 int nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES]; 538 /* to batch the mm stats */ 539 int mm_stats[NR_MM_STATS]; 540 /* total batched items */ 541 int batched; 542 int swappiness; 543 bool force_scan; 544 }; 545 546 /* 547 * For each node, memcgs are divided into two generations: the old and the 548 * young. For each generation, memcgs are randomly sharded into multiple bins 549 * to improve scalability. For each bin, the hlist_nulls is virtually divided 550 * into three segments: the head, the tail and the default. 551 * 552 * An onlining memcg is added to the tail of a random bin in the old generation. 553 * The eviction starts at the head of a random bin in the old generation. The 554 * per-node memcg generation counter, whose reminder (mod MEMCG_NR_GENS) indexes 555 * the old generation, is incremented when all its bins become empty. 556 * 557 * There are four operations: 558 * 1. MEMCG_LRU_HEAD, which moves a memcg to the head of a random bin in its 559 * current generation (old or young) and updates its "seg" to "head"; 560 * 2. MEMCG_LRU_TAIL, which moves a memcg to the tail of a random bin in its 561 * current generation (old or young) and updates its "seg" to "tail"; 562 * 3. MEMCG_LRU_OLD, which moves a memcg to the head of a random bin in the old 563 * generation, updates its "gen" to "old" and resets its "seg" to "default"; 564 * 4. MEMCG_LRU_YOUNG, which moves a memcg to the tail of a random bin in the 565 * young generation, updates its "gen" to "young" and resets its "seg" to 566 * "default". 567 * 568 * The events that trigger the above operations are: 569 * 1. Exceeding the soft limit, which triggers MEMCG_LRU_HEAD; 570 * 2. The first attempt to reclaim a memcg below low, which triggers 571 * MEMCG_LRU_TAIL; 572 * 3. The first attempt to reclaim a memcg offlined or below reclaimable size 573 * threshold, which triggers MEMCG_LRU_TAIL; 574 * 4. The second attempt to reclaim a memcg offlined or below reclaimable size 575 * threshold, which triggers MEMCG_LRU_YOUNG; 576 * 5. Attempting to reclaim a memcg below min, which triggers MEMCG_LRU_YOUNG; 577 * 6. Finishing the aging on the eviction path, which triggers MEMCG_LRU_YOUNG; 578 * 7. Offlining a memcg, which triggers MEMCG_LRU_OLD. 579 * 580 * Notes: 581 * 1. Memcg LRU only applies to global reclaim, and the round-robin incrementing 582 * of their max_seq counters ensures the eventual fairness to all eligible 583 * memcgs. For memcg reclaim, it still relies on mem_cgroup_iter(). 584 * 2. There are only two valid generations: old (seq) and young (seq+1). 585 * MEMCG_NR_GENS is set to three so that when reading the generation counter 586 * locklessly, a stale value (seq-1) does not wraparound to young. 587 */ 588 #define MEMCG_NR_GENS 3 589 #define MEMCG_NR_BINS 8 590 591 struct lru_gen_memcg { 592 /* the per-node memcg generation counter */ 593 unsigned long seq; 594 /* each memcg has one lru_gen_folio per node */ 595 unsigned long nr_memcgs[MEMCG_NR_GENS]; 596 /* per-node lru_gen_folio list for global reclaim */ 597 struct hlist_nulls_head fifo[MEMCG_NR_GENS][MEMCG_NR_BINS]; 598 /* protects the above */ 599 spinlock_t lock; 600 }; 601 602 void lru_gen_init_pgdat(struct pglist_data *pgdat); 603 void lru_gen_init_lruvec(struct lruvec *lruvec); 604 bool lru_gen_look_around(struct page_vma_mapped_walk *pvmw); 605 606 void lru_gen_init_memcg(struct mem_cgroup *memcg); 607 void lru_gen_exit_memcg(struct mem_cgroup *memcg); 608 void lru_gen_online_memcg(struct mem_cgroup *memcg); 609 void lru_gen_offline_memcg(struct mem_cgroup *memcg); 610 void lru_gen_release_memcg(struct mem_cgroup *memcg); 611 void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid); 612 613 #else /* !CONFIG_LRU_GEN */ 614 615 static inline void lru_gen_init_pgdat(struct pglist_data *pgdat) 616 { 617 } 618 619 static inline void lru_gen_init_lruvec(struct lruvec *lruvec) 620 { 621 } 622 623 static inline bool lru_gen_look_around(struct page_vma_mapped_walk *pvmw) 624 { 625 return false; 626 } 627 628 static inline void lru_gen_init_memcg(struct mem_cgroup *memcg) 629 { 630 } 631 632 static inline void lru_gen_exit_memcg(struct mem_cgroup *memcg) 633 { 634 } 635 636 static inline void lru_gen_online_memcg(struct mem_cgroup *memcg) 637 { 638 } 639 640 static inline void lru_gen_offline_memcg(struct mem_cgroup *memcg) 641 { 642 } 643 644 static inline void lru_gen_release_memcg(struct mem_cgroup *memcg) 645 { 646 } 647 648 static inline void lru_gen_soft_reclaim(struct mem_cgroup *memcg, int nid) 649 { 650 } 651 652 #endif /* CONFIG_LRU_GEN */ 653 654 struct lruvec { 655 struct list_head lists[NR_LRU_LISTS]; 656 /* per lruvec lru_lock for memcg */ 657 spinlock_t lru_lock; 658 /* 659 * These track the cost of reclaiming one LRU - file or anon - 660 * over the other. As the observed cost of reclaiming one LRU 661 * increases, the reclaim scan balance tips toward the other. 662 */ 663 unsigned long anon_cost; 664 unsigned long file_cost; 665 /* Non-resident age, driven by LRU movement */ 666 atomic_long_t nonresident_age; 667 /* Refaults at the time of last reclaim cycle */ 668 unsigned long refaults[ANON_AND_FILE]; 669 /* Various lruvec state flags (enum lruvec_flags) */ 670 unsigned long flags; 671 #ifdef CONFIG_LRU_GEN 672 /* evictable pages divided into generations */ 673 struct lru_gen_folio lrugen; 674 #ifdef CONFIG_LRU_GEN_WALKS_MMU 675 /* to concurrently iterate lru_gen_mm_list */ 676 struct lru_gen_mm_state mm_state; 677 #endif 678 #endif /* CONFIG_LRU_GEN */ 679 #ifdef CONFIG_MEMCG 680 struct pglist_data *pgdat; 681 #endif 682 struct zswap_lruvec_state zswap_lruvec_state; 683 }; 684 685 /* Isolate for asynchronous migration */ 686 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) 687 /* Isolate unevictable pages */ 688 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) 689 690 /* LRU Isolation modes. */ 691 typedef unsigned __bitwise isolate_mode_t; 692 693 enum zone_watermarks { 694 WMARK_MIN, 695 WMARK_LOW, 696 WMARK_HIGH, 697 WMARK_PROMO, 698 NR_WMARK 699 }; 700 701 /* 702 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER. Two additional lists 703 * are added for THP. One PCP list is used by GPF_MOVABLE, and the other PCP list 704 * is used by GFP_UNMOVABLE and GFP_RECLAIMABLE. 705 */ 706 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 707 #define NR_PCP_THP 2 708 #else 709 #define NR_PCP_THP 0 710 #endif 711 #define NR_LOWORDER_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1)) 712 #define NR_PCP_LISTS (NR_LOWORDER_PCP_LISTS + NR_PCP_THP) 713 714 /* 715 * Flags used in pcp->flags field. 716 * 717 * PCPF_PREV_FREE_HIGH_ORDER: a high-order page is freed in the 718 * previous page freeing. To avoid to drain PCP for an accident 719 * high-order page freeing. 720 * 721 * PCPF_FREE_HIGH_BATCH: preserve "pcp->batch" pages in PCP before 722 * draining PCP for consecutive high-order pages freeing without 723 * allocation if data cache slice of CPU is large enough. To reduce 724 * zone lock contention and keep cache-hot pages reusing. 725 */ 726 #define PCPF_PREV_FREE_HIGH_ORDER BIT(0) 727 #define PCPF_FREE_HIGH_BATCH BIT(1) 728 729 struct per_cpu_pages { 730 spinlock_t lock; /* Protects lists field */ 731 int count; /* number of pages in the list */ 732 int high; /* high watermark, emptying needed */ 733 int high_min; /* min high watermark */ 734 int high_max; /* max high watermark */ 735 int batch; /* chunk size for buddy add/remove */ 736 u8 flags; /* protected by pcp->lock */ 737 u8 alloc_factor; /* batch scaling factor during allocate */ 738 #ifdef CONFIG_NUMA 739 u8 expire; /* When 0, remote pagesets are drained */ 740 #endif 741 short free_count; /* consecutive free count */ 742 743 /* Lists of pages, one per migrate type stored on the pcp-lists */ 744 struct list_head lists[NR_PCP_LISTS]; 745 } ____cacheline_aligned_in_smp; 746 747 struct per_cpu_zonestat { 748 #ifdef CONFIG_SMP 749 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; 750 s8 stat_threshold; 751 #endif 752 #ifdef CONFIG_NUMA 753 /* 754 * Low priority inaccurate counters that are only folded 755 * on demand. Use a large type to avoid the overhead of 756 * folding during refresh_cpu_vm_stats. 757 */ 758 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS]; 759 #endif 760 }; 761 762 struct per_cpu_nodestat { 763 s8 stat_threshold; 764 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS]; 765 }; 766 767 #endif /* !__GENERATING_BOUNDS.H */ 768 769 enum zone_type { 770 /* 771 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able 772 * to DMA to all of the addressable memory (ZONE_NORMAL). 773 * On architectures where this area covers the whole 32 bit address 774 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller 775 * DMA addressing constraints. This distinction is important as a 32bit 776 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit 777 * platforms may need both zones as they support peripherals with 778 * different DMA addressing limitations. 779 */ 780 #ifdef CONFIG_ZONE_DMA 781 ZONE_DMA, 782 #endif 783 #ifdef CONFIG_ZONE_DMA32 784 ZONE_DMA32, 785 #endif 786 /* 787 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be 788 * performed on pages in ZONE_NORMAL if the DMA devices support 789 * transfers to all addressable memory. 790 */ 791 ZONE_NORMAL, 792 #ifdef CONFIG_HIGHMEM 793 /* 794 * A memory area that is only addressable by the kernel through 795 * mapping portions into its own address space. This is for example 796 * used by i386 to allow the kernel to address the memory beyond 797 * 900MB. The kernel will set up special mappings (page 798 * table entries on i386) for each page that the kernel needs to 799 * access. 800 */ 801 ZONE_HIGHMEM, 802 #endif 803 /* 804 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains 805 * movable pages with few exceptional cases described below. Main use 806 * cases for ZONE_MOVABLE are to make memory offlining/unplug more 807 * likely to succeed, and to locally limit unmovable allocations - e.g., 808 * to increase the number of THP/huge pages. Notable special cases are: 809 * 810 * 1. Pinned pages: (long-term) pinning of movable pages might 811 * essentially turn such pages unmovable. Therefore, we do not allow 812 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and 813 * faulted, they come from the right zone right away. However, it is 814 * still possible that address space already has pages in 815 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has 816 * touches that memory before pinning). In such case we migrate them 817 * to a different zone. When migration fails - pinning fails. 818 * 2. memblock allocations: kernelcore/movablecore setups might create 819 * situations where ZONE_MOVABLE contains unmovable allocations 820 * after boot. Memory offlining and allocations fail early. 821 * 3. Memory holes: kernelcore/movablecore setups might create very rare 822 * situations where ZONE_MOVABLE contains memory holes after boot, 823 * for example, if we have sections that are only partially 824 * populated. Memory offlining and allocations fail early. 825 * 4. PG_hwpoison pages: while poisoned pages can be skipped during 826 * memory offlining, such pages cannot be allocated. 827 * 5. Unmovable PG_offline pages: in paravirtualized environments, 828 * hotplugged memory blocks might only partially be managed by the 829 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The 830 * parts not manged by the buddy are unmovable PG_offline pages. In 831 * some cases (virtio-mem), such pages can be skipped during 832 * memory offlining, however, cannot be moved/allocated. These 833 * techniques might use alloc_contig_range() to hide previously 834 * exposed pages from the buddy again (e.g., to implement some sort 835 * of memory unplug in virtio-mem). 836 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create 837 * situations where ZERO_PAGE(0) which is allocated differently 838 * on different platforms may end up in a movable zone. ZERO_PAGE(0) 839 * cannot be migrated. 840 * 7. Memory-hotplug: when using memmap_on_memory and onlining the 841 * memory to the MOVABLE zone, the vmemmap pages are also placed in 842 * such zone. Such pages cannot be really moved around as they are 843 * self-stored in the range, but they are treated as movable when 844 * the range they describe is about to be offlined. 845 * 846 * In general, no unmovable allocations that degrade memory offlining 847 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range()) 848 * have to expect that migrating pages in ZONE_MOVABLE can fail (even 849 * if has_unmovable_pages() states that there are no unmovable pages, 850 * there can be false negatives). 851 */ 852 ZONE_MOVABLE, 853 #ifdef CONFIG_ZONE_DEVICE 854 ZONE_DEVICE, 855 #endif 856 __MAX_NR_ZONES 857 858 }; 859 860 #ifndef __GENERATING_BOUNDS_H 861 862 #define ASYNC_AND_SYNC 2 863 864 struct zone { 865 /* Read-mostly fields */ 866 867 /* zone watermarks, access with *_wmark_pages(zone) macros */ 868 unsigned long _watermark[NR_WMARK]; 869 unsigned long watermark_boost; 870 871 unsigned long nr_reserved_highatomic; 872 unsigned long nr_free_highatomic; 873 874 /* 875 * We don't know if the memory that we're going to allocate will be 876 * freeable or/and it will be released eventually, so to avoid totally 877 * wasting several GB of ram we must reserve some of the lower zone 878 * memory (otherwise we risk to run OOM on the lower zones despite 879 * there being tons of freeable ram on the higher zones). This array is 880 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl 881 * changes. 882 */ 883 long lowmem_reserve[MAX_NR_ZONES]; 884 885 #ifdef CONFIG_NUMA 886 int node; 887 #endif 888 struct pglist_data *zone_pgdat; 889 struct per_cpu_pages __percpu *per_cpu_pageset; 890 struct per_cpu_zonestat __percpu *per_cpu_zonestats; 891 /* 892 * the high and batch values are copied to individual pagesets for 893 * faster access 894 */ 895 int pageset_high_min; 896 int pageset_high_max; 897 int pageset_batch; 898 899 #ifndef CONFIG_SPARSEMEM 900 /* 901 * Flags for a pageblock_nr_pages block. See pageblock-flags.h. 902 * In SPARSEMEM, this map is stored in struct mem_section 903 */ 904 unsigned long *pageblock_flags; 905 #endif /* CONFIG_SPARSEMEM */ 906 907 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ 908 unsigned long zone_start_pfn; 909 910 /* 911 * spanned_pages is the total pages spanned by the zone, including 912 * holes, which is calculated as: 913 * spanned_pages = zone_end_pfn - zone_start_pfn; 914 * 915 * present_pages is physical pages existing within the zone, which 916 * is calculated as: 917 * present_pages = spanned_pages - absent_pages(pages in holes); 918 * 919 * present_early_pages is present pages existing within the zone 920 * located on memory available since early boot, excluding hotplugged 921 * memory. 922 * 923 * managed_pages is present pages managed by the buddy system, which 924 * is calculated as (reserved_pages includes pages allocated by the 925 * bootmem allocator): 926 * managed_pages = present_pages - reserved_pages; 927 * 928 * cma pages is present pages that are assigned for CMA use 929 * (MIGRATE_CMA). 930 * 931 * So present_pages may be used by memory hotplug or memory power 932 * management logic to figure out unmanaged pages by checking 933 * (present_pages - managed_pages). And managed_pages should be used 934 * by page allocator and vm scanner to calculate all kinds of watermarks 935 * and thresholds. 936 * 937 * Locking rules: 938 * 939 * zone_start_pfn and spanned_pages are protected by span_seqlock. 940 * It is a seqlock because it has to be read outside of zone->lock, 941 * and it is done in the main allocator path. But, it is written 942 * quite infrequently. 943 * 944 * The span_seq lock is declared along with zone->lock because it is 945 * frequently read in proximity to zone->lock. It's good to 946 * give them a chance of being in the same cacheline. 947 * 948 * Write access to present_pages at runtime should be protected by 949 * mem_hotplug_begin/done(). Any reader who can't tolerant drift of 950 * present_pages should use get_online_mems() to get a stable value. 951 */ 952 atomic_long_t managed_pages; 953 unsigned long spanned_pages; 954 unsigned long present_pages; 955 #if defined(CONFIG_MEMORY_HOTPLUG) 956 unsigned long present_early_pages; 957 #endif 958 #ifdef CONFIG_CMA 959 unsigned long cma_pages; 960 #endif 961 962 const char *name; 963 964 #ifdef CONFIG_MEMORY_ISOLATION 965 /* 966 * Number of isolated pageblock. It is used to solve incorrect 967 * freepage counting problem due to racy retrieving migratetype 968 * of pageblock. Protected by zone->lock. 969 */ 970 unsigned long nr_isolate_pageblock; 971 #endif 972 973 #ifdef CONFIG_MEMORY_HOTPLUG 974 /* see spanned/present_pages for more description */ 975 seqlock_t span_seqlock; 976 #endif 977 978 int initialized; 979 980 /* Write-intensive fields used from the page allocator */ 981 CACHELINE_PADDING(_pad1_); 982 983 /* free areas of different sizes */ 984 struct free_area free_area[NR_PAGE_ORDERS]; 985 986 #ifdef CONFIG_UNACCEPTED_MEMORY 987 /* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */ 988 struct list_head unaccepted_pages; 989 990 /* To be called once the last page in the zone is accepted */ 991 struct work_struct unaccepted_cleanup; 992 #endif 993 994 /* zone flags, see below */ 995 unsigned long flags; 996 997 /* Primarily protects free_area */ 998 spinlock_t lock; 999 1000 /* Pages to be freed when next trylock succeeds */ 1001 struct llist_head trylock_free_pages; 1002 1003 /* Write-intensive fields used by compaction and vmstats. */ 1004 CACHELINE_PADDING(_pad2_); 1005 1006 /* 1007 * When free pages are below this point, additional steps are taken 1008 * when reading the number of free pages to avoid per-cpu counter 1009 * drift allowing watermarks to be breached 1010 */ 1011 unsigned long percpu_drift_mark; 1012 1013 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 1014 /* pfn where compaction free scanner should start */ 1015 unsigned long compact_cached_free_pfn; 1016 /* pfn where compaction migration scanner should start */ 1017 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC]; 1018 unsigned long compact_init_migrate_pfn; 1019 unsigned long compact_init_free_pfn; 1020 #endif 1021 1022 #ifdef CONFIG_COMPACTION 1023 /* 1024 * On compaction failure, 1<<compact_defer_shift compactions 1025 * are skipped before trying again. The number attempted since 1026 * last failure is tracked with compact_considered. 1027 * compact_order_failed is the minimum compaction failed order. 1028 */ 1029 unsigned int compact_considered; 1030 unsigned int compact_defer_shift; 1031 int compact_order_failed; 1032 #endif 1033 1034 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 1035 /* Set to true when the PG_migrate_skip bits should be cleared */ 1036 bool compact_blockskip_flush; 1037 #endif 1038 1039 bool contiguous; 1040 1041 CACHELINE_PADDING(_pad3_); 1042 /* Zone statistics */ 1043 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; 1044 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS]; 1045 } ____cacheline_internodealigned_in_smp; 1046 1047 enum pgdat_flags { 1048 PGDAT_DIRTY, /* reclaim scanning has recently found 1049 * many dirty file pages at the tail 1050 * of the LRU. 1051 */ 1052 PGDAT_WRITEBACK, /* reclaim scanning has recently found 1053 * many pages under writeback 1054 */ 1055 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */ 1056 }; 1057 1058 enum zone_flags { 1059 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks. 1060 * Cleared when kswapd is woken. 1061 */ 1062 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */ 1063 ZONE_BELOW_HIGH, /* zone is below high watermark. */ 1064 }; 1065 1066 static inline unsigned long wmark_pages(const struct zone *z, 1067 enum zone_watermarks w) 1068 { 1069 return z->_watermark[w] + z->watermark_boost; 1070 } 1071 1072 static inline unsigned long min_wmark_pages(const struct zone *z) 1073 { 1074 return wmark_pages(z, WMARK_MIN); 1075 } 1076 1077 static inline unsigned long low_wmark_pages(const struct zone *z) 1078 { 1079 return wmark_pages(z, WMARK_LOW); 1080 } 1081 1082 static inline unsigned long high_wmark_pages(const struct zone *z) 1083 { 1084 return wmark_pages(z, WMARK_HIGH); 1085 } 1086 1087 static inline unsigned long promo_wmark_pages(const struct zone *z) 1088 { 1089 return wmark_pages(z, WMARK_PROMO); 1090 } 1091 1092 static inline unsigned long zone_managed_pages(struct zone *zone) 1093 { 1094 return (unsigned long)atomic_long_read(&zone->managed_pages); 1095 } 1096 1097 static inline unsigned long zone_cma_pages(struct zone *zone) 1098 { 1099 #ifdef CONFIG_CMA 1100 return zone->cma_pages; 1101 #else 1102 return 0; 1103 #endif 1104 } 1105 1106 static inline unsigned long zone_end_pfn(const struct zone *zone) 1107 { 1108 return zone->zone_start_pfn + zone->spanned_pages; 1109 } 1110 1111 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn) 1112 { 1113 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone); 1114 } 1115 1116 static inline bool zone_is_initialized(struct zone *zone) 1117 { 1118 return zone->initialized; 1119 } 1120 1121 static inline bool zone_is_empty(struct zone *zone) 1122 { 1123 return zone->spanned_pages == 0; 1124 } 1125 1126 #ifndef BUILD_VDSO32_64 1127 /* 1128 * The zone field is never updated after free_area_init_core() 1129 * sets it, so none of the operations on it need to be atomic. 1130 */ 1131 1132 /* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */ 1133 #define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH) 1134 #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH) 1135 #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH) 1136 #define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH) 1137 #define KASAN_TAG_PGOFF (LAST_CPUPID_PGOFF - KASAN_TAG_WIDTH) 1138 #define LRU_GEN_PGOFF (KASAN_TAG_PGOFF - LRU_GEN_WIDTH) 1139 #define LRU_REFS_PGOFF (LRU_GEN_PGOFF - LRU_REFS_WIDTH) 1140 1141 /* 1142 * Define the bit shifts to access each section. For non-existent 1143 * sections we define the shift as 0; that plus a 0 mask ensures 1144 * the compiler will optimise away reference to them. 1145 */ 1146 #define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0)) 1147 #define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0)) 1148 #define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0)) 1149 #define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0)) 1150 #define KASAN_TAG_PGSHIFT (KASAN_TAG_PGOFF * (KASAN_TAG_WIDTH != 0)) 1151 1152 /* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */ 1153 #ifdef NODE_NOT_IN_PAGE_FLAGS 1154 #define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT) 1155 #define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF) ? \ 1156 SECTIONS_PGOFF : ZONES_PGOFF) 1157 #else 1158 #define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT) 1159 #define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF) ? \ 1160 NODES_PGOFF : ZONES_PGOFF) 1161 #endif 1162 1163 #define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0)) 1164 1165 #define ZONES_MASK ((1UL << ZONES_WIDTH) - 1) 1166 #define NODES_MASK ((1UL << NODES_WIDTH) - 1) 1167 #define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1) 1168 #define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1) 1169 #define KASAN_TAG_MASK ((1UL << KASAN_TAG_WIDTH) - 1) 1170 #define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1) 1171 1172 static inline enum zone_type page_zonenum(const struct page *page) 1173 { 1174 ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT); 1175 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK; 1176 } 1177 1178 static inline enum zone_type folio_zonenum(const struct folio *folio) 1179 { 1180 return page_zonenum(&folio->page); 1181 } 1182 1183 #ifdef CONFIG_ZONE_DEVICE 1184 static inline bool is_zone_device_page(const struct page *page) 1185 { 1186 return page_zonenum(page) == ZONE_DEVICE; 1187 } 1188 1189 static inline struct dev_pagemap *page_pgmap(const struct page *page) 1190 { 1191 VM_WARN_ON_ONCE_PAGE(!is_zone_device_page(page), page); 1192 return page_folio(page)->pgmap; 1193 } 1194 1195 /* 1196 * Consecutive zone device pages should not be merged into the same sgl 1197 * or bvec segment with other types of pages or if they belong to different 1198 * pgmaps. Otherwise getting the pgmap of a given segment is not possible 1199 * without scanning the entire segment. This helper returns true either if 1200 * both pages are not zone device pages or both pages are zone device pages 1201 * with the same pgmap. 1202 */ 1203 static inline bool zone_device_pages_have_same_pgmap(const struct page *a, 1204 const struct page *b) 1205 { 1206 if (is_zone_device_page(a) != is_zone_device_page(b)) 1207 return false; 1208 if (!is_zone_device_page(a)) 1209 return true; 1210 return page_pgmap(a) == page_pgmap(b); 1211 } 1212 1213 extern void memmap_init_zone_device(struct zone *, unsigned long, 1214 unsigned long, struct dev_pagemap *); 1215 #else 1216 static inline bool is_zone_device_page(const struct page *page) 1217 { 1218 return false; 1219 } 1220 static inline bool zone_device_pages_have_same_pgmap(const struct page *a, 1221 const struct page *b) 1222 { 1223 return true; 1224 } 1225 static inline struct dev_pagemap *page_pgmap(const struct page *page) 1226 { 1227 return NULL; 1228 } 1229 #endif 1230 1231 static inline bool folio_is_zone_device(const struct folio *folio) 1232 { 1233 return is_zone_device_page(&folio->page); 1234 } 1235 1236 static inline bool is_zone_movable_page(const struct page *page) 1237 { 1238 return page_zonenum(page) == ZONE_MOVABLE; 1239 } 1240 1241 static inline bool folio_is_zone_movable(const struct folio *folio) 1242 { 1243 return folio_zonenum(folio) == ZONE_MOVABLE; 1244 } 1245 #endif 1246 1247 /* 1248 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty 1249 * intersection with the given zone 1250 */ 1251 static inline bool zone_intersects(struct zone *zone, 1252 unsigned long start_pfn, unsigned long nr_pages) 1253 { 1254 if (zone_is_empty(zone)) 1255 return false; 1256 if (start_pfn >= zone_end_pfn(zone) || 1257 start_pfn + nr_pages <= zone->zone_start_pfn) 1258 return false; 1259 1260 return true; 1261 } 1262 1263 /* 1264 * The "priority" of VM scanning is how much of the queues we will scan in one 1265 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the 1266 * queues ("queue_length >> 12") during an aging round. 1267 */ 1268 #define DEF_PRIORITY 12 1269 1270 /* Maximum number of zones on a zonelist */ 1271 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) 1272 1273 enum { 1274 ZONELIST_FALLBACK, /* zonelist with fallback */ 1275 #ifdef CONFIG_NUMA 1276 /* 1277 * The NUMA zonelists are doubled because we need zonelists that 1278 * restrict the allocations to a single node for __GFP_THISNODE. 1279 */ 1280 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */ 1281 #endif 1282 MAX_ZONELISTS 1283 }; 1284 1285 /* 1286 * This struct contains information about a zone in a zonelist. It is stored 1287 * here to avoid dereferences into large structures and lookups of tables 1288 */ 1289 struct zoneref { 1290 struct zone *zone; /* Pointer to actual zone */ 1291 int zone_idx; /* zone_idx(zoneref->zone) */ 1292 }; 1293 1294 /* 1295 * One allocation request operates on a zonelist. A zonelist 1296 * is a list of zones, the first one is the 'goal' of the 1297 * allocation, the other zones are fallback zones, in decreasing 1298 * priority. 1299 * 1300 * To speed the reading of the zonelist, the zonerefs contain the zone index 1301 * of the entry being read. Helper functions to access information given 1302 * a struct zoneref are 1303 * 1304 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs 1305 * zonelist_zone_idx() - Return the index of the zone for an entry 1306 * zonelist_node_idx() - Return the index of the node for an entry 1307 */ 1308 struct zonelist { 1309 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; 1310 }; 1311 1312 /* 1313 * The array of struct pages for flatmem. 1314 * It must be declared for SPARSEMEM as well because there are configurations 1315 * that rely on that. 1316 */ 1317 extern struct page *mem_map; 1318 1319 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1320 struct deferred_split { 1321 spinlock_t split_queue_lock; 1322 struct list_head split_queue; 1323 unsigned long split_queue_len; 1324 }; 1325 #endif 1326 1327 #ifdef CONFIG_MEMORY_FAILURE 1328 /* 1329 * Per NUMA node memory failure handling statistics. 1330 */ 1331 struct memory_failure_stats { 1332 /* 1333 * Number of raw pages poisoned. 1334 * Cases not accounted: memory outside kernel control, offline page, 1335 * arch-specific memory_failure (SGX), hwpoison_filter() filtered 1336 * error events, and unpoison actions from hwpoison_unpoison. 1337 */ 1338 unsigned long total; 1339 /* 1340 * Recovery results of poisoned raw pages handled by memory_failure, 1341 * in sync with mf_result. 1342 * total = ignored + failed + delayed + recovered. 1343 * total * PAGE_SIZE * #nodes = /proc/meminfo/HardwareCorrupted. 1344 */ 1345 unsigned long ignored; 1346 unsigned long failed; 1347 unsigned long delayed; 1348 unsigned long recovered; 1349 }; 1350 #endif 1351 1352 /* 1353 * On NUMA machines, each NUMA node would have a pg_data_t to describe 1354 * it's memory layout. On UMA machines there is a single pglist_data which 1355 * describes the whole memory. 1356 * 1357 * Memory statistics and page replacement data structures are maintained on a 1358 * per-zone basis. 1359 */ 1360 typedef struct pglist_data { 1361 /* 1362 * node_zones contains just the zones for THIS node. Not all of the 1363 * zones may be populated, but it is the full list. It is referenced by 1364 * this node's node_zonelists as well as other node's node_zonelists. 1365 */ 1366 struct zone node_zones[MAX_NR_ZONES]; 1367 1368 /* 1369 * node_zonelists contains references to all zones in all nodes. 1370 * Generally the first zones will be references to this node's 1371 * node_zones. 1372 */ 1373 struct zonelist node_zonelists[MAX_ZONELISTS]; 1374 1375 int nr_zones; /* number of populated zones in this node */ 1376 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */ 1377 struct page *node_mem_map; 1378 #ifdef CONFIG_PAGE_EXTENSION 1379 struct page_ext *node_page_ext; 1380 #endif 1381 #endif 1382 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT) 1383 /* 1384 * Must be held any time you expect node_start_pfn, 1385 * node_present_pages, node_spanned_pages or nr_zones to stay constant. 1386 * Also synchronizes pgdat->first_deferred_pfn during deferred page 1387 * init. 1388 * 1389 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to 1390 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG 1391 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT. 1392 * 1393 * Nests above zone->lock and zone->span_seqlock 1394 */ 1395 spinlock_t node_size_lock; 1396 #endif 1397 unsigned long node_start_pfn; 1398 unsigned long node_present_pages; /* total number of physical pages */ 1399 unsigned long node_spanned_pages; /* total size of physical page 1400 range, including holes */ 1401 int node_id; 1402 wait_queue_head_t kswapd_wait; 1403 wait_queue_head_t pfmemalloc_wait; 1404 1405 /* workqueues for throttling reclaim for different reasons. */ 1406 wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE]; 1407 1408 atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */ 1409 unsigned long nr_reclaim_start; /* nr pages written while throttled 1410 * when throttling started. */ 1411 #ifdef CONFIG_MEMORY_HOTPLUG 1412 struct mutex kswapd_lock; 1413 #endif 1414 struct task_struct *kswapd; /* Protected by kswapd_lock */ 1415 int kswapd_order; 1416 enum zone_type kswapd_highest_zoneidx; 1417 1418 int kswapd_failures; /* Number of 'reclaimed == 0' runs */ 1419 1420 #ifdef CONFIG_COMPACTION 1421 int kcompactd_max_order; 1422 enum zone_type kcompactd_highest_zoneidx; 1423 wait_queue_head_t kcompactd_wait; 1424 struct task_struct *kcompactd; 1425 bool proactive_compact_trigger; 1426 #endif 1427 /* 1428 * This is a per-node reserve of pages that are not available 1429 * to userspace allocations. 1430 */ 1431 unsigned long totalreserve_pages; 1432 1433 #ifdef CONFIG_NUMA 1434 /* 1435 * node reclaim becomes active if more unmapped pages exist. 1436 */ 1437 unsigned long min_unmapped_pages; 1438 unsigned long min_slab_pages; 1439 #endif /* CONFIG_NUMA */ 1440 1441 /* Write-intensive fields used by page reclaim */ 1442 CACHELINE_PADDING(_pad1_); 1443 1444 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1445 /* 1446 * If memory initialisation on large machines is deferred then this 1447 * is the first PFN that needs to be initialised. 1448 */ 1449 unsigned long first_deferred_pfn; 1450 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1451 1452 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1453 struct deferred_split deferred_split_queue; 1454 #endif 1455 1456 #ifdef CONFIG_NUMA_BALANCING 1457 /* start time in ms of current promote rate limit period */ 1458 unsigned int nbp_rl_start; 1459 /* number of promote candidate pages at start time of current rate limit period */ 1460 unsigned long nbp_rl_nr_cand; 1461 /* promote threshold in ms */ 1462 unsigned int nbp_threshold; 1463 /* start time in ms of current promote threshold adjustment period */ 1464 unsigned int nbp_th_start; 1465 /* 1466 * number of promote candidate pages at start time of current promote 1467 * threshold adjustment period 1468 */ 1469 unsigned long nbp_th_nr_cand; 1470 #endif 1471 /* Fields commonly accessed by the page reclaim scanner */ 1472 1473 /* 1474 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED. 1475 * 1476 * Use mem_cgroup_lruvec() to look up lruvecs. 1477 */ 1478 struct lruvec __lruvec; 1479 1480 unsigned long flags; 1481 1482 #ifdef CONFIG_LRU_GEN 1483 /* kswap mm walk data */ 1484 struct lru_gen_mm_walk mm_walk; 1485 /* lru_gen_folio list */ 1486 struct lru_gen_memcg memcg_lru; 1487 #endif 1488 1489 CACHELINE_PADDING(_pad2_); 1490 1491 /* Per-node vmstats */ 1492 struct per_cpu_nodestat __percpu *per_cpu_nodestats; 1493 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS]; 1494 #ifdef CONFIG_NUMA 1495 struct memory_tier __rcu *memtier; 1496 #endif 1497 #ifdef CONFIG_MEMORY_FAILURE 1498 struct memory_failure_stats mf_stats; 1499 #endif 1500 } pg_data_t; 1501 1502 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) 1503 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) 1504 1505 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) 1506 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid)) 1507 1508 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat) 1509 { 1510 return pgdat->node_start_pfn + pgdat->node_spanned_pages; 1511 } 1512 1513 #include <linux/memory_hotplug.h> 1514 1515 void build_all_zonelists(pg_data_t *pgdat); 1516 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order, 1517 enum zone_type highest_zoneidx); 1518 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, 1519 int highest_zoneidx, unsigned int alloc_flags, 1520 long free_pages); 1521 bool zone_watermark_ok(struct zone *z, unsigned int order, 1522 unsigned long mark, int highest_zoneidx, 1523 unsigned int alloc_flags); 1524 /* 1525 * Memory initialization context, use to differentiate memory added by 1526 * the platform statically or via memory hotplug interface. 1527 */ 1528 enum meminit_context { 1529 MEMINIT_EARLY, 1530 MEMINIT_HOTPLUG, 1531 }; 1532 1533 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, 1534 unsigned long size); 1535 1536 extern void lruvec_init(struct lruvec *lruvec); 1537 1538 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec) 1539 { 1540 #ifdef CONFIG_MEMCG 1541 return lruvec->pgdat; 1542 #else 1543 return container_of(lruvec, struct pglist_data, __lruvec); 1544 #endif 1545 } 1546 1547 #ifdef CONFIG_HAVE_MEMORYLESS_NODES 1548 int local_memory_node(int node_id); 1549 #else 1550 static inline int local_memory_node(int node_id) { return node_id; }; 1551 #endif 1552 1553 /* 1554 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. 1555 */ 1556 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) 1557 1558 #ifdef CONFIG_ZONE_DEVICE 1559 static inline bool zone_is_zone_device(struct zone *zone) 1560 { 1561 return zone_idx(zone) == ZONE_DEVICE; 1562 } 1563 #else 1564 static inline bool zone_is_zone_device(struct zone *zone) 1565 { 1566 return false; 1567 } 1568 #endif 1569 1570 /* 1571 * Returns true if a zone has pages managed by the buddy allocator. 1572 * All the reclaim decisions have to use this function rather than 1573 * populated_zone(). If the whole zone is reserved then we can easily 1574 * end up with populated_zone() && !managed_zone(). 1575 */ 1576 static inline bool managed_zone(struct zone *zone) 1577 { 1578 return zone_managed_pages(zone); 1579 } 1580 1581 /* Returns true if a zone has memory */ 1582 static inline bool populated_zone(struct zone *zone) 1583 { 1584 return zone->present_pages; 1585 } 1586 1587 #ifdef CONFIG_NUMA 1588 static inline int zone_to_nid(struct zone *zone) 1589 { 1590 return zone->node; 1591 } 1592 1593 static inline void zone_set_nid(struct zone *zone, int nid) 1594 { 1595 zone->node = nid; 1596 } 1597 #else 1598 static inline int zone_to_nid(struct zone *zone) 1599 { 1600 return 0; 1601 } 1602 1603 static inline void zone_set_nid(struct zone *zone, int nid) {} 1604 #endif 1605 1606 extern int movable_zone; 1607 1608 static inline int is_highmem_idx(enum zone_type idx) 1609 { 1610 #ifdef CONFIG_HIGHMEM 1611 return (idx == ZONE_HIGHMEM || 1612 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM)); 1613 #else 1614 return 0; 1615 #endif 1616 } 1617 1618 /** 1619 * is_highmem - helper function to quickly check if a struct zone is a 1620 * highmem zone or not. This is an attempt to keep references 1621 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. 1622 * @zone: pointer to struct zone variable 1623 * Return: 1 for a highmem zone, 0 otherwise 1624 */ 1625 static inline int is_highmem(struct zone *zone) 1626 { 1627 return is_highmem_idx(zone_idx(zone)); 1628 } 1629 1630 #ifdef CONFIG_ZONE_DMA 1631 bool has_managed_dma(void); 1632 #else 1633 static inline bool has_managed_dma(void) 1634 { 1635 return false; 1636 } 1637 #endif 1638 1639 1640 #ifndef CONFIG_NUMA 1641 1642 extern struct pglist_data contig_page_data; 1643 static inline struct pglist_data *NODE_DATA(int nid) 1644 { 1645 return &contig_page_data; 1646 } 1647 1648 #else /* CONFIG_NUMA */ 1649 1650 #include <asm/mmzone.h> 1651 1652 #endif /* !CONFIG_NUMA */ 1653 1654 extern struct pglist_data *first_online_pgdat(void); 1655 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); 1656 extern struct zone *next_zone(struct zone *zone); 1657 1658 /** 1659 * for_each_online_pgdat - helper macro to iterate over all online nodes 1660 * @pgdat: pointer to a pg_data_t variable 1661 */ 1662 #define for_each_online_pgdat(pgdat) \ 1663 for (pgdat = first_online_pgdat(); \ 1664 pgdat; \ 1665 pgdat = next_online_pgdat(pgdat)) 1666 /** 1667 * for_each_zone - helper macro to iterate over all memory zones 1668 * @zone: pointer to struct zone variable 1669 * 1670 * The user only needs to declare the zone variable, for_each_zone 1671 * fills it in. 1672 */ 1673 #define for_each_zone(zone) \ 1674 for (zone = (first_online_pgdat())->node_zones; \ 1675 zone; \ 1676 zone = next_zone(zone)) 1677 1678 #define for_each_populated_zone(zone) \ 1679 for (zone = (first_online_pgdat())->node_zones; \ 1680 zone; \ 1681 zone = next_zone(zone)) \ 1682 if (!populated_zone(zone)) \ 1683 ; /* do nothing */ \ 1684 else 1685 1686 static inline struct zone *zonelist_zone(struct zoneref *zoneref) 1687 { 1688 return zoneref->zone; 1689 } 1690 1691 static inline int zonelist_zone_idx(struct zoneref *zoneref) 1692 { 1693 return zoneref->zone_idx; 1694 } 1695 1696 static inline int zonelist_node_idx(struct zoneref *zoneref) 1697 { 1698 return zone_to_nid(zoneref->zone); 1699 } 1700 1701 struct zoneref *__next_zones_zonelist(struct zoneref *z, 1702 enum zone_type highest_zoneidx, 1703 nodemask_t *nodes); 1704 1705 /** 1706 * 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 1707 * @z: The cursor used as a starting point for the search 1708 * @highest_zoneidx: The zone index of the highest zone to return 1709 * @nodes: An optional nodemask to filter the zonelist with 1710 * 1711 * This function returns the next zone at or below a given zone index that is 1712 * within the allowed nodemask using a cursor as the starting point for the 1713 * search. The zoneref returned is a cursor that represents the current zone 1714 * being examined. It should be advanced by one before calling 1715 * next_zones_zonelist again. 1716 * 1717 * Return: the next zone at or below highest_zoneidx within the allowed 1718 * nodemask using a cursor within a zonelist as a starting point 1719 */ 1720 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z, 1721 enum zone_type highest_zoneidx, 1722 nodemask_t *nodes) 1723 { 1724 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx)) 1725 return z; 1726 return __next_zones_zonelist(z, highest_zoneidx, nodes); 1727 } 1728 1729 /** 1730 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist 1731 * @zonelist: The zonelist to search for a suitable zone 1732 * @highest_zoneidx: The zone index of the highest zone to return 1733 * @nodes: An optional nodemask to filter the zonelist with 1734 * 1735 * This function returns the first zone at or below a given zone index that is 1736 * within the allowed nodemask. The zoneref returned is a cursor that can be 1737 * used to iterate the zonelist with next_zones_zonelist by advancing it by 1738 * one before calling. 1739 * 1740 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is 1741 * never NULL). This may happen either genuinely, or due to concurrent nodemask 1742 * update due to cpuset modification. 1743 * 1744 * Return: Zoneref pointer for the first suitable zone found 1745 */ 1746 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, 1747 enum zone_type highest_zoneidx, 1748 nodemask_t *nodes) 1749 { 1750 return next_zones_zonelist(zonelist->_zonerefs, 1751 highest_zoneidx, nodes); 1752 } 1753 1754 /** 1755 * 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 1756 * @zone: The current zone in the iterator 1757 * @z: The current pointer within zonelist->_zonerefs being iterated 1758 * @zlist: The zonelist being iterated 1759 * @highidx: The zone index of the highest zone to return 1760 * @nodemask: Nodemask allowed by the allocator 1761 * 1762 * This iterator iterates though all zones at or below a given zone index and 1763 * within a given nodemask 1764 */ 1765 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ 1766 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \ 1767 zone; \ 1768 z = next_zones_zonelist(++z, highidx, nodemask), \ 1769 zone = zonelist_zone(z)) 1770 1771 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \ 1772 for (zone = zonelist_zone(z); \ 1773 zone; \ 1774 z = next_zones_zonelist(++z, highidx, nodemask), \ 1775 zone = zonelist_zone(z)) 1776 1777 1778 /** 1779 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index 1780 * @zone: The current zone in the iterator 1781 * @z: The current pointer within zonelist->zones being iterated 1782 * @zlist: The zonelist being iterated 1783 * @highidx: The zone index of the highest zone to return 1784 * 1785 * This iterator iterates though all zones at or below a given zone index. 1786 */ 1787 #define for_each_zone_zonelist(zone, z, zlist, highidx) \ 1788 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) 1789 1790 /* Whether the 'nodes' are all movable nodes */ 1791 static inline bool movable_only_nodes(nodemask_t *nodes) 1792 { 1793 struct zonelist *zonelist; 1794 struct zoneref *z; 1795 int nid; 1796 1797 if (nodes_empty(*nodes)) 1798 return false; 1799 1800 /* 1801 * We can chose arbitrary node from the nodemask to get a 1802 * zonelist as they are interlinked. We just need to find 1803 * at least one zone that can satisfy kernel allocations. 1804 */ 1805 nid = first_node(*nodes); 1806 zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK]; 1807 z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes); 1808 return (!zonelist_zone(z)) ? true : false; 1809 } 1810 1811 1812 #ifdef CONFIG_SPARSEMEM 1813 #include <asm/sparsemem.h> 1814 #endif 1815 1816 #ifdef CONFIG_FLATMEM 1817 #define pfn_to_nid(pfn) (0) 1818 #endif 1819 1820 #ifdef CONFIG_SPARSEMEM 1821 1822 /* 1823 * PA_SECTION_SHIFT physical address to/from section number 1824 * PFN_SECTION_SHIFT pfn to/from section number 1825 */ 1826 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) 1827 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) 1828 1829 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) 1830 1831 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) 1832 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) 1833 1834 #define SECTION_BLOCKFLAGS_BITS \ 1835 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) 1836 1837 #if (MAX_PAGE_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS 1838 #error Allocator MAX_PAGE_ORDER exceeds SECTION_SIZE 1839 #endif 1840 1841 static inline unsigned long pfn_to_section_nr(unsigned long pfn) 1842 { 1843 return pfn >> PFN_SECTION_SHIFT; 1844 } 1845 static inline unsigned long section_nr_to_pfn(unsigned long sec) 1846 { 1847 return sec << PFN_SECTION_SHIFT; 1848 } 1849 1850 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) 1851 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) 1852 1853 #define SUBSECTION_SHIFT 21 1854 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT) 1855 1856 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT) 1857 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT) 1858 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1)) 1859 1860 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS 1861 #error Subsection size exceeds section size 1862 #else 1863 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT)) 1864 #endif 1865 1866 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION) 1867 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK) 1868 1869 struct mem_section_usage { 1870 struct rcu_head rcu; 1871 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1872 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION); 1873 #endif 1874 /* See declaration of similar field in struct zone */ 1875 unsigned long pageblock_flags[0]; 1876 }; 1877 1878 void subsection_map_init(unsigned long pfn, unsigned long nr_pages); 1879 1880 struct page; 1881 struct page_ext; 1882 struct mem_section { 1883 /* 1884 * This is, logically, a pointer to an array of struct 1885 * pages. However, it is stored with some other magic. 1886 * (see sparse.c::sparse_init_one_section()) 1887 * 1888 * Additionally during early boot we encode node id of 1889 * the location of the section here to guide allocation. 1890 * (see sparse.c::memory_present()) 1891 * 1892 * Making it a UL at least makes someone do a cast 1893 * before using it wrong. 1894 */ 1895 unsigned long section_mem_map; 1896 1897 struct mem_section_usage *usage; 1898 #ifdef CONFIG_PAGE_EXTENSION 1899 /* 1900 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use 1901 * section. (see page_ext.h about this.) 1902 */ 1903 struct page_ext *page_ext; 1904 unsigned long pad; 1905 #endif 1906 /* 1907 * WARNING: mem_section must be a power-of-2 in size for the 1908 * calculation and use of SECTION_ROOT_MASK to make sense. 1909 */ 1910 }; 1911 1912 #ifdef CONFIG_SPARSEMEM_EXTREME 1913 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) 1914 #else 1915 #define SECTIONS_PER_ROOT 1 1916 #endif 1917 1918 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) 1919 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) 1920 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) 1921 1922 #ifdef CONFIG_SPARSEMEM_EXTREME 1923 extern struct mem_section **mem_section; 1924 #else 1925 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; 1926 #endif 1927 1928 static inline unsigned long *section_to_usemap(struct mem_section *ms) 1929 { 1930 return ms->usage->pageblock_flags; 1931 } 1932 1933 static inline struct mem_section *__nr_to_section(unsigned long nr) 1934 { 1935 unsigned long root = SECTION_NR_TO_ROOT(nr); 1936 1937 if (unlikely(root >= NR_SECTION_ROOTS)) 1938 return NULL; 1939 1940 #ifdef CONFIG_SPARSEMEM_EXTREME 1941 if (!mem_section || !mem_section[root]) 1942 return NULL; 1943 #endif 1944 return &mem_section[root][nr & SECTION_ROOT_MASK]; 1945 } 1946 extern size_t mem_section_usage_size(void); 1947 1948 /* 1949 * We use the lower bits of the mem_map pointer to store 1950 * a little bit of information. The pointer is calculated 1951 * as mem_map - section_nr_to_pfn(pnum). The result is 1952 * aligned to the minimum alignment of the two values: 1953 * 1. All mem_map arrays are page-aligned. 1954 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT 1955 * lowest bits. PFN_SECTION_SHIFT is arch-specific 1956 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the 1957 * worst combination is powerpc with 256k pages, 1958 * which results in PFN_SECTION_SHIFT equal 6. 1959 * To sum it up, at least 6 bits are available on all architectures. 1960 * However, we can exceed 6 bits on some other architectures except 1961 * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available 1962 * with the worst case of 64K pages on arm64) if we make sure the 1963 * exceeded bit is not applicable to powerpc. 1964 */ 1965 enum { 1966 SECTION_MARKED_PRESENT_BIT, 1967 SECTION_HAS_MEM_MAP_BIT, 1968 SECTION_IS_ONLINE_BIT, 1969 SECTION_IS_EARLY_BIT, 1970 #ifdef CONFIG_ZONE_DEVICE 1971 SECTION_TAINT_ZONE_DEVICE_BIT, 1972 #endif 1973 #ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT 1974 SECTION_IS_VMEMMAP_PREINIT_BIT, 1975 #endif 1976 SECTION_MAP_LAST_BIT, 1977 }; 1978 1979 #define SECTION_MARKED_PRESENT BIT(SECTION_MARKED_PRESENT_BIT) 1980 #define SECTION_HAS_MEM_MAP BIT(SECTION_HAS_MEM_MAP_BIT) 1981 #define SECTION_IS_ONLINE BIT(SECTION_IS_ONLINE_BIT) 1982 #define SECTION_IS_EARLY BIT(SECTION_IS_EARLY_BIT) 1983 #ifdef CONFIG_ZONE_DEVICE 1984 #define SECTION_TAINT_ZONE_DEVICE BIT(SECTION_TAINT_ZONE_DEVICE_BIT) 1985 #endif 1986 #ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT 1987 #define SECTION_IS_VMEMMAP_PREINIT BIT(SECTION_IS_VMEMMAP_PREINIT_BIT) 1988 #endif 1989 #define SECTION_MAP_MASK (~(BIT(SECTION_MAP_LAST_BIT) - 1)) 1990 #define SECTION_NID_SHIFT SECTION_MAP_LAST_BIT 1991 1992 static inline struct page *__section_mem_map_addr(struct mem_section *section) 1993 { 1994 unsigned long map = section->section_mem_map; 1995 map &= SECTION_MAP_MASK; 1996 return (struct page *)map; 1997 } 1998 1999 static inline int present_section(struct mem_section *section) 2000 { 2001 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); 2002 } 2003 2004 static inline int present_section_nr(unsigned long nr) 2005 { 2006 return present_section(__nr_to_section(nr)); 2007 } 2008 2009 static inline int valid_section(struct mem_section *section) 2010 { 2011 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); 2012 } 2013 2014 static inline int early_section(struct mem_section *section) 2015 { 2016 return (section && (section->section_mem_map & SECTION_IS_EARLY)); 2017 } 2018 2019 static inline int valid_section_nr(unsigned long nr) 2020 { 2021 return valid_section(__nr_to_section(nr)); 2022 } 2023 2024 static inline int online_section(struct mem_section *section) 2025 { 2026 return (section && (section->section_mem_map & SECTION_IS_ONLINE)); 2027 } 2028 2029 #ifdef CONFIG_ZONE_DEVICE 2030 static inline int online_device_section(struct mem_section *section) 2031 { 2032 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE; 2033 2034 return section && ((section->section_mem_map & flags) == flags); 2035 } 2036 #else 2037 static inline int online_device_section(struct mem_section *section) 2038 { 2039 return 0; 2040 } 2041 #endif 2042 2043 #ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT 2044 static inline int preinited_vmemmap_section(struct mem_section *section) 2045 { 2046 return (section && 2047 (section->section_mem_map & SECTION_IS_VMEMMAP_PREINIT)); 2048 } 2049 2050 void sparse_vmemmap_init_nid_early(int nid); 2051 void sparse_vmemmap_init_nid_late(int nid); 2052 2053 #else 2054 static inline int preinited_vmemmap_section(struct mem_section *section) 2055 { 2056 return 0; 2057 } 2058 static inline void sparse_vmemmap_init_nid_early(int nid) 2059 { 2060 } 2061 2062 static inline void sparse_vmemmap_init_nid_late(int nid) 2063 { 2064 } 2065 #endif 2066 2067 static inline int online_section_nr(unsigned long nr) 2068 { 2069 return online_section(__nr_to_section(nr)); 2070 } 2071 2072 #ifdef CONFIG_MEMORY_HOTPLUG 2073 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 2074 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 2075 #endif 2076 2077 static inline struct mem_section *__pfn_to_section(unsigned long pfn) 2078 { 2079 return __nr_to_section(pfn_to_section_nr(pfn)); 2080 } 2081 2082 extern unsigned long __highest_present_section_nr; 2083 2084 static inline int subsection_map_index(unsigned long pfn) 2085 { 2086 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION; 2087 } 2088 2089 #ifdef CONFIG_SPARSEMEM_VMEMMAP 2090 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 2091 { 2092 int idx = subsection_map_index(pfn); 2093 struct mem_section_usage *usage = READ_ONCE(ms->usage); 2094 2095 return usage ? test_bit(idx, usage->subsection_map) : 0; 2096 } 2097 2098 static inline bool pfn_section_first_valid(struct mem_section *ms, unsigned long *pfn) 2099 { 2100 struct mem_section_usage *usage = READ_ONCE(ms->usage); 2101 int idx = subsection_map_index(*pfn); 2102 unsigned long bit; 2103 2104 if (!usage) 2105 return false; 2106 2107 if (test_bit(idx, usage->subsection_map)) 2108 return true; 2109 2110 /* Find the next subsection that exists */ 2111 bit = find_next_bit(usage->subsection_map, SUBSECTIONS_PER_SECTION, idx); 2112 if (bit == SUBSECTIONS_PER_SECTION) 2113 return false; 2114 2115 *pfn = (*pfn & PAGE_SECTION_MASK) + (bit * PAGES_PER_SUBSECTION); 2116 return true; 2117 } 2118 #else 2119 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 2120 { 2121 return 1; 2122 } 2123 2124 static inline bool pfn_section_first_valid(struct mem_section *ms, unsigned long *pfn) 2125 { 2126 return true; 2127 } 2128 #endif 2129 2130 void sparse_init_early_section(int nid, struct page *map, unsigned long pnum, 2131 unsigned long flags); 2132 2133 #ifndef CONFIG_HAVE_ARCH_PFN_VALID 2134 /** 2135 * pfn_valid - check if there is a valid memory map entry for a PFN 2136 * @pfn: the page frame number to check 2137 * 2138 * Check if there is a valid memory map entry aka struct page for the @pfn. 2139 * Note, that availability of the memory map entry does not imply that 2140 * there is actual usable memory at that @pfn. The struct page may 2141 * represent a hole or an unusable page frame. 2142 * 2143 * Return: 1 for PFNs that have memory map entries and 0 otherwise 2144 */ 2145 static inline int pfn_valid(unsigned long pfn) 2146 { 2147 struct mem_section *ms; 2148 int ret; 2149 2150 /* 2151 * Ensure the upper PAGE_SHIFT bits are clear in the 2152 * pfn. Else it might lead to false positives when 2153 * some of the upper bits are set, but the lower bits 2154 * match a valid pfn. 2155 */ 2156 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn) 2157 return 0; 2158 2159 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 2160 return 0; 2161 ms = __pfn_to_section(pfn); 2162 rcu_read_lock_sched(); 2163 if (!valid_section(ms)) { 2164 rcu_read_unlock_sched(); 2165 return 0; 2166 } 2167 /* 2168 * Traditionally early sections always returned pfn_valid() for 2169 * the entire section-sized span. 2170 */ 2171 ret = early_section(ms) || pfn_section_valid(ms, pfn); 2172 rcu_read_unlock_sched(); 2173 2174 return ret; 2175 } 2176 2177 /* Returns end_pfn or higher if no valid PFN remaining in range */ 2178 static inline unsigned long first_valid_pfn(unsigned long pfn, unsigned long end_pfn) 2179 { 2180 unsigned long nr = pfn_to_section_nr(pfn); 2181 2182 rcu_read_lock_sched(); 2183 2184 while (nr <= __highest_present_section_nr && pfn < end_pfn) { 2185 struct mem_section *ms = __pfn_to_section(pfn); 2186 2187 if (valid_section(ms) && 2188 (early_section(ms) || pfn_section_first_valid(ms, &pfn))) { 2189 rcu_read_unlock_sched(); 2190 return pfn; 2191 } 2192 2193 /* Nothing left in this section? Skip to next section */ 2194 nr++; 2195 pfn = section_nr_to_pfn(nr); 2196 } 2197 2198 rcu_read_unlock_sched(); 2199 return end_pfn; 2200 } 2201 2202 static inline unsigned long next_valid_pfn(unsigned long pfn, unsigned long end_pfn) 2203 { 2204 pfn++; 2205 2206 if (pfn >= end_pfn) 2207 return end_pfn; 2208 2209 /* 2210 * Either every PFN within the section (or subsection for VMEMMAP) is 2211 * valid, or none of them are. So there's no point repeating the check 2212 * for every PFN; only call first_valid_pfn() again when crossing a 2213 * (sub)section boundary (i.e. !(pfn & ~PAGE_{SUB,}SECTION_MASK)). 2214 */ 2215 if (pfn & ~(IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP) ? 2216 PAGE_SUBSECTION_MASK : PAGE_SECTION_MASK)) 2217 return pfn; 2218 2219 return first_valid_pfn(pfn, end_pfn); 2220 } 2221 2222 2223 #define for_each_valid_pfn(_pfn, _start_pfn, _end_pfn) \ 2224 for ((_pfn) = first_valid_pfn((_start_pfn), (_end_pfn)); \ 2225 (_pfn) < (_end_pfn); \ 2226 (_pfn) = next_valid_pfn((_pfn), (_end_pfn))) 2227 2228 #endif 2229 2230 static inline int pfn_in_present_section(unsigned long pfn) 2231 { 2232 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 2233 return 0; 2234 return present_section(__pfn_to_section(pfn)); 2235 } 2236 2237 static inline unsigned long next_present_section_nr(unsigned long section_nr) 2238 { 2239 while (++section_nr <= __highest_present_section_nr) { 2240 if (present_section_nr(section_nr)) 2241 return section_nr; 2242 } 2243 2244 return -1; 2245 } 2246 2247 #define for_each_present_section_nr(start, section_nr) \ 2248 for (section_nr = next_present_section_nr(start - 1); \ 2249 section_nr != -1; \ 2250 section_nr = next_present_section_nr(section_nr)) 2251 2252 /* 2253 * These are _only_ used during initialisation, therefore they 2254 * can use __initdata ... They could have names to indicate 2255 * this restriction. 2256 */ 2257 #ifdef CONFIG_NUMA 2258 #define pfn_to_nid(pfn) \ 2259 ({ \ 2260 unsigned long __pfn_to_nid_pfn = (pfn); \ 2261 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ 2262 }) 2263 #else 2264 #define pfn_to_nid(pfn) (0) 2265 #endif 2266 2267 void sparse_init(void); 2268 #else 2269 #define sparse_init() do {} while (0) 2270 #define sparse_index_init(_sec, _nid) do {} while (0) 2271 #define sparse_vmemmap_init_nid_early(_nid, _use) do {} while (0) 2272 #define sparse_vmemmap_init_nid_late(_nid) do {} while (0) 2273 #define pfn_in_present_section pfn_valid 2274 #define subsection_map_init(_pfn, _nr_pages) do {} while (0) 2275 #endif /* CONFIG_SPARSEMEM */ 2276 2277 /* 2278 * Fallback case for when the architecture provides its own pfn_valid() but 2279 * not a corresponding for_each_valid_pfn(). 2280 */ 2281 #ifndef for_each_valid_pfn 2282 #define for_each_valid_pfn(_pfn, _start_pfn, _end_pfn) \ 2283 for ((_pfn) = (_start_pfn); (_pfn) < (_end_pfn); (_pfn)++) \ 2284 if (pfn_valid(_pfn)) 2285 #endif 2286 2287 #endif /* !__GENERATING_BOUNDS.H */ 2288 #endif /* !__ASSEMBLY__ */ 2289 #endif /* _LINUX_MMZONE_H */ 2290