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