1 /* SPDX-License-Identifier: GPL-2.0 */
5 #include <linux/spinlock.h>
6 #include <linux/linkage.h>
7 #include <linux/mmzone.h>
8 #include <linux/list.h>
9 #include <linux/memcontrol.h>
10 #include <linux/sched.h>
11 #include <linux/node.h>
13 #include <linux/pagemap.h>
14 #include <linux/atomic.h>
15 #include <linux/page-flags.h>
16 #include <uapi/linux/mempolicy.h>
19 struct notifier_block;
25 #define SWAP_FLAG_PREFER 0x8000 /* set if swap priority specified */
26 #define SWAP_FLAG_PRIO_MASK 0x7fff
27 #define SWAP_FLAG_PRIO_SHIFT 0
28 #define SWAP_FLAG_DISCARD 0x10000 /* enable discard for swap */
29 #define SWAP_FLAG_DISCARD_ONCE 0x20000 /* discard swap area at swapon-time */
30 #define SWAP_FLAG_DISCARD_PAGES 0x40000 /* discard page-clusters after use */
32 #define SWAP_FLAGS_VALID (SWAP_FLAG_PRIO_MASK | SWAP_FLAG_PREFER | \
33 SWAP_FLAG_DISCARD | SWAP_FLAG_DISCARD_ONCE | \
34 SWAP_FLAG_DISCARD_PAGES)
37 static inline int current_is_kswapd(void)
39 return current->flags & PF_KSWAPD;
43 * MAX_SWAPFILES defines the maximum number of swaptypes: things which can
44 * be swapped to. The swap type and the offset into that swap type are
45 * encoded into pte's and into pgoff_t's in the swapcache. Using five bits
46 * for the type means that the maximum number of swapcache pages is 27 bits
47 * on 32-bit-pgoff_t architectures. And that assumes that the architecture packs
48 * the type/offset into the pte as 5/27 as well.
50 #define MAX_SWAPFILES_SHIFT 5
53 * Use some of the swap files numbers for other purposes. This
54 * is a convenient way to hook into the VM to trigger special
59 * PTE markers are used to persist information onto PTEs that are mapped with
60 * file-backed memories. As its name "PTE" hints, it should only be applied to
61 * the leaves of pgtables.
63 #ifdef CONFIG_PTE_MARKER
64 #define SWP_PTE_MARKER_NUM 1
65 #define SWP_PTE_MARKER (MAX_SWAPFILES + SWP_HWPOISON_NUM + \
66 SWP_MIGRATION_NUM + SWP_DEVICE_NUM)
68 #define SWP_PTE_MARKER_NUM 0
72 * Unaddressable device memory support. See include/linux/hmm.h and
73 * Documentation/vm/hmm.rst. Short description is we need struct pages for
74 * device memory that is unaddressable (inaccessible) by CPU, so that we can
75 * migrate part of a process memory to device memory.
77 * When a page is migrated from CPU to device, we set the CPU page table entry
78 * to a special SWP_DEVICE_{READ|WRITE} entry.
80 * When a page is mapped by the device for exclusive access we set the CPU page
81 * table entries to special SWP_DEVICE_EXCLUSIVE_* entries.
83 #ifdef CONFIG_DEVICE_PRIVATE
84 #define SWP_DEVICE_NUM 4
85 #define SWP_DEVICE_WRITE (MAX_SWAPFILES+SWP_HWPOISON_NUM+SWP_MIGRATION_NUM)
86 #define SWP_DEVICE_READ (MAX_SWAPFILES+SWP_HWPOISON_NUM+SWP_MIGRATION_NUM+1)
87 #define SWP_DEVICE_EXCLUSIVE_WRITE (MAX_SWAPFILES+SWP_HWPOISON_NUM+SWP_MIGRATION_NUM+2)
88 #define SWP_DEVICE_EXCLUSIVE_READ (MAX_SWAPFILES+SWP_HWPOISON_NUM+SWP_MIGRATION_NUM+3)
90 #define SWP_DEVICE_NUM 0
94 * Page migration support.
96 * SWP_MIGRATION_READ_EXCLUSIVE is only applicable to anonymous pages and
97 * indicates that the referenced (part of) an anonymous page is exclusive to
98 * a single process. For SWP_MIGRATION_WRITE, that information is implicit:
99 * (part of) an anonymous page that are mapped writable are exclusive to a
102 #ifdef CONFIG_MIGRATION
103 #define SWP_MIGRATION_NUM 3
104 #define SWP_MIGRATION_READ (MAX_SWAPFILES + SWP_HWPOISON_NUM)
105 #define SWP_MIGRATION_READ_EXCLUSIVE (MAX_SWAPFILES + SWP_HWPOISON_NUM + 1)
106 #define SWP_MIGRATION_WRITE (MAX_SWAPFILES + SWP_HWPOISON_NUM + 2)
108 #define SWP_MIGRATION_NUM 0
112 * Handling of hardware poisoned pages with memory corruption.
114 #ifdef CONFIG_MEMORY_FAILURE
115 #define SWP_HWPOISON_NUM 1
116 #define SWP_HWPOISON MAX_SWAPFILES
118 #define SWP_HWPOISON_NUM 0
121 #define MAX_SWAPFILES \
122 ((1 << MAX_SWAPFILES_SHIFT) - SWP_DEVICE_NUM - \
123 SWP_MIGRATION_NUM - SWP_HWPOISON_NUM - SWP_PTE_MARKER_NUM)
126 * Magic header for a swap area. The first part of the union is
127 * what the swap magic looks like for the old (limited to 128MB)
128 * swap area format, the second part of the union adds - in the
129 * old reserved area - some extra information. Note that the first
130 * kilobyte is reserved for boot loader or disk label stuff...
132 * Having the magic at the end of the PAGE_SIZE makes detecting swap
133 * areas somewhat tricky on machines that support multiple page sizes.
134 * For 2.5 we'll probably want to move the magic to just beyond the
139 char reserved[PAGE_SIZE - 10];
140 char magic[10]; /* SWAP-SPACE or SWAPSPACE2 */
143 char bootbits[1024]; /* Space for disklabel etc. */
147 unsigned char sws_uuid[16];
148 unsigned char sws_volume[16];
155 * current->reclaim_state points to one of these when a task is running
158 struct reclaim_state {
159 unsigned long reclaimed_slab;
164 struct address_space;
166 struct writeback_control;
170 * A swap extent maps a range of a swapfile's PAGE_SIZE pages onto a range of
171 * disk blocks. A rbtree of swap extents maps the entire swapfile (Where the
172 * term `swapfile' refers to either a blockdevice or an IS_REG file). Apart
173 * from setup, they're handled identically.
175 * We always assume that blocks are of size PAGE_SIZE.
178 struct rb_node rb_node;
181 sector_t start_block;
185 * Max bad pages in the new format..
187 #define MAX_SWAP_BADPAGES \
188 ((offsetof(union swap_header, magic.magic) - \
189 offsetof(union swap_header, info.badpages)) / sizeof(int))
192 SWP_USED = (1 << 0), /* is slot in swap_info[] used? */
193 SWP_WRITEOK = (1 << 1), /* ok to write to this swap? */
194 SWP_DISCARDABLE = (1 << 2), /* blkdev support discard */
195 SWP_DISCARDING = (1 << 3), /* now discarding a free cluster */
196 SWP_SOLIDSTATE = (1 << 4), /* blkdev seeks are cheap */
197 SWP_CONTINUED = (1 << 5), /* swap_map has count continuation */
198 SWP_BLKDEV = (1 << 6), /* its a block device */
199 SWP_ACTIVATED = (1 << 7), /* set after swap_activate success */
200 SWP_FS_OPS = (1 << 8), /* swapfile operations go through fs */
201 SWP_AREA_DISCARD = (1 << 9), /* single-time swap area discards */
202 SWP_PAGE_DISCARD = (1 << 10), /* freed swap page-cluster discards */
203 SWP_STABLE_WRITES = (1 << 11), /* no overwrite PG_writeback pages */
204 SWP_SYNCHRONOUS_IO = (1 << 12), /* synchronous IO is efficient */
205 /* add others here before... */
206 SWP_SCANNING = (1 << 14), /* refcount in scan_swap_map */
209 #define SWAP_CLUSTER_MAX 32UL
210 #define COMPACT_CLUSTER_MAX SWAP_CLUSTER_MAX
212 /* Bit flag in swap_map */
213 #define SWAP_HAS_CACHE 0x40 /* Flag page is cached, in first swap_map */
214 #define COUNT_CONTINUED 0x80 /* Flag swap_map continuation for full count */
216 /* Special value in first swap_map */
217 #define SWAP_MAP_MAX 0x3e /* Max count */
218 #define SWAP_MAP_BAD 0x3f /* Note page is bad */
219 #define SWAP_MAP_SHMEM 0xbf /* Owned by shmem/tmpfs */
221 /* Special value in each swap_map continuation */
222 #define SWAP_CONT_MAX 0x7f /* Max count */
225 * We use this to track usage of a cluster. A cluster is a block of swap disk
226 * space with SWAPFILE_CLUSTER pages long and naturally aligns in disk. All
227 * free clusters are organized into a list. We fetch an entry from the list to
228 * get a free cluster.
230 * The data field stores next cluster if the cluster is free or cluster usage
231 * counter otherwise. The flags field determines if a cluster is free. This is
232 * protected by swap_info_struct.lock.
234 struct swap_cluster_info {
236 * Protect swap_cluster_info fields
237 * and swap_info_struct->swap_map
238 * elements correspond to the swap
241 unsigned int data:24;
242 unsigned int flags:8;
244 #define CLUSTER_FLAG_FREE 1 /* This cluster is free */
245 #define CLUSTER_FLAG_NEXT_NULL 2 /* This cluster has no next cluster */
246 #define CLUSTER_FLAG_HUGE 4 /* This cluster is backing a transparent huge page */
249 * We assign a cluster to each CPU, so each CPU can allocate swap entry from
250 * its own cluster and swapout sequentially. The purpose is to optimize swapout
253 struct percpu_cluster {
254 struct swap_cluster_info index; /* Current cluster index */
255 unsigned int next; /* Likely next allocation offset */
258 struct swap_cluster_list {
259 struct swap_cluster_info head;
260 struct swap_cluster_info tail;
264 * The in-memory structure used to track swap areas.
266 struct swap_info_struct {
267 struct percpu_ref users; /* indicate and keep swap device valid. */
268 unsigned long flags; /* SWP_USED etc: see above */
269 signed short prio; /* swap priority of this type */
270 struct plist_node list; /* entry in swap_active_head */
271 signed char type; /* strange name for an index */
272 unsigned int max; /* extent of the swap_map */
273 unsigned char *swap_map; /* vmalloc'ed array of usage counts */
274 struct swap_cluster_info *cluster_info; /* cluster info. Only for SSD */
275 struct swap_cluster_list free_clusters; /* free clusters list */
276 unsigned int lowest_bit; /* index of first free in swap_map */
277 unsigned int highest_bit; /* index of last free in swap_map */
278 unsigned int pages; /* total of usable pages of swap */
279 unsigned int inuse_pages; /* number of those currently in use */
280 unsigned int cluster_next; /* likely index for next allocation */
281 unsigned int cluster_nr; /* countdown to next cluster search */
282 unsigned int __percpu *cluster_next_cpu; /*percpu index for next allocation */
283 struct percpu_cluster __percpu *percpu_cluster; /* per cpu's swap location */
284 struct rb_root swap_extent_root;/* root of the swap extent rbtree */
285 struct block_device *bdev; /* swap device or bdev of swap file */
286 struct file *swap_file; /* seldom referenced */
287 unsigned int old_block_size; /* seldom referenced */
288 struct completion comp; /* seldom referenced */
289 #ifdef CONFIG_FRONTSWAP
290 unsigned long *frontswap_map; /* frontswap in-use, one bit per page */
291 atomic_t frontswap_pages; /* frontswap pages in-use counter */
294 * protect map scan related fields like
295 * swap_map, lowest_bit, highest_bit,
296 * inuse_pages, cluster_next,
297 * cluster_nr, lowest_alloc,
298 * highest_alloc, free/discard cluster
299 * list. other fields are only changed
300 * at swapon/swapoff, so are protected
301 * by swap_lock. changing flags need
302 * hold this lock and swap_lock. If
303 * both locks need hold, hold swap_lock
306 spinlock_t cont_lock; /*
307 * protect swap count continuation page
310 struct work_struct discard_work; /* discard worker */
311 struct swap_cluster_list discard_clusters; /* discard clusters list */
312 struct plist_node avail_lists[]; /*
313 * entries in swap_avail_heads, one
315 * Must be last as the number of the
316 * array is nr_node_ids, which is not
317 * a fixed value so have to allocate
319 * And it has to be an array so that
320 * plist_for_each_* can work.
325 #define SWAP_RA_ORDER_CEILING 5
327 /* Avoid stack overflow, because we need to save part of page table */
328 #define SWAP_RA_ORDER_CEILING 3
329 #define SWAP_RA_PTE_CACHE_SIZE (1 << SWAP_RA_ORDER_CEILING)
332 struct vma_swap_readahead {
334 unsigned short offset;
335 unsigned short nr_pte;
339 pte_t ptes[SWAP_RA_PTE_CACHE_SIZE];
343 static inline swp_entry_t folio_swap_entry(struct folio *folio)
345 swp_entry_t entry = { .val = page_private(&folio->page) };
349 /* linux/mm/workingset.c */
350 void workingset_age_nonresident(struct lruvec *lruvec, unsigned long nr_pages);
351 void *workingset_eviction(struct folio *folio, struct mem_cgroup *target_memcg);
352 void workingset_refault(struct folio *folio, void *shadow);
353 void workingset_activation(struct folio *folio);
355 /* Only track the nodes of mappings with shadow entries */
356 void workingset_update_node(struct xa_node *node);
357 extern struct list_lru shadow_nodes;
358 #define mapping_set_update(xas, mapping) do { \
359 if (!dax_mapping(mapping) && !shmem_mapping(mapping)) { \
360 xas_set_update(xas, workingset_update_node); \
361 xas_set_lru(xas, &shadow_nodes); \
365 /* linux/mm/page_alloc.c */
366 extern unsigned long totalreserve_pages;
368 /* Definition of global_zone_page_state not available yet */
369 #define nr_free_pages() global_zone_page_state(NR_FREE_PAGES)
372 /* linux/mm/swap.c */
373 extern void lru_note_cost(struct lruvec *lruvec, bool file,
374 unsigned int nr_pages);
375 extern void lru_note_cost_folio(struct folio *);
376 extern void folio_add_lru(struct folio *);
377 extern void lru_cache_add(struct page *);
378 void mark_page_accessed(struct page *);
379 void folio_mark_accessed(struct folio *);
381 extern atomic_t lru_disable_count;
383 static inline bool lru_cache_disabled(void)
385 return atomic_read(&lru_disable_count);
388 static inline void lru_cache_enable(void)
390 atomic_dec(&lru_disable_count);
393 extern void lru_cache_disable(void);
394 extern void lru_add_drain(void);
395 extern void lru_add_drain_cpu(int cpu);
396 extern void lru_add_drain_cpu_zone(struct zone *zone);
397 extern void lru_add_drain_all(void);
398 extern void deactivate_page(struct page *page);
399 extern void mark_page_lazyfree(struct page *page);
400 extern void swap_setup(void);
402 extern void lru_cache_add_inactive_or_unevictable(struct page *page,
403 struct vm_area_struct *vma);
405 /* linux/mm/vmscan.c */
406 extern unsigned long zone_reclaimable_pages(struct zone *zone);
407 extern unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
408 gfp_t gfp_mask, nodemask_t *mask);
409 extern unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
410 unsigned long nr_pages,
413 extern unsigned long mem_cgroup_shrink_node(struct mem_cgroup *mem,
414 gfp_t gfp_mask, bool noswap,
416 unsigned long *nr_scanned);
417 extern unsigned long shrink_all_memory(unsigned long nr_pages);
418 extern int vm_swappiness;
419 long remove_mapping(struct address_space *mapping, struct folio *folio);
421 extern unsigned long reclaim_pages(struct list_head *page_list);
423 extern int node_reclaim_mode;
424 extern int sysctl_min_unmapped_ratio;
425 extern int sysctl_min_slab_ratio;
427 #define node_reclaim_mode 0
430 static inline bool node_reclaim_enabled(void)
432 /* Is any node_reclaim_mode bit set? */
433 return node_reclaim_mode & (RECLAIM_ZONE|RECLAIM_WRITE|RECLAIM_UNMAP);
436 extern void check_move_unevictable_pages(struct pagevec *pvec);
438 extern void kswapd_run(int nid);
439 extern void kswapd_stop(int nid);
443 int add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
444 unsigned long nr_pages, sector_t start_block);
445 int generic_swapfile_activate(struct swap_info_struct *, struct file *,
448 static inline unsigned long total_swapcache_pages(void)
450 return global_node_page_state(NR_SWAPCACHE);
453 extern void free_page_and_swap_cache(struct page *);
454 extern void free_pages_and_swap_cache(struct page **, int);
455 /* linux/mm/swapfile.c */
456 extern atomic_long_t nr_swap_pages;
457 extern long total_swap_pages;
458 extern atomic_t nr_rotate_swap;
459 extern bool has_usable_swap(void);
461 /* Swap 50% full? Release swapcache more aggressively.. */
462 static inline bool vm_swap_full(void)
464 return atomic_long_read(&nr_swap_pages) * 2 < total_swap_pages;
467 static inline long get_nr_swap_pages(void)
469 return atomic_long_read(&nr_swap_pages);
472 extern void si_swapinfo(struct sysinfo *);
473 swp_entry_t folio_alloc_swap(struct folio *folio);
474 extern void put_swap_page(struct page *page, swp_entry_t entry);
475 extern swp_entry_t get_swap_page_of_type(int);
476 extern int get_swap_pages(int n, swp_entry_t swp_entries[], int entry_size);
477 extern int add_swap_count_continuation(swp_entry_t, gfp_t);
478 extern void swap_shmem_alloc(swp_entry_t);
479 extern int swap_duplicate(swp_entry_t);
480 extern int swapcache_prepare(swp_entry_t);
481 extern void swap_free(swp_entry_t);
482 extern void swapcache_free_entries(swp_entry_t *entries, int n);
483 extern int free_swap_and_cache(swp_entry_t);
484 int swap_type_of(dev_t device, sector_t offset);
485 int find_first_swap(dev_t *device);
486 extern unsigned int count_swap_pages(int, int);
487 extern sector_t swapdev_block(int, pgoff_t);
488 extern int __swap_count(swp_entry_t entry);
489 extern int __swp_swapcount(swp_entry_t entry);
490 extern int swp_swapcount(swp_entry_t entry);
491 extern struct swap_info_struct *page_swap_info(struct page *);
492 extern struct swap_info_struct *swp_swap_info(swp_entry_t entry);
493 extern int try_to_free_swap(struct page *);
494 struct backing_dev_info;
495 extern int init_swap_address_space(unsigned int type, unsigned long nr_pages);
496 extern void exit_swap_address_space(unsigned int type);
497 extern struct swap_info_struct *get_swap_device(swp_entry_t entry);
498 sector_t swap_page_sector(struct page *page);
500 static inline void put_swap_device(struct swap_info_struct *si)
502 percpu_ref_put(&si->users);
505 #else /* CONFIG_SWAP */
506 static inline struct swap_info_struct *swp_swap_info(swp_entry_t entry)
511 static inline struct swap_info_struct *get_swap_device(swp_entry_t entry)
516 static inline void put_swap_device(struct swap_info_struct *si)
520 #define get_nr_swap_pages() 0L
521 #define total_swap_pages 0L
522 #define total_swapcache_pages() 0UL
523 #define vm_swap_full() 0
525 #define si_swapinfo(val) \
526 do { (val)->freeswap = (val)->totalswap = 0; } while (0)
527 /* only sparc can not include linux/pagemap.h in this file
528 * so leave put_page and release_pages undeclared... */
529 #define free_page_and_swap_cache(page) \
531 #define free_pages_and_swap_cache(pages, nr) \
532 release_pages((pages), (nr));
534 /* used to sanity check ptes in zap_pte_range when CONFIG_SWAP=0 */
535 #define free_swap_and_cache(e) is_pfn_swap_entry(e)
537 static inline int add_swap_count_continuation(swp_entry_t swp, gfp_t gfp_mask)
542 static inline void swap_shmem_alloc(swp_entry_t swp)
546 static inline int swap_duplicate(swp_entry_t swp)
551 static inline void swap_free(swp_entry_t swp)
555 static inline void put_swap_page(struct page *page, swp_entry_t swp)
559 static inline int __swap_count(swp_entry_t entry)
564 static inline int __swp_swapcount(swp_entry_t entry)
569 static inline int swp_swapcount(swp_entry_t entry)
574 static inline int try_to_free_swap(struct page *page)
579 static inline swp_entry_t folio_alloc_swap(struct folio *folio)
586 static inline int add_swap_extent(struct swap_info_struct *sis,
587 unsigned long start_page,
588 unsigned long nr_pages, sector_t start_block)
592 #endif /* CONFIG_SWAP */
594 #ifdef CONFIG_THP_SWAP
595 extern int split_swap_cluster(swp_entry_t entry);
597 static inline int split_swap_cluster(swp_entry_t entry)
604 static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
606 /* Cgroup2 doesn't have per-cgroup swappiness */
607 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
608 return vm_swappiness;
611 if (mem_cgroup_disabled() || mem_cgroup_is_root(memcg))
612 return vm_swappiness;
614 return memcg->swappiness;
617 static inline int mem_cgroup_swappiness(struct mem_cgroup *mem)
619 return vm_swappiness;
624 extern u64 zswap_pool_total_size;
625 extern atomic_t zswap_stored_pages;
628 #if defined(CONFIG_SWAP) && defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
629 extern void __cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask);
630 static inline void cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask)
632 if (mem_cgroup_disabled())
634 __cgroup_throttle_swaprate(page, gfp_mask);
637 static inline void cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask)
641 static inline void folio_throttle_swaprate(struct folio *folio, gfp_t gfp)
643 cgroup_throttle_swaprate(&folio->page, gfp);
646 #ifdef CONFIG_MEMCG_SWAP
647 void mem_cgroup_swapout(struct folio *folio, swp_entry_t entry);
648 int __mem_cgroup_try_charge_swap(struct folio *folio, swp_entry_t entry);
649 static inline int mem_cgroup_try_charge_swap(struct folio *folio,
652 if (mem_cgroup_disabled())
654 return __mem_cgroup_try_charge_swap(folio, entry);
657 extern void __mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages);
658 static inline void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages)
660 if (mem_cgroup_disabled())
662 __mem_cgroup_uncharge_swap(entry, nr_pages);
665 extern long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg);
666 extern bool mem_cgroup_swap_full(struct page *page);
668 static inline void mem_cgroup_swapout(struct folio *folio, swp_entry_t entry)
672 static inline int mem_cgroup_try_charge_swap(struct folio *folio,
678 static inline void mem_cgroup_uncharge_swap(swp_entry_t entry,
679 unsigned int nr_pages)
683 static inline long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
685 return get_nr_swap_pages();
688 static inline bool mem_cgroup_swap_full(struct page *page)
690 return vm_swap_full();
694 #endif /* __KERNEL__*/
695 #endif /* _LINUX_SWAP_H */