| 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | #ifndef _MM_SWAP_H |
| 3 | #define _MM_SWAP_H |
| 4 | |
| 5 | #include <linux/atomic.h> /* for atomic_long_t */ |
| 6 | struct mempolicy; |
| 7 | struct swap_iocb; |
| 8 | |
| 9 | extern int page_cluster; |
| 10 | |
| 11 | #ifdef CONFIG_THP_SWAP |
| 12 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR |
| 13 | #define swap_entry_order(order) (order) |
| 14 | #else |
| 15 | #define SWAPFILE_CLUSTER 256 |
| 16 | #define swap_entry_order(order) 0 |
| 17 | #endif |
| 18 | |
| 19 | extern struct swap_info_struct *swap_info[]; |
| 20 | |
| 21 | /* |
| 22 | * We use this to track usage of a cluster. A cluster is a block of swap disk |
| 23 | * space with SWAPFILE_CLUSTER pages long and naturally aligns in disk. All |
| 24 | * free clusters are organized into a list. We fetch an entry from the list to |
| 25 | * get a free cluster. |
| 26 | * |
| 27 | * The flags field determines if a cluster is free. This is |
| 28 | * protected by cluster lock. |
| 29 | */ |
| 30 | struct swap_cluster_info { |
| 31 | spinlock_t lock; /* |
| 32 | * Protect swap_cluster_info fields |
| 33 | * other than list, and swap_info_struct->swap_map |
| 34 | * elements corresponding to the swap cluster. |
| 35 | */ |
| 36 | u16 count; |
| 37 | u8 flags; |
| 38 | u8 order; |
| 39 | atomic_long_t __rcu *table; /* Swap table entries, see mm/swap_table.h */ |
| 40 | struct list_head list; |
| 41 | }; |
| 42 | |
| 43 | /* All on-list cluster must have a non-zero flag. */ |
| 44 | enum swap_cluster_flags { |
| 45 | CLUSTER_FLAG_NONE = 0, /* For temporary off-list cluster */ |
| 46 | CLUSTER_FLAG_FREE, |
| 47 | CLUSTER_FLAG_NONFULL, |
| 48 | CLUSTER_FLAG_FRAG, |
| 49 | /* Clusters with flags above are allocatable */ |
| 50 | CLUSTER_FLAG_USABLE = CLUSTER_FLAG_FRAG, |
| 51 | CLUSTER_FLAG_FULL, |
| 52 | CLUSTER_FLAG_DISCARD, |
| 53 | CLUSTER_FLAG_MAX, |
| 54 | }; |
| 55 | |
| 56 | #ifdef CONFIG_SWAP |
| 57 | #include <linux/swapops.h> /* for swp_offset */ |
| 58 | #include <linux/blk_types.h> /* for bio_end_io_t */ |
| 59 | |
| 60 | static inline unsigned int swp_cluster_offset(swp_entry_t entry) |
| 61 | { |
| 62 | return swp_offset(entry) % SWAPFILE_CLUSTER; |
| 63 | } |
| 64 | |
| 65 | /* |
| 66 | * Callers of all helpers below must ensure the entry, type, or offset is |
| 67 | * valid, and protect the swap device with reference count or locks. |
| 68 | */ |
| 69 | static inline struct swap_info_struct *__swap_type_to_info(int type) |
| 70 | { |
| 71 | struct swap_info_struct *si; |
| 72 | |
| 73 | si = READ_ONCE(swap_info[type]); /* rcu_dereference() */ |
| 74 | VM_WARN_ON_ONCE(percpu_ref_is_zero(&si->users)); /* race with swapoff */ |
| 75 | return si; |
| 76 | } |
| 77 | |
| 78 | static inline struct swap_info_struct *__swap_entry_to_info(swp_entry_t entry) |
| 79 | { |
| 80 | return __swap_type_to_info(type: swp_type(entry)); |
| 81 | } |
| 82 | |
| 83 | static inline struct swap_cluster_info *__swap_offset_to_cluster( |
| 84 | struct swap_info_struct *si, pgoff_t offset) |
| 85 | { |
| 86 | VM_WARN_ON_ONCE(percpu_ref_is_zero(&si->users)); /* race with swapoff */ |
| 87 | VM_WARN_ON_ONCE(offset >= si->max); |
| 88 | return &si->cluster_info[offset / SWAPFILE_CLUSTER]; |
| 89 | } |
| 90 | |
| 91 | static inline struct swap_cluster_info *__swap_entry_to_cluster(swp_entry_t entry) |
| 92 | { |
| 93 | return __swap_offset_to_cluster(si: __swap_entry_to_info(entry), |
| 94 | offset: swp_offset(entry)); |
| 95 | } |
| 96 | |
| 97 | static __always_inline struct swap_cluster_info *__swap_cluster_lock( |
| 98 | struct swap_info_struct *si, unsigned long offset, bool irq) |
| 99 | { |
| 100 | struct swap_cluster_info *ci = __swap_offset_to_cluster(si, offset); |
| 101 | |
| 102 | /* |
| 103 | * Nothing modifies swap cache in an IRQ context. All access to |
| 104 | * swap cache is wrapped by swap_cache_* helpers, and swap cache |
| 105 | * writeback is handled outside of IRQs. Swapin or swapout never |
| 106 | * occurs in IRQ, and neither does in-place split or replace. |
| 107 | * |
| 108 | * Besides, modifying swap cache requires synchronization with |
| 109 | * swap_map, which was never IRQ safe. |
| 110 | */ |
| 111 | VM_WARN_ON_ONCE(!in_task()); |
| 112 | VM_WARN_ON_ONCE(percpu_ref_is_zero(&si->users)); /* race with swapoff */ |
| 113 | if (irq) |
| 114 | spin_lock_irq(lock: &ci->lock); |
| 115 | else |
| 116 | spin_lock(lock: &ci->lock); |
| 117 | return ci; |
| 118 | } |
| 119 | |
| 120 | /** |
| 121 | * swap_cluster_lock - Lock and return the swap cluster of given offset. |
| 122 | * @si: swap device the cluster belongs to. |
| 123 | * @offset: the swap entry offset, pointing to a valid slot. |
| 124 | * |
| 125 | * Context: The caller must ensure the offset is in the valid range and |
| 126 | * protect the swap device with reference count or locks. |
| 127 | */ |
| 128 | static inline struct swap_cluster_info *swap_cluster_lock( |
| 129 | struct swap_info_struct *si, unsigned long offset) |
| 130 | { |
| 131 | return __swap_cluster_lock(si, offset, irq: false); |
| 132 | } |
| 133 | |
| 134 | static inline struct swap_cluster_info *__swap_cluster_get_and_lock( |
| 135 | const struct folio *folio, bool irq) |
| 136 | { |
| 137 | VM_WARN_ON_ONCE_FOLIO(!folio_test_locked(folio), folio); |
| 138 | VM_WARN_ON_ONCE_FOLIO(!folio_test_swapcache(folio), folio); |
| 139 | return __swap_cluster_lock(si: __swap_entry_to_info(entry: folio->swap), |
| 140 | offset: swp_offset(entry: folio->swap), irq); |
| 141 | } |
| 142 | |
| 143 | /* |
| 144 | * swap_cluster_get_and_lock - Locks the cluster that holds a folio's entries. |
| 145 | * @folio: The folio. |
| 146 | * |
| 147 | * This locks and returns the swap cluster that contains a folio's swap |
| 148 | * entries. The swap entries of a folio are always in one single cluster. |
| 149 | * The folio has to be locked so its swap entries won't change and the |
| 150 | * cluster won't be freed. |
| 151 | * |
| 152 | * Context: Caller must ensure the folio is locked and in the swap cache. |
| 153 | * Return: Pointer to the swap cluster. |
| 154 | */ |
| 155 | static inline struct swap_cluster_info *swap_cluster_get_and_lock( |
| 156 | const struct folio *folio) |
| 157 | { |
| 158 | return __swap_cluster_get_and_lock(folio, irq: false); |
| 159 | } |
| 160 | |
| 161 | /* |
| 162 | * swap_cluster_get_and_lock_irq - Locks the cluster that holds a folio's entries. |
| 163 | * @folio: The folio. |
| 164 | * |
| 165 | * Same as swap_cluster_get_and_lock but also disable IRQ. |
| 166 | * |
| 167 | * Context: Caller must ensure the folio is locked and in the swap cache. |
| 168 | * Return: Pointer to the swap cluster. |
| 169 | */ |
| 170 | static inline struct swap_cluster_info *swap_cluster_get_and_lock_irq( |
| 171 | const struct folio *folio) |
| 172 | { |
| 173 | return __swap_cluster_get_and_lock(folio, irq: true); |
| 174 | } |
| 175 | |
| 176 | static inline void swap_cluster_unlock(struct swap_cluster_info *ci) |
| 177 | { |
| 178 | spin_unlock(lock: &ci->lock); |
| 179 | } |
| 180 | |
| 181 | static inline void swap_cluster_unlock_irq(struct swap_cluster_info *ci) |
| 182 | { |
| 183 | spin_unlock_irq(lock: &ci->lock); |
| 184 | } |
| 185 | |
| 186 | /* linux/mm/page_io.c */ |
| 187 | int sio_pool_init(void); |
| 188 | struct swap_iocb; |
| 189 | void swap_read_folio(struct folio *folio, struct swap_iocb **plug); |
| 190 | void __swap_read_unplug(struct swap_iocb *plug); |
| 191 | static inline void swap_read_unplug(struct swap_iocb *plug) |
| 192 | { |
| 193 | if (unlikely(plug)) |
| 194 | __swap_read_unplug(plug); |
| 195 | } |
| 196 | void swap_write_unplug(struct swap_iocb *sio); |
| 197 | int swap_writeout(struct folio *folio, struct swap_iocb **swap_plug); |
| 198 | void __swap_writepage(struct folio *folio, struct swap_iocb **swap_plug); |
| 199 | |
| 200 | /* linux/mm/swap_state.c */ |
| 201 | extern struct address_space swap_space __ro_after_init; |
| 202 | static inline struct address_space *swap_address_space(swp_entry_t entry) |
| 203 | { |
| 204 | return &swap_space; |
| 205 | } |
| 206 | |
| 207 | /* |
| 208 | * Return the swap device position of the swap entry. |
| 209 | */ |
| 210 | static inline loff_t swap_dev_pos(swp_entry_t entry) |
| 211 | { |
| 212 | return ((loff_t)swp_offset(entry)) << PAGE_SHIFT; |
| 213 | } |
| 214 | |
| 215 | /** |
| 216 | * folio_matches_swap_entry - Check if a folio matches a given swap entry. |
| 217 | * @folio: The folio. |
| 218 | * @entry: The swap entry to check against. |
| 219 | * |
| 220 | * Context: The caller should have the folio locked to ensure it's stable |
| 221 | * and nothing will move it in or out of the swap cache. |
| 222 | * Return: true or false. |
| 223 | */ |
| 224 | static inline bool folio_matches_swap_entry(const struct folio *folio, |
| 225 | swp_entry_t entry) |
| 226 | { |
| 227 | swp_entry_t folio_entry = folio->swap; |
| 228 | long nr_pages = folio_nr_pages(folio); |
| 229 | |
| 230 | VM_WARN_ON_ONCE_FOLIO(!folio_test_locked(folio), folio); |
| 231 | if (!folio_test_swapcache(folio)) |
| 232 | return false; |
| 233 | VM_WARN_ON_ONCE_FOLIO(!IS_ALIGNED(folio_entry.val, nr_pages), folio); |
| 234 | return folio_entry.val == round_down(entry.val, nr_pages); |
| 235 | } |
| 236 | |
| 237 | /* |
| 238 | * All swap cache helpers below require the caller to ensure the swap entries |
| 239 | * used are valid and stablize the device by any of the following ways: |
| 240 | * - Hold a reference by get_swap_device(): this ensures a single entry is |
| 241 | * valid and increases the swap device's refcount. |
| 242 | * - Locking a folio in the swap cache: this ensures the folio's swap entries |
| 243 | * are valid and pinned, also implies reference to the device. |
| 244 | * - Locking anything referencing the swap entry: e.g. PTL that protects |
| 245 | * swap entries in the page table, similar to locking swap cache folio. |
| 246 | * - See the comment of get_swap_device() for more complex usage. |
| 247 | */ |
| 248 | struct folio *swap_cache_get_folio(swp_entry_t entry); |
| 249 | void *swap_cache_get_shadow(swp_entry_t entry); |
| 250 | void swap_cache_add_folio(struct folio *folio, swp_entry_t entry, void **shadow); |
| 251 | void swap_cache_del_folio(struct folio *folio); |
| 252 | /* Below helpers require the caller to lock and pass in the swap cluster. */ |
| 253 | void __swap_cache_del_folio(struct swap_cluster_info *ci, |
| 254 | struct folio *folio, swp_entry_t entry, void *shadow); |
| 255 | void __swap_cache_replace_folio(struct swap_cluster_info *ci, |
| 256 | struct folio *old, struct folio *new); |
| 257 | void __swap_cache_clear_shadow(swp_entry_t entry, int nr_ents); |
| 258 | |
| 259 | void show_swap_cache_info(void); |
| 260 | void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry, int nr); |
| 261 | struct folio *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, |
| 262 | struct vm_area_struct *vma, unsigned long addr, |
| 263 | struct swap_iocb **plug); |
| 264 | struct folio *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_flags, |
| 265 | struct mempolicy *mpol, pgoff_t ilx, bool *new_page_allocated, |
| 266 | bool skip_if_exists); |
| 267 | struct folio *swap_cluster_readahead(swp_entry_t entry, gfp_t flag, |
| 268 | struct mempolicy *mpol, pgoff_t ilx); |
| 269 | struct folio *swapin_readahead(swp_entry_t entry, gfp_t flag, |
| 270 | struct vm_fault *vmf); |
| 271 | void swap_update_readahead(struct folio *folio, struct vm_area_struct *vma, |
| 272 | unsigned long addr); |
| 273 | |
| 274 | static inline unsigned int folio_swap_flags(struct folio *folio) |
| 275 | { |
| 276 | return __swap_entry_to_info(entry: folio->swap)->flags; |
| 277 | } |
| 278 | |
| 279 | /* |
| 280 | * Return the count of contiguous swap entries that share the same |
| 281 | * zeromap status as the starting entry. If is_zeromap is not NULL, |
| 282 | * it will return the zeromap status of the starting entry. |
| 283 | */ |
| 284 | static inline int swap_zeromap_batch(swp_entry_t entry, int max_nr, |
| 285 | bool *is_zeromap) |
| 286 | { |
| 287 | struct swap_info_struct *sis = __swap_entry_to_info(entry); |
| 288 | unsigned long start = swp_offset(entry); |
| 289 | unsigned long end = start + max_nr; |
| 290 | bool first_bit; |
| 291 | |
| 292 | first_bit = test_bit(start, sis->zeromap); |
| 293 | if (is_zeromap) |
| 294 | *is_zeromap = first_bit; |
| 295 | |
| 296 | if (max_nr <= 1) |
| 297 | return max_nr; |
| 298 | if (first_bit) |
| 299 | return find_next_zero_bit(addr: sis->zeromap, size: end, offset: start) - start; |
| 300 | else |
| 301 | return find_next_bit(addr: sis->zeromap, size: end, offset: start) - start; |
| 302 | } |
| 303 | |
| 304 | static inline int non_swapcache_batch(swp_entry_t entry, int max_nr) |
| 305 | { |
| 306 | struct swap_info_struct *si = __swap_entry_to_info(entry); |
| 307 | pgoff_t offset = swp_offset(entry); |
| 308 | int i; |
| 309 | |
| 310 | /* |
| 311 | * While allocating a large folio and doing mTHP swapin, we need to |
| 312 | * ensure all entries are not cached, otherwise, the mTHP folio will |
| 313 | * be in conflict with the folio in swap cache. |
| 314 | */ |
| 315 | for (i = 0; i < max_nr; i++) { |
| 316 | if ((si->swap_map[offset + i] & SWAP_HAS_CACHE)) |
| 317 | return i; |
| 318 | } |
| 319 | |
| 320 | return i; |
| 321 | } |
| 322 | |
| 323 | #else /* CONFIG_SWAP */ |
| 324 | struct swap_iocb; |
| 325 | static inline struct swap_cluster_info *swap_cluster_lock( |
| 326 | struct swap_info_struct *si, pgoff_t offset, bool irq) |
| 327 | { |
| 328 | return NULL; |
| 329 | } |
| 330 | |
| 331 | static inline struct swap_cluster_info *swap_cluster_get_and_lock( |
| 332 | struct folio *folio) |
| 333 | { |
| 334 | return NULL; |
| 335 | } |
| 336 | |
| 337 | static inline struct swap_cluster_info *swap_cluster_get_and_lock_irq( |
| 338 | struct folio *folio) |
| 339 | { |
| 340 | return NULL; |
| 341 | } |
| 342 | |
| 343 | static inline void swap_cluster_unlock(struct swap_cluster_info *ci) |
| 344 | { |
| 345 | } |
| 346 | |
| 347 | static inline void swap_cluster_unlock_irq(struct swap_cluster_info *ci) |
| 348 | { |
| 349 | } |
| 350 | |
| 351 | static inline struct swap_info_struct *__swap_entry_to_info(swp_entry_t entry) |
| 352 | { |
| 353 | return NULL; |
| 354 | } |
| 355 | |
| 356 | static inline void swap_read_folio(struct folio *folio, struct swap_iocb **plug) |
| 357 | { |
| 358 | } |
| 359 | static inline void swap_write_unplug(struct swap_iocb *sio) |
| 360 | { |
| 361 | } |
| 362 | |
| 363 | static inline struct address_space *swap_address_space(swp_entry_t entry) |
| 364 | { |
| 365 | return NULL; |
| 366 | } |
| 367 | |
| 368 | static inline bool folio_matches_swap_entry(const struct folio *folio, swp_entry_t entry) |
| 369 | { |
| 370 | return false; |
| 371 | } |
| 372 | |
| 373 | static inline void show_swap_cache_info(void) |
| 374 | { |
| 375 | } |
| 376 | |
| 377 | static inline struct folio *swap_cluster_readahead(swp_entry_t entry, |
| 378 | gfp_t gfp_mask, struct mempolicy *mpol, pgoff_t ilx) |
| 379 | { |
| 380 | return NULL; |
| 381 | } |
| 382 | |
| 383 | static inline struct folio *swapin_readahead(swp_entry_t swp, gfp_t gfp_mask, |
| 384 | struct vm_fault *vmf) |
| 385 | { |
| 386 | return NULL; |
| 387 | } |
| 388 | |
| 389 | static inline void swap_update_readahead(struct folio *folio, |
| 390 | struct vm_area_struct *vma, unsigned long addr) |
| 391 | { |
| 392 | } |
| 393 | |
| 394 | static inline int swap_writeout(struct folio *folio, |
| 395 | struct swap_iocb **swap_plug) |
| 396 | { |
| 397 | return 0; |
| 398 | } |
| 399 | |
| 400 | static inline void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry, int nr) |
| 401 | { |
| 402 | } |
| 403 | |
| 404 | static inline struct folio *swap_cache_get_folio(swp_entry_t entry) |
| 405 | { |
| 406 | return NULL; |
| 407 | } |
| 408 | |
| 409 | static inline void *swap_cache_get_shadow(swp_entry_t entry) |
| 410 | { |
| 411 | return NULL; |
| 412 | } |
| 413 | |
| 414 | static inline void swap_cache_add_folio(struct folio *folio, swp_entry_t entry, void **shadow) |
| 415 | { |
| 416 | } |
| 417 | |
| 418 | static inline void swap_cache_del_folio(struct folio *folio) |
| 419 | { |
| 420 | } |
| 421 | |
| 422 | static inline void __swap_cache_del_folio(struct swap_cluster_info *ci, |
| 423 | struct folio *folio, swp_entry_t entry, void *shadow) |
| 424 | { |
| 425 | } |
| 426 | |
| 427 | static inline void __swap_cache_replace_folio(struct swap_cluster_info *ci, |
| 428 | struct folio *old, struct folio *new) |
| 429 | { |
| 430 | } |
| 431 | |
| 432 | static inline unsigned int folio_swap_flags(struct folio *folio) |
| 433 | { |
| 434 | return 0; |
| 435 | } |
| 436 | |
| 437 | static inline int swap_zeromap_batch(swp_entry_t entry, int max_nr, |
| 438 | bool *has_zeromap) |
| 439 | { |
| 440 | return 0; |
| 441 | } |
| 442 | |
| 443 | static inline int non_swapcache_batch(swp_entry_t entry, int max_nr) |
| 444 | { |
| 445 | return 0; |
| 446 | } |
| 447 | #endif /* CONFIG_SWAP */ |
| 448 | |
| 449 | /** |
| 450 | * folio_index - File index of a folio. |
| 451 | * @folio: The folio. |
| 452 | * |
| 453 | * For a folio which is either in the page cache or the swap cache, |
| 454 | * return its index within the address_space it belongs to. If you know |
| 455 | * the folio is definitely in the page cache, you can look at the folio's |
| 456 | * index directly. |
| 457 | * |
| 458 | * Return: The index (offset in units of pages) of a folio in its file. |
| 459 | */ |
| 460 | static inline pgoff_t folio_index(struct folio *folio) |
| 461 | { |
| 462 | #ifdef CONFIG_SWAP |
| 463 | if (unlikely(folio_test_swapcache(folio))) |
| 464 | return swp_offset(entry: folio->swap); |
| 465 | #endif |
| 466 | return folio->index; |
| 467 | } |
| 468 | |
| 469 | #endif /* _MM_SWAP_H */ |
| 470 | |