| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * linux/mm/swap.c |
| 4 | * |
| 5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 6 | */ |
| 7 | |
| 8 | /* |
| 9 | * This file contains the default values for the operation of the |
| 10 | * Linux VM subsystem. Fine-tuning documentation can be found in |
| 11 | * Documentation/admin-guide/sysctl/vm.rst. |
| 12 | * Started 18.12.91 |
| 13 | * Swap aging added 23.2.95, Stephen Tweedie. |
| 14 | * Buffermem limits added 12.3.98, Rik van Riel. |
| 15 | */ |
| 16 | |
| 17 | #include <linux/mm.h> |
| 18 | #include <linux/sched.h> |
| 19 | #include <linux/kernel_stat.h> |
| 20 | #include <linux/swap.h> |
| 21 | #include <linux/mman.h> |
| 22 | #include <linux/pagemap.h> |
| 23 | #include <linux/pagevec.h> |
| 24 | #include <linux/init.h> |
| 25 | #include <linux/export.h> |
| 26 | #include <linux/mm_inline.h> |
| 27 | #include <linux/percpu_counter.h> |
| 28 | #include <linux/memremap.h> |
| 29 | #include <linux/percpu.h> |
| 30 | #include <linux/cpu.h> |
| 31 | #include <linux/notifier.h> |
| 32 | #include <linux/backing-dev.h> |
| 33 | #include <linux/memcontrol.h> |
| 34 | #include <linux/gfp.h> |
| 35 | #include <linux/uio.h> |
| 36 | #include <linux/hugetlb.h> |
| 37 | #include <linux/page_idle.h> |
| 38 | #include <linux/local_lock.h> |
| 39 | #include <linux/buffer_head.h> |
| 40 | |
| 41 | #include "internal.h" |
| 42 | |
| 43 | #define CREATE_TRACE_POINTS |
| 44 | #include <trace/events/pagemap.h> |
| 45 | |
| 46 | /* How many pages do we try to swap or page in/out together? As a power of 2 */ |
| 47 | int page_cluster; |
| 48 | const int page_cluster_max = 31; |
| 49 | |
| 50 | struct cpu_fbatches { |
| 51 | /* |
| 52 | * The following folio batches are grouped together because they are protected |
| 53 | * by disabling preemption (and interrupts remain enabled). |
| 54 | */ |
| 55 | local_lock_t lock; |
| 56 | struct folio_batch lru_add; |
| 57 | struct folio_batch lru_deactivate_file; |
| 58 | struct folio_batch lru_deactivate; |
| 59 | struct folio_batch lru_lazyfree; |
| 60 | #ifdef CONFIG_SMP |
| 61 | struct folio_batch lru_activate; |
| 62 | #endif |
| 63 | /* Protecting the following batches which require disabling interrupts */ |
| 64 | local_lock_t lock_irq; |
| 65 | struct folio_batch lru_move_tail; |
| 66 | }; |
| 67 | |
| 68 | static DEFINE_PER_CPU(struct cpu_fbatches, cpu_fbatches) = { |
| 69 | .lock = INIT_LOCAL_LOCK(lock), |
| 70 | .lock_irq = INIT_LOCAL_LOCK(lock_irq), |
| 71 | }; |
| 72 | |
| 73 | static void __page_cache_release(struct folio *folio, struct lruvec **lruvecp, |
| 74 | unsigned long *flagsp) |
| 75 | { |
| 76 | if (folio_test_lru(folio)) { |
| 77 | folio_lruvec_relock_irqsave(folio, lruvecp, flagsp); |
| 78 | lruvec_del_folio(*lruvecp, folio); |
| 79 | __folio_clear_lru_flags(folio); |
| 80 | } |
| 81 | |
| 82 | /* |
| 83 | * In rare cases, when truncation or holepunching raced with |
| 84 | * munlock after VM_LOCKED was cleared, Mlocked may still be |
| 85 | * found set here. This does not indicate a problem, unless |
| 86 | * "unevictable_pgs_cleared" appears worryingly large. |
| 87 | */ |
| 88 | if (unlikely(folio_test_mlocked(folio))) { |
| 89 | long nr_pages = folio_nr_pages(folio); |
| 90 | |
| 91 | __folio_clear_mlocked(folio); |
| 92 | zone_stat_mod_folio(folio, NR_MLOCK, -nr_pages); |
| 93 | count_vm_events(UNEVICTABLE_PGCLEARED, nr_pages); |
| 94 | } |
| 95 | } |
| 96 | |
| 97 | /* |
| 98 | * This path almost never happens for VM activity - pages are normally freed |
| 99 | * in batches. But it gets used by networking - and for compound pages. |
| 100 | */ |
| 101 | static void page_cache_release(struct folio *folio) |
| 102 | { |
| 103 | struct lruvec *lruvec = NULL; |
| 104 | unsigned long flags; |
| 105 | |
| 106 | __page_cache_release(folio, &lruvec, &flags); |
| 107 | if (lruvec) |
| 108 | unlock_page_lruvec_irqrestore(lruvec, flags); |
| 109 | } |
| 110 | |
| 111 | void __folio_put(struct folio *folio) |
| 112 | { |
| 113 | if (unlikely(folio_is_zone_device(folio))) { |
| 114 | free_zone_device_folio(folio); |
| 115 | return; |
| 116 | } |
| 117 | |
| 118 | if (folio_test_hugetlb(folio)) { |
| 119 | free_huge_folio(folio); |
| 120 | return; |
| 121 | } |
| 122 | |
| 123 | page_cache_release(folio); |
| 124 | folio_undo_large_rmappable(folio); |
| 125 | mem_cgroup_uncharge(folio); |
| 126 | free_unref_page(&folio->page, folio_order(folio)); |
| 127 | } |
| 128 | EXPORT_SYMBOL(__folio_put); |
| 129 | |
| 130 | /** |
| 131 | * put_pages_list() - release a list of pages |
| 132 | * @pages: list of pages threaded on page->lru |
| 133 | * |
| 134 | * Release a list of pages which are strung together on page.lru. |
| 135 | */ |
| 136 | void put_pages_list(struct list_head *pages) |
| 137 | { |
| 138 | struct folio_batch fbatch; |
| 139 | struct folio *folio, *next; |
| 140 | |
| 141 | folio_batch_init(&fbatch); |
| 142 | list_for_each_entry_safe(folio, next, pages, lru) { |
| 143 | if (!folio_put_testzero(folio)) |
| 144 | continue; |
| 145 | if (folio_test_hugetlb(folio)) { |
| 146 | free_huge_folio(folio); |
| 147 | continue; |
| 148 | } |
| 149 | /* LRU flag must be clear because it's passed using the lru */ |
| 150 | if (folio_batch_add(&fbatch, folio) > 0) |
| 151 | continue; |
| 152 | free_unref_folios(&fbatch); |
| 153 | } |
| 154 | |
| 155 | if (fbatch.nr) |
| 156 | free_unref_folios(&fbatch); |
| 157 | INIT_LIST_HEAD(pages); |
| 158 | } |
| 159 | EXPORT_SYMBOL(put_pages_list); |
| 160 | |
| 161 | typedef void (*move_fn_t)(struct lruvec *lruvec, struct folio *folio); |
| 162 | |
| 163 | static void lru_add(struct lruvec *lruvec, struct folio *folio) |
| 164 | { |
| 165 | int was_unevictable = folio_test_clear_unevictable(folio); |
| 166 | long nr_pages = folio_nr_pages(folio); |
| 167 | |
| 168 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
| 169 | |
| 170 | /* |
| 171 | * Is an smp_mb__after_atomic() still required here, before |
| 172 | * folio_evictable() tests the mlocked flag, to rule out the possibility |
| 173 | * of stranding an evictable folio on an unevictable LRU? I think |
| 174 | * not, because __munlock_folio() only clears the mlocked flag |
| 175 | * while the LRU lock is held. |
| 176 | * |
| 177 | * (That is not true of __page_cache_release(), and not necessarily |
| 178 | * true of folios_put(): but those only clear the mlocked flag after |
| 179 | * folio_put_testzero() has excluded any other users of the folio.) |
| 180 | */ |
| 181 | if (folio_evictable(folio)) { |
| 182 | if (was_unevictable) |
| 183 | __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages); |
| 184 | } else { |
| 185 | folio_clear_active(folio); |
| 186 | folio_set_unevictable(folio); |
| 187 | /* |
| 188 | * folio->mlock_count = !!folio_test_mlocked(folio)? |
| 189 | * But that leaves __mlock_folio() in doubt whether another |
| 190 | * actor has already counted the mlock or not. Err on the |
| 191 | * safe side, underestimate, let page reclaim fix it, rather |
| 192 | * than leaving a page on the unevictable LRU indefinitely. |
| 193 | */ |
| 194 | folio->mlock_count = 0; |
| 195 | if (!was_unevictable) |
| 196 | __count_vm_events(UNEVICTABLE_PGCULLED, nr_pages); |
| 197 | } |
| 198 | |
| 199 | lruvec_add_folio(lruvec, folio); |
| 200 | trace_mm_lru_insertion(folio); |
| 201 | } |
| 202 | |
| 203 | static void folio_batch_move_lru(struct folio_batch *fbatch, move_fn_t move_fn) |
| 204 | { |
| 205 | int i; |
| 206 | struct lruvec *lruvec = NULL; |
| 207 | unsigned long flags = 0; |
| 208 | |
| 209 | for (i = 0; i < folio_batch_count(fbatch); i++) { |
| 210 | struct folio *folio = fbatch->folios[i]; |
| 211 | |
| 212 | folio_lruvec_relock_irqsave(folio, &lruvec, &flags); |
| 213 | move_fn(lruvec, folio); |
| 214 | |
| 215 | folio_set_lru(folio); |
| 216 | } |
| 217 | |
| 218 | if (lruvec) |
| 219 | unlock_page_lruvec_irqrestore(lruvec, flags); |
| 220 | folios_put(fbatch); |
| 221 | } |
| 222 | |
| 223 | static void __folio_batch_add_and_move(struct folio_batch __percpu *fbatch, |
| 224 | struct folio *folio, move_fn_t move_fn, |
| 225 | bool on_lru, bool disable_irq) |
| 226 | { |
| 227 | unsigned long flags; |
| 228 | |
| 229 | if (on_lru && !folio_test_clear_lru(folio)) |
| 230 | return; |
| 231 | |
| 232 | folio_get(folio); |
| 233 | |
| 234 | if (disable_irq) |
| 235 | local_lock_irqsave(&cpu_fbatches.lock_irq, flags); |
| 236 | else |
| 237 | local_lock(&cpu_fbatches.lock); |
| 238 | |
| 239 | if (!folio_batch_add(this_cpu_ptr(fbatch), folio) || folio_test_large(folio) || |
| 240 | lru_cache_disabled()) |
| 241 | folio_batch_move_lru(this_cpu_ptr(fbatch), move_fn); |
| 242 | |
| 243 | if (disable_irq) |
| 244 | local_unlock_irqrestore(&cpu_fbatches.lock_irq, flags); |
| 245 | else |
| 246 | local_unlock(&cpu_fbatches.lock); |
| 247 | } |
| 248 | |
| 249 | #define folio_batch_add_and_move(folio, op, on_lru) \ |
| 250 | __folio_batch_add_and_move( \ |
| 251 | &cpu_fbatches.op, \ |
| 252 | folio, \ |
| 253 | op, \ |
| 254 | on_lru, \ |
| 255 | offsetof(struct cpu_fbatches, op) >= offsetof(struct cpu_fbatches, lock_irq) \ |
| 256 | ) |
| 257 | |
| 258 | static void lru_move_tail(struct lruvec *lruvec, struct folio *folio) |
| 259 | { |
| 260 | if (folio_test_unevictable(folio)) |
| 261 | return; |
| 262 | |
| 263 | lruvec_del_folio(lruvec, folio); |
| 264 | folio_clear_active(folio); |
| 265 | lruvec_add_folio_tail(lruvec, folio); |
| 266 | __count_vm_events(PGROTATED, folio_nr_pages(folio)); |
| 267 | } |
| 268 | |
| 269 | /* |
| 270 | * Writeback is about to end against a folio which has been marked for |
| 271 | * immediate reclaim. If it still appears to be reclaimable, move it |
| 272 | * to the tail of the inactive list. |
| 273 | * |
| 274 | * folio_rotate_reclaimable() must disable IRQs, to prevent nasty races. |
| 275 | */ |
| 276 | void folio_rotate_reclaimable(struct folio *folio) |
| 277 | { |
| 278 | if (folio_test_locked(folio) || folio_test_dirty(folio) || |
| 279 | folio_test_unevictable(folio)) |
| 280 | return; |
| 281 | |
| 282 | folio_batch_add_and_move(folio, lru_move_tail, true); |
| 283 | } |
| 284 | |
| 285 | void lru_note_cost(struct lruvec *lruvec, bool file, |
| 286 | unsigned int nr_io, unsigned int nr_rotated) |
| 287 | { |
| 288 | unsigned long cost; |
| 289 | |
| 290 | /* |
| 291 | * Reflect the relative cost of incurring IO and spending CPU |
| 292 | * time on rotations. This doesn't attempt to make a precise |
| 293 | * comparison, it just says: if reloads are about comparable |
| 294 | * between the LRU lists, or rotations are overwhelmingly |
| 295 | * different between them, adjust scan balance for CPU work. |
| 296 | */ |
| 297 | cost = nr_io * SWAP_CLUSTER_MAX + nr_rotated; |
| 298 | |
| 299 | do { |
| 300 | unsigned long lrusize; |
| 301 | |
| 302 | /* |
| 303 | * Hold lruvec->lru_lock is safe here, since |
| 304 | * 1) The pinned lruvec in reclaim, or |
| 305 | * 2) From a pre-LRU page during refault (which also holds the |
| 306 | * rcu lock, so would be safe even if the page was on the LRU |
| 307 | * and could move simultaneously to a new lruvec). |
| 308 | */ |
| 309 | spin_lock_irq(&lruvec->lru_lock); |
| 310 | /* Record cost event */ |
| 311 | if (file) |
| 312 | lruvec->file_cost += cost; |
| 313 | else |
| 314 | lruvec->anon_cost += cost; |
| 315 | |
| 316 | /* |
| 317 | * Decay previous events |
| 318 | * |
| 319 | * Because workloads change over time (and to avoid |
| 320 | * overflow) we keep these statistics as a floating |
| 321 | * average, which ends up weighing recent refaults |
| 322 | * more than old ones. |
| 323 | */ |
| 324 | lrusize = lruvec_page_state(lruvec, NR_INACTIVE_ANON) + |
| 325 | lruvec_page_state(lruvec, NR_ACTIVE_ANON) + |
| 326 | lruvec_page_state(lruvec, NR_INACTIVE_FILE) + |
| 327 | lruvec_page_state(lruvec, NR_ACTIVE_FILE); |
| 328 | |
| 329 | if (lruvec->file_cost + lruvec->anon_cost > lrusize / 4) { |
| 330 | lruvec->file_cost /= 2; |
| 331 | lruvec->anon_cost /= 2; |
| 332 | } |
| 333 | spin_unlock_irq(&lruvec->lru_lock); |
| 334 | } while ((lruvec = parent_lruvec(lruvec))); |
| 335 | } |
| 336 | |
| 337 | void lru_note_cost_refault(struct folio *folio) |
| 338 | { |
| 339 | lru_note_cost(folio_lruvec(folio), folio_is_file_lru(folio), |
| 340 | folio_nr_pages(folio), 0); |
| 341 | } |
| 342 | |
| 343 | static void lru_activate(struct lruvec *lruvec, struct folio *folio) |
| 344 | { |
| 345 | long nr_pages = folio_nr_pages(folio); |
| 346 | |
| 347 | if (folio_test_active(folio) || folio_test_unevictable(folio)) |
| 348 | return; |
| 349 | |
| 350 | |
| 351 | lruvec_del_folio(lruvec, folio); |
| 352 | folio_set_active(folio); |
| 353 | lruvec_add_folio(lruvec, folio); |
| 354 | trace_mm_lru_activate(folio); |
| 355 | |
| 356 | __count_vm_events(PGACTIVATE, nr_pages); |
| 357 | __count_memcg_events(lruvec_memcg(lruvec), PGACTIVATE, nr_pages); |
| 358 | } |
| 359 | |
| 360 | #ifdef CONFIG_SMP |
| 361 | static void folio_activate_drain(int cpu) |
| 362 | { |
| 363 | struct folio_batch *fbatch = &per_cpu(cpu_fbatches.lru_activate, cpu); |
| 364 | |
| 365 | if (folio_batch_count(fbatch)) |
| 366 | folio_batch_move_lru(fbatch, lru_activate); |
| 367 | } |
| 368 | |
| 369 | void folio_activate(struct folio *folio) |
| 370 | { |
| 371 | if (folio_test_active(folio) || folio_test_unevictable(folio)) |
| 372 | return; |
| 373 | |
| 374 | folio_batch_add_and_move(folio, lru_activate, true); |
| 375 | } |
| 376 | |
| 377 | #else |
| 378 | static inline void folio_activate_drain(int cpu) |
| 379 | { |
| 380 | } |
| 381 | |
| 382 | void folio_activate(struct folio *folio) |
| 383 | { |
| 384 | struct lruvec *lruvec; |
| 385 | |
| 386 | if (!folio_test_clear_lru(folio)) |
| 387 | return; |
| 388 | |
| 389 | lruvec = folio_lruvec_lock_irq(folio); |
| 390 | lru_activate(lruvec, folio); |
| 391 | unlock_page_lruvec_irq(lruvec); |
| 392 | folio_set_lru(folio); |
| 393 | } |
| 394 | #endif |
| 395 | |
| 396 | static void __lru_cache_activate_folio(struct folio *folio) |
| 397 | { |
| 398 | struct folio_batch *fbatch; |
| 399 | int i; |
| 400 | |
| 401 | local_lock(&cpu_fbatches.lock); |
| 402 | fbatch = this_cpu_ptr(&cpu_fbatches.lru_add); |
| 403 | |
| 404 | /* |
| 405 | * Search backwards on the optimistic assumption that the folio being |
| 406 | * activated has just been added to this batch. Note that only |
| 407 | * the local batch is examined as a !LRU folio could be in the |
| 408 | * process of being released, reclaimed, migrated or on a remote |
| 409 | * batch that is currently being drained. Furthermore, marking |
| 410 | * a remote batch's folio active potentially hits a race where |
| 411 | * a folio is marked active just after it is added to the inactive |
| 412 | * list causing accounting errors and BUG_ON checks to trigger. |
| 413 | */ |
| 414 | for (i = folio_batch_count(fbatch) - 1; i >= 0; i--) { |
| 415 | struct folio *batch_folio = fbatch->folios[i]; |
| 416 | |
| 417 | if (batch_folio == folio) { |
| 418 | folio_set_active(folio); |
| 419 | break; |
| 420 | } |
| 421 | } |
| 422 | |
| 423 | local_unlock(&cpu_fbatches.lock); |
| 424 | } |
| 425 | |
| 426 | #ifdef CONFIG_LRU_GEN |
| 427 | static void folio_inc_refs(struct folio *folio) |
| 428 | { |
| 429 | unsigned long new_flags, old_flags = READ_ONCE(folio->flags); |
| 430 | |
| 431 | if (folio_test_unevictable(folio)) |
| 432 | return; |
| 433 | |
| 434 | if (!folio_test_referenced(folio)) { |
| 435 | folio_set_referenced(folio); |
| 436 | return; |
| 437 | } |
| 438 | |
| 439 | if (!folio_test_workingset(folio)) { |
| 440 | folio_set_workingset(folio); |
| 441 | return; |
| 442 | } |
| 443 | |
| 444 | /* see the comment on MAX_NR_TIERS */ |
| 445 | do { |
| 446 | new_flags = old_flags & LRU_REFS_MASK; |
| 447 | if (new_flags == LRU_REFS_MASK) |
| 448 | break; |
| 449 | |
| 450 | new_flags += BIT(LRU_REFS_PGOFF); |
| 451 | new_flags |= old_flags & ~LRU_REFS_MASK; |
| 452 | } while (!try_cmpxchg(&folio->flags, &old_flags, new_flags)); |
| 453 | } |
| 454 | #else |
| 455 | static void folio_inc_refs(struct folio *folio) |
| 456 | { |
| 457 | } |
| 458 | #endif /* CONFIG_LRU_GEN */ |
| 459 | |
| 460 | /** |
| 461 | * folio_mark_accessed - Mark a folio as having seen activity. |
| 462 | * @folio: The folio to mark. |
| 463 | * |
| 464 | * This function will perform one of the following transitions: |
| 465 | * |
| 466 | * * inactive,unreferenced -> inactive,referenced |
| 467 | * * inactive,referenced -> active,unreferenced |
| 468 | * * active,unreferenced -> active,referenced |
| 469 | * |
| 470 | * When a newly allocated folio is not yet visible, so safe for non-atomic ops, |
| 471 | * __folio_set_referenced() may be substituted for folio_mark_accessed(). |
| 472 | */ |
| 473 | void folio_mark_accessed(struct folio *folio) |
| 474 | { |
| 475 | if (lru_gen_enabled()) { |
| 476 | folio_inc_refs(folio); |
| 477 | return; |
| 478 | } |
| 479 | |
| 480 | if (!folio_test_referenced(folio)) { |
| 481 | folio_set_referenced(folio); |
| 482 | } else if (folio_test_unevictable(folio)) { |
| 483 | /* |
| 484 | * Unevictable pages are on the "LRU_UNEVICTABLE" list. But, |
| 485 | * this list is never rotated or maintained, so marking an |
| 486 | * unevictable page accessed has no effect. |
| 487 | */ |
| 488 | } else if (!folio_test_active(folio)) { |
| 489 | /* |
| 490 | * If the folio is on the LRU, queue it for activation via |
| 491 | * cpu_fbatches.lru_activate. Otherwise, assume the folio is in a |
| 492 | * folio_batch, mark it active and it'll be moved to the active |
| 493 | * LRU on the next drain. |
| 494 | */ |
| 495 | if (folio_test_lru(folio)) |
| 496 | folio_activate(folio); |
| 497 | else |
| 498 | __lru_cache_activate_folio(folio); |
| 499 | folio_clear_referenced(folio); |
| 500 | workingset_activation(folio); |
| 501 | } |
| 502 | if (folio_test_idle(folio)) |
| 503 | folio_clear_idle(folio); |
| 504 | } |
| 505 | EXPORT_SYMBOL(folio_mark_accessed); |
| 506 | |
| 507 | /** |
| 508 | * folio_add_lru - Add a folio to an LRU list. |
| 509 | * @folio: The folio to be added to the LRU. |
| 510 | * |
| 511 | * Queue the folio for addition to the LRU. The decision on whether |
| 512 | * to add the page to the [in]active [file|anon] list is deferred until the |
| 513 | * folio_batch is drained. This gives a chance for the caller of folio_add_lru() |
| 514 | * have the folio added to the active list using folio_mark_accessed(). |
| 515 | */ |
| 516 | void folio_add_lru(struct folio *folio) |
| 517 | { |
| 518 | VM_BUG_ON_FOLIO(folio_test_active(folio) && |
| 519 | folio_test_unevictable(folio), folio); |
| 520 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
| 521 | |
| 522 | /* see the comment in lru_gen_add_folio() */ |
| 523 | if (lru_gen_enabled() && !folio_test_unevictable(folio) && |
| 524 | lru_gen_in_fault() && !(current->flags & PF_MEMALLOC)) |
| 525 | folio_set_active(folio); |
| 526 | |
| 527 | folio_batch_add_and_move(folio, lru_add, false); |
| 528 | } |
| 529 | EXPORT_SYMBOL(folio_add_lru); |
| 530 | |
| 531 | /** |
| 532 | * folio_add_lru_vma() - Add a folio to the appropate LRU list for this VMA. |
| 533 | * @folio: The folio to be added to the LRU. |
| 534 | * @vma: VMA in which the folio is mapped. |
| 535 | * |
| 536 | * If the VMA is mlocked, @folio is added to the unevictable list. |
| 537 | * Otherwise, it is treated the same way as folio_add_lru(). |
| 538 | */ |
| 539 | void folio_add_lru_vma(struct folio *folio, struct vm_area_struct *vma) |
| 540 | { |
| 541 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
| 542 | |
| 543 | if (unlikely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) == VM_LOCKED)) |
| 544 | mlock_new_folio(folio); |
| 545 | else |
| 546 | folio_add_lru(folio); |
| 547 | } |
| 548 | |
| 549 | /* |
| 550 | * If the folio cannot be invalidated, it is moved to the |
| 551 | * inactive list to speed up its reclaim. It is moved to the |
| 552 | * head of the list, rather than the tail, to give the flusher |
| 553 | * threads some time to write it out, as this is much more |
| 554 | * effective than the single-page writeout from reclaim. |
| 555 | * |
| 556 | * If the folio isn't mapped and dirty/writeback, the folio |
| 557 | * could be reclaimed asap using the reclaim flag. |
| 558 | * |
| 559 | * 1. active, mapped folio -> none |
| 560 | * 2. active, dirty/writeback folio -> inactive, head, reclaim |
| 561 | * 3. inactive, mapped folio -> none |
| 562 | * 4. inactive, dirty/writeback folio -> inactive, head, reclaim |
| 563 | * 5. inactive, clean -> inactive, tail |
| 564 | * 6. Others -> none |
| 565 | * |
| 566 | * In 4, it moves to the head of the inactive list so the folio is |
| 567 | * written out by flusher threads as this is much more efficient |
| 568 | * than the single-page writeout from reclaim. |
| 569 | */ |
| 570 | static void lru_deactivate_file(struct lruvec *lruvec, struct folio *folio) |
| 571 | { |
| 572 | bool active = folio_test_active(folio); |
| 573 | long nr_pages = folio_nr_pages(folio); |
| 574 | |
| 575 | if (folio_test_unevictable(folio)) |
| 576 | return; |
| 577 | |
| 578 | /* Some processes are using the folio */ |
| 579 | if (folio_mapped(folio)) |
| 580 | return; |
| 581 | |
| 582 | lruvec_del_folio(lruvec, folio); |
| 583 | folio_clear_active(folio); |
| 584 | folio_clear_referenced(folio); |
| 585 | |
| 586 | if (folio_test_writeback(folio) || folio_test_dirty(folio)) { |
| 587 | /* |
| 588 | * Setting the reclaim flag could race with |
| 589 | * folio_end_writeback() and confuse readahead. But the |
| 590 | * race window is _really_ small and it's not a critical |
| 591 | * problem. |
| 592 | */ |
| 593 | lruvec_add_folio(lruvec, folio); |
| 594 | folio_set_reclaim(folio); |
| 595 | } else { |
| 596 | /* |
| 597 | * The folio's writeback ended while it was in the batch. |
| 598 | * We move that folio to the tail of the inactive list. |
| 599 | */ |
| 600 | lruvec_add_folio_tail(lruvec, folio); |
| 601 | __count_vm_events(PGROTATED, nr_pages); |
| 602 | } |
| 603 | |
| 604 | if (active) { |
| 605 | __count_vm_events(PGDEACTIVATE, nr_pages); |
| 606 | __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, |
| 607 | nr_pages); |
| 608 | } |
| 609 | } |
| 610 | |
| 611 | static void lru_deactivate(struct lruvec *lruvec, struct folio *folio) |
| 612 | { |
| 613 | long nr_pages = folio_nr_pages(folio); |
| 614 | |
| 615 | if (folio_test_unevictable(folio) || !(folio_test_active(folio) || lru_gen_enabled())) |
| 616 | return; |
| 617 | |
| 618 | lruvec_del_folio(lruvec, folio); |
| 619 | folio_clear_active(folio); |
| 620 | folio_clear_referenced(folio); |
| 621 | lruvec_add_folio(lruvec, folio); |
| 622 | |
| 623 | __count_vm_events(PGDEACTIVATE, nr_pages); |
| 624 | __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, nr_pages); |
| 625 | } |
| 626 | |
| 627 | static void lru_lazyfree(struct lruvec *lruvec, struct folio *folio) |
| 628 | { |
| 629 | long nr_pages = folio_nr_pages(folio); |
| 630 | |
| 631 | if (!folio_test_anon(folio) || !folio_test_swapbacked(folio) || |
| 632 | folio_test_swapcache(folio) || folio_test_unevictable(folio)) |
| 633 | return; |
| 634 | |
| 635 | lruvec_del_folio(lruvec, folio); |
| 636 | folio_clear_active(folio); |
| 637 | folio_clear_referenced(folio); |
| 638 | /* |
| 639 | * Lazyfree folios are clean anonymous folios. They have |
| 640 | * the swapbacked flag cleared, to distinguish them from normal |
| 641 | * anonymous folios |
| 642 | */ |
| 643 | folio_clear_swapbacked(folio); |
| 644 | lruvec_add_folio(lruvec, folio); |
| 645 | |
| 646 | __count_vm_events(PGLAZYFREE, nr_pages); |
| 647 | __count_memcg_events(lruvec_memcg(lruvec), PGLAZYFREE, nr_pages); |
| 648 | } |
| 649 | |
| 650 | /* |
| 651 | * Drain pages out of the cpu's folio_batch. |
| 652 | * Either "cpu" is the current CPU, and preemption has already been |
| 653 | * disabled; or "cpu" is being hot-unplugged, and is already dead. |
| 654 | */ |
| 655 | void lru_add_drain_cpu(int cpu) |
| 656 | { |
| 657 | struct cpu_fbatches *fbatches = &per_cpu(cpu_fbatches, cpu); |
| 658 | struct folio_batch *fbatch = &fbatches->lru_add; |
| 659 | |
| 660 | if (folio_batch_count(fbatch)) |
| 661 | folio_batch_move_lru(fbatch, lru_add); |
| 662 | |
| 663 | fbatch = &fbatches->lru_move_tail; |
| 664 | /* Disabling interrupts below acts as a compiler barrier. */ |
| 665 | if (data_race(folio_batch_count(fbatch))) { |
| 666 | unsigned long flags; |
| 667 | |
| 668 | /* No harm done if a racing interrupt already did this */ |
| 669 | local_lock_irqsave(&cpu_fbatches.lock_irq, flags); |
| 670 | folio_batch_move_lru(fbatch, lru_move_tail); |
| 671 | local_unlock_irqrestore(&cpu_fbatches.lock_irq, flags); |
| 672 | } |
| 673 | |
| 674 | fbatch = &fbatches->lru_deactivate_file; |
| 675 | if (folio_batch_count(fbatch)) |
| 676 | folio_batch_move_lru(fbatch, lru_deactivate_file); |
| 677 | |
| 678 | fbatch = &fbatches->lru_deactivate; |
| 679 | if (folio_batch_count(fbatch)) |
| 680 | folio_batch_move_lru(fbatch, lru_deactivate); |
| 681 | |
| 682 | fbatch = &fbatches->lru_lazyfree; |
| 683 | if (folio_batch_count(fbatch)) |
| 684 | folio_batch_move_lru(fbatch, lru_lazyfree); |
| 685 | |
| 686 | folio_activate_drain(cpu); |
| 687 | } |
| 688 | |
| 689 | /** |
| 690 | * deactivate_file_folio() - Deactivate a file folio. |
| 691 | * @folio: Folio to deactivate. |
| 692 | * |
| 693 | * This function hints to the VM that @folio is a good reclaim candidate, |
| 694 | * for example if its invalidation fails due to the folio being dirty |
| 695 | * or under writeback. |
| 696 | * |
| 697 | * Context: Caller holds a reference on the folio. |
| 698 | */ |
| 699 | void deactivate_file_folio(struct folio *folio) |
| 700 | { |
| 701 | /* Deactivating an unevictable folio will not accelerate reclaim */ |
| 702 | if (folio_test_unevictable(folio)) |
| 703 | return; |
| 704 | |
| 705 | folio_batch_add_and_move(folio, lru_deactivate_file, true); |
| 706 | } |
| 707 | |
| 708 | /* |
| 709 | * folio_deactivate - deactivate a folio |
| 710 | * @folio: folio to deactivate |
| 711 | * |
| 712 | * folio_deactivate() moves @folio to the inactive list if @folio was on the |
| 713 | * active list and was not unevictable. This is done to accelerate the |
| 714 | * reclaim of @folio. |
| 715 | */ |
| 716 | void folio_deactivate(struct folio *folio) |
| 717 | { |
| 718 | if (folio_test_unevictable(folio) || !(folio_test_active(folio) || lru_gen_enabled())) |
| 719 | return; |
| 720 | |
| 721 | folio_batch_add_and_move(folio, lru_deactivate, true); |
| 722 | } |
| 723 | |
| 724 | /** |
| 725 | * folio_mark_lazyfree - make an anon folio lazyfree |
| 726 | * @folio: folio to deactivate |
| 727 | * |
| 728 | * folio_mark_lazyfree() moves @folio to the inactive file list. |
| 729 | * This is done to accelerate the reclaim of @folio. |
| 730 | */ |
| 731 | void folio_mark_lazyfree(struct folio *folio) |
| 732 | { |
| 733 | if (!folio_test_anon(folio) || !folio_test_swapbacked(folio) || |
| 734 | folio_test_swapcache(folio) || folio_test_unevictable(folio)) |
| 735 | return; |
| 736 | |
| 737 | folio_batch_add_and_move(folio, lru_lazyfree, true); |
| 738 | } |
| 739 | |
| 740 | void lru_add_drain(void) |
| 741 | { |
| 742 | local_lock(&cpu_fbatches.lock); |
| 743 | lru_add_drain_cpu(smp_processor_id()); |
| 744 | local_unlock(&cpu_fbatches.lock); |
| 745 | mlock_drain_local(); |
| 746 | } |
| 747 | |
| 748 | /* |
| 749 | * It's called from per-cpu workqueue context in SMP case so |
| 750 | * lru_add_drain_cpu and invalidate_bh_lrus_cpu should run on |
| 751 | * the same cpu. It shouldn't be a problem in !SMP case since |
| 752 | * the core is only one and the locks will disable preemption. |
| 753 | */ |
| 754 | static void lru_add_and_bh_lrus_drain(void) |
| 755 | { |
| 756 | local_lock(&cpu_fbatches.lock); |
| 757 | lru_add_drain_cpu(smp_processor_id()); |
| 758 | local_unlock(&cpu_fbatches.lock); |
| 759 | invalidate_bh_lrus_cpu(); |
| 760 | mlock_drain_local(); |
| 761 | } |
| 762 | |
| 763 | void lru_add_drain_cpu_zone(struct zone *zone) |
| 764 | { |
| 765 | local_lock(&cpu_fbatches.lock); |
| 766 | lru_add_drain_cpu(smp_processor_id()); |
| 767 | drain_local_pages(zone); |
| 768 | local_unlock(&cpu_fbatches.lock); |
| 769 | mlock_drain_local(); |
| 770 | } |
| 771 | |
| 772 | #ifdef CONFIG_SMP |
| 773 | |
| 774 | static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); |
| 775 | |
| 776 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
| 777 | { |
| 778 | lru_add_and_bh_lrus_drain(); |
| 779 | } |
| 780 | |
| 781 | static bool cpu_needs_drain(unsigned int cpu) |
| 782 | { |
| 783 | struct cpu_fbatches *fbatches = &per_cpu(cpu_fbatches, cpu); |
| 784 | |
| 785 | /* Check these in order of likelihood that they're not zero */ |
| 786 | return folio_batch_count(&fbatches->lru_add) || |
| 787 | folio_batch_count(&fbatches->lru_move_tail) || |
| 788 | folio_batch_count(&fbatches->lru_deactivate_file) || |
| 789 | folio_batch_count(&fbatches->lru_deactivate) || |
| 790 | folio_batch_count(&fbatches->lru_lazyfree) || |
| 791 | folio_batch_count(&fbatches->lru_activate) || |
| 792 | need_mlock_drain(cpu) || |
| 793 | has_bh_in_lru(cpu, NULL); |
| 794 | } |
| 795 | |
| 796 | /* |
| 797 | * Doesn't need any cpu hotplug locking because we do rely on per-cpu |
| 798 | * kworkers being shut down before our page_alloc_cpu_dead callback is |
| 799 | * executed on the offlined cpu. |
| 800 | * Calling this function with cpu hotplug locks held can actually lead |
| 801 | * to obscure indirect dependencies via WQ context. |
| 802 | */ |
| 803 | static inline void __lru_add_drain_all(bool force_all_cpus) |
| 804 | { |
| 805 | /* |
| 806 | * lru_drain_gen - Global pages generation number |
| 807 | * |
| 808 | * (A) Definition: global lru_drain_gen = x implies that all generations |
| 809 | * 0 < n <= x are already *scheduled* for draining. |
| 810 | * |
| 811 | * This is an optimization for the highly-contended use case where a |
| 812 | * user space workload keeps constantly generating a flow of pages for |
| 813 | * each CPU. |
| 814 | */ |
| 815 | static unsigned int lru_drain_gen; |
| 816 | static struct cpumask has_work; |
| 817 | static DEFINE_MUTEX(lock); |
| 818 | unsigned cpu, this_gen; |
| 819 | |
| 820 | /* |
| 821 | * Make sure nobody triggers this path before mm_percpu_wq is fully |
| 822 | * initialized. |
| 823 | */ |
| 824 | if (WARN_ON(!mm_percpu_wq)) |
| 825 | return; |
| 826 | |
| 827 | /* |
| 828 | * Guarantee folio_batch counter stores visible by this CPU |
| 829 | * are visible to other CPUs before loading the current drain |
| 830 | * generation. |
| 831 | */ |
| 832 | smp_mb(); |
| 833 | |
| 834 | /* |
| 835 | * (B) Locally cache global LRU draining generation number |
| 836 | * |
| 837 | * The read barrier ensures that the counter is loaded before the mutex |
| 838 | * is taken. It pairs with smp_mb() inside the mutex critical section |
| 839 | * at (D). |
| 840 | */ |
| 841 | this_gen = smp_load_acquire(&lru_drain_gen); |
| 842 | |
| 843 | mutex_lock(&lock); |
| 844 | |
| 845 | /* |
| 846 | * (C) Exit the draining operation if a newer generation, from another |
| 847 | * lru_add_drain_all(), was already scheduled for draining. Check (A). |
| 848 | */ |
| 849 | if (unlikely(this_gen != lru_drain_gen && !force_all_cpus)) |
| 850 | goto done; |
| 851 | |
| 852 | /* |
| 853 | * (D) Increment global generation number |
| 854 | * |
| 855 | * Pairs with smp_load_acquire() at (B), outside of the critical |
| 856 | * section. Use a full memory barrier to guarantee that the |
| 857 | * new global drain generation number is stored before loading |
| 858 | * folio_batch counters. |
| 859 | * |
| 860 | * This pairing must be done here, before the for_each_online_cpu loop |
| 861 | * below which drains the page vectors. |
| 862 | * |
| 863 | * Let x, y, and z represent some system CPU numbers, where x < y < z. |
| 864 | * Assume CPU #z is in the middle of the for_each_online_cpu loop |
| 865 | * below and has already reached CPU #y's per-cpu data. CPU #x comes |
| 866 | * along, adds some pages to its per-cpu vectors, then calls |
| 867 | * lru_add_drain_all(). |
| 868 | * |
| 869 | * If the paired barrier is done at any later step, e.g. after the |
| 870 | * loop, CPU #x will just exit at (C) and miss flushing out all of its |
| 871 | * added pages. |
| 872 | */ |
| 873 | WRITE_ONCE(lru_drain_gen, lru_drain_gen + 1); |
| 874 | smp_mb(); |
| 875 | |
| 876 | cpumask_clear(&has_work); |
| 877 | for_each_online_cpu(cpu) { |
| 878 | struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); |
| 879 | |
| 880 | if (cpu_needs_drain(cpu)) { |
| 881 | INIT_WORK(work, lru_add_drain_per_cpu); |
| 882 | queue_work_on(cpu, mm_percpu_wq, work); |
| 883 | __cpumask_set_cpu(cpu, &has_work); |
| 884 | } |
| 885 | } |
| 886 | |
| 887 | for_each_cpu(cpu, &has_work) |
| 888 | flush_work(&per_cpu(lru_add_drain_work, cpu)); |
| 889 | |
| 890 | done: |
| 891 | mutex_unlock(&lock); |
| 892 | } |
| 893 | |
| 894 | void lru_add_drain_all(void) |
| 895 | { |
| 896 | __lru_add_drain_all(false); |
| 897 | } |
| 898 | #else |
| 899 | void lru_add_drain_all(void) |
| 900 | { |
| 901 | lru_add_drain(); |
| 902 | } |
| 903 | #endif /* CONFIG_SMP */ |
| 904 | |
| 905 | atomic_t lru_disable_count = ATOMIC_INIT(0); |
| 906 | |
| 907 | /* |
| 908 | * lru_cache_disable() needs to be called before we start compiling |
| 909 | * a list of folios to be migrated using folio_isolate_lru(). |
| 910 | * It drains folios on LRU cache and then disable on all cpus until |
| 911 | * lru_cache_enable is called. |
| 912 | * |
| 913 | * Must be paired with a call to lru_cache_enable(). |
| 914 | */ |
| 915 | void lru_cache_disable(void) |
| 916 | { |
| 917 | atomic_inc(&lru_disable_count); |
| 918 | /* |
| 919 | * Readers of lru_disable_count are protected by either disabling |
| 920 | * preemption or rcu_read_lock: |
| 921 | * |
| 922 | * preempt_disable, local_irq_disable [bh_lru_lock()] |
| 923 | * rcu_read_lock [rt_spin_lock CONFIG_PREEMPT_RT] |
| 924 | * preempt_disable [local_lock !CONFIG_PREEMPT_RT] |
| 925 | * |
| 926 | * Since v5.1 kernel, synchronize_rcu() is guaranteed to wait on |
| 927 | * preempt_disable() regions of code. So any CPU which sees |
| 928 | * lru_disable_count = 0 will have exited the critical |
| 929 | * section when synchronize_rcu() returns. |
| 930 | */ |
| 931 | synchronize_rcu_expedited(); |
| 932 | #ifdef CONFIG_SMP |
| 933 | __lru_add_drain_all(true); |
| 934 | #else |
| 935 | lru_add_and_bh_lrus_drain(); |
| 936 | #endif |
| 937 | } |
| 938 | |
| 939 | /** |
| 940 | * folios_put_refs - Reduce the reference count on a batch of folios. |
| 941 | * @folios: The folios. |
| 942 | * @refs: The number of refs to subtract from each folio. |
| 943 | * |
| 944 | * Like folio_put(), but for a batch of folios. This is more efficient |
| 945 | * than writing the loop yourself as it will optimise the locks which need |
| 946 | * to be taken if the folios are freed. The folios batch is returned |
| 947 | * empty and ready to be reused for another batch; there is no need |
| 948 | * to reinitialise it. If @refs is NULL, we subtract one from each |
| 949 | * folio refcount. |
| 950 | * |
| 951 | * Context: May be called in process or interrupt context, but not in NMI |
| 952 | * context. May be called while holding a spinlock. |
| 953 | */ |
| 954 | void folios_put_refs(struct folio_batch *folios, unsigned int *refs) |
| 955 | { |
| 956 | int i, j; |
| 957 | struct lruvec *lruvec = NULL; |
| 958 | unsigned long flags = 0; |
| 959 | |
| 960 | for (i = 0, j = 0; i < folios->nr; i++) { |
| 961 | struct folio *folio = folios->folios[i]; |
| 962 | unsigned int nr_refs = refs ? refs[i] : 1; |
| 963 | |
| 964 | if (is_huge_zero_folio(folio)) |
| 965 | continue; |
| 966 | |
| 967 | if (folio_is_zone_device(folio)) { |
| 968 | if (lruvec) { |
| 969 | unlock_page_lruvec_irqrestore(lruvec, flags); |
| 970 | lruvec = NULL; |
| 971 | } |
| 972 | if (put_devmap_managed_folio_refs(folio, nr_refs)) |
| 973 | continue; |
| 974 | if (folio_ref_sub_and_test(folio, nr_refs)) |
| 975 | free_zone_device_folio(folio); |
| 976 | continue; |
| 977 | } |
| 978 | |
| 979 | if (!folio_ref_sub_and_test(folio, nr_refs)) |
| 980 | continue; |
| 981 | |
| 982 | /* hugetlb has its own memcg */ |
| 983 | if (folio_test_hugetlb(folio)) { |
| 984 | if (lruvec) { |
| 985 | unlock_page_lruvec_irqrestore(lruvec, flags); |
| 986 | lruvec = NULL; |
| 987 | } |
| 988 | free_huge_folio(folio); |
| 989 | continue; |
| 990 | } |
| 991 | folio_undo_large_rmappable(folio); |
| 992 | __page_cache_release(folio, &lruvec, &flags); |
| 993 | |
| 994 | if (j != i) |
| 995 | folios->folios[j] = folio; |
| 996 | j++; |
| 997 | } |
| 998 | if (lruvec) |
| 999 | unlock_page_lruvec_irqrestore(lruvec, flags); |
| 1000 | if (!j) { |
| 1001 | folio_batch_reinit(folios); |
| 1002 | return; |
| 1003 | } |
| 1004 | |
| 1005 | folios->nr = j; |
| 1006 | mem_cgroup_uncharge_folios(folios); |
| 1007 | free_unref_folios(folios); |
| 1008 | } |
| 1009 | EXPORT_SYMBOL(folios_put_refs); |
| 1010 | |
| 1011 | /** |
| 1012 | * release_pages - batched put_page() |
| 1013 | * @arg: array of pages to release |
| 1014 | * @nr: number of pages |
| 1015 | * |
| 1016 | * Decrement the reference count on all the pages in @arg. If it |
| 1017 | * fell to zero, remove the page from the LRU and free it. |
| 1018 | * |
| 1019 | * Note that the argument can be an array of pages, encoded pages, |
| 1020 | * or folio pointers. We ignore any encoded bits, and turn any of |
| 1021 | * them into just a folio that gets free'd. |
| 1022 | */ |
| 1023 | void release_pages(release_pages_arg arg, int nr) |
| 1024 | { |
| 1025 | struct folio_batch fbatch; |
| 1026 | int refs[PAGEVEC_SIZE]; |
| 1027 | struct encoded_page **encoded = arg.encoded_pages; |
| 1028 | int i; |
| 1029 | |
| 1030 | folio_batch_init(&fbatch); |
| 1031 | for (i = 0; i < nr; i++) { |
| 1032 | /* Turn any of the argument types into a folio */ |
| 1033 | struct folio *folio = page_folio(encoded_page_ptr(encoded[i])); |
| 1034 | |
| 1035 | /* Is our next entry actually "nr_pages" -> "nr_refs" ? */ |
| 1036 | refs[fbatch.nr] = 1; |
| 1037 | if (unlikely(encoded_page_flags(encoded[i]) & |
| 1038 | ENCODED_PAGE_BIT_NR_PAGES_NEXT)) |
| 1039 | refs[fbatch.nr] = encoded_nr_pages(encoded[++i]); |
| 1040 | |
| 1041 | if (folio_batch_add(&fbatch, folio) > 0) |
| 1042 | continue; |
| 1043 | folios_put_refs(&fbatch, refs); |
| 1044 | } |
| 1045 | |
| 1046 | if (fbatch.nr) |
| 1047 | folios_put_refs(&fbatch, refs); |
| 1048 | } |
| 1049 | EXPORT_SYMBOL(release_pages); |
| 1050 | |
| 1051 | /* |
| 1052 | * The folios which we're about to release may be in the deferred lru-addition |
| 1053 | * queues. That would prevent them from really being freed right now. That's |
| 1054 | * OK from a correctness point of view but is inefficient - those folios may be |
| 1055 | * cache-warm and we want to give them back to the page allocator ASAP. |
| 1056 | * |
| 1057 | * So __folio_batch_release() will drain those queues here. |
| 1058 | * folio_batch_move_lru() calls folios_put() directly to avoid |
| 1059 | * mutual recursion. |
| 1060 | */ |
| 1061 | void __folio_batch_release(struct folio_batch *fbatch) |
| 1062 | { |
| 1063 | if (!fbatch->percpu_pvec_drained) { |
| 1064 | lru_add_drain(); |
| 1065 | fbatch->percpu_pvec_drained = true; |
| 1066 | } |
| 1067 | folios_put(fbatch); |
| 1068 | } |
| 1069 | EXPORT_SYMBOL(__folio_batch_release); |
| 1070 | |
| 1071 | /** |
| 1072 | * folio_batch_remove_exceptionals() - Prune non-folios from a batch. |
| 1073 | * @fbatch: The batch to prune |
| 1074 | * |
| 1075 | * find_get_entries() fills a batch with both folios and shadow/swap/DAX |
| 1076 | * entries. This function prunes all the non-folio entries from @fbatch |
| 1077 | * without leaving holes, so that it can be passed on to folio-only batch |
| 1078 | * operations. |
| 1079 | */ |
| 1080 | void folio_batch_remove_exceptionals(struct folio_batch *fbatch) |
| 1081 | { |
| 1082 | unsigned int i, j; |
| 1083 | |
| 1084 | for (i = 0, j = 0; i < folio_batch_count(fbatch); i++) { |
| 1085 | struct folio *folio = fbatch->folios[i]; |
| 1086 | if (!xa_is_value(folio)) |
| 1087 | fbatch->folios[j++] = folio; |
| 1088 | } |
| 1089 | fbatch->nr = j; |
| 1090 | } |
| 1091 | |
| 1092 | /* |
| 1093 | * Perform any setup for the swap system |
| 1094 | */ |
| 1095 | void __init swap_setup(void) |
| 1096 | { |
| 1097 | unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT); |
| 1098 | |
| 1099 | /* Use a smaller cluster for small-memory machines */ |
| 1100 | if (megs < 16) |
| 1101 | page_cluster = 2; |
| 1102 | else |
| 1103 | page_cluster = 3; |
| 1104 | /* |
| 1105 | * Right now other parts of the system means that we |
| 1106 | * _really_ don't want to cluster much more |
| 1107 | */ |
| 1108 | } |