1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef LINUX_MM_INLINE_H 3 #define LINUX_MM_INLINE_H 4 5 #include <linux/atomic.h> 6 #include <linux/huge_mm.h> 7 #include <linux/swap.h> 8 #include <linux/string.h> 9 #include <linux/userfaultfd_k.h> 10 #include <linux/swapops.h> 11 12 /** 13 * folio_is_file_lru - Should the folio be on a file LRU or anon LRU? 14 * @folio: The folio to test. 15 * 16 * We would like to get this info without a page flag, but the state 17 * needs to survive until the folio is last deleted from the LRU, which 18 * could be as far down as __page_cache_release. 19 * 20 * Return: An integer (not a boolean!) used to sort a folio onto the 21 * right LRU list and to account folios correctly. 22 * 1 if @folio is a regular filesystem backed page cache folio 23 * or a lazily freed anonymous folio (e.g. via MADV_FREE). 24 * 0 if @folio is a normal anonymous folio, a tmpfs folio or otherwise 25 * ram or swap backed folio. 26 */ 27 static inline int folio_is_file_lru(struct folio *folio) 28 { 29 return !folio_test_swapbacked(folio); 30 } 31 32 static inline int page_is_file_lru(struct page *page) 33 { 34 return folio_is_file_lru(page_folio(page)); 35 } 36 37 static __always_inline void __update_lru_size(struct lruvec *lruvec, 38 enum lru_list lru, enum zone_type zid, 39 long nr_pages) 40 { 41 struct pglist_data *pgdat = lruvec_pgdat(lruvec); 42 43 lockdep_assert_held(&lruvec->lru_lock); 44 WARN_ON_ONCE(nr_pages != (int)nr_pages); 45 46 __mod_lruvec_state(lruvec, NR_LRU_BASE + lru, nr_pages); 47 __mod_zone_page_state(&pgdat->node_zones[zid], 48 NR_ZONE_LRU_BASE + lru, nr_pages); 49 } 50 51 static __always_inline void update_lru_size(struct lruvec *lruvec, 52 enum lru_list lru, enum zone_type zid, 53 long nr_pages) 54 { 55 __update_lru_size(lruvec, lru, zid, nr_pages); 56 #ifdef CONFIG_MEMCG 57 mem_cgroup_update_lru_size(lruvec, lru, zid, nr_pages); 58 #endif 59 } 60 61 /** 62 * __folio_clear_lru_flags - Clear page lru flags before releasing a page. 63 * @folio: The folio that was on lru and now has a zero reference. 64 */ 65 static __always_inline void __folio_clear_lru_flags(struct folio *folio) 66 { 67 VM_BUG_ON_FOLIO(!folio_test_lru(folio), folio); 68 69 __folio_clear_lru(folio); 70 71 /* this shouldn't happen, so leave the flags to bad_page() */ 72 if (folio_test_active(folio) && folio_test_unevictable(folio)) 73 return; 74 75 __folio_clear_active(folio); 76 __folio_clear_unevictable(folio); 77 } 78 79 /** 80 * folio_lru_list - Which LRU list should a folio be on? 81 * @folio: The folio to test. 82 * 83 * Return: The LRU list a folio should be on, as an index 84 * into the array of LRU lists. 85 */ 86 static __always_inline enum lru_list folio_lru_list(struct folio *folio) 87 { 88 enum lru_list lru; 89 90 VM_BUG_ON_FOLIO(folio_test_active(folio) && folio_test_unevictable(folio), folio); 91 92 if (folio_test_unevictable(folio)) 93 return LRU_UNEVICTABLE; 94 95 lru = folio_is_file_lru(folio) ? LRU_INACTIVE_FILE : LRU_INACTIVE_ANON; 96 if (folio_test_active(folio)) 97 lru += LRU_ACTIVE; 98 99 return lru; 100 } 101 102 #ifdef CONFIG_LRU_GEN 103 104 #ifdef CONFIG_LRU_GEN_ENABLED 105 static inline bool lru_gen_enabled(void) 106 { 107 DECLARE_STATIC_KEY_TRUE(lru_gen_caps[NR_LRU_GEN_CAPS]); 108 109 return static_branch_likely(&lru_gen_caps[LRU_GEN_CORE]); 110 } 111 #else 112 static inline bool lru_gen_enabled(void) 113 { 114 DECLARE_STATIC_KEY_FALSE(lru_gen_caps[NR_LRU_GEN_CAPS]); 115 116 return static_branch_unlikely(&lru_gen_caps[LRU_GEN_CORE]); 117 } 118 #endif 119 120 static inline bool lru_gen_in_fault(void) 121 { 122 return current->in_lru_fault; 123 } 124 125 static inline int lru_gen_from_seq(unsigned long seq) 126 { 127 return seq % MAX_NR_GENS; 128 } 129 130 static inline int lru_hist_from_seq(unsigned long seq) 131 { 132 return seq % NR_HIST_GENS; 133 } 134 135 static inline int lru_tier_from_refs(int refs) 136 { 137 VM_WARN_ON_ONCE(refs > BIT(LRU_REFS_WIDTH)); 138 139 /* see the comment in folio_lru_refs() */ 140 return order_base_2(refs + 1); 141 } 142 143 static inline int folio_lru_refs(struct folio *folio) 144 { 145 unsigned long flags = READ_ONCE(folio->flags); 146 bool workingset = flags & BIT(PG_workingset); 147 148 /* 149 * Return the number of accesses beyond PG_referenced, i.e., N-1 if the 150 * total number of accesses is N>1, since N=0,1 both map to the first 151 * tier. lru_tier_from_refs() will account for this off-by-one. Also see 152 * the comment on MAX_NR_TIERS. 153 */ 154 return ((flags & LRU_REFS_MASK) >> LRU_REFS_PGOFF) + workingset; 155 } 156 157 static inline int folio_lru_gen(struct folio *folio) 158 { 159 unsigned long flags = READ_ONCE(folio->flags); 160 161 return ((flags & LRU_GEN_MASK) >> LRU_GEN_PGOFF) - 1; 162 } 163 164 static inline bool lru_gen_is_active(struct lruvec *lruvec, int gen) 165 { 166 unsigned long max_seq = lruvec->lrugen.max_seq; 167 168 VM_WARN_ON_ONCE(gen >= MAX_NR_GENS); 169 170 /* see the comment on MIN_NR_GENS */ 171 return gen == lru_gen_from_seq(max_seq) || gen == lru_gen_from_seq(max_seq - 1); 172 } 173 174 static inline void lru_gen_update_size(struct lruvec *lruvec, struct folio *folio, 175 int old_gen, int new_gen) 176 { 177 int type = folio_is_file_lru(folio); 178 int zone = folio_zonenum(folio); 179 int delta = folio_nr_pages(folio); 180 enum lru_list lru = type * LRU_INACTIVE_FILE; 181 struct lru_gen_folio *lrugen = &lruvec->lrugen; 182 183 VM_WARN_ON_ONCE(old_gen != -1 && old_gen >= MAX_NR_GENS); 184 VM_WARN_ON_ONCE(new_gen != -1 && new_gen >= MAX_NR_GENS); 185 VM_WARN_ON_ONCE(old_gen == -1 && new_gen == -1); 186 187 if (old_gen >= 0) 188 WRITE_ONCE(lrugen->nr_pages[old_gen][type][zone], 189 lrugen->nr_pages[old_gen][type][zone] - delta); 190 if (new_gen >= 0) 191 WRITE_ONCE(lrugen->nr_pages[new_gen][type][zone], 192 lrugen->nr_pages[new_gen][type][zone] + delta); 193 194 /* addition */ 195 if (old_gen < 0) { 196 if (lru_gen_is_active(lruvec, new_gen)) 197 lru += LRU_ACTIVE; 198 __update_lru_size(lruvec, lru, zone, delta); 199 return; 200 } 201 202 /* deletion */ 203 if (new_gen < 0) { 204 if (lru_gen_is_active(lruvec, old_gen)) 205 lru += LRU_ACTIVE; 206 __update_lru_size(lruvec, lru, zone, -delta); 207 return; 208 } 209 210 /* promotion */ 211 if (!lru_gen_is_active(lruvec, old_gen) && lru_gen_is_active(lruvec, new_gen)) { 212 __update_lru_size(lruvec, lru, zone, -delta); 213 __update_lru_size(lruvec, lru + LRU_ACTIVE, zone, delta); 214 } 215 216 /* demotion requires isolation, e.g., lru_deactivate_fn() */ 217 VM_WARN_ON_ONCE(lru_gen_is_active(lruvec, old_gen) && !lru_gen_is_active(lruvec, new_gen)); 218 } 219 220 static inline bool lru_gen_add_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming) 221 { 222 unsigned long seq; 223 unsigned long flags; 224 int gen = folio_lru_gen(folio); 225 int type = folio_is_file_lru(folio); 226 int zone = folio_zonenum(folio); 227 struct lru_gen_folio *lrugen = &lruvec->lrugen; 228 229 VM_WARN_ON_ONCE_FOLIO(gen != -1, folio); 230 231 if (folio_test_unevictable(folio) || !lrugen->enabled) 232 return false; 233 /* 234 * There are three common cases for this page: 235 * 1. If it's hot, e.g., freshly faulted in or previously hot and 236 * migrated, add it to the youngest generation. 237 * 2. If it's cold but can't be evicted immediately, i.e., an anon page 238 * not in swapcache or a dirty page pending writeback, add it to the 239 * second oldest generation. 240 * 3. Everything else (clean, cold) is added to the oldest generation. 241 */ 242 if (folio_test_active(folio)) 243 seq = lrugen->max_seq; 244 else if ((type == LRU_GEN_ANON && !folio_test_swapcache(folio)) || 245 (folio_test_reclaim(folio) && 246 (folio_test_dirty(folio) || folio_test_writeback(folio)))) 247 seq = lrugen->min_seq[type] + 1; 248 else 249 seq = lrugen->min_seq[type]; 250 251 gen = lru_gen_from_seq(seq); 252 flags = (gen + 1UL) << LRU_GEN_PGOFF; 253 /* see the comment on MIN_NR_GENS about PG_active */ 254 set_mask_bits(&folio->flags, LRU_GEN_MASK | BIT(PG_active), flags); 255 256 lru_gen_update_size(lruvec, folio, -1, gen); 257 /* for folio_rotate_reclaimable() */ 258 if (reclaiming) 259 list_add_tail(&folio->lru, &lrugen->folios[gen][type][zone]); 260 else 261 list_add(&folio->lru, &lrugen->folios[gen][type][zone]); 262 263 return true; 264 } 265 266 static inline bool lru_gen_del_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming) 267 { 268 unsigned long flags; 269 int gen = folio_lru_gen(folio); 270 271 if (gen < 0) 272 return false; 273 274 VM_WARN_ON_ONCE_FOLIO(folio_test_active(folio), folio); 275 VM_WARN_ON_ONCE_FOLIO(folio_test_unevictable(folio), folio); 276 277 /* for folio_migrate_flags() */ 278 flags = !reclaiming && lru_gen_is_active(lruvec, gen) ? BIT(PG_active) : 0; 279 flags = set_mask_bits(&folio->flags, LRU_GEN_MASK, flags); 280 gen = ((flags & LRU_GEN_MASK) >> LRU_GEN_PGOFF) - 1; 281 282 lru_gen_update_size(lruvec, folio, gen, -1); 283 list_del(&folio->lru); 284 285 return true; 286 } 287 288 #else /* !CONFIG_LRU_GEN */ 289 290 static inline bool lru_gen_enabled(void) 291 { 292 return false; 293 } 294 295 static inline bool lru_gen_in_fault(void) 296 { 297 return false; 298 } 299 300 static inline bool lru_gen_add_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming) 301 { 302 return false; 303 } 304 305 static inline bool lru_gen_del_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming) 306 { 307 return false; 308 } 309 310 #endif /* CONFIG_LRU_GEN */ 311 312 static __always_inline 313 void lruvec_add_folio(struct lruvec *lruvec, struct folio *folio) 314 { 315 enum lru_list lru = folio_lru_list(folio); 316 317 if (lru_gen_add_folio(lruvec, folio, false)) 318 return; 319 320 update_lru_size(lruvec, lru, folio_zonenum(folio), 321 folio_nr_pages(folio)); 322 if (lru != LRU_UNEVICTABLE) 323 list_add(&folio->lru, &lruvec->lists[lru]); 324 } 325 326 static __always_inline void add_page_to_lru_list(struct page *page, 327 struct lruvec *lruvec) 328 { 329 lruvec_add_folio(lruvec, page_folio(page)); 330 } 331 332 static __always_inline 333 void lruvec_add_folio_tail(struct lruvec *lruvec, struct folio *folio) 334 { 335 enum lru_list lru = folio_lru_list(folio); 336 337 if (lru_gen_add_folio(lruvec, folio, true)) 338 return; 339 340 update_lru_size(lruvec, lru, folio_zonenum(folio), 341 folio_nr_pages(folio)); 342 /* This is not expected to be used on LRU_UNEVICTABLE */ 343 list_add_tail(&folio->lru, &lruvec->lists[lru]); 344 } 345 346 static __always_inline 347 void lruvec_del_folio(struct lruvec *lruvec, struct folio *folio) 348 { 349 enum lru_list lru = folio_lru_list(folio); 350 351 if (lru_gen_del_folio(lruvec, folio, false)) 352 return; 353 354 if (lru != LRU_UNEVICTABLE) 355 list_del(&folio->lru); 356 update_lru_size(lruvec, lru, folio_zonenum(folio), 357 -folio_nr_pages(folio)); 358 } 359 360 static __always_inline void del_page_from_lru_list(struct page *page, 361 struct lruvec *lruvec) 362 { 363 lruvec_del_folio(lruvec, page_folio(page)); 364 } 365 366 #ifdef CONFIG_ANON_VMA_NAME 367 /* 368 * mmap_lock should be read-locked when calling anon_vma_name(). Caller should 369 * either keep holding the lock while using the returned pointer or it should 370 * raise anon_vma_name refcount before releasing the lock. 371 */ 372 extern struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma); 373 extern struct anon_vma_name *anon_vma_name_alloc(const char *name); 374 extern void anon_vma_name_free(struct kref *kref); 375 376 /* mmap_lock should be read-locked */ 377 static inline void anon_vma_name_get(struct anon_vma_name *anon_name) 378 { 379 if (anon_name) 380 kref_get(&anon_name->kref); 381 } 382 383 static inline void anon_vma_name_put(struct anon_vma_name *anon_name) 384 { 385 if (anon_name) 386 kref_put(&anon_name->kref, anon_vma_name_free); 387 } 388 389 static inline 390 struct anon_vma_name *anon_vma_name_reuse(struct anon_vma_name *anon_name) 391 { 392 /* Prevent anon_name refcount saturation early on */ 393 if (kref_read(&anon_name->kref) < REFCOUNT_MAX) { 394 anon_vma_name_get(anon_name); 395 return anon_name; 396 397 } 398 return anon_vma_name_alloc(anon_name->name); 399 } 400 401 static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma, 402 struct vm_area_struct *new_vma) 403 { 404 struct anon_vma_name *anon_name = anon_vma_name(orig_vma); 405 406 if (anon_name) 407 new_vma->anon_name = anon_vma_name_reuse(anon_name); 408 } 409 410 static inline void free_anon_vma_name(struct vm_area_struct *vma) 411 { 412 /* 413 * Not using anon_vma_name because it generates a warning if mmap_lock 414 * is not held, which might be the case here. 415 */ 416 anon_vma_name_put(vma->anon_name); 417 } 418 419 static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1, 420 struct anon_vma_name *anon_name2) 421 { 422 if (anon_name1 == anon_name2) 423 return true; 424 425 return anon_name1 && anon_name2 && 426 !strcmp(anon_name1->name, anon_name2->name); 427 } 428 429 #else /* CONFIG_ANON_VMA_NAME */ 430 static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma) 431 { 432 return NULL; 433 } 434 435 static inline struct anon_vma_name *anon_vma_name_alloc(const char *name) 436 { 437 return NULL; 438 } 439 440 static inline void anon_vma_name_get(struct anon_vma_name *anon_name) {} 441 static inline void anon_vma_name_put(struct anon_vma_name *anon_name) {} 442 static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma, 443 struct vm_area_struct *new_vma) {} 444 static inline void free_anon_vma_name(struct vm_area_struct *vma) {} 445 446 static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1, 447 struct anon_vma_name *anon_name2) 448 { 449 return true; 450 } 451 452 #endif /* CONFIG_ANON_VMA_NAME */ 453 454 static inline void init_tlb_flush_pending(struct mm_struct *mm) 455 { 456 atomic_set(&mm->tlb_flush_pending, 0); 457 } 458 459 static inline void inc_tlb_flush_pending(struct mm_struct *mm) 460 { 461 atomic_inc(&mm->tlb_flush_pending); 462 /* 463 * The only time this value is relevant is when there are indeed pages 464 * to flush. And we'll only flush pages after changing them, which 465 * requires the PTL. 466 * 467 * So the ordering here is: 468 * 469 * atomic_inc(&mm->tlb_flush_pending); 470 * spin_lock(&ptl); 471 * ... 472 * set_pte_at(); 473 * spin_unlock(&ptl); 474 * 475 * spin_lock(&ptl) 476 * mm_tlb_flush_pending(); 477 * .... 478 * spin_unlock(&ptl); 479 * 480 * flush_tlb_range(); 481 * atomic_dec(&mm->tlb_flush_pending); 482 * 483 * Where the increment if constrained by the PTL unlock, it thus 484 * ensures that the increment is visible if the PTE modification is 485 * visible. After all, if there is no PTE modification, nobody cares 486 * about TLB flushes either. 487 * 488 * This very much relies on users (mm_tlb_flush_pending() and 489 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and 490 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc 491 * locks (PPC) the unlock of one doesn't order against the lock of 492 * another PTL. 493 * 494 * The decrement is ordered by the flush_tlb_range(), such that 495 * mm_tlb_flush_pending() will not return false unless all flushes have 496 * completed. 497 */ 498 } 499 500 static inline void dec_tlb_flush_pending(struct mm_struct *mm) 501 { 502 /* 503 * See inc_tlb_flush_pending(). 504 * 505 * This cannot be smp_mb__before_atomic() because smp_mb() simply does 506 * not order against TLB invalidate completion, which is what we need. 507 * 508 * Therefore we must rely on tlb_flush_*() to guarantee order. 509 */ 510 atomic_dec(&mm->tlb_flush_pending); 511 } 512 513 static inline bool mm_tlb_flush_pending(struct mm_struct *mm) 514 { 515 /* 516 * Must be called after having acquired the PTL; orders against that 517 * PTLs release and therefore ensures that if we observe the modified 518 * PTE we must also observe the increment from inc_tlb_flush_pending(). 519 * 520 * That is, it only guarantees to return true if there is a flush 521 * pending for _this_ PTL. 522 */ 523 return atomic_read(&mm->tlb_flush_pending); 524 } 525 526 static inline bool mm_tlb_flush_nested(struct mm_struct *mm) 527 { 528 /* 529 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL 530 * for which there is a TLB flush pending in order to guarantee 531 * we've seen both that PTE modification and the increment. 532 * 533 * (no requirement on actually still holding the PTL, that is irrelevant) 534 */ 535 return atomic_read(&mm->tlb_flush_pending) > 1; 536 } 537 538 /* 539 * If this pte is wr-protected by uffd-wp in any form, arm the special pte to 540 * replace a none pte. NOTE! This should only be called when *pte is already 541 * cleared so we will never accidentally replace something valuable. Meanwhile 542 * none pte also means we are not demoting the pte so tlb flushed is not needed. 543 * E.g., when pte cleared the caller should have taken care of the tlb flush. 544 * 545 * Must be called with pgtable lock held so that no thread will see the none 546 * pte, and if they see it, they'll fault and serialize at the pgtable lock. 547 * 548 * This function is a no-op if PTE_MARKER_UFFD_WP is not enabled. 549 */ 550 static inline void 551 pte_install_uffd_wp_if_needed(struct vm_area_struct *vma, unsigned long addr, 552 pte_t *pte, pte_t pteval) 553 { 554 #ifdef CONFIG_PTE_MARKER_UFFD_WP 555 bool arm_uffd_pte = false; 556 557 /* The current status of the pte should be "cleared" before calling */ 558 WARN_ON_ONCE(!pte_none(*pte)); 559 560 /* 561 * NOTE: userfaultfd_wp_unpopulated() doesn't need this whole 562 * thing, because when zapping either it means it's dropping the 563 * page, or in TTU where the present pte will be quickly replaced 564 * with a swap pte. There's no way of leaking the bit. 565 */ 566 if (vma_is_anonymous(vma) || !userfaultfd_wp(vma)) 567 return; 568 569 /* A uffd-wp wr-protected normal pte */ 570 if (unlikely(pte_present(pteval) && pte_uffd_wp(pteval))) 571 arm_uffd_pte = true; 572 573 /* 574 * A uffd-wp wr-protected swap pte. Note: this should even cover an 575 * existing pte marker with uffd-wp bit set. 576 */ 577 if (unlikely(pte_swp_uffd_wp_any(pteval))) 578 arm_uffd_pte = true; 579 580 if (unlikely(arm_uffd_pte)) 581 set_pte_at(vma->vm_mm, addr, pte, 582 make_pte_marker(PTE_MARKER_UFFD_WP)); 583 #endif 584 } 585 586 static inline bool vma_has_recency(struct vm_area_struct *vma) 587 { 588 if (vma->vm_flags & (VM_SEQ_READ | VM_RAND_READ)) 589 return false; 590 591 if (vma->vm_file && (vma->vm_file->f_mode & FMODE_NOREUSE)) 592 return false; 593 594 return true; 595 } 596 597 #endif 598