1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2013 Red Hat, Inc. and Parallels Inc. All rights reserved. 4 * Authors: David Chinner and Glauber Costa 5 * 6 * Generic LRU infrastructure 7 */ 8 #include <linux/kernel.h> 9 #include <linux/module.h> 10 #include <linux/mm.h> 11 #include <linux/list_lru.h> 12 #include <linux/slab.h> 13 #include <linux/mutex.h> 14 #include <linux/memcontrol.h> 15 #include "slab.h" 16 17 #ifdef CONFIG_MEMCG_KMEM 18 static LIST_HEAD(list_lrus); 19 static DEFINE_MUTEX(list_lrus_mutex); 20 21 static void list_lru_register(struct list_lru *lru) 22 { 23 mutex_lock(&list_lrus_mutex); 24 list_add(&lru->list, &list_lrus); 25 mutex_unlock(&list_lrus_mutex); 26 } 27 28 static void list_lru_unregister(struct list_lru *lru) 29 { 30 mutex_lock(&list_lrus_mutex); 31 list_del(&lru->list); 32 mutex_unlock(&list_lrus_mutex); 33 } 34 35 static int lru_shrinker_id(struct list_lru *lru) 36 { 37 return lru->shrinker_id; 38 } 39 40 static inline bool list_lru_memcg_aware(struct list_lru *lru) 41 { 42 return lru->memcg_aware; 43 } 44 45 static inline struct list_lru_one * 46 list_lru_from_memcg_idx(struct list_lru_node *nlru, int idx) 47 { 48 struct list_lru_memcg *memcg_lrus; 49 /* 50 * Either lock or RCU protects the array of per cgroup lists 51 * from relocation (see memcg_update_list_lru_node). 52 */ 53 memcg_lrus = rcu_dereference_check(nlru->memcg_lrus, 54 lockdep_is_held(&nlru->lock)); 55 if (memcg_lrus && idx >= 0) 56 return memcg_lrus->lru[idx]; 57 return &nlru->lru; 58 } 59 60 static inline struct list_lru_one * 61 list_lru_from_kmem(struct list_lru_node *nlru, void *ptr, 62 struct mem_cgroup **memcg_ptr) 63 { 64 struct list_lru_one *l = &nlru->lru; 65 struct mem_cgroup *memcg = NULL; 66 67 if (!nlru->memcg_lrus) 68 goto out; 69 70 memcg = mem_cgroup_from_obj(ptr); 71 if (!memcg) 72 goto out; 73 74 l = list_lru_from_memcg_idx(nlru, memcg_cache_id(memcg)); 75 out: 76 if (memcg_ptr) 77 *memcg_ptr = memcg; 78 return l; 79 } 80 #else 81 static void list_lru_register(struct list_lru *lru) 82 { 83 } 84 85 static void list_lru_unregister(struct list_lru *lru) 86 { 87 } 88 89 static int lru_shrinker_id(struct list_lru *lru) 90 { 91 return -1; 92 } 93 94 static inline bool list_lru_memcg_aware(struct list_lru *lru) 95 { 96 return false; 97 } 98 99 static inline struct list_lru_one * 100 list_lru_from_memcg_idx(struct list_lru_node *nlru, int idx) 101 { 102 return &nlru->lru; 103 } 104 105 static inline struct list_lru_one * 106 list_lru_from_kmem(struct list_lru_node *nlru, void *ptr, 107 struct mem_cgroup **memcg_ptr) 108 { 109 if (memcg_ptr) 110 *memcg_ptr = NULL; 111 return &nlru->lru; 112 } 113 #endif /* CONFIG_MEMCG_KMEM */ 114 115 bool list_lru_add(struct list_lru *lru, struct list_head *item) 116 { 117 int nid = page_to_nid(virt_to_page(item)); 118 struct list_lru_node *nlru = &lru->node[nid]; 119 struct mem_cgroup *memcg; 120 struct list_lru_one *l; 121 122 spin_lock(&nlru->lock); 123 if (list_empty(item)) { 124 l = list_lru_from_kmem(nlru, item, &memcg); 125 list_add_tail(item, &l->list); 126 /* Set shrinker bit if the first element was added */ 127 if (!l->nr_items++) 128 set_shrinker_bit(memcg, nid, 129 lru_shrinker_id(lru)); 130 nlru->nr_items++; 131 spin_unlock(&nlru->lock); 132 return true; 133 } 134 spin_unlock(&nlru->lock); 135 return false; 136 } 137 EXPORT_SYMBOL_GPL(list_lru_add); 138 139 bool list_lru_del(struct list_lru *lru, struct list_head *item) 140 { 141 int nid = page_to_nid(virt_to_page(item)); 142 struct list_lru_node *nlru = &lru->node[nid]; 143 struct list_lru_one *l; 144 145 spin_lock(&nlru->lock); 146 if (!list_empty(item)) { 147 l = list_lru_from_kmem(nlru, item, NULL); 148 list_del_init(item); 149 l->nr_items--; 150 nlru->nr_items--; 151 spin_unlock(&nlru->lock); 152 return true; 153 } 154 spin_unlock(&nlru->lock); 155 return false; 156 } 157 EXPORT_SYMBOL_GPL(list_lru_del); 158 159 void list_lru_isolate(struct list_lru_one *list, struct list_head *item) 160 { 161 list_del_init(item); 162 list->nr_items--; 163 } 164 EXPORT_SYMBOL_GPL(list_lru_isolate); 165 166 void list_lru_isolate_move(struct list_lru_one *list, struct list_head *item, 167 struct list_head *head) 168 { 169 list_move(item, head); 170 list->nr_items--; 171 } 172 EXPORT_SYMBOL_GPL(list_lru_isolate_move); 173 174 unsigned long list_lru_count_one(struct list_lru *lru, 175 int nid, struct mem_cgroup *memcg) 176 { 177 struct list_lru_node *nlru = &lru->node[nid]; 178 struct list_lru_one *l; 179 unsigned long count; 180 181 rcu_read_lock(); 182 l = list_lru_from_memcg_idx(nlru, memcg_cache_id(memcg)); 183 count = READ_ONCE(l->nr_items); 184 rcu_read_unlock(); 185 186 return count; 187 } 188 EXPORT_SYMBOL_GPL(list_lru_count_one); 189 190 unsigned long list_lru_count_node(struct list_lru *lru, int nid) 191 { 192 struct list_lru_node *nlru; 193 194 nlru = &lru->node[nid]; 195 return nlru->nr_items; 196 } 197 EXPORT_SYMBOL_GPL(list_lru_count_node); 198 199 static unsigned long 200 __list_lru_walk_one(struct list_lru_node *nlru, int memcg_idx, 201 list_lru_walk_cb isolate, void *cb_arg, 202 unsigned long *nr_to_walk) 203 { 204 205 struct list_lru_one *l; 206 struct list_head *item, *n; 207 unsigned long isolated = 0; 208 209 l = list_lru_from_memcg_idx(nlru, memcg_idx); 210 restart: 211 list_for_each_safe(item, n, &l->list) { 212 enum lru_status ret; 213 214 /* 215 * decrement nr_to_walk first so that we don't livelock if we 216 * get stuck on large numbers of LRU_RETRY items 217 */ 218 if (!*nr_to_walk) 219 break; 220 --*nr_to_walk; 221 222 ret = isolate(item, l, &nlru->lock, cb_arg); 223 switch (ret) { 224 case LRU_REMOVED_RETRY: 225 assert_spin_locked(&nlru->lock); 226 fallthrough; 227 case LRU_REMOVED: 228 isolated++; 229 nlru->nr_items--; 230 /* 231 * If the lru lock has been dropped, our list 232 * traversal is now invalid and so we have to 233 * restart from scratch. 234 */ 235 if (ret == LRU_REMOVED_RETRY) 236 goto restart; 237 break; 238 case LRU_ROTATE: 239 list_move_tail(item, &l->list); 240 break; 241 case LRU_SKIP: 242 break; 243 case LRU_RETRY: 244 /* 245 * The lru lock has been dropped, our list traversal is 246 * now invalid and so we have to restart from scratch. 247 */ 248 assert_spin_locked(&nlru->lock); 249 goto restart; 250 default: 251 BUG(); 252 } 253 } 254 return isolated; 255 } 256 257 unsigned long 258 list_lru_walk_one(struct list_lru *lru, int nid, struct mem_cgroup *memcg, 259 list_lru_walk_cb isolate, void *cb_arg, 260 unsigned long *nr_to_walk) 261 { 262 struct list_lru_node *nlru = &lru->node[nid]; 263 unsigned long ret; 264 265 spin_lock(&nlru->lock); 266 ret = __list_lru_walk_one(nlru, memcg_cache_id(memcg), isolate, cb_arg, 267 nr_to_walk); 268 spin_unlock(&nlru->lock); 269 return ret; 270 } 271 EXPORT_SYMBOL_GPL(list_lru_walk_one); 272 273 unsigned long 274 list_lru_walk_one_irq(struct list_lru *lru, int nid, struct mem_cgroup *memcg, 275 list_lru_walk_cb isolate, void *cb_arg, 276 unsigned long *nr_to_walk) 277 { 278 struct list_lru_node *nlru = &lru->node[nid]; 279 unsigned long ret; 280 281 spin_lock_irq(&nlru->lock); 282 ret = __list_lru_walk_one(nlru, memcg_cache_id(memcg), isolate, cb_arg, 283 nr_to_walk); 284 spin_unlock_irq(&nlru->lock); 285 return ret; 286 } 287 288 unsigned long list_lru_walk_node(struct list_lru *lru, int nid, 289 list_lru_walk_cb isolate, void *cb_arg, 290 unsigned long *nr_to_walk) 291 { 292 long isolated = 0; 293 int memcg_idx; 294 295 isolated += list_lru_walk_one(lru, nid, NULL, isolate, cb_arg, 296 nr_to_walk); 297 if (*nr_to_walk > 0 && list_lru_memcg_aware(lru)) { 298 for_each_memcg_cache_index(memcg_idx) { 299 struct list_lru_node *nlru = &lru->node[nid]; 300 301 spin_lock(&nlru->lock); 302 isolated += __list_lru_walk_one(nlru, memcg_idx, 303 isolate, cb_arg, 304 nr_to_walk); 305 spin_unlock(&nlru->lock); 306 307 if (*nr_to_walk <= 0) 308 break; 309 } 310 } 311 return isolated; 312 } 313 EXPORT_SYMBOL_GPL(list_lru_walk_node); 314 315 static void init_one_lru(struct list_lru_one *l) 316 { 317 INIT_LIST_HEAD(&l->list); 318 l->nr_items = 0; 319 } 320 321 #ifdef CONFIG_MEMCG_KMEM 322 static void __memcg_destroy_list_lru_node(struct list_lru_memcg *memcg_lrus, 323 int begin, int end) 324 { 325 int i; 326 327 for (i = begin; i < end; i++) 328 kfree(memcg_lrus->lru[i]); 329 } 330 331 static int __memcg_init_list_lru_node(struct list_lru_memcg *memcg_lrus, 332 int begin, int end) 333 { 334 int i; 335 336 for (i = begin; i < end; i++) { 337 struct list_lru_one *l; 338 339 l = kmalloc(sizeof(struct list_lru_one), GFP_KERNEL); 340 if (!l) 341 goto fail; 342 343 init_one_lru(l); 344 memcg_lrus->lru[i] = l; 345 } 346 return 0; 347 fail: 348 __memcg_destroy_list_lru_node(memcg_lrus, begin, i); 349 return -ENOMEM; 350 } 351 352 static int memcg_init_list_lru_node(struct list_lru_node *nlru) 353 { 354 struct list_lru_memcg *memcg_lrus; 355 int size = memcg_nr_cache_ids; 356 357 memcg_lrus = kvmalloc(sizeof(*memcg_lrus) + 358 size * sizeof(void *), GFP_KERNEL); 359 if (!memcg_lrus) 360 return -ENOMEM; 361 362 if (__memcg_init_list_lru_node(memcg_lrus, 0, size)) { 363 kvfree(memcg_lrus); 364 return -ENOMEM; 365 } 366 RCU_INIT_POINTER(nlru->memcg_lrus, memcg_lrus); 367 368 return 0; 369 } 370 371 static void memcg_destroy_list_lru_node(struct list_lru_node *nlru) 372 { 373 struct list_lru_memcg *memcg_lrus; 374 /* 375 * This is called when shrinker has already been unregistered, 376 * and nobody can use it. So, there is no need to use kvfree_rcu(). 377 */ 378 memcg_lrus = rcu_dereference_protected(nlru->memcg_lrus, true); 379 __memcg_destroy_list_lru_node(memcg_lrus, 0, memcg_nr_cache_ids); 380 kvfree(memcg_lrus); 381 } 382 383 static int memcg_update_list_lru_node(struct list_lru_node *nlru, 384 int old_size, int new_size) 385 { 386 struct list_lru_memcg *old, *new; 387 388 BUG_ON(old_size > new_size); 389 390 old = rcu_dereference_protected(nlru->memcg_lrus, 391 lockdep_is_held(&list_lrus_mutex)); 392 new = kvmalloc(sizeof(*new) + new_size * sizeof(void *), GFP_KERNEL); 393 if (!new) 394 return -ENOMEM; 395 396 if (__memcg_init_list_lru_node(new, old_size, new_size)) { 397 kvfree(new); 398 return -ENOMEM; 399 } 400 401 memcpy(&new->lru, &old->lru, old_size * sizeof(void *)); 402 403 /* 404 * The locking below allows readers that hold nlru->lock avoid taking 405 * rcu_read_lock (see list_lru_from_memcg_idx). 406 * 407 * Since list_lru_{add,del} may be called under an IRQ-safe lock, 408 * we have to use IRQ-safe primitives here to avoid deadlock. 409 */ 410 spin_lock_irq(&nlru->lock); 411 rcu_assign_pointer(nlru->memcg_lrus, new); 412 spin_unlock_irq(&nlru->lock); 413 414 kvfree_rcu(old, rcu); 415 return 0; 416 } 417 418 static void memcg_cancel_update_list_lru_node(struct list_lru_node *nlru, 419 int old_size, int new_size) 420 { 421 struct list_lru_memcg *memcg_lrus; 422 423 memcg_lrus = rcu_dereference_protected(nlru->memcg_lrus, 424 lockdep_is_held(&list_lrus_mutex)); 425 /* do not bother shrinking the array back to the old size, because we 426 * cannot handle allocation failures here */ 427 __memcg_destroy_list_lru_node(memcg_lrus, old_size, new_size); 428 } 429 430 static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware) 431 { 432 int i; 433 434 lru->memcg_aware = memcg_aware; 435 436 if (!memcg_aware) 437 return 0; 438 439 for_each_node(i) { 440 if (memcg_init_list_lru_node(&lru->node[i])) 441 goto fail; 442 } 443 return 0; 444 fail: 445 for (i = i - 1; i >= 0; i--) { 446 if (!lru->node[i].memcg_lrus) 447 continue; 448 memcg_destroy_list_lru_node(&lru->node[i]); 449 } 450 return -ENOMEM; 451 } 452 453 static void memcg_destroy_list_lru(struct list_lru *lru) 454 { 455 int i; 456 457 if (!list_lru_memcg_aware(lru)) 458 return; 459 460 for_each_node(i) 461 memcg_destroy_list_lru_node(&lru->node[i]); 462 } 463 464 static int memcg_update_list_lru(struct list_lru *lru, 465 int old_size, int new_size) 466 { 467 int i; 468 469 if (!list_lru_memcg_aware(lru)) 470 return 0; 471 472 for_each_node(i) { 473 if (memcg_update_list_lru_node(&lru->node[i], 474 old_size, new_size)) 475 goto fail; 476 } 477 return 0; 478 fail: 479 for (i = i - 1; i >= 0; i--) { 480 if (!lru->node[i].memcg_lrus) 481 continue; 482 483 memcg_cancel_update_list_lru_node(&lru->node[i], 484 old_size, new_size); 485 } 486 return -ENOMEM; 487 } 488 489 static void memcg_cancel_update_list_lru(struct list_lru *lru, 490 int old_size, int new_size) 491 { 492 int i; 493 494 if (!list_lru_memcg_aware(lru)) 495 return; 496 497 for_each_node(i) 498 memcg_cancel_update_list_lru_node(&lru->node[i], 499 old_size, new_size); 500 } 501 502 int memcg_update_all_list_lrus(int new_size) 503 { 504 int ret = 0; 505 struct list_lru *lru; 506 int old_size = memcg_nr_cache_ids; 507 508 mutex_lock(&list_lrus_mutex); 509 list_for_each_entry(lru, &list_lrus, list) { 510 ret = memcg_update_list_lru(lru, old_size, new_size); 511 if (ret) 512 goto fail; 513 } 514 out: 515 mutex_unlock(&list_lrus_mutex); 516 return ret; 517 fail: 518 list_for_each_entry_continue_reverse(lru, &list_lrus, list) 519 memcg_cancel_update_list_lru(lru, old_size, new_size); 520 goto out; 521 } 522 523 static void memcg_drain_list_lru_node(struct list_lru *lru, int nid, 524 int src_idx, struct mem_cgroup *dst_memcg) 525 { 526 struct list_lru_node *nlru = &lru->node[nid]; 527 int dst_idx = dst_memcg->kmemcg_id; 528 struct list_lru_one *src, *dst; 529 530 /* 531 * Since list_lru_{add,del} may be called under an IRQ-safe lock, 532 * we have to use IRQ-safe primitives here to avoid deadlock. 533 */ 534 spin_lock_irq(&nlru->lock); 535 536 src = list_lru_from_memcg_idx(nlru, src_idx); 537 dst = list_lru_from_memcg_idx(nlru, dst_idx); 538 539 list_splice_init(&src->list, &dst->list); 540 541 if (src->nr_items) { 542 dst->nr_items += src->nr_items; 543 set_shrinker_bit(dst_memcg, nid, lru_shrinker_id(lru)); 544 src->nr_items = 0; 545 } 546 547 spin_unlock_irq(&nlru->lock); 548 } 549 550 static void memcg_drain_list_lru(struct list_lru *lru, 551 int src_idx, struct mem_cgroup *dst_memcg) 552 { 553 int i; 554 555 if (!list_lru_memcg_aware(lru)) 556 return; 557 558 for_each_node(i) 559 memcg_drain_list_lru_node(lru, i, src_idx, dst_memcg); 560 } 561 562 void memcg_drain_all_list_lrus(int src_idx, struct mem_cgroup *dst_memcg) 563 { 564 struct list_lru *lru; 565 566 mutex_lock(&list_lrus_mutex); 567 list_for_each_entry(lru, &list_lrus, list) 568 memcg_drain_list_lru(lru, src_idx, dst_memcg); 569 mutex_unlock(&list_lrus_mutex); 570 } 571 #else 572 static int memcg_init_list_lru(struct list_lru *lru, bool memcg_aware) 573 { 574 return 0; 575 } 576 577 static void memcg_destroy_list_lru(struct list_lru *lru) 578 { 579 } 580 #endif /* CONFIG_MEMCG_KMEM */ 581 582 int __list_lru_init(struct list_lru *lru, bool memcg_aware, 583 struct lock_class_key *key, struct shrinker *shrinker) 584 { 585 int i; 586 int err = -ENOMEM; 587 588 #ifdef CONFIG_MEMCG_KMEM 589 if (shrinker) 590 lru->shrinker_id = shrinker->id; 591 else 592 lru->shrinker_id = -1; 593 #endif 594 memcg_get_cache_ids(); 595 596 lru->node = kcalloc(nr_node_ids, sizeof(*lru->node), GFP_KERNEL); 597 if (!lru->node) 598 goto out; 599 600 for_each_node(i) { 601 spin_lock_init(&lru->node[i].lock); 602 if (key) 603 lockdep_set_class(&lru->node[i].lock, key); 604 init_one_lru(&lru->node[i].lru); 605 } 606 607 err = memcg_init_list_lru(lru, memcg_aware); 608 if (err) { 609 kfree(lru->node); 610 /* Do this so a list_lru_destroy() doesn't crash: */ 611 lru->node = NULL; 612 goto out; 613 } 614 615 list_lru_register(lru); 616 out: 617 memcg_put_cache_ids(); 618 return err; 619 } 620 EXPORT_SYMBOL_GPL(__list_lru_init); 621 622 void list_lru_destroy(struct list_lru *lru) 623 { 624 /* Already destroyed or not yet initialized? */ 625 if (!lru->node) 626 return; 627 628 memcg_get_cache_ids(); 629 630 list_lru_unregister(lru); 631 632 memcg_destroy_list_lru(lru); 633 kfree(lru->node); 634 lru->node = NULL; 635 636 #ifdef CONFIG_MEMCG_KMEM 637 lru->shrinker_id = -1; 638 #endif 639 memcg_put_cache_ids(); 640 } 641 EXPORT_SYMBOL_GPL(list_lru_destroy); 642