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