mm/list_lru.c

   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 #include "internal.h"
  17
  18 #ifdef CONFIG_MEMCG_KMEM
  19 static LIST_HEAD(memcg_list_lrus);
  20 static DEFINE_MUTEX(list_lrus_mutex);
  21
  22 static inline bool list_lru_memcg_aware(struct list_lru *lru)
  23 {
  24         return lru->memcg_aware;
  25 }
  26
  27 static void list_lru_register(struct list_lru *lru)
  28 {
  29         if (!list_lru_memcg_aware(lru))
  30                 return;
  31
  32         mutex_lock(&list_lrus_mutex);
  33         list_add(&lru->list, &memcg_list_lrus);
  34         mutex_unlock(&list_lrus_mutex);
  35 }
  36
  37 static void list_lru_unregister(struct list_lru *lru)
  38 {
  39         if (!list_lru_memcg_aware(lru))
  40                 return;
  41
  42         mutex_lock(&list_lrus_mutex);
  43         list_del(&lru->list);
  44         mutex_unlock(&list_lrus_mutex);
  45 }
  46
  47 static int lru_shrinker_id(struct list_lru *lru)
  48 {
  49         return lru->shrinker_id;
  50 }
  51
  52 static inline struct list_lru_one *
  53 list_lru_from_memcg_idx(struct list_lru *lru, int nid, int idx)
  54 {
  55         if (list_lru_memcg_aware(lru) && idx >= 0) {
  56                 struct list_lru_memcg *mlru = xa_load(&lru->xa, idx);
  57
  58                 return mlru ? &mlru->node[nid] : NULL;
  59         }
  60         return &lru->node[nid].lru;
  61 }
  62
  63 static inline struct list_lru_one *
  64 list_lru_from_kmem(struct list_lru *lru, int nid, void *ptr,
  65                    struct mem_cgroup **memcg_ptr)
  66 {
  67         struct list_lru_node *nlru = &lru->node[nid];
  68         struct list_lru_one *l = &nlru->lru;
  69         struct mem_cgroup *memcg = NULL;
  70
  71         if (!list_lru_memcg_aware(lru))
  72                 goto out;
  73
  74         memcg = mem_cgroup_from_slab_obj(ptr);
  75         if (!memcg)
  76                 goto out;
  77
  78         l = list_lru_from_memcg_idx(lru, nid, memcg_kmem_id(memcg));
  79 out:
  80         if (memcg_ptr)
  81                 *memcg_ptr = memcg;
  82         return l;
  83 }
  84 #else
  85 static void list_lru_register(struct list_lru *lru)
  86 {
  87 }
  88
  89 static void list_lru_unregister(struct list_lru *lru)
  90 {
  91 }
  92
  93 static int lru_shrinker_id(struct list_lru *lru)
  94 {
  95         return -1;
  96 }
  97
  98 static inline bool list_lru_memcg_aware(struct list_lru *lru)
  99 {
 100         return false;
 101 }
 102
 103 static inline struct list_lru_one *
 104 list_lru_from_memcg_idx(struct list_lru *lru, int nid, int idx)
 105 {
 106         return &lru->node[nid].lru;
 107 }
 108
 109 static inline struct list_lru_one *
 110 list_lru_from_kmem(struct list_lru *lru, int nid, void *ptr,
 111                    struct mem_cgroup **memcg_ptr)
 112 {
 113         if (memcg_ptr)
 114                 *memcg_ptr = NULL;
 115         return &lru->node[nid].lru;
 116 }
 117 #endif /* CONFIG_MEMCG_KMEM */
 118
 119 bool list_lru_add(struct list_lru *lru, struct list_head *item)
 120 {
 121         int nid = page_to_nid(virt_to_page(item));
 122         struct list_lru_node *nlru = &lru->node[nid];
 123         struct mem_cgroup *memcg;
 124         struct list_lru_one *l;
 125
 126         spin_lock(&nlru->lock);
 127         if (list_empty(item)) {
 128                 l = list_lru_from_kmem(lru, nid, item, &memcg);
 129                 list_add_tail(item, &l->list);
 130                 /* Set shrinker bit if the first element was added */
 131                 if (!l->nr_items++)
 132                         set_shrinker_bit(memcg, nid,
 133                                          lru_shrinker_id(lru));
 134                 nlru->nr_items++;
 135                 spin_unlock(&nlru->lock);
 136                 return true;
 137         }
 138         spin_unlock(&nlru->lock);
 139         return false;
 140 }
 141 EXPORT_SYMBOL_GPL(list_lru_add);
 142
 143 bool list_lru_del(struct list_lru *lru, struct list_head *item)
 144 {
 145         int nid = page_to_nid(virt_to_page(item));
 146         struct list_lru_node *nlru = &lru->node[nid];
 147         struct list_lru_one *l;
 148
 149         spin_lock(&nlru->lock);
 150         if (!list_empty(item)) {
 151                 l = list_lru_from_kmem(lru, nid, item, NULL);
 152                 list_del_init(item);
 153                 l->nr_items--;
 154                 nlru->nr_items--;
 155                 spin_unlock(&nlru->lock);
 156                 return true;
 157         }
 158         spin_unlock(&nlru->lock);
 159         return false;
 160 }
 161 EXPORT_SYMBOL_GPL(list_lru_del);
 162
 163 void list_lru_isolate(struct list_lru_one *list, struct list_head *item)
 164 {
 165         list_del_init(item);
 166         list->nr_items--;
 167 }
 168 EXPORT_SYMBOL_GPL(list_lru_isolate);
 169
 170 void list_lru_isolate_move(struct list_lru_one *list, struct list_head *item,
 171                            struct list_head *head)
 172 {
 173         list_move(item, head);
 174         list->nr_items--;
 175 }
 176 EXPORT_SYMBOL_GPL(list_lru_isolate_move);
 177
 178 unsigned long list_lru_count_one(struct list_lru *lru,
 179                                  int nid, struct mem_cgroup *memcg)
 180 {
 181         struct list_lru_one *l;
 182         long count;
 183
 184         rcu_read_lock();
 185         l = list_lru_from_memcg_idx(lru, nid, memcg_kmem_id(memcg));
 186         count = l ? READ_ONCE(l->nr_items) : 0;
 187         rcu_read_unlock();
 188
 189         if (unlikely(count < 0))
 190                 count = 0;
 191
 192         return count;
 193 }
 194 EXPORT_SYMBOL_GPL(list_lru_count_one);
 195
 196 unsigned long list_lru_count_node(struct list_lru *lru, int nid)
 197 {
 198         struct list_lru_node *nlru;
 199
 200         nlru = &lru->node[nid];
 201         return nlru->nr_items;
 202 }
 203 EXPORT_SYMBOL_GPL(list_lru_count_node);
 204
 205 static unsigned long
 206 __list_lru_walk_one(struct list_lru *lru, int nid, int memcg_idx,
 207                     list_lru_walk_cb isolate, void *cb_arg,
 208                     unsigned long *nr_to_walk)
 209 {
 210         struct list_lru_node *nlru = &lru->node[nid];
 211         struct list_lru_one *l;
 212         struct list_head *item, *n;
 213         unsigned long isolated = 0;
 214
 215 restart:
 216         l = list_lru_from_memcg_idx(lru, nid, memcg_idx);
 217         if (!l)
 218                 goto out;
 219
 220         list_for_each_safe(item, n, &l->list) {
 221                 enum lru_status ret;
 222
 223                 /*
 224                  * decrement nr_to_walk first so that we don't livelock if we
 225                  * get stuck on large numbers of LRU_RETRY items
 226                  */
 227                 if (!*nr_to_walk)
 228                         break;
 229                 --*nr_to_walk;
 230
 231                 ret = isolate(item, l, &nlru->lock, cb_arg);
 232                 switch (ret) {
 233                 case LRU_REMOVED_RETRY:
 234                         assert_spin_locked(&nlru->lock);
 235                         fallthrough;
 236                 case LRU_REMOVED:
 237                         isolated++;
 238                         nlru->nr_items--;
 239                         /*
 240                          * If the lru lock has been dropped, our list
 241                          * traversal is now invalid and so we have to
 242                          * restart from scratch.
 243                          */
 244                         if (ret == LRU_REMOVED_RETRY)
 245                                 goto restart;
 246                         break;
 247                 case LRU_ROTATE:
 248                         list_move_tail(item, &l->list);
 249                         break;
 250                 case LRU_SKIP:
 251                         break;
 252                 case LRU_RETRY:
 253                         /*
 254                          * The lru lock has been dropped, our list traversal is
 255                          * now invalid and so we have to restart from scratch.
 256                          */
 257                         assert_spin_locked(&nlru->lock);
 258                         goto restart;
 259                 default:
 260                         BUG();
 261                 }
 262         }
 263 out:
 264         return isolated;
 265 }
 266
 267 unsigned long
 268 list_lru_walk_one(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
 269                   list_lru_walk_cb isolate, void *cb_arg,
 270                   unsigned long *nr_to_walk)
 271 {
 272         struct list_lru_node *nlru = &lru->node[nid];
 273         unsigned long ret;
 274
 275         spin_lock(&nlru->lock);
 276         ret = __list_lru_walk_one(lru, nid, memcg_kmem_id(memcg), isolate,
 277                                   cb_arg, nr_to_walk);
 278         spin_unlock(&nlru->lock);
 279         return ret;
 280 }
 281 EXPORT_SYMBOL_GPL(list_lru_walk_one);
 282
 283 unsigned long
 284 list_lru_walk_one_irq(struct list_lru *lru, int nid, struct mem_cgroup *memcg,
 285                       list_lru_walk_cb isolate, void *cb_arg,
 286                       unsigned long *nr_to_walk)
 287 {
 288         struct list_lru_node *nlru = &lru->node[nid];
 289         unsigned long ret;
 290
 291         spin_lock_irq(&nlru->lock);
 292         ret = __list_lru_walk_one(lru, nid, memcg_kmem_id(memcg), isolate,
 293                                   cb_arg, nr_to_walk);
 294         spin_unlock_irq(&nlru->lock);
 295         return ret;
 296 }
 297
 298 unsigned long list_lru_walk_node(struct list_lru *lru, int nid,
 299                                  list_lru_walk_cb isolate, void *cb_arg,
 300                                  unsigned long *nr_to_walk)
 301 {
 302         long isolated = 0;
 303
 304         isolated += list_lru_walk_one(lru, nid, NULL, isolate, cb_arg,
 305                                       nr_to_walk);
 306
 307 #ifdef CONFIG_MEMCG_KMEM
 308         if (*nr_to_walk > 0 && list_lru_memcg_aware(lru)) {
 309                 struct list_lru_memcg *mlru;
 310                 unsigned long index;
 311
 312                 xa_for_each(&lru->xa, index, mlru) {
 313                         struct list_lru_node *nlru = &lru->node[nid];
 314
 315                         spin_lock(&nlru->lock);
 316                         isolated += __list_lru_walk_one(lru, nid, index,
 317                                                         isolate, cb_arg,
 318                                                         nr_to_walk);
 319                         spin_unlock(&nlru->lock);
 320
 321                         if (*nr_to_walk <= 0)
 322                                 break;
 323                 }
 324         }
 325 #endif
 326
 327         return isolated;
 328 }
 329 EXPORT_SYMBOL_GPL(list_lru_walk_node);
 330
 331 static void init_one_lru(struct list_lru_one *l)
 332 {
 333         INIT_LIST_HEAD(&l->list);
 334         l->nr_items = 0;
 335 }
 336
 337 #ifdef CONFIG_MEMCG_KMEM
 338 static struct list_lru_memcg *memcg_init_list_lru_one(gfp_t gfp)
 339 {
 340         int nid;
 341         struct list_lru_memcg *mlru;
 342
 343         mlru = kmalloc(struct_size(mlru, node, nr_node_ids), gfp);
 344         if (!mlru)
 345                 return NULL;
 346
 347         for_each_node(nid)
 348                 init_one_lru(&mlru->node[nid]);
 349
 350         return mlru;
 351 }
 352
 353 static void memcg_list_lru_free(struct list_lru *lru, int src_idx)
 354 {
 355         struct list_lru_memcg *mlru = xa_erase_irq(&lru->xa, src_idx);
 356
 357         /*
 358          * The __list_lru_walk_one() can walk the list of this node.
 359          * We need kvfree_rcu() here. And the walking of the list
 360          * is under lru->node[nid]->lock, which can serve as a RCU
 361          * read-side critical section.
 362          */
 363         if (mlru)
 364                 kvfree_rcu(mlru, rcu);
 365 }
 366
 367 static inline void memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
 368 {
 369         if (memcg_aware)
 370                 xa_init_flags(&lru->xa, XA_FLAGS_LOCK_IRQ);
 371         lru->memcg_aware = memcg_aware;
 372 }
 373
 374 static void memcg_destroy_list_lru(struct list_lru *lru)
 375 {
 376         XA_STATE(xas, &lru->xa, 0);
 377         struct list_lru_memcg *mlru;
 378
 379         if (!list_lru_memcg_aware(lru))
 380                 return;
 381
 382         xas_lock_irq(&xas);
 383         xas_for_each(&xas, mlru, ULONG_MAX) {
 384                 kfree(mlru);
 385                 xas_store(&xas, NULL);
 386         }
 387         xas_unlock_irq(&xas);
 388 }
 389
 390 static void memcg_reparent_list_lru_node(struct list_lru *lru, int nid,
 391                                          int src_idx, struct mem_cgroup *dst_memcg)
 392 {
 393         struct list_lru_node *nlru = &lru->node[nid];
 394         int dst_idx = dst_memcg->kmemcg_id;
 395         struct list_lru_one *src, *dst;
 396
 397         /*
 398          * Since list_lru_{add,del} may be called under an IRQ-safe lock,
 399          * we have to use IRQ-safe primitives here to avoid deadlock.
 400          */
 401         spin_lock_irq(&nlru->lock);
 402
 403         src = list_lru_from_memcg_idx(lru, nid, src_idx);
 404         if (!src)
 405                 goto out;
 406         dst = list_lru_from_memcg_idx(lru, nid, dst_idx);
 407
 408         list_splice_init(&src->list, &dst->list);
 409
 410         if (src->nr_items) {
 411                 dst->nr_items += src->nr_items;
 412                 set_shrinker_bit(dst_memcg, nid, lru_shrinker_id(lru));
 413                 src->nr_items = 0;
 414         }
 415 out:
 416         spin_unlock_irq(&nlru->lock);
 417 }
 418
 419 static void memcg_reparent_list_lru(struct list_lru *lru,
 420                                     int src_idx, struct mem_cgroup *dst_memcg)
 421 {
 422         int i;
 423
 424         for_each_node(i)
 425                 memcg_reparent_list_lru_node(lru, i, src_idx, dst_memcg);
 426
 427         memcg_list_lru_free(lru, src_idx);
 428 }
 429
 430 void memcg_reparent_list_lrus(struct mem_cgroup *memcg, struct mem_cgroup *parent)
 431 {
 432         struct cgroup_subsys_state *css;
 433         struct list_lru *lru;
 434         int src_idx = memcg->kmemcg_id;
 435
 436         /*
 437          * Change kmemcg_id of this cgroup and all its descendants to the
 438          * parent's id, and then move all entries from this cgroup's list_lrus
 439          * to ones of the parent.
 440          *
 441          * After we have finished, all list_lrus corresponding to this cgroup
 442          * are guaranteed to remain empty. So we can safely free this cgroup's
 443          * list lrus in memcg_list_lru_free().
 444          *
 445          * Changing ->kmemcg_id to the parent can prevent memcg_list_lru_alloc()
 446          * from allocating list lrus for this cgroup after memcg_list_lru_free()
 447          * call.
 448          */
 449         rcu_read_lock();
 450         css_for_each_descendant_pre(css, &memcg->css) {
 451                 struct mem_cgroup *child;
 452
 453                 child = mem_cgroup_from_css(css);
 454                 WRITE_ONCE(child->kmemcg_id, parent->kmemcg_id);
 455         }
 456         rcu_read_unlock();
 457
 458         mutex_lock(&list_lrus_mutex);
 459         list_for_each_entry(lru, &memcg_list_lrus, list)
 460                 memcg_reparent_list_lru(lru, src_idx, parent);
 461         mutex_unlock(&list_lrus_mutex);
 462 }
 463
 464 static inline bool memcg_list_lru_allocated(struct mem_cgroup *memcg,
 465                                             struct list_lru *lru)
 466 {
 467         int idx = memcg->kmemcg_id;
 468
 469         return idx < 0 || xa_load(&lru->xa, idx);
 470 }
 471
 472 int memcg_list_lru_alloc(struct mem_cgroup *memcg, struct list_lru *lru,
 473                          gfp_t gfp)
 474 {
 475         int i;
 476         unsigned long flags;
 477         struct list_lru_memcg_table {
 478                 struct list_lru_memcg *mlru;
 479                 struct mem_cgroup *memcg;
 480         } *table;
 481         XA_STATE(xas, &lru->xa, 0);
 482
 483         if (!list_lru_memcg_aware(lru) || memcg_list_lru_allocated(memcg, lru))
 484                 return 0;
 485
 486         gfp &= GFP_RECLAIM_MASK;
 487         table = kmalloc_array(memcg->css.cgroup->level, sizeof(*table), gfp);
 488         if (!table)
 489                 return -ENOMEM;
 490
 491         /*
 492          * Because the list_lru can be reparented to the parent cgroup's
 493          * list_lru, we should make sure that this cgroup and all its
 494          * ancestors have allocated list_lru_memcg.
 495          */
 496         for (i = 0; memcg; memcg = parent_mem_cgroup(memcg), i++) {
 497                 if (memcg_list_lru_allocated(memcg, lru))
 498                         break;
 499
 500                 table[i].memcg = memcg;
 501                 table[i].mlru = memcg_init_list_lru_one(gfp);
 502                 if (!table[i].mlru) {
 503                         while (i--)
 504                                 kfree(table[i].mlru);
 505                         kfree(table);
 506                         return -ENOMEM;
 507                 }
 508         }
 509
 510         xas_lock_irqsave(&xas, flags);
 511         while (i--) {
 512                 int index = READ_ONCE(table[i].memcg->kmemcg_id);
 513                 struct list_lru_memcg *mlru = table[i].mlru;
 514
 515                 xas_set(&xas, index);
 516 retry:
 517                 if (unlikely(index < 0 || xas_error(&xas) || xas_load(&xas))) {
 518                         kfree(mlru);
 519                 } else {
 520                         xas_store(&xas, mlru);
 521                         if (xas_error(&xas) == -ENOMEM) {
 522                                 xas_unlock_irqrestore(&xas, flags);
 523                                 if (xas_nomem(&xas, gfp))
 524                                         xas_set_err(&xas, 0);
 525                                 xas_lock_irqsave(&xas, flags);
 526                                 /*
 527                                  * The xas lock has been released, this memcg
 528                                  * can be reparented before us. So reload
 529                                  * memcg id. More details see the comments
 530                                  * in memcg_reparent_list_lrus().
 531                                  */
 532                                 index = READ_ONCE(table[i].memcg->kmemcg_id);
 533                                 if (index < 0)
 534                                         xas_set_err(&xas, 0);
 535                                 else if (!xas_error(&xas) && index != xas.xa_index)
 536                                         xas_set(&xas, index);
 537                                 goto retry;
 538                         }
 539                 }
 540         }
 541         /* xas_nomem() is used to free memory instead of memory allocation. */
 542         if (xas.xa_alloc)
 543                 xas_nomem(&xas, gfp);
 544         xas_unlock_irqrestore(&xas, flags);
 545         kfree(table);
 546
 547         return xas_error(&xas);
 548 }
 549 #else
 550 static inline void memcg_init_list_lru(struct list_lru *lru, bool memcg_aware)
 551 {
 552 }
 553
 554 static void memcg_destroy_list_lru(struct list_lru *lru)
 555 {
 556 }
 557 #endif /* CONFIG_MEMCG_KMEM */
 558
 559 int __list_lru_init(struct list_lru *lru, bool memcg_aware,
 560                     struct lock_class_key *key, struct shrinker *shrinker)
 561 {
 562         int i;
 563
 564 #ifdef CONFIG_MEMCG_KMEM
 565         if (shrinker)
 566                 lru->shrinker_id = shrinker->id;
 567         else
 568                 lru->shrinker_id = -1;
 569 #endif
 570
 571         lru->node = kcalloc(nr_node_ids, sizeof(*lru->node), GFP_KERNEL);
 572         if (!lru->node)
 573                 return -ENOMEM;
 574
 575         for_each_node(i) {
 576                 spin_lock_init(&lru->node[i].lock);
 577                 if (key)
 578                         lockdep_set_class(&lru->node[i].lock, key);
 579                 init_one_lru(&lru->node[i].lru);
 580         }
 581
 582         memcg_init_list_lru(lru, memcg_aware);
 583         list_lru_register(lru);
 584
 585         return 0;
 586 }
 587 EXPORT_SYMBOL_GPL(__list_lru_init);
 588
 589 void list_lru_destroy(struct list_lru *lru)
 590 {
 591         /* Already destroyed or not yet initialized? */
 592         if (!lru->node)
 593                 return;
 594
 595         list_lru_unregister(lru);
 596
 597         memcg_destroy_list_lru(lru);
 598         kfree(lru->node);
 599         lru->node = NULL;
 600
 601 #ifdef CONFIG_MEMCG_KMEM
 602         lru->shrinker_id = -1;
 603 #endif
 604 }
 605 EXPORT_SYMBOL_GPL(list_lru_destroy);