mm/zswap.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * zswap.c - zswap driver file
   4  *
   5  * zswap is a backend for frontswap that takes pages that are in the process
   6  * of being swapped out and attempts to compress and store them in a
   7  * RAM-based memory pool.  This can result in a significant I/O reduction on
   8  * the swap device and, in the case where decompressing from RAM is faster
   9  * than reading from the swap device, can also improve workload performance.
  10  *
  11  * Copyright (C) 2012  Seth Jennings <sjenning@linux.vnet.ibm.com>
  12 */
  13
  14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  15
  16 #include <linux/module.h>
  17 #include <linux/cpu.h>
  18 #include <linux/highmem.h>
  19 #include <linux/slab.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/types.h>
  22 #include <linux/atomic.h>
  23 #include <linux/frontswap.h>
  24 #include <linux/rbtree.h>
  25 #include <linux/swap.h>
  26 #include <linux/crypto.h>
  27 #include <linux/mempool.h>
  28 #include <linux/zpool.h>
  29
  30 #include <linux/mm_types.h>
  31 #include <linux/page-flags.h>
  32 #include <linux/swapops.h>
  33 #include <linux/writeback.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/workqueue.h>
  36
  37 /*********************************
  38 * statistics
  39 **********************************/
  40 /* Total bytes used by the compressed storage */
  41 static u64 zswap_pool_total_size;
  42 /* The number of compressed pages currently stored in zswap */
  43 static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
  44 /* The number of same-value filled pages currently stored in zswap */
  45 static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
  46
  47 /*
  48  * The statistics below are not protected from concurrent access for
  49  * performance reasons so they may not be a 100% accurate.  However,
  50  * they do provide useful information on roughly how many times a
  51  * certain event is occurring.
  52 */
  53
  54 /* Pool limit was hit (see zswap_max_pool_percent) */
  55 static u64 zswap_pool_limit_hit;
  56 /* Pages written back when pool limit was reached */
  57 static u64 zswap_written_back_pages;
  58 /* Store failed due to a reclaim failure after pool limit was reached */
  59 static u64 zswap_reject_reclaim_fail;
  60 /* Compressed page was too big for the allocator to (optimally) store */
  61 static u64 zswap_reject_compress_poor;
  62 /* Store failed because underlying allocator could not get memory */
  63 static u64 zswap_reject_alloc_fail;
  64 /* Store failed because the entry metadata could not be allocated (rare) */
  65 static u64 zswap_reject_kmemcache_fail;
  66 /* Duplicate store was encountered (rare) */
  67 static u64 zswap_duplicate_entry;
  68
  69 /* Shrinker work queue */
  70 static struct workqueue_struct *shrink_wq;
  71 /* Pool limit was hit, we need to calm down */
  72 static bool zswap_pool_reached_full;
  73
  74 /*********************************
  75 * tunables
  76 **********************************/
  77
  78 #define ZSWAP_PARAM_UNSET ""
  79
  80 /* Enable/disable zswap */
  81 static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
  82 static int zswap_enabled_param_set(const char *,
  83                                    const struct kernel_param *);
  84 static struct kernel_param_ops zswap_enabled_param_ops = {
  85         .set =          zswap_enabled_param_set,
  86         .get =          param_get_bool,
  87 };
  88 module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
  89
  90 /* Crypto compressor to use */
  91 static char *zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
  92 static int zswap_compressor_param_set(const char *,
  93                                       const struct kernel_param *);
  94 static struct kernel_param_ops zswap_compressor_param_ops = {
  95         .set =          zswap_compressor_param_set,
  96         .get =          param_get_charp,
  97         .free =         param_free_charp,
  98 };
  99 module_param_cb(compressor, &zswap_compressor_param_ops,
 100                 &zswap_compressor, 0644);
 101
 102 /* Compressed storage zpool to use */
 103 static char *zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
 104 static int zswap_zpool_param_set(const char *, const struct kernel_param *);
 105 static struct kernel_param_ops zswap_zpool_param_ops = {
 106         .set =          zswap_zpool_param_set,
 107         .get =          param_get_charp,
 108         .free =         param_free_charp,
 109 };
 110 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
 111
 112 /* The maximum percentage of memory that the compressed pool can occupy */
 113 static unsigned int zswap_max_pool_percent = 20;
 114 module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
 115
 116 /* The threshold for accepting new pages after the max_pool_percent was hit */
 117 static unsigned int zswap_accept_thr_percent = 90; /* of max pool size */
 118 module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
 119                    uint, 0644);
 120
 121 /* Enable/disable handling same-value filled pages (enabled by default) */
 122 static bool zswap_same_filled_pages_enabled = true;
 123 module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
 124                    bool, 0644);
 125
 126 /*********************************
 127 * data structures
 128 **********************************/
 129
 130 struct zswap_pool {
 131         struct zpool *zpool;
 132         struct crypto_comp * __percpu *tfm;
 133         struct kref kref;
 134         struct list_head list;
 135         struct work_struct release_work;
 136         struct work_struct shrink_work;
 137         struct hlist_node node;
 138         char tfm_name[CRYPTO_MAX_ALG_NAME];
 139 };
 140
 141 /*
 142  * struct zswap_entry
 143  *
 144  * This structure contains the metadata for tracking a single compressed
 145  * page within zswap.
 146  *
 147  * rbnode - links the entry into red-black tree for the appropriate swap type
 148  * offset - the swap offset for the entry.  Index into the red-black tree.
 149  * refcount - the number of outstanding reference to the entry. This is needed
 150  *            to protect against premature freeing of the entry by code
 151  *            concurrent calls to load, invalidate, and writeback.  The lock
 152  *            for the zswap_tree structure that contains the entry must
 153  *            be held while changing the refcount.  Since the lock must
 154  *            be held, there is no reason to also make refcount atomic.
 155  * length - the length in bytes of the compressed page data.  Needed during
 156  *          decompression. For a same value filled page length is 0.
 157  * pool - the zswap_pool the entry's data is in
 158  * handle - zpool allocation handle that stores the compressed page data
 159  * value - value of the same-value filled pages which have same content
 160  */
 161 struct zswap_entry {
 162         struct rb_node rbnode;
 163         pgoff_t offset;
 164         int refcount;
 165         unsigned int length;
 166         struct zswap_pool *pool;
 167         union {
 168                 unsigned long handle;
 169                 unsigned long value;
 170         };
 171 };
 172
 173 struct zswap_header {
 174         swp_entry_t swpentry;
 175 };
 176
 177 /*
 178  * The tree lock in the zswap_tree struct protects a few things:
 179  * - the rbtree
 180  * - the refcount field of each entry in the tree
 181  */
 182 struct zswap_tree {
 183         struct rb_root rbroot;
 184         spinlock_t lock;
 185 };
 186
 187 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
 188
 189 /* RCU-protected iteration */
 190 static LIST_HEAD(zswap_pools);
 191 /* protects zswap_pools list modification */
 192 static DEFINE_SPINLOCK(zswap_pools_lock);
 193 /* pool counter to provide unique names to zpool */
 194 static atomic_t zswap_pools_count = ATOMIC_INIT(0);
 195
 196 /* used by param callback function */
 197 static bool zswap_init_started;
 198
 199 /* fatal error during init */
 200 static bool zswap_init_failed;
 201
 202 /* init completed, but couldn't create the initial pool */
 203 static bool zswap_has_pool;
 204
 205 /*********************************
 206 * helpers and fwd declarations
 207 **********************************/
 208
 209 #define zswap_pool_debug(msg, p)                                \
 210         pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name,         \
 211                  zpool_get_type((p)->zpool))
 212
 213 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
 214 static int zswap_pool_get(struct zswap_pool *pool);
 215 static void zswap_pool_put(struct zswap_pool *pool);
 216
 217 static const struct zpool_ops zswap_zpool_ops = {
 218         .evict = zswap_writeback_entry
 219 };
 220
 221 static bool zswap_is_full(void)
 222 {
 223         return totalram_pages() * zswap_max_pool_percent / 100 <
 224                         DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
 225 }
 226
 227 static bool zswap_can_accept(void)
 228 {
 229         return totalram_pages() * zswap_accept_thr_percent / 100 *
 230                                 zswap_max_pool_percent / 100 >
 231                         DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
 232 }
 233
 234 static void zswap_update_total_size(void)
 235 {
 236         struct zswap_pool *pool;
 237         u64 total = 0;
 238
 239         rcu_read_lock();
 240
 241         list_for_each_entry_rcu(pool, &zswap_pools, list)
 242                 total += zpool_get_total_size(pool->zpool);
 243
 244         rcu_read_unlock();
 245
 246         zswap_pool_total_size = total;
 247 }
 248
 249 /*********************************
 250 * zswap entry functions
 251 **********************************/
 252 static struct kmem_cache *zswap_entry_cache;
 253
 254 static int __init zswap_entry_cache_create(void)
 255 {
 256         zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
 257         return zswap_entry_cache == NULL;
 258 }
 259
 260 static void __init zswap_entry_cache_destroy(void)
 261 {
 262         kmem_cache_destroy(zswap_entry_cache);
 263 }
 264
 265 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
 266 {
 267         struct zswap_entry *entry;
 268         entry = kmem_cache_alloc(zswap_entry_cache, gfp);
 269         if (!entry)
 270                 return NULL;
 271         entry->refcount = 1;
 272         RB_CLEAR_NODE(&entry->rbnode);
 273         return entry;
 274 }
 275
 276 static void zswap_entry_cache_free(struct zswap_entry *entry)
 277 {
 278         kmem_cache_free(zswap_entry_cache, entry);
 279 }
 280
 281 /*********************************
 282 * rbtree functions
 283 **********************************/
 284 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
 285 {
 286         struct rb_node *node = root->rb_node;
 287         struct zswap_entry *entry;
 288
 289         while (node) {
 290                 entry = rb_entry(node, struct zswap_entry, rbnode);
 291                 if (entry->offset > offset)
 292                         node = node->rb_left;
 293                 else if (entry->offset < offset)
 294                         node = node->rb_right;
 295                 else
 296                         return entry;
 297         }
 298         return NULL;
 299 }
 300
 301 /*
 302  * In the case that a entry with the same offset is found, a pointer to
 303  * the existing entry is stored in dupentry and the function returns -EEXIST
 304  */
 305 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
 306                         struct zswap_entry **dupentry)
 307 {
 308         struct rb_node **link = &root->rb_node, *parent = NULL;
 309         struct zswap_entry *myentry;
 310
 311         while (*link) {
 312                 parent = *link;
 313                 myentry = rb_entry(parent, struct zswap_entry, rbnode);
 314                 if (myentry->offset > entry->offset)
 315                         link = &(*link)->rb_left;
 316                 else if (myentry->offset < entry->offset)
 317                         link = &(*link)->rb_right;
 318                 else {
 319                         *dupentry = myentry;
 320                         return -EEXIST;
 321                 }
 322         }
 323         rb_link_node(&entry->rbnode, parent, link);
 324         rb_insert_color(&entry->rbnode, root);
 325         return 0;
 326 }
 327
 328 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
 329 {
 330         if (!RB_EMPTY_NODE(&entry->rbnode)) {
 331                 rb_erase(&entry->rbnode, root);
 332                 RB_CLEAR_NODE(&entry->rbnode);
 333         }
 334 }
 335
 336 /*
 337  * Carries out the common pattern of freeing and entry's zpool allocation,
 338  * freeing the entry itself, and decrementing the number of stored pages.
 339  */
 340 static void zswap_free_entry(struct zswap_entry *entry)
 341 {
 342         if (!entry->length)
 343                 atomic_dec(&zswap_same_filled_pages);
 344         else {
 345                 zpool_free(entry->pool->zpool, entry->handle);
 346                 zswap_pool_put(entry->pool);
 347         }
 348         zswap_entry_cache_free(entry);
 349         atomic_dec(&zswap_stored_pages);
 350         zswap_update_total_size();
 351 }
 352
 353 /* caller must hold the tree lock */
 354 static void zswap_entry_get(struct zswap_entry *entry)
 355 {
 356         entry->refcount++;
 357 }
 358
 359 /* caller must hold the tree lock
 360 * remove from the tree and free it, if nobody reference the entry
 361 */
 362 static void zswap_entry_put(struct zswap_tree *tree,
 363                         struct zswap_entry *entry)
 364 {
 365         int refcount = --entry->refcount;
 366
 367         BUG_ON(refcount < 0);
 368         if (refcount == 0) {
 369                 zswap_rb_erase(&tree->rbroot, entry);
 370                 zswap_free_entry(entry);
 371         }
 372 }
 373
 374 /* caller must hold the tree lock */
 375 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
 376                                 pgoff_t offset)
 377 {
 378         struct zswap_entry *entry;
 379
 380         entry = zswap_rb_search(root, offset);
 381         if (entry)
 382                 zswap_entry_get(entry);
 383
 384         return entry;
 385 }
 386
 387 /*********************************
 388 * per-cpu code
 389 **********************************/
 390 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
 391
 392 static int zswap_dstmem_prepare(unsigned int cpu)
 393 {
 394         u8 *dst;
 395
 396         dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
 397         if (!dst)
 398                 return -ENOMEM;
 399
 400         per_cpu(zswap_dstmem, cpu) = dst;
 401         return 0;
 402 }
 403
 404 static int zswap_dstmem_dead(unsigned int cpu)
 405 {
 406         u8 *dst;
 407
 408         dst = per_cpu(zswap_dstmem, cpu);
 409         kfree(dst);
 410         per_cpu(zswap_dstmem, cpu) = NULL;
 411
 412         return 0;
 413 }
 414
 415 static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
 416 {
 417         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
 418         struct crypto_comp *tfm;
 419
 420         if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu)))
 421                 return 0;
 422
 423         tfm = crypto_alloc_comp(pool->tfm_name, 0, 0);
 424         if (IS_ERR_OR_NULL(tfm)) {
 425                 pr_err("could not alloc crypto comp %s : %ld\n",
 426                        pool->tfm_name, PTR_ERR(tfm));
 427                 return -ENOMEM;
 428         }
 429         *per_cpu_ptr(pool->tfm, cpu) = tfm;
 430         return 0;
 431 }
 432
 433 static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
 434 {
 435         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
 436         struct crypto_comp *tfm;
 437
 438         tfm = *per_cpu_ptr(pool->tfm, cpu);
 439         if (!IS_ERR_OR_NULL(tfm))
 440                 crypto_free_comp(tfm);
 441         *per_cpu_ptr(pool->tfm, cpu) = NULL;
 442         return 0;
 443 }
 444
 445 /*********************************
 446 * pool functions
 447 **********************************/
 448
 449 static struct zswap_pool *__zswap_pool_current(void)
 450 {
 451         struct zswap_pool *pool;
 452
 453         pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
 454         WARN_ONCE(!pool && zswap_has_pool,
 455                   "%s: no page storage pool!\n", __func__);
 456
 457         return pool;
 458 }
 459
 460 static struct zswap_pool *zswap_pool_current(void)
 461 {
 462         assert_spin_locked(&zswap_pools_lock);
 463
 464         return __zswap_pool_current();
 465 }
 466
 467 static struct zswap_pool *zswap_pool_current_get(void)
 468 {
 469         struct zswap_pool *pool;
 470
 471         rcu_read_lock();
 472
 473         pool = __zswap_pool_current();
 474         if (!zswap_pool_get(pool))
 475                 pool = NULL;
 476
 477         rcu_read_unlock();
 478
 479         return pool;
 480 }
 481
 482 static struct zswap_pool *zswap_pool_last_get(void)
 483 {
 484         struct zswap_pool *pool, *last = NULL;
 485
 486         rcu_read_lock();
 487
 488         list_for_each_entry_rcu(pool, &zswap_pools, list)
 489                 last = pool;
 490         WARN_ONCE(!last && zswap_has_pool,
 491                   "%s: no page storage pool!\n", __func__);
 492         if (!zswap_pool_get(last))
 493                 last = NULL;
 494
 495         rcu_read_unlock();
 496
 497         return last;
 498 }
 499
 500 /* type and compressor must be null-terminated */
 501 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
 502 {
 503         struct zswap_pool *pool;
 504
 505         assert_spin_locked(&zswap_pools_lock);
 506
 507         list_for_each_entry_rcu(pool, &zswap_pools, list) {
 508                 if (strcmp(pool->tfm_name, compressor))
 509                         continue;
 510                 if (strcmp(zpool_get_type(pool->zpool), type))
 511                         continue;
 512                 /* if we can't get it, it's about to be destroyed */
 513                 if (!zswap_pool_get(pool))
 514                         continue;
 515                 return pool;
 516         }
 517
 518         return NULL;
 519 }
 520
 521 static void shrink_worker(struct work_struct *w)
 522 {
 523         struct zswap_pool *pool = container_of(w, typeof(*pool),
 524                                                 shrink_work);
 525
 526         if (zpool_shrink(pool->zpool, 1, NULL))
 527                 zswap_reject_reclaim_fail++;
 528         zswap_pool_put(pool);
 529 }
 530
 531 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
 532 {
 533         struct zswap_pool *pool;
 534         char name[38]; /* 'zswap' + 32 char (max) num + \0 */
 535         gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
 536         int ret;
 537
 538         if (!zswap_has_pool) {
 539                 /* if either are unset, pool initialization failed, and we
 540                  * need both params to be set correctly before trying to
 541                  * create a pool.
 542                  */
 543                 if (!strcmp(type, ZSWAP_PARAM_UNSET))
 544                         return NULL;
 545                 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
 546                         return NULL;
 547         }
 548
 549         pool = kzalloc(sizeof(*pool), GFP_KERNEL);
 550         if (!pool)
 551                 return NULL;
 552
 553         /* unique name for each pool specifically required by zsmalloc */
 554         snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
 555
 556         pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
 557         if (!pool->zpool) {
 558                 pr_err("%s zpool not available\n", type);
 559                 goto error;
 560         }
 561         pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
 562
 563         strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
 564         pool->tfm = alloc_percpu(struct crypto_comp *);
 565         if (!pool->tfm) {
 566                 pr_err("percpu alloc failed\n");
 567                 goto error;
 568         }
 569
 570         ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
 571                                        &pool->node);
 572         if (ret)
 573                 goto error;
 574         pr_debug("using %s compressor\n", pool->tfm_name);
 575
 576         /* being the current pool takes 1 ref; this func expects the
 577          * caller to always add the new pool as the current pool
 578          */
 579         kref_init(&pool->kref);
 580         INIT_LIST_HEAD(&pool->list);
 581         INIT_WORK(&pool->shrink_work, shrink_worker);
 582
 583         zswap_pool_debug("created", pool);
 584
 585         return pool;
 586
 587 error:
 588         free_percpu(pool->tfm);
 589         if (pool->zpool)
 590                 zpool_destroy_pool(pool->zpool);
 591         kfree(pool);
 592         return NULL;
 593 }
 594
 595 static bool zswap_try_pool_create(void)
 596 {
 597         struct zswap_pool *pool;
 598         bool has_comp, has_zpool;
 599
 600         has_comp = crypto_has_comp(zswap_compressor, 0, 0);
 601         if (!has_comp && strcmp(zswap_compressor,
 602                                 CONFIG_ZSWAP_COMPRESSOR_DEFAULT)) {
 603                 pr_err("compressor %s not available, using default %s\n",
 604                        zswap_compressor, CONFIG_ZSWAP_COMPRESSOR_DEFAULT);
 605                 param_free_charp(&zswap_compressor);
 606                 zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
 607                 has_comp = crypto_has_comp(zswap_compressor, 0, 0);
 608         }
 609         if (!has_comp) {
 610                 pr_err("default compressor %s not available\n",
 611                        zswap_compressor);
 612                 param_free_charp(&zswap_compressor);
 613                 zswap_compressor = ZSWAP_PARAM_UNSET;
 614         }
 615
 616         has_zpool = zpool_has_pool(zswap_zpool_type);
 617         if (!has_zpool && strcmp(zswap_zpool_type,
 618                                  CONFIG_ZSWAP_ZPOOL_DEFAULT)) {
 619                 pr_err("zpool %s not available, using default %s\n",
 620                        zswap_zpool_type, CONFIG_ZSWAP_ZPOOL_DEFAULT);
 621                 param_free_charp(&zswap_zpool_type);
 622                 zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
 623                 has_zpool = zpool_has_pool(zswap_zpool_type);
 624         }
 625         if (!has_zpool) {
 626                 pr_err("default zpool %s not available\n",
 627                        zswap_zpool_type);
 628                 param_free_charp(&zswap_zpool_type);
 629                 zswap_zpool_type = ZSWAP_PARAM_UNSET;
 630         }
 631
 632         if (!has_comp || !has_zpool)
 633                 return false;
 634
 635         pool = zswap_pool_create(zswap_zpool_type, zswap_compressor);
 636
 637         if (pool) {
 638                 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
 639                         zpool_get_type(pool->zpool));
 640                 list_add(&pool->list, &zswap_pools);
 641                 zswap_has_pool = true;
 642         } else {
 643                 pr_err("pool creation failed\n");
 644                 zswap_enabled = false;
 645         }
 646
 647         return zswap_enabled;
 648 }
 649
 650 static void zswap_pool_destroy(struct zswap_pool *pool)
 651 {
 652         zswap_pool_debug("destroying", pool);
 653
 654         cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
 655         free_percpu(pool->tfm);
 656         zpool_destroy_pool(pool->zpool);
 657         kfree(pool);
 658 }
 659
 660 static int __must_check zswap_pool_get(struct zswap_pool *pool)
 661 {
 662         if (!pool)
 663                 return 0;
 664
 665         return kref_get_unless_zero(&pool->kref);
 666 }
 667
 668 static void __zswap_pool_release(struct work_struct *work)
 669 {
 670         struct zswap_pool *pool = container_of(work, typeof(*pool),
 671                                                 release_work);
 672
 673         synchronize_rcu();
 674
 675         /* nobody should have been able to get a kref... */
 676         WARN_ON(kref_get_unless_zero(&pool->kref));
 677
 678         /* pool is now off zswap_pools list and has no references. */
 679         zswap_pool_destroy(pool);
 680 }
 681
 682 static void __zswap_pool_empty(struct kref *kref)
 683 {
 684         struct zswap_pool *pool;
 685
 686         pool = container_of(kref, typeof(*pool), kref);
 687
 688         spin_lock(&zswap_pools_lock);
 689
 690         WARN_ON(pool == zswap_pool_current());
 691
 692         list_del_rcu(&pool->list);
 693
 694         INIT_WORK(&pool->release_work, __zswap_pool_release);
 695         schedule_work(&pool->release_work);
 696
 697         spin_unlock(&zswap_pools_lock);
 698 }
 699
 700 static void zswap_pool_put(struct zswap_pool *pool)
 701 {
 702         kref_put(&pool->kref, __zswap_pool_empty);
 703 }
 704
 705 /*********************************
 706 * param callbacks
 707 **********************************/
 708
 709 /* val must be a null-terminated string */
 710 static int __zswap_param_set(const char *val, const struct kernel_param *kp,
 711                              char *type, char *compressor)
 712 {
 713         struct zswap_pool *pool, *put_pool = NULL;
 714         char *s = strstrip((char *)val);
 715         int ret;
 716
 717         if (zswap_init_failed) {
 718                 pr_err("can't set param, initialization failed\n");
 719                 return -ENODEV;
 720         }
 721
 722         /* no change required */
 723         if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
 724                 return 0;
 725
 726         /* if this is load-time (pre-init) param setting,
 727          * don't create a pool; that's done during init.
 728          */
 729         if (!zswap_init_started)
 730                 return param_set_charp(s, kp);
 731
 732         if (!type) {
 733                 if (!zpool_has_pool(s)) {
 734                         pr_err("zpool %s not available\n", s);
 735                         return -ENOENT;
 736                 }
 737                 type = s;
 738         } else if (!compressor) {
 739                 if (!crypto_has_comp(s, 0, 0)) {
 740                         pr_err("compressor %s not available\n", s);
 741                         return -ENOENT;
 742                 }
 743                 compressor = s;
 744         } else {
 745                 WARN_ON(1);
 746                 return -EINVAL;
 747         }
 748
 749         spin_lock(&zswap_pools_lock);
 750
 751         pool = zswap_pool_find_get(type, compressor);
 752         if (pool) {
 753                 zswap_pool_debug("using existing", pool);
 754                 WARN_ON(pool == zswap_pool_current());
 755                 list_del_rcu(&pool->list);
 756         }
 757
 758         spin_unlock(&zswap_pools_lock);
 759
 760         if (!pool)
 761                 pool = zswap_pool_create(type, compressor);
 762
 763         if (pool)
 764                 ret = param_set_charp(s, kp);
 765         else
 766                 ret = -EINVAL;
 767
 768         spin_lock(&zswap_pools_lock);
 769
 770         if (!ret) {
 771                 put_pool = zswap_pool_current();
 772                 list_add_rcu(&pool->list, &zswap_pools);
 773                 zswap_has_pool = true;
 774         } else if (pool) {
 775                 /* add the possibly pre-existing pool to the end of the pools
 776                  * list; if it's new (and empty) then it'll be removed and
 777                  * destroyed by the put after we drop the lock
 778                  */
 779                 list_add_tail_rcu(&pool->list, &zswap_pools);
 780                 put_pool = pool;
 781         }
 782
 783         spin_unlock(&zswap_pools_lock);
 784
 785         if (!zswap_has_pool && !pool) {
 786                 /* if initial pool creation failed, and this pool creation also
 787                  * failed, maybe both compressor and zpool params were bad.
 788                  * Allow changing this param, so pool creation will succeed
 789                  * when the other param is changed. We already verified this
 790                  * param is ok in the zpool_has_pool() or crypto_has_comp()
 791                  * checks above.
 792                  */
 793                 ret = param_set_charp(s, kp);
 794         }
 795
 796         /* drop the ref from either the old current pool,
 797          * or the new pool we failed to add
 798          */
 799         if (put_pool)
 800                 zswap_pool_put(put_pool);
 801
 802         return ret;
 803 }
 804
 805 static int zswap_compressor_param_set(const char *val,
 806                                       const struct kernel_param *kp)
 807 {
 808         return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
 809 }
 810
 811 static int zswap_zpool_param_set(const char *val,
 812                                  const struct kernel_param *kp)
 813 {
 814         return __zswap_param_set(val, kp, NULL, zswap_compressor);
 815 }
 816
 817 static int zswap_enabled_param_set(const char *val,
 818                                    const struct kernel_param *kp)
 819 {
 820         int ret;
 821
 822         if (zswap_init_failed) {
 823                 pr_err("can't enable, initialization failed\n");
 824                 return -ENODEV;
 825         }
 826
 827         ret = param_set_bool(val, kp);
 828         if (!ret && zswap_enabled && zswap_init_started && !zswap_has_pool)
 829                 if (!zswap_try_pool_create())
 830                         ret = -ENODEV;
 831
 832         return ret;
 833 }
 834
 835 /*********************************
 836 * writeback code
 837 **********************************/
 838 /* return enum for zswap_get_swap_cache_page */
 839 enum zswap_get_swap_ret {
 840         ZSWAP_SWAPCACHE_NEW,
 841         ZSWAP_SWAPCACHE_EXIST,
 842         ZSWAP_SWAPCACHE_FAIL,
 843 };
 844
 845 /*
 846  * zswap_get_swap_cache_page
 847  *
 848  * This is an adaption of read_swap_cache_async()
 849  *
 850  * This function tries to find a page with the given swap entry
 851  * in the swapper_space address space (the swap cache).  If the page
 852  * is found, it is returned in retpage.  Otherwise, a page is allocated,
 853  * added to the swap cache, and returned in retpage.
 854  *
 855  * If success, the swap cache page is returned in retpage
 856  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
 857  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
 858  *     the new page is added to swapcache and locked
 859  * Returns ZSWAP_SWAPCACHE_FAIL on error
 860  */
 861 static int zswap_get_swap_cache_page(swp_entry_t entry,
 862                                 struct page **retpage)
 863 {
 864         bool page_was_allocated;
 865
 866         *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
 867                         NULL, 0, &page_was_allocated);
 868         if (page_was_allocated)
 869                 return ZSWAP_SWAPCACHE_NEW;
 870         if (!*retpage)
 871                 return ZSWAP_SWAPCACHE_FAIL;
 872         return ZSWAP_SWAPCACHE_EXIST;
 873 }
 874
 875 /*
 876  * Attempts to free an entry by adding a page to the swap cache,
 877  * decompressing the entry data into the page, and issuing a
 878  * bio write to write the page back to the swap device.
 879  *
 880  * This can be thought of as a "resumed writeback" of the page
 881  * to the swap device.  We are basically resuming the same swap
 882  * writeback path that was intercepted with the frontswap_store()
 883  * in the first place.  After the page has been decompressed into
 884  * the swap cache, the compressed version stored by zswap can be
 885  * freed.
 886  */
 887 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
 888 {
 889         struct zswap_header *zhdr;
 890         swp_entry_t swpentry;
 891         struct zswap_tree *tree;
 892         pgoff_t offset;
 893         struct zswap_entry *entry;
 894         struct page *page;
 895         struct crypto_comp *tfm;
 896         u8 *src, *dst;
 897         unsigned int dlen;
 898         int ret;
 899         struct writeback_control wbc = {
 900                 .sync_mode = WB_SYNC_NONE,
 901         };
 902
 903         /* extract swpentry from data */
 904         zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
 905         swpentry = zhdr->swpentry; /* here */
 906         tree = zswap_trees[swp_type(swpentry)];
 907         offset = swp_offset(swpentry);
 908
 909         /* find and ref zswap entry */
 910         spin_lock(&tree->lock);
 911         entry = zswap_entry_find_get(&tree->rbroot, offset);
 912         if (!entry) {
 913                 /* entry was invalidated */
 914                 spin_unlock(&tree->lock);
 915                 zpool_unmap_handle(pool, handle);
 916                 return 0;
 917         }
 918         spin_unlock(&tree->lock);
 919         BUG_ON(offset != entry->offset);
 920
 921         /* try to allocate swap cache page */
 922         switch (zswap_get_swap_cache_page(swpentry, &page)) {
 923         case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
 924                 ret = -ENOMEM;
 925                 goto fail;
 926
 927         case ZSWAP_SWAPCACHE_EXIST:
 928                 /* page is already in the swap cache, ignore for now */
 929                 put_page(page);
 930                 ret = -EEXIST;
 931                 goto fail;
 932
 933         case ZSWAP_SWAPCACHE_NEW: /* page is locked */
 934                 /* decompress */
 935                 dlen = PAGE_SIZE;
 936                 src = (u8 *)zhdr + sizeof(struct zswap_header);
 937                 dst = kmap_atomic(page);
 938                 tfm = *get_cpu_ptr(entry->pool->tfm);
 939                 ret = crypto_comp_decompress(tfm, src, entry->length,
 940                                              dst, &dlen);
 941                 put_cpu_ptr(entry->pool->tfm);
 942                 kunmap_atomic(dst);
 943                 BUG_ON(ret);
 944                 BUG_ON(dlen != PAGE_SIZE);
 945
 946                 /* page is up to date */
 947                 SetPageUptodate(page);
 948         }
 949
 950         /* move it to the tail of the inactive list after end_writeback */
 951         SetPageReclaim(page);
 952
 953         /* start writeback */
 954         __swap_writepage(page, &wbc, end_swap_bio_write);
 955         put_page(page);
 956         zswap_written_back_pages++;
 957
 958         spin_lock(&tree->lock);
 959         /* drop local reference */
 960         zswap_entry_put(tree, entry);
 961
 962         /*
 963         * There are two possible situations for entry here:
 964         * (1) refcount is 1(normal case),  entry is valid and on the tree
 965         * (2) refcount is 0, entry is freed and not on the tree
 966         *     because invalidate happened during writeback
 967         *  search the tree and free the entry if find entry
 968         */
 969         if (entry == zswap_rb_search(&tree->rbroot, offset))
 970                 zswap_entry_put(tree, entry);
 971         spin_unlock(&tree->lock);
 972
 973         goto end;
 974
 975         /*
 976         * if we get here due to ZSWAP_SWAPCACHE_EXIST
 977         * a load may happening concurrently
 978         * it is safe and okay to not free the entry
 979         * if we free the entry in the following put
 980         * it it either okay to return !0
 981         */
 982 fail:
 983         spin_lock(&tree->lock);
 984         zswap_entry_put(tree, entry);
 985         spin_unlock(&tree->lock);
 986
 987 end:
 988         zpool_unmap_handle(pool, handle);
 989         return ret;
 990 }
 991
 992 static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
 993 {
 994         unsigned int pos;
 995         unsigned long *page;
 996
 997         page = (unsigned long *)ptr;
 998         for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
 999                 if (page[pos] != page[0])
1000                         return 0;
1001         }
1002         *value = page[0];
1003         return 1;
1004 }
1005
1006 static void zswap_fill_page(void *ptr, unsigned long value)
1007 {
1008         unsigned long *page;
1009
1010         page = (unsigned long *)ptr;
1011         memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
1012 }
1013
1014 /*********************************
1015 * frontswap hooks
1016 **********************************/
1017 /* attempts to compress and store an single page */
1018 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1019                                 struct page *page)
1020 {
1021         struct zswap_tree *tree = zswap_trees[type];
1022         struct zswap_entry *entry, *dupentry;
1023         struct crypto_comp *tfm;
1024         int ret;
1025         unsigned int hlen, dlen = PAGE_SIZE;
1026         unsigned long handle, value;
1027         char *buf;
1028         u8 *src, *dst;
1029         struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1030         gfp_t gfp;
1031
1032         /* THP isn't supported */
1033         if (PageTransHuge(page)) {
1034                 ret = -EINVAL;
1035                 goto reject;
1036         }
1037
1038         if (!zswap_enabled || !tree) {
1039                 ret = -ENODEV;
1040                 goto reject;
1041         }
1042
1043         /* reclaim space if needed */
1044         if (zswap_is_full()) {
1045                 struct zswap_pool *pool;
1046
1047                 zswap_pool_limit_hit++;
1048                 zswap_pool_reached_full = true;
1049                 pool = zswap_pool_last_get();
1050                 if (pool)
1051                         queue_work(shrink_wq, &pool->shrink_work);
1052                 ret = -ENOMEM;
1053                 goto reject;
1054         }
1055
1056         if (zswap_pool_reached_full) {
1057                if (!zswap_can_accept()) {
1058                         ret = -ENOMEM;
1059                         goto reject;
1060                 } else
1061                         zswap_pool_reached_full = false;
1062         }
1063
1064         /* allocate entry */
1065         entry = zswap_entry_cache_alloc(GFP_KERNEL);
1066         if (!entry) {
1067                 zswap_reject_kmemcache_fail++;
1068                 ret = -ENOMEM;
1069                 goto reject;
1070         }
1071
1072         if (zswap_same_filled_pages_enabled) {
1073                 src = kmap_atomic(page);
1074                 if (zswap_is_page_same_filled(src, &value)) {
1075                         kunmap_atomic(src);
1076                         entry->offset = offset;
1077                         entry->length = 0;
1078                         entry->value = value;
1079                         atomic_inc(&zswap_same_filled_pages);
1080                         goto insert_entry;
1081                 }
1082                 kunmap_atomic(src);
1083         }
1084
1085         /* if entry is successfully added, it keeps the reference */
1086         entry->pool = zswap_pool_current_get();
1087         if (!entry->pool) {
1088                 ret = -EINVAL;
1089                 goto freepage;
1090         }
1091
1092         /* compress */
1093         dst = get_cpu_var(zswap_dstmem);
1094         tfm = *get_cpu_ptr(entry->pool->tfm);
1095         src = kmap_atomic(page);
1096         ret = crypto_comp_compress(tfm, src, PAGE_SIZE, dst, &dlen);
1097         kunmap_atomic(src);
1098         put_cpu_ptr(entry->pool->tfm);
1099         if (ret) {
1100                 ret = -EINVAL;
1101                 goto put_dstmem;
1102         }
1103
1104         /* store */
1105         hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1106         gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1107         if (zpool_malloc_support_movable(entry->pool->zpool))
1108                 gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1109         ret = zpool_malloc(entry->pool->zpool, hlen + dlen, gfp, &handle);
1110         if (ret == -ENOSPC) {
1111                 zswap_reject_compress_poor++;
1112                 goto put_dstmem;
1113         }
1114         if (ret) {
1115                 zswap_reject_alloc_fail++;
1116                 goto put_dstmem;
1117         }
1118         buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW);
1119         memcpy(buf, &zhdr, hlen);
1120         memcpy(buf + hlen, dst, dlen);
1121         zpool_unmap_handle(entry->pool->zpool, handle);
1122         put_cpu_var(zswap_dstmem);
1123
1124         /* populate entry */
1125         entry->offset = offset;
1126         entry->handle = handle;
1127         entry->length = dlen;
1128
1129 insert_entry:
1130         /* map */
1131         spin_lock(&tree->lock);
1132         do {
1133                 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1134                 if (ret == -EEXIST) {
1135                         zswap_duplicate_entry++;
1136                         /* remove from rbtree */
1137                         zswap_rb_erase(&tree->rbroot, dupentry);
1138                         zswap_entry_put(tree, dupentry);
1139                 }
1140         } while (ret == -EEXIST);
1141         spin_unlock(&tree->lock);
1142
1143         /* update stats */
1144         atomic_inc(&zswap_stored_pages);
1145         zswap_update_total_size();
1146
1147         return 0;
1148
1149 put_dstmem:
1150         put_cpu_var(zswap_dstmem);
1151         zswap_pool_put(entry->pool);
1152 freepage:
1153         zswap_entry_cache_free(entry);
1154 reject:
1155         return ret;
1156 }
1157
1158 /*
1159  * returns 0 if the page was successfully decompressed
1160  * return -1 on entry not found or error
1161 */
1162 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1163                                 struct page *page)
1164 {
1165         struct zswap_tree *tree = zswap_trees[type];
1166         struct zswap_entry *entry;
1167         struct crypto_comp *tfm;
1168         u8 *src, *dst;
1169         unsigned int dlen;
1170         int ret;
1171
1172         /* find */
1173         spin_lock(&tree->lock);
1174         entry = zswap_entry_find_get(&tree->rbroot, offset);
1175         if (!entry) {
1176                 /* entry was written back */
1177                 spin_unlock(&tree->lock);
1178                 return -1;
1179         }
1180         spin_unlock(&tree->lock);
1181
1182         if (!entry->length) {
1183                 dst = kmap_atomic(page);
1184                 zswap_fill_page(dst, entry->value);
1185                 kunmap_atomic(dst);
1186                 goto freeentry;
1187         }
1188
1189         /* decompress */
1190         dlen = PAGE_SIZE;
1191         src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1192         if (zpool_evictable(entry->pool->zpool))
1193                 src += sizeof(struct zswap_header);
1194         dst = kmap_atomic(page);
1195         tfm = *get_cpu_ptr(entry->pool->tfm);
1196         ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen);
1197         put_cpu_ptr(entry->pool->tfm);
1198         kunmap_atomic(dst);
1199         zpool_unmap_handle(entry->pool->zpool, entry->handle);
1200         BUG_ON(ret);
1201
1202 freeentry:
1203         spin_lock(&tree->lock);
1204         zswap_entry_put(tree, entry);
1205         spin_unlock(&tree->lock);
1206
1207         return 0;
1208 }
1209
1210 /* frees an entry in zswap */
1211 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1212 {
1213         struct zswap_tree *tree = zswap_trees[type];
1214         struct zswap_entry *entry;
1215
1216         /* find */
1217         spin_lock(&tree->lock);
1218         entry = zswap_rb_search(&tree->rbroot, offset);
1219         if (!entry) {
1220                 /* entry was written back */
1221                 spin_unlock(&tree->lock);
1222                 return;
1223         }
1224
1225         /* remove from rbtree */
1226         zswap_rb_erase(&tree->rbroot, entry);
1227
1228         /* drop the initial reference from entry creation */
1229         zswap_entry_put(tree, entry);
1230
1231         spin_unlock(&tree->lock);
1232 }
1233
1234 /* frees all zswap entries for the given swap type */
1235 static void zswap_frontswap_invalidate_area(unsigned type)
1236 {
1237         struct zswap_tree *tree = zswap_trees[type];
1238         struct zswap_entry *entry, *n;
1239
1240         if (!tree)
1241                 return;
1242
1243         /* walk the tree and free everything */
1244         spin_lock(&tree->lock);
1245         rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1246                 zswap_free_entry(entry);
1247         tree->rbroot = RB_ROOT;
1248         spin_unlock(&tree->lock);
1249         kfree(tree);
1250         zswap_trees[type] = NULL;
1251 }
1252
1253 static void zswap_frontswap_init(unsigned type)
1254 {
1255         struct zswap_tree *tree;
1256
1257         tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1258         if (!tree) {
1259                 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1260                 return;
1261         }
1262
1263         tree->rbroot = RB_ROOT;
1264         spin_lock_init(&tree->lock);
1265         zswap_trees[type] = tree;
1266 }
1267
1268 static struct frontswap_ops zswap_frontswap_ops = {
1269         .store = zswap_frontswap_store,
1270         .load = zswap_frontswap_load,
1271         .invalidate_page = zswap_frontswap_invalidate_page,
1272         .invalidate_area = zswap_frontswap_invalidate_area,
1273         .init = zswap_frontswap_init
1274 };
1275
1276 /*********************************
1277 * debugfs functions
1278 **********************************/
1279 #ifdef CONFIG_DEBUG_FS
1280 #include <linux/debugfs.h>
1281
1282 static struct dentry *zswap_debugfs_root;
1283
1284 static int __init zswap_debugfs_init(void)
1285 {
1286         if (!debugfs_initialized())
1287                 return -ENODEV;
1288
1289         zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1290
1291         debugfs_create_u64("pool_limit_hit", 0444,
1292                            zswap_debugfs_root, &zswap_pool_limit_hit);
1293         debugfs_create_u64("reject_reclaim_fail", 0444,
1294                            zswap_debugfs_root, &zswap_reject_reclaim_fail);
1295         debugfs_create_u64("reject_alloc_fail", 0444,
1296                            zswap_debugfs_root, &zswap_reject_alloc_fail);
1297         debugfs_create_u64("reject_kmemcache_fail", 0444,
1298                            zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1299         debugfs_create_u64("reject_compress_poor", 0444,
1300                            zswap_debugfs_root, &zswap_reject_compress_poor);
1301         debugfs_create_u64("written_back_pages", 0444,
1302                            zswap_debugfs_root, &zswap_written_back_pages);
1303         debugfs_create_u64("duplicate_entry", 0444,
1304                            zswap_debugfs_root, &zswap_duplicate_entry);
1305         debugfs_create_u64("pool_total_size", 0444,
1306                            zswap_debugfs_root, &zswap_pool_total_size);
1307         debugfs_create_atomic_t("stored_pages", 0444,
1308                                 zswap_debugfs_root, &zswap_stored_pages);
1309         debugfs_create_atomic_t("same_filled_pages", 0444,
1310                                 zswap_debugfs_root, &zswap_same_filled_pages);
1311
1312         return 0;
1313 }
1314
1315 static void __exit zswap_debugfs_exit(void)
1316 {
1317         debugfs_remove_recursive(zswap_debugfs_root);
1318 }
1319 #else
1320 static int __init zswap_debugfs_init(void)
1321 {
1322         return 0;
1323 }
1324
1325 static void __exit zswap_debugfs_exit(void) { }
1326 #endif
1327
1328 /*********************************
1329 * module init and exit
1330 **********************************/
1331 static int __init init_zswap(void)
1332 {
1333         int ret;
1334
1335         zswap_init_started = true;
1336
1337         if (zswap_entry_cache_create()) {
1338                 pr_err("entry cache creation failed\n");
1339                 goto cache_fail;
1340         }
1341
1342         ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1343                                 zswap_dstmem_prepare, zswap_dstmem_dead);
1344         if (ret) {
1345                 pr_err("dstmem alloc failed\n");
1346                 goto dstmem_fail;
1347         }
1348
1349         ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1350                                       "mm/zswap_pool:prepare",
1351                                       zswap_cpu_comp_prepare,
1352                                       zswap_cpu_comp_dead);
1353         if (ret)
1354                 goto hp_fail;
1355
1356         shrink_wq = create_workqueue("zswap-shrink");
1357         if (!shrink_wq)
1358                 goto hp_fail;
1359
1360         frontswap_register_ops(&zswap_frontswap_ops);
1361         if (zswap_debugfs_init())
1362                 pr_warn("debugfs initialization failed\n");
1363
1364         if (zswap_enabled)
1365                 zswap_try_pool_create();
1366
1367         return 0;
1368
1369 hp_fail:
1370         cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1371 dstmem_fail:
1372         zswap_entry_cache_destroy();
1373 cache_fail:
1374         /* if built-in, we aren't unloaded on failure; don't allow use */
1375         zswap_init_failed = true;
1376         zswap_enabled = false;
1377         return -ENOMEM;
1378 }
1379 /* must be late so crypto has time to come up */
1380 late_initcall(init_zswap);
1381
1382 MODULE_LICENSE("GPL");
1383 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1384 MODULE_DESCRIPTION("Compressed cache for swap pages");