Merge remote-tracking branch 'stable/linux-5.15.y' into rpi-5.15.y
[platform/kernel/linux-rpi.git] / 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/scatterlist.h>
28 #include <linux/mempool.h>
29 #include <linux/zpool.h>
30 #include <crypto/acompress.h>
31
32 #include <linux/mm_types.h>
33 #include <linux/page-flags.h>
34 #include <linux/swapops.h>
35 #include <linux/writeback.h>
36 #include <linux/pagemap.h>
37 #include <linux/workqueue.h>
38
39 /*********************************
40 * statistics
41 **********************************/
42 /* Total bytes used by the compressed storage */
43 static u64 zswap_pool_total_size;
44 /* The number of compressed pages currently stored in zswap */
45 static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
46 /* The number of same-value filled pages currently stored in zswap */
47 static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
48
49 /*
50  * The statistics below are not protected from concurrent access for
51  * performance reasons so they may not be a 100% accurate.  However,
52  * they do provide useful information on roughly how many times a
53  * certain event is occurring.
54 */
55
56 /* Pool limit was hit (see zswap_max_pool_percent) */
57 static u64 zswap_pool_limit_hit;
58 /* Pages written back when pool limit was reached */
59 static u64 zswap_written_back_pages;
60 /* Store failed due to a reclaim failure after pool limit was reached */
61 static u64 zswap_reject_reclaim_fail;
62 /* Compressed page was too big for the allocator to (optimally) store */
63 static u64 zswap_reject_compress_poor;
64 /* Store failed because underlying allocator could not get memory */
65 static u64 zswap_reject_alloc_fail;
66 /* Store failed because the entry metadata could not be allocated (rare) */
67 static u64 zswap_reject_kmemcache_fail;
68 /* Duplicate store was encountered (rare) */
69 static u64 zswap_duplicate_entry;
70
71 /* Shrinker work queue */
72 static struct workqueue_struct *shrink_wq;
73 /* Pool limit was hit, we need to calm down */
74 static bool zswap_pool_reached_full;
75
76 /*********************************
77 * tunables
78 **********************************/
79
80 #define ZSWAP_PARAM_UNSET ""
81
82 /* Enable/disable zswap */
83 static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
84 static int zswap_enabled_param_set(const char *,
85                                    const struct kernel_param *);
86 static const struct kernel_param_ops zswap_enabled_param_ops = {
87         .set =          zswap_enabled_param_set,
88         .get =          param_get_bool,
89 };
90 module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
91
92 /* Crypto compressor to use */
93 static char *zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
94 static int zswap_compressor_param_set(const char *,
95                                       const struct kernel_param *);
96 static const struct kernel_param_ops zswap_compressor_param_ops = {
97         .set =          zswap_compressor_param_set,
98         .get =          param_get_charp,
99         .free =         param_free_charp,
100 };
101 module_param_cb(compressor, &zswap_compressor_param_ops,
102                 &zswap_compressor, 0644);
103
104 /* Compressed storage zpool to use */
105 static char *zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
106 static int zswap_zpool_param_set(const char *, const struct kernel_param *);
107 static const struct kernel_param_ops zswap_zpool_param_ops = {
108         .set =          zswap_zpool_param_set,
109         .get =          param_get_charp,
110         .free =         param_free_charp,
111 };
112 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
113
114 /* The maximum percentage of memory that the compressed pool can occupy */
115 static unsigned int zswap_max_pool_percent = 20;
116 module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
117
118 /* The threshold for accepting new pages after the max_pool_percent was hit */
119 static unsigned int zswap_accept_thr_percent = 90; /* of max pool size */
120 module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
121                    uint, 0644);
122
123 /* Enable/disable handling same-value filled pages (enabled by default) */
124 static bool zswap_same_filled_pages_enabled = true;
125 module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
126                    bool, 0644);
127
128 /*********************************
129 * data structures
130 **********************************/
131
132 struct crypto_acomp_ctx {
133         struct crypto_acomp *acomp;
134         struct acomp_req *req;
135         struct crypto_wait wait;
136         u8 *dstmem;
137         struct mutex *mutex;
138 };
139
140 struct zswap_pool {
141         struct zpool *zpool;
142         struct crypto_acomp_ctx __percpu *acomp_ctx;
143         struct kref kref;
144         struct list_head list;
145         struct work_struct release_work;
146         struct work_struct shrink_work;
147         struct hlist_node node;
148         char tfm_name[CRYPTO_MAX_ALG_NAME];
149 };
150
151 /*
152  * struct zswap_entry
153  *
154  * This structure contains the metadata for tracking a single compressed
155  * page within zswap.
156  *
157  * rbnode - links the entry into red-black tree for the appropriate swap type
158  * offset - the swap offset for the entry.  Index into the red-black tree.
159  * refcount - the number of outstanding reference to the entry. This is needed
160  *            to protect against premature freeing of the entry by code
161  *            concurrent calls to load, invalidate, and writeback.  The lock
162  *            for the zswap_tree structure that contains the entry must
163  *            be held while changing the refcount.  Since the lock must
164  *            be held, there is no reason to also make refcount atomic.
165  * length - the length in bytes of the compressed page data.  Needed during
166  *          decompression. For a same value filled page length is 0.
167  * pool - the zswap_pool the entry's data is in
168  * handle - zpool allocation handle that stores the compressed page data
169  * value - value of the same-value filled pages which have same content
170  */
171 struct zswap_entry {
172         struct rb_node rbnode;
173         pgoff_t offset;
174         int refcount;
175         unsigned int length;
176         struct zswap_pool *pool;
177         union {
178                 unsigned long handle;
179                 unsigned long value;
180         };
181 };
182
183 struct zswap_header {
184         swp_entry_t swpentry;
185 };
186
187 /*
188  * The tree lock in the zswap_tree struct protects a few things:
189  * - the rbtree
190  * - the refcount field of each entry in the tree
191  */
192 struct zswap_tree {
193         struct rb_root rbroot;
194         spinlock_t lock;
195 };
196
197 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
198
199 /* RCU-protected iteration */
200 static LIST_HEAD(zswap_pools);
201 /* protects zswap_pools list modification */
202 static DEFINE_SPINLOCK(zswap_pools_lock);
203 /* pool counter to provide unique names to zpool */
204 static atomic_t zswap_pools_count = ATOMIC_INIT(0);
205
206 /* used by param callback function */
207 static bool zswap_init_started;
208
209 /* fatal error during init */
210 static bool zswap_init_failed;
211
212 /* init completed, but couldn't create the initial pool */
213 static bool zswap_has_pool;
214
215 /*********************************
216 * helpers and fwd declarations
217 **********************************/
218
219 #define zswap_pool_debug(msg, p)                                \
220         pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name,         \
221                  zpool_get_type((p)->zpool))
222
223 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
224 static int zswap_pool_get(struct zswap_pool *pool);
225 static void zswap_pool_put(struct zswap_pool *pool);
226
227 static const struct zpool_ops zswap_zpool_ops = {
228         .evict = zswap_writeback_entry
229 };
230
231 static bool zswap_is_full(void)
232 {
233         return totalram_pages() * zswap_max_pool_percent / 100 <
234                         DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
235 }
236
237 static bool zswap_can_accept(void)
238 {
239         return totalram_pages() * zswap_accept_thr_percent / 100 *
240                                 zswap_max_pool_percent / 100 >
241                         DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
242 }
243
244 static void zswap_update_total_size(void)
245 {
246         struct zswap_pool *pool;
247         u64 total = 0;
248
249         rcu_read_lock();
250
251         list_for_each_entry_rcu(pool, &zswap_pools, list)
252                 total += zpool_get_total_size(pool->zpool);
253
254         rcu_read_unlock();
255
256         zswap_pool_total_size = total;
257 }
258
259 /*********************************
260 * zswap entry functions
261 **********************************/
262 static struct kmem_cache *zswap_entry_cache;
263
264 static int __init zswap_entry_cache_create(void)
265 {
266         zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
267         return zswap_entry_cache == NULL;
268 }
269
270 static void __init zswap_entry_cache_destroy(void)
271 {
272         kmem_cache_destroy(zswap_entry_cache);
273 }
274
275 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
276 {
277         struct zswap_entry *entry;
278         entry = kmem_cache_alloc(zswap_entry_cache, gfp);
279         if (!entry)
280                 return NULL;
281         entry->refcount = 1;
282         RB_CLEAR_NODE(&entry->rbnode);
283         return entry;
284 }
285
286 static void zswap_entry_cache_free(struct zswap_entry *entry)
287 {
288         kmem_cache_free(zswap_entry_cache, entry);
289 }
290
291 /*********************************
292 * rbtree functions
293 **********************************/
294 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
295 {
296         struct rb_node *node = root->rb_node;
297         struct zswap_entry *entry;
298
299         while (node) {
300                 entry = rb_entry(node, struct zswap_entry, rbnode);
301                 if (entry->offset > offset)
302                         node = node->rb_left;
303                 else if (entry->offset < offset)
304                         node = node->rb_right;
305                 else
306                         return entry;
307         }
308         return NULL;
309 }
310
311 /*
312  * In the case that a entry with the same offset is found, a pointer to
313  * the existing entry is stored in dupentry and the function returns -EEXIST
314  */
315 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
316                         struct zswap_entry **dupentry)
317 {
318         struct rb_node **link = &root->rb_node, *parent = NULL;
319         struct zswap_entry *myentry;
320
321         while (*link) {
322                 parent = *link;
323                 myentry = rb_entry(parent, struct zswap_entry, rbnode);
324                 if (myentry->offset > entry->offset)
325                         link = &(*link)->rb_left;
326                 else if (myentry->offset < entry->offset)
327                         link = &(*link)->rb_right;
328                 else {
329                         *dupentry = myentry;
330                         return -EEXIST;
331                 }
332         }
333         rb_link_node(&entry->rbnode, parent, link);
334         rb_insert_color(&entry->rbnode, root);
335         return 0;
336 }
337
338 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
339 {
340         if (!RB_EMPTY_NODE(&entry->rbnode)) {
341                 rb_erase(&entry->rbnode, root);
342                 RB_CLEAR_NODE(&entry->rbnode);
343         }
344 }
345
346 /*
347  * Carries out the common pattern of freeing and entry's zpool allocation,
348  * freeing the entry itself, and decrementing the number of stored pages.
349  */
350 static void zswap_free_entry(struct zswap_entry *entry)
351 {
352         if (!entry->length)
353                 atomic_dec(&zswap_same_filled_pages);
354         else {
355                 zpool_free(entry->pool->zpool, entry->handle);
356                 zswap_pool_put(entry->pool);
357         }
358         zswap_entry_cache_free(entry);
359         atomic_dec(&zswap_stored_pages);
360         zswap_update_total_size();
361 }
362
363 /* caller must hold the tree lock */
364 static void zswap_entry_get(struct zswap_entry *entry)
365 {
366         entry->refcount++;
367 }
368
369 /* caller must hold the tree lock
370 * remove from the tree and free it, if nobody reference the entry
371 */
372 static void zswap_entry_put(struct zswap_tree *tree,
373                         struct zswap_entry *entry)
374 {
375         int refcount = --entry->refcount;
376
377         BUG_ON(refcount < 0);
378         if (refcount == 0) {
379                 zswap_rb_erase(&tree->rbroot, entry);
380                 zswap_free_entry(entry);
381         }
382 }
383
384 /* caller must hold the tree lock */
385 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
386                                 pgoff_t offset)
387 {
388         struct zswap_entry *entry;
389
390         entry = zswap_rb_search(root, offset);
391         if (entry)
392                 zswap_entry_get(entry);
393
394         return entry;
395 }
396
397 /*********************************
398 * per-cpu code
399 **********************************/
400 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
401 /*
402  * If users dynamically change the zpool type and compressor at runtime, i.e.
403  * zswap is running, zswap can have more than one zpool on one cpu, but they
404  * are sharing dtsmem. So we need this mutex to be per-cpu.
405  */
406 static DEFINE_PER_CPU(struct mutex *, zswap_mutex);
407
408 static int zswap_dstmem_prepare(unsigned int cpu)
409 {
410         struct mutex *mutex;
411         u8 *dst;
412
413         dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
414         if (!dst)
415                 return -ENOMEM;
416
417         mutex = kmalloc_node(sizeof(*mutex), GFP_KERNEL, cpu_to_node(cpu));
418         if (!mutex) {
419                 kfree(dst);
420                 return -ENOMEM;
421         }
422
423         mutex_init(mutex);
424         per_cpu(zswap_dstmem, cpu) = dst;
425         per_cpu(zswap_mutex, cpu) = mutex;
426         return 0;
427 }
428
429 static int zswap_dstmem_dead(unsigned int cpu)
430 {
431         struct mutex *mutex;
432         u8 *dst;
433
434         mutex = per_cpu(zswap_mutex, cpu);
435         kfree(mutex);
436         per_cpu(zswap_mutex, cpu) = NULL;
437
438         dst = per_cpu(zswap_dstmem, cpu);
439         kfree(dst);
440         per_cpu(zswap_dstmem, cpu) = NULL;
441
442         return 0;
443 }
444
445 static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
446 {
447         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
448         struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
449         struct crypto_acomp *acomp;
450         struct acomp_req *req;
451
452         acomp = crypto_alloc_acomp_node(pool->tfm_name, 0, 0, cpu_to_node(cpu));
453         if (IS_ERR(acomp)) {
454                 pr_err("could not alloc crypto acomp %s : %ld\n",
455                                 pool->tfm_name, PTR_ERR(acomp));
456                 return PTR_ERR(acomp);
457         }
458         acomp_ctx->acomp = acomp;
459
460         req = acomp_request_alloc(acomp_ctx->acomp);
461         if (!req) {
462                 pr_err("could not alloc crypto acomp_request %s\n",
463                        pool->tfm_name);
464                 crypto_free_acomp(acomp_ctx->acomp);
465                 return -ENOMEM;
466         }
467         acomp_ctx->req = req;
468
469         crypto_init_wait(&acomp_ctx->wait);
470         /*
471          * if the backend of acomp is async zip, crypto_req_done() will wakeup
472          * crypto_wait_req(); if the backend of acomp is scomp, the callback
473          * won't be called, crypto_wait_req() will return without blocking.
474          */
475         acomp_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
476                                    crypto_req_done, &acomp_ctx->wait);
477
478         acomp_ctx->mutex = per_cpu(zswap_mutex, cpu);
479         acomp_ctx->dstmem = per_cpu(zswap_dstmem, cpu);
480
481         return 0;
482 }
483
484 static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
485 {
486         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
487         struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
488
489         if (!IS_ERR_OR_NULL(acomp_ctx)) {
490                 if (!IS_ERR_OR_NULL(acomp_ctx->req))
491                         acomp_request_free(acomp_ctx->req);
492                 if (!IS_ERR_OR_NULL(acomp_ctx->acomp))
493                         crypto_free_acomp(acomp_ctx->acomp);
494         }
495
496         return 0;
497 }
498
499 /*********************************
500 * pool functions
501 **********************************/
502
503 static struct zswap_pool *__zswap_pool_current(void)
504 {
505         struct zswap_pool *pool;
506
507         pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
508         WARN_ONCE(!pool && zswap_has_pool,
509                   "%s: no page storage pool!\n", __func__);
510
511         return pool;
512 }
513
514 static struct zswap_pool *zswap_pool_current(void)
515 {
516         assert_spin_locked(&zswap_pools_lock);
517
518         return __zswap_pool_current();
519 }
520
521 static struct zswap_pool *zswap_pool_current_get(void)
522 {
523         struct zswap_pool *pool;
524
525         rcu_read_lock();
526
527         pool = __zswap_pool_current();
528         if (!zswap_pool_get(pool))
529                 pool = NULL;
530
531         rcu_read_unlock();
532
533         return pool;
534 }
535
536 static struct zswap_pool *zswap_pool_last_get(void)
537 {
538         struct zswap_pool *pool, *last = NULL;
539
540         rcu_read_lock();
541
542         list_for_each_entry_rcu(pool, &zswap_pools, list)
543                 last = pool;
544         WARN_ONCE(!last && zswap_has_pool,
545                   "%s: no page storage pool!\n", __func__);
546         if (!zswap_pool_get(last))
547                 last = NULL;
548
549         rcu_read_unlock();
550
551         return last;
552 }
553
554 /* type and compressor must be null-terminated */
555 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
556 {
557         struct zswap_pool *pool;
558
559         assert_spin_locked(&zswap_pools_lock);
560
561         list_for_each_entry_rcu(pool, &zswap_pools, list) {
562                 if (strcmp(pool->tfm_name, compressor))
563                         continue;
564                 if (strcmp(zpool_get_type(pool->zpool), type))
565                         continue;
566                 /* if we can't get it, it's about to be destroyed */
567                 if (!zswap_pool_get(pool))
568                         continue;
569                 return pool;
570         }
571
572         return NULL;
573 }
574
575 static void shrink_worker(struct work_struct *w)
576 {
577         struct zswap_pool *pool = container_of(w, typeof(*pool),
578                                                 shrink_work);
579
580         if (zpool_shrink(pool->zpool, 1, NULL))
581                 zswap_reject_reclaim_fail++;
582         zswap_pool_put(pool);
583 }
584
585 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
586 {
587         struct zswap_pool *pool;
588         char name[38]; /* 'zswap' + 32 char (max) num + \0 */
589         gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
590         int ret;
591
592         if (!zswap_has_pool) {
593                 /* if either are unset, pool initialization failed, and we
594                  * need both params to be set correctly before trying to
595                  * create a pool.
596                  */
597                 if (!strcmp(type, ZSWAP_PARAM_UNSET))
598                         return NULL;
599                 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
600                         return NULL;
601         }
602
603         pool = kzalloc(sizeof(*pool), GFP_KERNEL);
604         if (!pool)
605                 return NULL;
606
607         /* unique name for each pool specifically required by zsmalloc */
608         snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
609
610         pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
611         if (!pool->zpool) {
612                 pr_err("%s zpool not available\n", type);
613                 goto error;
614         }
615         pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
616
617         strscpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
618
619         pool->acomp_ctx = alloc_percpu(*pool->acomp_ctx);
620         if (!pool->acomp_ctx) {
621                 pr_err("percpu alloc failed\n");
622                 goto error;
623         }
624
625         ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
626                                        &pool->node);
627         if (ret)
628                 goto error;
629         pr_debug("using %s compressor\n", pool->tfm_name);
630
631         /* being the current pool takes 1 ref; this func expects the
632          * caller to always add the new pool as the current pool
633          */
634         kref_init(&pool->kref);
635         INIT_LIST_HEAD(&pool->list);
636         INIT_WORK(&pool->shrink_work, shrink_worker);
637
638         zswap_pool_debug("created", pool);
639
640         return pool;
641
642 error:
643         if (pool->acomp_ctx)
644                 free_percpu(pool->acomp_ctx);
645         if (pool->zpool)
646                 zpool_destroy_pool(pool->zpool);
647         kfree(pool);
648         return NULL;
649 }
650
651 static bool zswap_try_pool_create(void)
652 {
653         struct zswap_pool *pool;
654         bool has_comp, has_zpool;
655
656         has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
657         if (!has_comp && strcmp(zswap_compressor,
658                                 CONFIG_ZSWAP_COMPRESSOR_DEFAULT)) {
659                 pr_err("compressor %s not available, using default %s\n",
660                        zswap_compressor, CONFIG_ZSWAP_COMPRESSOR_DEFAULT);
661                 param_free_charp(&zswap_compressor);
662                 zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
663                 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
664         }
665         if (!has_comp) {
666                 pr_err("default compressor %s not available\n",
667                        zswap_compressor);
668                 param_free_charp(&zswap_compressor);
669                 zswap_compressor = ZSWAP_PARAM_UNSET;
670         }
671
672         has_zpool = zpool_has_pool(zswap_zpool_type);
673         if (!has_zpool && strcmp(zswap_zpool_type,
674                                  CONFIG_ZSWAP_ZPOOL_DEFAULT)) {
675                 pr_err("zpool %s not available, using default %s\n",
676                        zswap_zpool_type, CONFIG_ZSWAP_ZPOOL_DEFAULT);
677                 param_free_charp(&zswap_zpool_type);
678                 zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
679                 has_zpool = zpool_has_pool(zswap_zpool_type);
680         }
681         if (!has_zpool) {
682                 pr_err("default zpool %s not available\n",
683                        zswap_zpool_type);
684                 param_free_charp(&zswap_zpool_type);
685                 zswap_zpool_type = ZSWAP_PARAM_UNSET;
686         }
687
688         if (!has_comp || !has_zpool)
689                 return false;
690
691         pool = zswap_pool_create(zswap_zpool_type, zswap_compressor);
692
693         if (pool) {
694                 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
695                         zpool_get_type(pool->zpool));
696                 list_add(&pool->list, &zswap_pools);
697                 zswap_has_pool = true;
698         } else {
699                 pr_err("pool creation failed\n");
700                 zswap_enabled = false;
701         }
702
703         return zswap_enabled;
704 }
705
706 static void zswap_pool_destroy(struct zswap_pool *pool)
707 {
708         zswap_pool_debug("destroying", pool);
709
710         cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
711         free_percpu(pool->acomp_ctx);
712         zpool_destroy_pool(pool->zpool);
713         kfree(pool);
714 }
715
716 static int __must_check zswap_pool_get(struct zswap_pool *pool)
717 {
718         if (!pool)
719                 return 0;
720
721         return kref_get_unless_zero(&pool->kref);
722 }
723
724 static void __zswap_pool_release(struct work_struct *work)
725 {
726         struct zswap_pool *pool = container_of(work, typeof(*pool),
727                                                 release_work);
728
729         synchronize_rcu();
730
731         /* nobody should have been able to get a kref... */
732         WARN_ON(kref_get_unless_zero(&pool->kref));
733
734         /* pool is now off zswap_pools list and has no references. */
735         zswap_pool_destroy(pool);
736 }
737
738 static void __zswap_pool_empty(struct kref *kref)
739 {
740         struct zswap_pool *pool;
741
742         pool = container_of(kref, typeof(*pool), kref);
743
744         spin_lock(&zswap_pools_lock);
745
746         WARN_ON(pool == zswap_pool_current());
747
748         list_del_rcu(&pool->list);
749
750         INIT_WORK(&pool->release_work, __zswap_pool_release);
751         schedule_work(&pool->release_work);
752
753         spin_unlock(&zswap_pools_lock);
754 }
755
756 static void zswap_pool_put(struct zswap_pool *pool)
757 {
758         kref_put(&pool->kref, __zswap_pool_empty);
759 }
760
761 /*********************************
762 * param callbacks
763 **********************************/
764
765 /* val must be a null-terminated string */
766 static int __zswap_param_set(const char *val, const struct kernel_param *kp,
767                              char *type, char *compressor)
768 {
769         struct zswap_pool *pool, *put_pool = NULL;
770         char *s = strstrip((char *)val);
771         int ret;
772
773         if (zswap_init_failed) {
774                 pr_err("can't set param, initialization failed\n");
775                 return -ENODEV;
776         }
777
778         /* no change required */
779         if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
780                 return 0;
781
782         /* if this is load-time (pre-init) param setting,
783          * don't create a pool; that's done during init.
784          */
785         if (!zswap_init_started)
786                 return param_set_charp(s, kp);
787
788         if (!type) {
789                 if (!zpool_has_pool(s)) {
790                         pr_err("zpool %s not available\n", s);
791                         return -ENOENT;
792                 }
793                 type = s;
794         } else if (!compressor) {
795                 if (!crypto_has_acomp(s, 0, 0)) {
796                         pr_err("compressor %s not available\n", s);
797                         return -ENOENT;
798                 }
799                 compressor = s;
800         } else {
801                 WARN_ON(1);
802                 return -EINVAL;
803         }
804
805         spin_lock(&zswap_pools_lock);
806
807         pool = zswap_pool_find_get(type, compressor);
808         if (pool) {
809                 zswap_pool_debug("using existing", pool);
810                 WARN_ON(pool == zswap_pool_current());
811                 list_del_rcu(&pool->list);
812         }
813
814         spin_unlock(&zswap_pools_lock);
815
816         if (!pool)
817                 pool = zswap_pool_create(type, compressor);
818
819         if (pool)
820                 ret = param_set_charp(s, kp);
821         else
822                 ret = -EINVAL;
823
824         spin_lock(&zswap_pools_lock);
825
826         if (!ret) {
827                 put_pool = zswap_pool_current();
828                 list_add_rcu(&pool->list, &zswap_pools);
829                 zswap_has_pool = true;
830         } else if (pool) {
831                 /* add the possibly pre-existing pool to the end of the pools
832                  * list; if it's new (and empty) then it'll be removed and
833                  * destroyed by the put after we drop the lock
834                  */
835                 list_add_tail_rcu(&pool->list, &zswap_pools);
836                 put_pool = pool;
837         }
838
839         spin_unlock(&zswap_pools_lock);
840
841         if (!zswap_has_pool && !pool) {
842                 /* if initial pool creation failed, and this pool creation also
843                  * failed, maybe both compressor and zpool params were bad.
844                  * Allow changing this param, so pool creation will succeed
845                  * when the other param is changed. We already verified this
846                  * param is ok in the zpool_has_pool() or crypto_has_acomp()
847                  * checks above.
848                  */
849                 ret = param_set_charp(s, kp);
850         }
851
852         /* drop the ref from either the old current pool,
853          * or the new pool we failed to add
854          */
855         if (put_pool)
856                 zswap_pool_put(put_pool);
857
858         return ret;
859 }
860
861 static int zswap_compressor_param_set(const char *val,
862                                       const struct kernel_param *kp)
863 {
864         return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
865 }
866
867 static int zswap_zpool_param_set(const char *val,
868                                  const struct kernel_param *kp)
869 {
870         return __zswap_param_set(val, kp, NULL, zswap_compressor);
871 }
872
873 static int zswap_enabled_param_set(const char *val,
874                                    const struct kernel_param *kp)
875 {
876         int ret;
877
878         if (zswap_init_failed) {
879                 pr_err("can't enable, initialization failed\n");
880                 return -ENODEV;
881         }
882
883         ret = param_set_bool(val, kp);
884         if (!ret && zswap_enabled && zswap_init_started && !zswap_has_pool)
885                 if (!zswap_try_pool_create())
886                         ret = -ENODEV;
887
888         return ret;
889 }
890
891 /*********************************
892 * writeback code
893 **********************************/
894 /* return enum for zswap_get_swap_cache_page */
895 enum zswap_get_swap_ret {
896         ZSWAP_SWAPCACHE_NEW,
897         ZSWAP_SWAPCACHE_EXIST,
898         ZSWAP_SWAPCACHE_FAIL,
899 };
900
901 /*
902  * zswap_get_swap_cache_page
903  *
904  * This is an adaption of read_swap_cache_async()
905  *
906  * This function tries to find a page with the given swap entry
907  * in the swapper_space address space (the swap cache).  If the page
908  * is found, it is returned in retpage.  Otherwise, a page is allocated,
909  * added to the swap cache, and returned in retpage.
910  *
911  * If success, the swap cache page is returned in retpage
912  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
913  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
914  *     the new page is added to swapcache and locked
915  * Returns ZSWAP_SWAPCACHE_FAIL on error
916  */
917 static int zswap_get_swap_cache_page(swp_entry_t entry,
918                                 struct page **retpage)
919 {
920         bool page_was_allocated;
921
922         *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
923                         NULL, 0, &page_was_allocated);
924         if (page_was_allocated)
925                 return ZSWAP_SWAPCACHE_NEW;
926         if (!*retpage)
927                 return ZSWAP_SWAPCACHE_FAIL;
928         return ZSWAP_SWAPCACHE_EXIST;
929 }
930
931 /*
932  * Attempts to free an entry by adding a page to the swap cache,
933  * decompressing the entry data into the page, and issuing a
934  * bio write to write the page back to the swap device.
935  *
936  * This can be thought of as a "resumed writeback" of the page
937  * to the swap device.  We are basically resuming the same swap
938  * writeback path that was intercepted with the frontswap_store()
939  * in the first place.  After the page has been decompressed into
940  * the swap cache, the compressed version stored by zswap can be
941  * freed.
942  */
943 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
944 {
945         struct zswap_header *zhdr;
946         swp_entry_t swpentry;
947         struct zswap_tree *tree;
948         pgoff_t offset;
949         struct zswap_entry *entry;
950         struct page *page;
951         struct scatterlist input, output;
952         struct crypto_acomp_ctx *acomp_ctx;
953
954         u8 *src, *tmp = NULL;
955         unsigned int dlen;
956         int ret;
957         struct writeback_control wbc = {
958                 .sync_mode = WB_SYNC_NONE,
959         };
960
961         if (!zpool_can_sleep_mapped(pool)) {
962                 tmp = kmalloc(PAGE_SIZE, GFP_ATOMIC);
963                 if (!tmp)
964                         return -ENOMEM;
965         }
966
967         /* extract swpentry from data */
968         zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
969         swpentry = zhdr->swpentry; /* here */
970         tree = zswap_trees[swp_type(swpentry)];
971         offset = swp_offset(swpentry);
972
973         /* find and ref zswap entry */
974         spin_lock(&tree->lock);
975         entry = zswap_entry_find_get(&tree->rbroot, offset);
976         if (!entry) {
977                 /* entry was invalidated */
978                 spin_unlock(&tree->lock);
979                 zpool_unmap_handle(pool, handle);
980                 kfree(tmp);
981                 return 0;
982         }
983         spin_unlock(&tree->lock);
984         BUG_ON(offset != entry->offset);
985
986         src = (u8 *)zhdr + sizeof(struct zswap_header);
987         if (!zpool_can_sleep_mapped(pool)) {
988                 memcpy(tmp, src, entry->length);
989                 src = tmp;
990                 zpool_unmap_handle(pool, handle);
991         }
992
993         /* try to allocate swap cache page */
994         switch (zswap_get_swap_cache_page(swpentry, &page)) {
995         case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
996                 ret = -ENOMEM;
997                 goto fail;
998
999         case ZSWAP_SWAPCACHE_EXIST:
1000                 /* page is already in the swap cache, ignore for now */
1001                 put_page(page);
1002                 ret = -EEXIST;
1003                 goto fail;
1004
1005         case ZSWAP_SWAPCACHE_NEW: /* page is locked */
1006                 /* decompress */
1007                 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1008                 dlen = PAGE_SIZE;
1009
1010                 mutex_lock(acomp_ctx->mutex);
1011                 sg_init_one(&input, src, entry->length);
1012                 sg_init_table(&output, 1);
1013                 sg_set_page(&output, page, PAGE_SIZE, 0);
1014                 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1015                 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1016                 dlen = acomp_ctx->req->dlen;
1017                 mutex_unlock(acomp_ctx->mutex);
1018
1019                 BUG_ON(ret);
1020                 BUG_ON(dlen != PAGE_SIZE);
1021
1022                 /* page is up to date */
1023                 SetPageUptodate(page);
1024         }
1025
1026         /* move it to the tail of the inactive list after end_writeback */
1027         SetPageReclaim(page);
1028
1029         /* start writeback */
1030         __swap_writepage(page, &wbc, end_swap_bio_write);
1031         put_page(page);
1032         zswap_written_back_pages++;
1033
1034         spin_lock(&tree->lock);
1035         /* drop local reference */
1036         zswap_entry_put(tree, entry);
1037
1038         /*
1039         * There are two possible situations for entry here:
1040         * (1) refcount is 1(normal case),  entry is valid and on the tree
1041         * (2) refcount is 0, entry is freed and not on the tree
1042         *     because invalidate happened during writeback
1043         *  search the tree and free the entry if find entry
1044         */
1045         if (entry == zswap_rb_search(&tree->rbroot, offset))
1046                 zswap_entry_put(tree, entry);
1047         spin_unlock(&tree->lock);
1048
1049         goto end;
1050
1051         /*
1052         * if we get here due to ZSWAP_SWAPCACHE_EXIST
1053         * a load may be happening concurrently.
1054         * it is safe and okay to not free the entry.
1055         * if we free the entry in the following put
1056         * it is also okay to return !0
1057         */
1058 fail:
1059         spin_lock(&tree->lock);
1060         zswap_entry_put(tree, entry);
1061         spin_unlock(&tree->lock);
1062
1063 end:
1064         if (zpool_can_sleep_mapped(pool))
1065                 zpool_unmap_handle(pool, handle);
1066         else
1067                 kfree(tmp);
1068
1069         return ret;
1070 }
1071
1072 static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
1073 {
1074         unsigned int pos;
1075         unsigned long *page;
1076
1077         page = (unsigned long *)ptr;
1078         for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
1079                 if (page[pos] != page[0])
1080                         return 0;
1081         }
1082         *value = page[0];
1083         return 1;
1084 }
1085
1086 static void zswap_fill_page(void *ptr, unsigned long value)
1087 {
1088         unsigned long *page;
1089
1090         page = (unsigned long *)ptr;
1091         memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
1092 }
1093
1094 /*********************************
1095 * frontswap hooks
1096 **********************************/
1097 /* attempts to compress and store an single page */
1098 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1099                                 struct page *page)
1100 {
1101         struct zswap_tree *tree = zswap_trees[type];
1102         struct zswap_entry *entry, *dupentry;
1103         struct scatterlist input, output;
1104         struct crypto_acomp_ctx *acomp_ctx;
1105         int ret;
1106         unsigned int hlen, dlen = PAGE_SIZE;
1107         unsigned long handle, value;
1108         char *buf;
1109         u8 *src, *dst;
1110         struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1111         gfp_t gfp;
1112
1113         /* THP isn't supported */
1114         if (PageTransHuge(page)) {
1115                 ret = -EINVAL;
1116                 goto reject;
1117         }
1118
1119         if (!zswap_enabled || !tree) {
1120                 ret = -ENODEV;
1121                 goto reject;
1122         }
1123
1124         /* reclaim space if needed */
1125         if (zswap_is_full()) {
1126                 struct zswap_pool *pool;
1127
1128                 zswap_pool_limit_hit++;
1129                 zswap_pool_reached_full = true;
1130                 pool = zswap_pool_last_get();
1131                 if (pool)
1132                         queue_work(shrink_wq, &pool->shrink_work);
1133                 ret = -ENOMEM;
1134                 goto reject;
1135         }
1136
1137         if (zswap_pool_reached_full) {
1138                if (!zswap_can_accept()) {
1139                         ret = -ENOMEM;
1140                         goto reject;
1141                 } else
1142                         zswap_pool_reached_full = false;
1143         }
1144
1145         /* allocate entry */
1146         entry = zswap_entry_cache_alloc(GFP_KERNEL);
1147         if (!entry) {
1148                 zswap_reject_kmemcache_fail++;
1149                 ret = -ENOMEM;
1150                 goto reject;
1151         }
1152
1153         if (zswap_same_filled_pages_enabled) {
1154                 src = kmap_atomic(page);
1155                 if (zswap_is_page_same_filled(src, &value)) {
1156                         kunmap_atomic(src);
1157                         entry->offset = offset;
1158                         entry->length = 0;
1159                         entry->value = value;
1160                         atomic_inc(&zswap_same_filled_pages);
1161                         goto insert_entry;
1162                 }
1163                 kunmap_atomic(src);
1164         }
1165
1166         /* if entry is successfully added, it keeps the reference */
1167         entry->pool = zswap_pool_current_get();
1168         if (!entry->pool) {
1169                 ret = -EINVAL;
1170                 goto freepage;
1171         }
1172
1173         /* compress */
1174         acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1175
1176         mutex_lock(acomp_ctx->mutex);
1177
1178         dst = acomp_ctx->dstmem;
1179         sg_init_table(&input, 1);
1180         sg_set_page(&input, page, PAGE_SIZE, 0);
1181
1182         /* zswap_dstmem is of size (PAGE_SIZE * 2). Reflect same in sg_list */
1183         sg_init_one(&output, dst, PAGE_SIZE * 2);
1184         acomp_request_set_params(acomp_ctx->req, &input, &output, PAGE_SIZE, dlen);
1185         /*
1186          * it maybe looks a little bit silly that we send an asynchronous request,
1187          * then wait for its completion synchronously. This makes the process look
1188          * synchronous in fact.
1189          * Theoretically, acomp supports users send multiple acomp requests in one
1190          * acomp instance, then get those requests done simultaneously. but in this
1191          * case, frontswap actually does store and load page by page, there is no
1192          * existing method to send the second page before the first page is done
1193          * in one thread doing frontswap.
1194          * but in different threads running on different cpu, we have different
1195          * acomp instance, so multiple threads can do (de)compression in parallel.
1196          */
1197         ret = crypto_wait_req(crypto_acomp_compress(acomp_ctx->req), &acomp_ctx->wait);
1198         dlen = acomp_ctx->req->dlen;
1199
1200         if (ret) {
1201                 ret = -EINVAL;
1202                 goto put_dstmem;
1203         }
1204
1205         /* store */
1206         hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1207         gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1208         if (zpool_malloc_support_movable(entry->pool->zpool))
1209                 gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1210         ret = zpool_malloc(entry->pool->zpool, hlen + dlen, gfp, &handle);
1211         if (ret == -ENOSPC) {
1212                 zswap_reject_compress_poor++;
1213                 goto put_dstmem;
1214         }
1215         if (ret) {
1216                 zswap_reject_alloc_fail++;
1217                 goto put_dstmem;
1218         }
1219         buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_WO);
1220         memcpy(buf, &zhdr, hlen);
1221         memcpy(buf + hlen, dst, dlen);
1222         zpool_unmap_handle(entry->pool->zpool, handle);
1223         mutex_unlock(acomp_ctx->mutex);
1224
1225         /* populate entry */
1226         entry->offset = offset;
1227         entry->handle = handle;
1228         entry->length = dlen;
1229
1230 insert_entry:
1231         /* map */
1232         spin_lock(&tree->lock);
1233         do {
1234                 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1235                 if (ret == -EEXIST) {
1236                         zswap_duplicate_entry++;
1237                         /* remove from rbtree */
1238                         zswap_rb_erase(&tree->rbroot, dupentry);
1239                         zswap_entry_put(tree, dupentry);
1240                 }
1241         } while (ret == -EEXIST);
1242         spin_unlock(&tree->lock);
1243
1244         /* update stats */
1245         atomic_inc(&zswap_stored_pages);
1246         zswap_update_total_size();
1247
1248         return 0;
1249
1250 put_dstmem:
1251         mutex_unlock(acomp_ctx->mutex);
1252         zswap_pool_put(entry->pool);
1253 freepage:
1254         zswap_entry_cache_free(entry);
1255 reject:
1256         return ret;
1257 }
1258
1259 /*
1260  * returns 0 if the page was successfully decompressed
1261  * return -1 on entry not found or error
1262 */
1263 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1264                                 struct page *page)
1265 {
1266         struct zswap_tree *tree = zswap_trees[type];
1267         struct zswap_entry *entry;
1268         struct scatterlist input, output;
1269         struct crypto_acomp_ctx *acomp_ctx;
1270         u8 *src, *dst, *tmp;
1271         unsigned int dlen;
1272         int ret;
1273
1274         /* find */
1275         spin_lock(&tree->lock);
1276         entry = zswap_entry_find_get(&tree->rbroot, offset);
1277         if (!entry) {
1278                 /* entry was written back */
1279                 spin_unlock(&tree->lock);
1280                 return -1;
1281         }
1282         spin_unlock(&tree->lock);
1283
1284         if (!entry->length) {
1285                 dst = kmap_atomic(page);
1286                 zswap_fill_page(dst, entry->value);
1287                 kunmap_atomic(dst);
1288                 ret = 0;
1289                 goto freeentry;
1290         }
1291
1292         if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1293
1294                 tmp = kmalloc(entry->length, GFP_ATOMIC);
1295                 if (!tmp) {
1296                         ret = -ENOMEM;
1297                         goto freeentry;
1298                 }
1299         }
1300
1301         /* decompress */
1302         dlen = PAGE_SIZE;
1303         src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1304         if (zpool_evictable(entry->pool->zpool))
1305                 src += sizeof(struct zswap_header);
1306
1307         if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1308
1309                 memcpy(tmp, src, entry->length);
1310                 src = tmp;
1311
1312                 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1313         }
1314
1315         acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1316         mutex_lock(acomp_ctx->mutex);
1317         sg_init_one(&input, src, entry->length);
1318         sg_init_table(&output, 1);
1319         sg_set_page(&output, page, PAGE_SIZE, 0);
1320         acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1321         ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1322         mutex_unlock(acomp_ctx->mutex);
1323
1324         if (zpool_can_sleep_mapped(entry->pool->zpool))
1325                 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1326         else
1327                 kfree(tmp);
1328
1329         BUG_ON(ret);
1330
1331 freeentry:
1332         spin_lock(&tree->lock);
1333         zswap_entry_put(tree, entry);
1334         spin_unlock(&tree->lock);
1335
1336         return ret;
1337 }
1338
1339 /* frees an entry in zswap */
1340 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1341 {
1342         struct zswap_tree *tree = zswap_trees[type];
1343         struct zswap_entry *entry;
1344
1345         /* find */
1346         spin_lock(&tree->lock);
1347         entry = zswap_rb_search(&tree->rbroot, offset);
1348         if (!entry) {
1349                 /* entry was written back */
1350                 spin_unlock(&tree->lock);
1351                 return;
1352         }
1353
1354         /* remove from rbtree */
1355         zswap_rb_erase(&tree->rbroot, entry);
1356
1357         /* drop the initial reference from entry creation */
1358         zswap_entry_put(tree, entry);
1359
1360         spin_unlock(&tree->lock);
1361 }
1362
1363 /* frees all zswap entries for the given swap type */
1364 static void zswap_frontswap_invalidate_area(unsigned type)
1365 {
1366         struct zswap_tree *tree = zswap_trees[type];
1367         struct zswap_entry *entry, *n;
1368
1369         if (!tree)
1370                 return;
1371
1372         /* walk the tree and free everything */
1373         spin_lock(&tree->lock);
1374         rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1375                 zswap_free_entry(entry);
1376         tree->rbroot = RB_ROOT;
1377         spin_unlock(&tree->lock);
1378         kfree(tree);
1379         zswap_trees[type] = NULL;
1380 }
1381
1382 static void zswap_frontswap_init(unsigned type)
1383 {
1384         struct zswap_tree *tree;
1385
1386         tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1387         if (!tree) {
1388                 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1389                 return;
1390         }
1391
1392         tree->rbroot = RB_ROOT;
1393         spin_lock_init(&tree->lock);
1394         zswap_trees[type] = tree;
1395 }
1396
1397 static struct frontswap_ops zswap_frontswap_ops = {
1398         .store = zswap_frontswap_store,
1399         .load = zswap_frontswap_load,
1400         .invalidate_page = zswap_frontswap_invalidate_page,
1401         .invalidate_area = zswap_frontswap_invalidate_area,
1402         .init = zswap_frontswap_init
1403 };
1404
1405 /*********************************
1406 * debugfs functions
1407 **********************************/
1408 #ifdef CONFIG_DEBUG_FS
1409 #include <linux/debugfs.h>
1410
1411 static struct dentry *zswap_debugfs_root;
1412
1413 static int __init zswap_debugfs_init(void)
1414 {
1415         if (!debugfs_initialized())
1416                 return -ENODEV;
1417
1418         zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1419
1420         debugfs_create_u64("pool_limit_hit", 0444,
1421                            zswap_debugfs_root, &zswap_pool_limit_hit);
1422         debugfs_create_u64("reject_reclaim_fail", 0444,
1423                            zswap_debugfs_root, &zswap_reject_reclaim_fail);
1424         debugfs_create_u64("reject_alloc_fail", 0444,
1425                            zswap_debugfs_root, &zswap_reject_alloc_fail);
1426         debugfs_create_u64("reject_kmemcache_fail", 0444,
1427                            zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1428         debugfs_create_u64("reject_compress_poor", 0444,
1429                            zswap_debugfs_root, &zswap_reject_compress_poor);
1430         debugfs_create_u64("written_back_pages", 0444,
1431                            zswap_debugfs_root, &zswap_written_back_pages);
1432         debugfs_create_u64("duplicate_entry", 0444,
1433                            zswap_debugfs_root, &zswap_duplicate_entry);
1434         debugfs_create_u64("pool_total_size", 0444,
1435                            zswap_debugfs_root, &zswap_pool_total_size);
1436         debugfs_create_atomic_t("stored_pages", 0444,
1437                                 zswap_debugfs_root, &zswap_stored_pages);
1438         debugfs_create_atomic_t("same_filled_pages", 0444,
1439                                 zswap_debugfs_root, &zswap_same_filled_pages);
1440
1441         return 0;
1442 }
1443 #else
1444 static int __init zswap_debugfs_init(void)
1445 {
1446         return 0;
1447 }
1448 #endif
1449
1450 /*********************************
1451 * module init and exit
1452 **********************************/
1453 static int __init init_zswap(void)
1454 {
1455         int ret;
1456
1457         zswap_init_started = true;
1458
1459         if (zswap_entry_cache_create()) {
1460                 pr_err("entry cache creation failed\n");
1461                 goto cache_fail;
1462         }
1463
1464         ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1465                                 zswap_dstmem_prepare, zswap_dstmem_dead);
1466         if (ret) {
1467                 pr_err("dstmem alloc failed\n");
1468                 goto dstmem_fail;
1469         }
1470
1471         ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1472                                       "mm/zswap_pool:prepare",
1473                                       zswap_cpu_comp_prepare,
1474                                       zswap_cpu_comp_dead);
1475         if (ret)
1476                 goto hp_fail;
1477
1478         shrink_wq = create_workqueue("zswap-shrink");
1479         if (!shrink_wq)
1480                 goto hp_fail;
1481
1482         frontswap_register_ops(&zswap_frontswap_ops);
1483         if (zswap_debugfs_init())
1484                 pr_warn("debugfs initialization failed\n");
1485
1486         if (zswap_enabled)
1487                 zswap_try_pool_create();
1488
1489         return 0;
1490
1491 hp_fail:
1492         cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1493 dstmem_fail:
1494         zswap_entry_cache_destroy();
1495 cache_fail:
1496         /* if built-in, we aren't unloaded on failure; don't allow use */
1497         zswap_init_failed = true;
1498         zswap_enabled = false;
1499         return -ENOMEM;
1500 }
1501 /* must be late so crypto has time to come up */
1502 late_initcall(init_zswap);
1503
1504 MODULE_LICENSE("GPL");
1505 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1506 MODULE_DESCRIPTION("Compressed cache for swap pages");