1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * zswap.c - zswap driver file
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.
11 * Copyright (C) 2012 Seth Jennings <sjenning@linux.vnet.ibm.com>
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
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>
37 /*********************************
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);
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.
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;
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;
74 /*********************************
76 **********************************/
78 #define ZSWAP_PARAM_UNSET ""
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,
88 module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
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,
99 module_param_cb(compressor, &zswap_compressor_param_ops,
100 &zswap_compressor, 0644);
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,
110 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
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);
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,
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,
126 /*********************************
128 **********************************/
132 struct crypto_comp * __percpu *tfm;
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];
144 * This structure contains the metadata for tracking a single compressed
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
162 struct rb_node rbnode;
166 struct zswap_pool *pool;
168 unsigned long handle;
173 struct zswap_header {
174 swp_entry_t swpentry;
178 * The tree lock in the zswap_tree struct protects a few things:
180 * - the refcount field of each entry in the tree
183 struct rb_root rbroot;
187 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
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);
196 /* used by param callback function */
197 static bool zswap_init_started;
199 /* fatal error during init */
200 static bool zswap_init_failed;
202 /* init completed, but couldn't create the initial pool */
203 static bool zswap_has_pool;
205 /*********************************
206 * helpers and fwd declarations
207 **********************************/
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))
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);
217 static const struct zpool_ops zswap_zpool_ops = {
218 .evict = zswap_writeback_entry
221 static bool zswap_is_full(void)
223 return totalram_pages() * zswap_max_pool_percent / 100 <
224 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
227 static bool zswap_can_accept(void)
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);
234 static void zswap_update_total_size(void)
236 struct zswap_pool *pool;
241 list_for_each_entry_rcu(pool, &zswap_pools, list)
242 total += zpool_get_total_size(pool->zpool);
246 zswap_pool_total_size = total;
249 /*********************************
250 * zswap entry functions
251 **********************************/
252 static struct kmem_cache *zswap_entry_cache;
254 static int __init zswap_entry_cache_create(void)
256 zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
257 return zswap_entry_cache == NULL;
260 static void __init zswap_entry_cache_destroy(void)
262 kmem_cache_destroy(zswap_entry_cache);
265 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
267 struct zswap_entry *entry;
268 entry = kmem_cache_alloc(zswap_entry_cache, gfp);
272 RB_CLEAR_NODE(&entry->rbnode);
276 static void zswap_entry_cache_free(struct zswap_entry *entry)
278 kmem_cache_free(zswap_entry_cache, entry);
281 /*********************************
283 **********************************/
284 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
286 struct rb_node *node = root->rb_node;
287 struct zswap_entry *entry;
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;
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
305 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
306 struct zswap_entry **dupentry)
308 struct rb_node **link = &root->rb_node, *parent = NULL;
309 struct zswap_entry *myentry;
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;
323 rb_link_node(&entry->rbnode, parent, link);
324 rb_insert_color(&entry->rbnode, root);
328 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
330 if (!RB_EMPTY_NODE(&entry->rbnode)) {
331 rb_erase(&entry->rbnode, root);
332 RB_CLEAR_NODE(&entry->rbnode);
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.
340 static void zswap_free_entry(struct zswap_entry *entry)
343 atomic_dec(&zswap_same_filled_pages);
345 zpool_free(entry->pool->zpool, entry->handle);
346 zswap_pool_put(entry->pool);
348 zswap_entry_cache_free(entry);
349 atomic_dec(&zswap_stored_pages);
350 zswap_update_total_size();
353 /* caller must hold the tree lock */
354 static void zswap_entry_get(struct zswap_entry *entry)
359 /* caller must hold the tree lock
360 * remove from the tree and free it, if nobody reference the entry
362 static void zswap_entry_put(struct zswap_tree *tree,
363 struct zswap_entry *entry)
365 int refcount = --entry->refcount;
367 BUG_ON(refcount < 0);
369 zswap_rb_erase(&tree->rbroot, entry);
370 zswap_free_entry(entry);
374 /* caller must hold the tree lock */
375 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
378 struct zswap_entry *entry;
380 entry = zswap_rb_search(root, offset);
382 zswap_entry_get(entry);
387 /*********************************
389 **********************************/
390 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
392 static int zswap_dstmem_prepare(unsigned int cpu)
396 dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
400 per_cpu(zswap_dstmem, cpu) = dst;
404 static int zswap_dstmem_dead(unsigned int cpu)
408 dst = per_cpu(zswap_dstmem, cpu);
410 per_cpu(zswap_dstmem, cpu) = NULL;
415 static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
417 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
418 struct crypto_comp *tfm;
420 if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu)))
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));
429 *per_cpu_ptr(pool->tfm, cpu) = tfm;
433 static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
435 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
436 struct crypto_comp *tfm;
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;
445 /*********************************
447 **********************************/
449 static struct zswap_pool *__zswap_pool_current(void)
451 struct zswap_pool *pool;
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__);
460 static struct zswap_pool *zswap_pool_current(void)
462 assert_spin_locked(&zswap_pools_lock);
464 return __zswap_pool_current();
467 static struct zswap_pool *zswap_pool_current_get(void)
469 struct zswap_pool *pool;
473 pool = __zswap_pool_current();
474 if (!zswap_pool_get(pool))
482 static struct zswap_pool *zswap_pool_last_get(void)
484 struct zswap_pool *pool, *last = NULL;
488 list_for_each_entry_rcu(pool, &zswap_pools, list)
490 WARN_ONCE(!last && zswap_has_pool,
491 "%s: no page storage pool!\n", __func__);
492 if (!zswap_pool_get(last))
500 /* type and compressor must be null-terminated */
501 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
503 struct zswap_pool *pool;
505 assert_spin_locked(&zswap_pools_lock);
507 list_for_each_entry_rcu(pool, &zswap_pools, list) {
508 if (strcmp(pool->tfm_name, compressor))
510 if (strcmp(zpool_get_type(pool->zpool), type))
512 /* if we can't get it, it's about to be destroyed */
513 if (!zswap_pool_get(pool))
521 static void shrink_worker(struct work_struct *w)
523 struct zswap_pool *pool = container_of(w, typeof(*pool),
526 if (zpool_shrink(pool->zpool, 1, NULL))
527 zswap_reject_reclaim_fail++;
528 zswap_pool_put(pool);
531 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
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;
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
543 if (!strcmp(type, ZSWAP_PARAM_UNSET))
545 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
549 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
553 /* unique name for each pool specifically required by zsmalloc */
554 snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
556 pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
558 pr_err("%s zpool not available\n", type);
561 pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
563 strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
564 pool->tfm = alloc_percpu(struct crypto_comp *);
566 pr_err("percpu alloc failed\n");
570 ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
574 pr_debug("using %s compressor\n", pool->tfm_name);
576 /* being the current pool takes 1 ref; this func expects the
577 * caller to always add the new pool as the current pool
579 kref_init(&pool->kref);
580 INIT_LIST_HEAD(&pool->list);
581 INIT_WORK(&pool->shrink_work, shrink_worker);
583 zswap_pool_debug("created", pool);
588 free_percpu(pool->tfm);
590 zpool_destroy_pool(pool->zpool);
595 static bool zswap_try_pool_create(void)
597 struct zswap_pool *pool;
598 bool has_comp, has_zpool;
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);
610 pr_err("default compressor %s not available\n",
612 param_free_charp(&zswap_compressor);
613 zswap_compressor = ZSWAP_PARAM_UNSET;
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);
626 pr_err("default zpool %s not available\n",
628 param_free_charp(&zswap_zpool_type);
629 zswap_zpool_type = ZSWAP_PARAM_UNSET;
632 if (!has_comp || !has_zpool)
635 pool = zswap_pool_create(zswap_zpool_type, zswap_compressor);
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;
643 pr_err("pool creation failed\n");
644 zswap_enabled = false;
647 return zswap_enabled;
650 static void zswap_pool_destroy(struct zswap_pool *pool)
652 zswap_pool_debug("destroying", pool);
654 cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
655 free_percpu(pool->tfm);
656 zpool_destroy_pool(pool->zpool);
660 static int __must_check zswap_pool_get(struct zswap_pool *pool)
665 return kref_get_unless_zero(&pool->kref);
668 static void __zswap_pool_release(struct work_struct *work)
670 struct zswap_pool *pool = container_of(work, typeof(*pool),
675 /* nobody should have been able to get a kref... */
676 WARN_ON(kref_get_unless_zero(&pool->kref));
678 /* pool is now off zswap_pools list and has no references. */
679 zswap_pool_destroy(pool);
682 static void __zswap_pool_empty(struct kref *kref)
684 struct zswap_pool *pool;
686 pool = container_of(kref, typeof(*pool), kref);
688 spin_lock(&zswap_pools_lock);
690 WARN_ON(pool == zswap_pool_current());
692 list_del_rcu(&pool->list);
694 INIT_WORK(&pool->release_work, __zswap_pool_release);
695 schedule_work(&pool->release_work);
697 spin_unlock(&zswap_pools_lock);
700 static void zswap_pool_put(struct zswap_pool *pool)
702 kref_put(&pool->kref, __zswap_pool_empty);
705 /*********************************
707 **********************************/
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)
713 struct zswap_pool *pool, *put_pool = NULL;
714 char *s = strstrip((char *)val);
717 if (zswap_init_failed) {
718 pr_err("can't set param, initialization failed\n");
722 /* no change required */
723 if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
726 /* if this is load-time (pre-init) param setting,
727 * don't create a pool; that's done during init.
729 if (!zswap_init_started)
730 return param_set_charp(s, kp);
733 if (!zpool_has_pool(s)) {
734 pr_err("zpool %s not available\n", s);
738 } else if (!compressor) {
739 if (!crypto_has_comp(s, 0, 0)) {
740 pr_err("compressor %s not available\n", s);
749 spin_lock(&zswap_pools_lock);
751 pool = zswap_pool_find_get(type, compressor);
753 zswap_pool_debug("using existing", pool);
754 WARN_ON(pool == zswap_pool_current());
755 list_del_rcu(&pool->list);
758 spin_unlock(&zswap_pools_lock);
761 pool = zswap_pool_create(type, compressor);
764 ret = param_set_charp(s, kp);
768 spin_lock(&zswap_pools_lock);
771 put_pool = zswap_pool_current();
772 list_add_rcu(&pool->list, &zswap_pools);
773 zswap_has_pool = true;
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
779 list_add_tail_rcu(&pool->list, &zswap_pools);
783 spin_unlock(&zswap_pools_lock);
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()
793 ret = param_set_charp(s, kp);
796 /* drop the ref from either the old current pool,
797 * or the new pool we failed to add
800 zswap_pool_put(put_pool);
805 static int zswap_compressor_param_set(const char *val,
806 const struct kernel_param *kp)
808 return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
811 static int zswap_zpool_param_set(const char *val,
812 const struct kernel_param *kp)
814 return __zswap_param_set(val, kp, NULL, zswap_compressor);
817 static int zswap_enabled_param_set(const char *val,
818 const struct kernel_param *kp)
822 if (zswap_init_failed) {
823 pr_err("can't enable, initialization failed\n");
827 ret = param_set_bool(val, kp);
828 if (!ret && zswap_enabled && zswap_init_started && !zswap_has_pool)
829 if (!zswap_try_pool_create())
835 /*********************************
837 **********************************/
838 /* return enum for zswap_get_swap_cache_page */
839 enum zswap_get_swap_ret {
841 ZSWAP_SWAPCACHE_EXIST,
842 ZSWAP_SWAPCACHE_FAIL,
846 * zswap_get_swap_cache_page
848 * This is an adaption of read_swap_cache_async()
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.
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
861 static int zswap_get_swap_cache_page(swp_entry_t entry,
862 struct page **retpage)
864 bool page_was_allocated;
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;
871 return ZSWAP_SWAPCACHE_FAIL;
872 return ZSWAP_SWAPCACHE_EXIST;
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.
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
887 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
889 struct zswap_header *zhdr;
890 swp_entry_t swpentry;
891 struct zswap_tree *tree;
893 struct zswap_entry *entry;
895 struct crypto_comp *tfm;
899 struct writeback_control wbc = {
900 .sync_mode = WB_SYNC_NONE,
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);
909 /* find and ref zswap entry */
910 spin_lock(&tree->lock);
911 entry = zswap_entry_find_get(&tree->rbroot, offset);
913 /* entry was invalidated */
914 spin_unlock(&tree->lock);
915 zpool_unmap_handle(pool, handle);
918 spin_unlock(&tree->lock);
919 BUG_ON(offset != entry->offset);
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 */
927 case ZSWAP_SWAPCACHE_EXIST:
928 /* page is already in the swap cache, ignore for now */
933 case ZSWAP_SWAPCACHE_NEW: /* page is locked */
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,
941 put_cpu_ptr(entry->pool->tfm);
944 BUG_ON(dlen != PAGE_SIZE);
946 /* page is up to date */
947 SetPageUptodate(page);
950 /* move it to the tail of the inactive list after end_writeback */
951 SetPageReclaim(page);
953 /* start writeback */
954 __swap_writepage(page, &wbc, end_swap_bio_write);
956 zswap_written_back_pages++;
958 spin_lock(&tree->lock);
959 /* drop local reference */
960 zswap_entry_put(tree, entry);
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
969 if (entry == zswap_rb_search(&tree->rbroot, offset))
970 zswap_entry_put(tree, entry);
971 spin_unlock(&tree->lock);
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
983 spin_lock(&tree->lock);
984 zswap_entry_put(tree, entry);
985 spin_unlock(&tree->lock);
988 zpool_unmap_handle(pool, handle);
992 static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
997 page = (unsigned long *)ptr;
998 for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
999 if (page[pos] != page[0])
1006 static void zswap_fill_page(void *ptr, unsigned long value)
1008 unsigned long *page;
1010 page = (unsigned long *)ptr;
1011 memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
1014 /*********************************
1016 **********************************/
1017 /* attempts to compress and store an single page */
1018 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1021 struct zswap_tree *tree = zswap_trees[type];
1022 struct zswap_entry *entry, *dupentry;
1023 struct crypto_comp *tfm;
1025 unsigned int hlen, dlen = PAGE_SIZE;
1026 unsigned long handle, value;
1029 struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1032 /* THP isn't supported */
1033 if (PageTransHuge(page)) {
1038 if (!zswap_enabled || !tree) {
1043 /* reclaim space if needed */
1044 if (zswap_is_full()) {
1045 struct zswap_pool *pool;
1047 zswap_pool_limit_hit++;
1048 zswap_pool_reached_full = true;
1049 pool = zswap_pool_last_get();
1051 queue_work(shrink_wq, &pool->shrink_work);
1056 if (zswap_pool_reached_full) {
1057 if (!zswap_can_accept()) {
1061 zswap_pool_reached_full = false;
1064 /* allocate entry */
1065 entry = zswap_entry_cache_alloc(GFP_KERNEL);
1067 zswap_reject_kmemcache_fail++;
1072 if (zswap_same_filled_pages_enabled) {
1073 src = kmap_atomic(page);
1074 if (zswap_is_page_same_filled(src, &value)) {
1076 entry->offset = offset;
1078 entry->value = value;
1079 atomic_inc(&zswap_same_filled_pages);
1085 /* if entry is successfully added, it keeps the reference */
1086 entry->pool = zswap_pool_current_get();
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);
1098 put_cpu_ptr(entry->pool->tfm);
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++;
1115 zswap_reject_alloc_fail++;
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);
1124 /* populate entry */
1125 entry->offset = offset;
1126 entry->handle = handle;
1127 entry->length = dlen;
1131 spin_lock(&tree->lock);
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);
1140 } while (ret == -EEXIST);
1141 spin_unlock(&tree->lock);
1144 atomic_inc(&zswap_stored_pages);
1145 zswap_update_total_size();
1150 put_cpu_var(zswap_dstmem);
1151 zswap_pool_put(entry->pool);
1153 zswap_entry_cache_free(entry);
1159 * returns 0 if the page was successfully decompressed
1160 * return -1 on entry not found or error
1162 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1165 struct zswap_tree *tree = zswap_trees[type];
1166 struct zswap_entry *entry;
1167 struct crypto_comp *tfm;
1173 spin_lock(&tree->lock);
1174 entry = zswap_entry_find_get(&tree->rbroot, offset);
1176 /* entry was written back */
1177 spin_unlock(&tree->lock);
1180 spin_unlock(&tree->lock);
1182 if (!entry->length) {
1183 dst = kmap_atomic(page);
1184 zswap_fill_page(dst, entry->value);
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);
1199 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1203 spin_lock(&tree->lock);
1204 zswap_entry_put(tree, entry);
1205 spin_unlock(&tree->lock);
1210 /* frees an entry in zswap */
1211 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1213 struct zswap_tree *tree = zswap_trees[type];
1214 struct zswap_entry *entry;
1217 spin_lock(&tree->lock);
1218 entry = zswap_rb_search(&tree->rbroot, offset);
1220 /* entry was written back */
1221 spin_unlock(&tree->lock);
1225 /* remove from rbtree */
1226 zswap_rb_erase(&tree->rbroot, entry);
1228 /* drop the initial reference from entry creation */
1229 zswap_entry_put(tree, entry);
1231 spin_unlock(&tree->lock);
1234 /* frees all zswap entries for the given swap type */
1235 static void zswap_frontswap_invalidate_area(unsigned type)
1237 struct zswap_tree *tree = zswap_trees[type];
1238 struct zswap_entry *entry, *n;
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);
1250 zswap_trees[type] = NULL;
1253 static void zswap_frontswap_init(unsigned type)
1255 struct zswap_tree *tree;
1257 tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1259 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1263 tree->rbroot = RB_ROOT;
1264 spin_lock_init(&tree->lock);
1265 zswap_trees[type] = tree;
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
1276 /*********************************
1278 **********************************/
1279 #ifdef CONFIG_DEBUG_FS
1280 #include <linux/debugfs.h>
1282 static struct dentry *zswap_debugfs_root;
1284 static int __init zswap_debugfs_init(void)
1286 if (!debugfs_initialized())
1289 zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
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);
1315 static void __exit zswap_debugfs_exit(void)
1317 debugfs_remove_recursive(zswap_debugfs_root);
1320 static int __init zswap_debugfs_init(void)
1325 static void __exit zswap_debugfs_exit(void) { }
1328 /*********************************
1329 * module init and exit
1330 **********************************/
1331 static int __init init_zswap(void)
1335 zswap_init_started = true;
1337 if (zswap_entry_cache_create()) {
1338 pr_err("entry cache creation failed\n");
1342 ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1343 zswap_dstmem_prepare, zswap_dstmem_dead);
1345 pr_err("dstmem alloc failed\n");
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);
1356 shrink_wq = create_workqueue("zswap-shrink");
1360 frontswap_register_ops(&zswap_frontswap_ops);
1361 if (zswap_debugfs_init())
1362 pr_warn("debugfs initialization failed\n");
1365 zswap_try_pool_create();
1370 cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1372 zswap_entry_cache_destroy();
1374 /* if built-in, we aren't unloaded on failure; don't allow use */
1375 zswap_init_failed = true;
1376 zswap_enabled = false;
1379 /* must be late so crypto has time to come up */
1380 late_initcall(init_zswap);
1382 MODULE_LICENSE("GPL");
1383 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1384 MODULE_DESCRIPTION("Compressed cache for swap pages");