1 // SPDX-License-Identifier: GPL-2.0-only
5 * Author: Vitaly Wool <vitaly.wool@konsulko.com>
6 * Copyright (C) 2016, Sony Mobile Communications Inc.
8 * This implementation is based on zbud written by Seth Jennings.
10 * z3fold is an special purpose allocator for storing compressed pages. It
11 * can store up to three compressed pages per page which improves the
12 * compression ratio of zbud while retaining its main concepts (e. g. always
13 * storing an integral number of objects per page) and simplicity.
14 * It still has simple and deterministic reclaim properties that make it
15 * preferable to a higher density approach (with no requirement on integral
16 * number of object per page) when reclaim is used.
18 * As in zbud, pages are divided into "chunks". The size of the chunks is
19 * fixed at compile time and is determined by NCHUNKS_ORDER below.
21 * z3fold doesn't export any API and is meant to be used via zpool API.
24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
26 #include <linux/atomic.h>
27 #include <linux/sched.h>
28 #include <linux/cpumask.h>
29 #include <linux/list.h>
31 #include <linux/module.h>
32 #include <linux/page-flags.h>
33 #include <linux/migrate.h>
34 #include <linux/node.h>
35 #include <linux/compaction.h>
36 #include <linux/percpu.h>
37 #include <linux/mount.h>
38 #include <linux/pseudo_fs.h>
40 #include <linux/preempt.h>
41 #include <linux/workqueue.h>
42 #include <linux/slab.h>
43 #include <linux/spinlock.h>
44 #include <linux/zpool.h>
45 #include <linux/magic.h>
46 #include <linux/kmemleak.h>
49 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
50 * adjusting internal fragmentation. It also determines the number of
51 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
52 * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks
53 * in the beginning of an allocated page are occupied by z3fold header, so
54 * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y),
55 * which shows the max number of free chunks in z3fold page, also there will
56 * be 63, or 62, respectively, freelists per pool.
58 #define NCHUNKS_ORDER 6
60 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER)
61 #define CHUNK_SIZE (1 << CHUNK_SHIFT)
62 #define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE)
63 #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT)
64 #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SHIFT)
65 #define NCHUNKS (TOTAL_CHUNKS - ZHDR_CHUNKS)
67 #define BUDDY_MASK (0x3)
69 #define SLOTS_ALIGN (0x40)
76 int (*evict)(struct z3fold_pool *pool, unsigned long handle);
87 struct z3fold_buddy_slots {
89 * we are using BUDDY_MASK in handle_to_buddy etc. so there should
90 * be enough slots to hold all possible variants
92 unsigned long slot[BUDDY_MASK + 1];
93 unsigned long pool; /* back link */
96 #define HANDLE_FLAG_MASK (0x03)
99 * struct z3fold_header - z3fold page metadata occupying first chunks of each
100 * z3fold page, except for HEADLESS pages
101 * @buddy: links the z3fold page into the relevant list in the
103 * @page_lock: per-page lock
104 * @refcount: reference count for the z3fold page
105 * @work: work_struct for page layout optimization
106 * @slots: pointer to the structure holding buddy slots
107 * @pool: pointer to the containing pool
108 * @cpu: CPU which this page "belongs" to
109 * @first_chunks: the size of the first buddy in chunks, 0 if free
110 * @middle_chunks: the size of the middle buddy in chunks, 0 if free
111 * @last_chunks: the size of the last buddy in chunks, 0 if free
112 * @first_num: the starting number (for the first handle)
113 * @mapped_count: the number of objects currently mapped
115 struct z3fold_header {
116 struct list_head buddy;
117 spinlock_t page_lock;
118 struct kref refcount;
119 struct work_struct work;
120 struct z3fold_buddy_slots *slots;
121 struct z3fold_pool *pool;
123 unsigned short first_chunks;
124 unsigned short middle_chunks;
125 unsigned short last_chunks;
126 unsigned short start_middle;
127 unsigned short first_num:2;
128 unsigned short mapped_count:2;
129 unsigned short foreign_handles:2;
133 * struct z3fold_pool - stores metadata for each z3fold pool
135 * @lock: protects pool unbuddied/lru lists
136 * @stale_lock: protects pool stale page list
137 * @unbuddied: per-cpu array of lists tracking z3fold pages that contain 2-
138 * buddies; the list each z3fold page is added to depends on
139 * the size of its free region.
140 * @lru: list tracking the z3fold pages in LRU order by most recently
142 * @stale: list of pages marked for freeing
143 * @pages_nr: number of z3fold pages in the pool.
144 * @c_handle: cache for z3fold_buddy_slots allocation
145 * @ops: pointer to a structure of user defined operations specified at
146 * pool creation time.
147 * @compact_wq: workqueue for page layout background optimization
148 * @release_wq: workqueue for safe page release
149 * @work: work_struct for safe page release
150 * @inode: inode for z3fold pseudo filesystem
152 * This structure is allocated at pool creation time and maintains metadata
153 * pertaining to a particular z3fold pool.
158 spinlock_t stale_lock;
159 struct list_head *unbuddied;
160 struct list_head lru;
161 struct list_head stale;
163 struct kmem_cache *c_handle;
164 const struct z3fold_ops *ops;
166 const struct zpool_ops *zpool_ops;
167 struct workqueue_struct *compact_wq;
168 struct workqueue_struct *release_wq;
169 struct work_struct work;
174 * Internal z3fold page flags
176 enum z3fold_page_flags {
181 PAGE_CLAIMED, /* by either reclaim or free */
185 * handle flags, go under HANDLE_FLAG_MASK
187 enum z3fold_handle_flags {
192 * Forward declarations
194 static struct z3fold_header *__z3fold_alloc(struct z3fold_pool *, size_t, bool);
195 static void compact_page_work(struct work_struct *w);
201 /* Converts an allocation size in bytes to size in z3fold chunks */
202 static int size_to_chunks(size_t size)
204 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
207 #define for_each_unbuddied_list(_iter, _begin) \
208 for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
210 static inline struct z3fold_buddy_slots *alloc_slots(struct z3fold_pool *pool,
213 struct z3fold_buddy_slots *slots;
215 slots = kmem_cache_zalloc(pool->c_handle,
216 (gfp & ~(__GFP_HIGHMEM | __GFP_MOVABLE)));
219 /* It will be freed separately in free_handle(). */
220 kmemleak_not_leak(slots);
221 slots->pool = (unsigned long)pool;
222 rwlock_init(&slots->lock);
228 static inline struct z3fold_pool *slots_to_pool(struct z3fold_buddy_slots *s)
230 return (struct z3fold_pool *)(s->pool & ~HANDLE_FLAG_MASK);
233 static inline struct z3fold_buddy_slots *handle_to_slots(unsigned long handle)
235 return (struct z3fold_buddy_slots *)(handle & ~(SLOTS_ALIGN - 1));
238 /* Lock a z3fold page */
239 static inline void z3fold_page_lock(struct z3fold_header *zhdr)
241 spin_lock(&zhdr->page_lock);
244 /* Try to lock a z3fold page */
245 static inline int z3fold_page_trylock(struct z3fold_header *zhdr)
247 return spin_trylock(&zhdr->page_lock);
250 /* Unlock a z3fold page */
251 static inline void z3fold_page_unlock(struct z3fold_header *zhdr)
253 spin_unlock(&zhdr->page_lock);
256 /* return locked z3fold page if it's not headless */
257 static inline struct z3fold_header *get_z3fold_header(unsigned long handle)
259 struct z3fold_buddy_slots *slots;
260 struct z3fold_header *zhdr;
263 if (!(handle & (1 << PAGE_HEADLESS))) {
264 slots = handle_to_slots(handle);
268 read_lock(&slots->lock);
269 addr = *(unsigned long *)handle;
270 zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
271 locked = z3fold_page_trylock(zhdr);
272 read_unlock(&slots->lock);
278 zhdr = (struct z3fold_header *)(handle & PAGE_MASK);
284 static inline void put_z3fold_header(struct z3fold_header *zhdr)
286 struct page *page = virt_to_page(zhdr);
288 if (!test_bit(PAGE_HEADLESS, &page->private))
289 z3fold_page_unlock(zhdr);
292 static inline void free_handle(unsigned long handle, struct z3fold_header *zhdr)
294 struct z3fold_buddy_slots *slots;
298 if (handle & (1 << PAGE_HEADLESS))
301 if (WARN_ON(*(unsigned long *)handle == 0))
304 slots = handle_to_slots(handle);
305 write_lock(&slots->lock);
306 *(unsigned long *)handle = 0;
308 if (test_bit(HANDLES_NOFREE, &slots->pool)) {
309 write_unlock(&slots->lock);
310 return; /* simple case, nothing else to do */
313 if (zhdr->slots != slots)
314 zhdr->foreign_handles--;
317 for (i = 0; i <= BUDDY_MASK; i++) {
318 if (slots->slot[i]) {
323 write_unlock(&slots->lock);
326 struct z3fold_pool *pool = slots_to_pool(slots);
328 if (zhdr->slots == slots)
330 kmem_cache_free(pool->c_handle, slots);
334 static int z3fold_init_fs_context(struct fs_context *fc)
336 return init_pseudo(fc, Z3FOLD_MAGIC) ? 0 : -ENOMEM;
339 static struct file_system_type z3fold_fs = {
341 .init_fs_context = z3fold_init_fs_context,
342 .kill_sb = kill_anon_super,
345 static struct vfsmount *z3fold_mnt;
346 static int z3fold_mount(void)
350 z3fold_mnt = kern_mount(&z3fold_fs);
351 if (IS_ERR(z3fold_mnt))
352 ret = PTR_ERR(z3fold_mnt);
357 static void z3fold_unmount(void)
359 kern_unmount(z3fold_mnt);
362 static const struct address_space_operations z3fold_aops;
363 static int z3fold_register_migration(struct z3fold_pool *pool)
365 pool->inode = alloc_anon_inode(z3fold_mnt->mnt_sb);
366 if (IS_ERR(pool->inode)) {
371 pool->inode->i_mapping->private_data = pool;
372 pool->inode->i_mapping->a_ops = &z3fold_aops;
376 static void z3fold_unregister_migration(struct z3fold_pool *pool)
382 /* Initializes the z3fold header of a newly allocated z3fold page */
383 static struct z3fold_header *init_z3fold_page(struct page *page, bool headless,
384 struct z3fold_pool *pool, gfp_t gfp)
386 struct z3fold_header *zhdr = page_address(page);
387 struct z3fold_buddy_slots *slots;
389 INIT_LIST_HEAD(&page->lru);
390 clear_bit(PAGE_HEADLESS, &page->private);
391 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
392 clear_bit(NEEDS_COMPACTING, &page->private);
393 clear_bit(PAGE_STALE, &page->private);
394 clear_bit(PAGE_CLAIMED, &page->private);
398 slots = alloc_slots(pool, gfp);
402 memset(zhdr, 0, sizeof(*zhdr));
403 spin_lock_init(&zhdr->page_lock);
404 kref_init(&zhdr->refcount);
408 INIT_LIST_HEAD(&zhdr->buddy);
409 INIT_WORK(&zhdr->work, compact_page_work);
413 /* Resets the struct page fields and frees the page */
414 static void free_z3fold_page(struct page *page, bool headless)
418 __ClearPageMovable(page);
421 ClearPagePrivate(page);
425 /* Helper function to build the index */
426 static inline int __idx(struct z3fold_header *zhdr, enum buddy bud)
428 return (bud + zhdr->first_num) & BUDDY_MASK;
432 * Encodes the handle of a particular buddy within a z3fold page
433 * Pool lock should be held as this function accesses first_num
435 static unsigned long __encode_handle(struct z3fold_header *zhdr,
436 struct z3fold_buddy_slots *slots,
439 unsigned long h = (unsigned long)zhdr;
443 * For a headless page, its handle is its pointer with the extra
444 * PAGE_HEADLESS bit set
447 return h | (1 << PAGE_HEADLESS);
449 /* otherwise, return pointer to encoded handle */
450 idx = __idx(zhdr, bud);
453 h |= (zhdr->last_chunks << BUDDY_SHIFT);
455 write_lock(&slots->lock);
456 slots->slot[idx] = h;
457 write_unlock(&slots->lock);
458 return (unsigned long)&slots->slot[idx];
461 static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud)
463 return __encode_handle(zhdr, zhdr->slots, bud);
466 /* only for LAST bud, returns zero otherwise */
467 static unsigned short handle_to_chunks(unsigned long handle)
469 struct z3fold_buddy_slots *slots = handle_to_slots(handle);
472 read_lock(&slots->lock);
473 addr = *(unsigned long *)handle;
474 read_unlock(&slots->lock);
475 return (addr & ~PAGE_MASK) >> BUDDY_SHIFT;
479 * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle
480 * but that doesn't matter. because the masking will result in the
481 * correct buddy number.
483 static enum buddy handle_to_buddy(unsigned long handle)
485 struct z3fold_header *zhdr;
486 struct z3fold_buddy_slots *slots = handle_to_slots(handle);
489 read_lock(&slots->lock);
490 WARN_ON(handle & (1 << PAGE_HEADLESS));
491 addr = *(unsigned long *)handle;
492 read_unlock(&slots->lock);
493 zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
494 return (addr - zhdr->first_num) & BUDDY_MASK;
497 static inline struct z3fold_pool *zhdr_to_pool(struct z3fold_header *zhdr)
502 static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked)
504 struct page *page = virt_to_page(zhdr);
505 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
507 WARN_ON(!list_empty(&zhdr->buddy));
508 set_bit(PAGE_STALE, &page->private);
509 clear_bit(NEEDS_COMPACTING, &page->private);
510 spin_lock(&pool->lock);
511 if (!list_empty(&page->lru))
512 list_del_init(&page->lru);
513 spin_unlock(&pool->lock);
516 z3fold_page_unlock(zhdr);
518 spin_lock(&pool->stale_lock);
519 list_add(&zhdr->buddy, &pool->stale);
520 queue_work(pool->release_wq, &pool->work);
521 spin_unlock(&pool->stale_lock);
524 static void release_z3fold_page(struct kref *ref)
526 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
528 __release_z3fold_page(zhdr, false);
531 static void release_z3fold_page_locked(struct kref *ref)
533 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
535 WARN_ON(z3fold_page_trylock(zhdr));
536 __release_z3fold_page(zhdr, true);
539 static void release_z3fold_page_locked_list(struct kref *ref)
541 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
543 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
545 spin_lock(&pool->lock);
546 list_del_init(&zhdr->buddy);
547 spin_unlock(&pool->lock);
549 WARN_ON(z3fold_page_trylock(zhdr));
550 __release_z3fold_page(zhdr, true);
553 static void free_pages_work(struct work_struct *w)
555 struct z3fold_pool *pool = container_of(w, struct z3fold_pool, work);
557 spin_lock(&pool->stale_lock);
558 while (!list_empty(&pool->stale)) {
559 struct z3fold_header *zhdr = list_first_entry(&pool->stale,
560 struct z3fold_header, buddy);
561 struct page *page = virt_to_page(zhdr);
563 list_del(&zhdr->buddy);
564 if (WARN_ON(!test_bit(PAGE_STALE, &page->private)))
566 spin_unlock(&pool->stale_lock);
567 cancel_work_sync(&zhdr->work);
568 free_z3fold_page(page, false);
570 spin_lock(&pool->stale_lock);
572 spin_unlock(&pool->stale_lock);
576 * Returns the number of free chunks in a z3fold page.
577 * NB: can't be used with HEADLESS pages.
579 static int num_free_chunks(struct z3fold_header *zhdr)
583 * If there is a middle object, pick up the bigger free space
584 * either before or after it. Otherwise just subtract the number
585 * of chunks occupied by the first and the last objects.
587 if (zhdr->middle_chunks != 0) {
588 int nfree_before = zhdr->first_chunks ?
589 0 : zhdr->start_middle - ZHDR_CHUNKS;
590 int nfree_after = zhdr->last_chunks ?
592 (zhdr->start_middle + zhdr->middle_chunks);
593 nfree = max(nfree_before, nfree_after);
595 nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
599 /* Add to the appropriate unbuddied list */
600 static inline void add_to_unbuddied(struct z3fold_pool *pool,
601 struct z3fold_header *zhdr)
603 if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 ||
604 zhdr->middle_chunks == 0) {
605 struct list_head *unbuddied;
606 int freechunks = num_free_chunks(zhdr);
609 unbuddied = this_cpu_ptr(pool->unbuddied);
610 spin_lock(&pool->lock);
611 list_add(&zhdr->buddy, &unbuddied[freechunks]);
612 spin_unlock(&pool->lock);
613 zhdr->cpu = smp_processor_id();
618 static inline enum buddy get_free_buddy(struct z3fold_header *zhdr, int chunks)
620 enum buddy bud = HEADLESS;
622 if (zhdr->middle_chunks) {
623 if (!zhdr->first_chunks &&
624 chunks <= zhdr->start_middle - ZHDR_CHUNKS)
626 else if (!zhdr->last_chunks)
629 if (!zhdr->first_chunks)
631 else if (!zhdr->last_chunks)
640 static inline void *mchunk_memmove(struct z3fold_header *zhdr,
641 unsigned short dst_chunk)
644 return memmove(beg + (dst_chunk << CHUNK_SHIFT),
645 beg + (zhdr->start_middle << CHUNK_SHIFT),
646 zhdr->middle_chunks << CHUNK_SHIFT);
649 static inline bool buddy_single(struct z3fold_header *zhdr)
651 return !((zhdr->first_chunks && zhdr->middle_chunks) ||
652 (zhdr->first_chunks && zhdr->last_chunks) ||
653 (zhdr->middle_chunks && zhdr->last_chunks));
656 static struct z3fold_header *compact_single_buddy(struct z3fold_header *zhdr)
658 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
660 unsigned long old_handle = 0;
662 struct z3fold_header *new_zhdr = NULL;
663 int first_idx = __idx(zhdr, FIRST);
664 int middle_idx = __idx(zhdr, MIDDLE);
665 int last_idx = __idx(zhdr, LAST);
666 unsigned short *moved_chunks = NULL;
669 * No need to protect slots here -- all the slots are "local" and
670 * the page lock is already taken
672 if (zhdr->first_chunks && zhdr->slots->slot[first_idx]) {
673 p += ZHDR_SIZE_ALIGNED;
674 sz = zhdr->first_chunks << CHUNK_SHIFT;
675 old_handle = (unsigned long)&zhdr->slots->slot[first_idx];
676 moved_chunks = &zhdr->first_chunks;
677 } else if (zhdr->middle_chunks && zhdr->slots->slot[middle_idx]) {
678 p += zhdr->start_middle << CHUNK_SHIFT;
679 sz = zhdr->middle_chunks << CHUNK_SHIFT;
680 old_handle = (unsigned long)&zhdr->slots->slot[middle_idx];
681 moved_chunks = &zhdr->middle_chunks;
682 } else if (zhdr->last_chunks && zhdr->slots->slot[last_idx]) {
683 p += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
684 sz = zhdr->last_chunks << CHUNK_SHIFT;
685 old_handle = (unsigned long)&zhdr->slots->slot[last_idx];
686 moved_chunks = &zhdr->last_chunks;
690 enum buddy new_bud = HEADLESS;
691 short chunks = size_to_chunks(sz);
694 new_zhdr = __z3fold_alloc(pool, sz, false);
698 if (WARN_ON(new_zhdr == zhdr))
701 new_bud = get_free_buddy(new_zhdr, chunks);
705 new_zhdr->first_chunks = chunks;
706 q += ZHDR_SIZE_ALIGNED;
709 new_zhdr->middle_chunks = chunks;
710 new_zhdr->start_middle =
711 new_zhdr->first_chunks + ZHDR_CHUNKS;
712 q += new_zhdr->start_middle << CHUNK_SHIFT;
715 new_zhdr->last_chunks = chunks;
716 q += PAGE_SIZE - (new_zhdr->last_chunks << CHUNK_SHIFT);
721 new_zhdr->foreign_handles++;
723 write_lock(&zhdr->slots->lock);
724 *(unsigned long *)old_handle = (unsigned long)new_zhdr +
725 __idx(new_zhdr, new_bud);
727 *(unsigned long *)old_handle |=
728 (new_zhdr->last_chunks << BUDDY_SHIFT);
729 write_unlock(&zhdr->slots->lock);
730 add_to_unbuddied(pool, new_zhdr);
731 z3fold_page_unlock(new_zhdr);
740 if (kref_put(&new_zhdr->refcount, release_z3fold_page_locked))
741 atomic64_dec(&pool->pages_nr);
743 add_to_unbuddied(pool, new_zhdr);
744 z3fold_page_unlock(new_zhdr);
751 #define BIG_CHUNK_GAP 3
752 /* Has to be called with lock held */
753 static int z3fold_compact_page(struct z3fold_header *zhdr)
755 struct page *page = virt_to_page(zhdr);
757 if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private))
758 return 0; /* can't move middle chunk, it's used */
760 if (unlikely(PageIsolated(page)))
763 if (zhdr->middle_chunks == 0)
764 return 0; /* nothing to compact */
766 if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
767 /* move to the beginning */
768 mchunk_memmove(zhdr, ZHDR_CHUNKS);
769 zhdr->first_chunks = zhdr->middle_chunks;
770 zhdr->middle_chunks = 0;
771 zhdr->start_middle = 0;
777 * moving data is expensive, so let's only do that if
778 * there's substantial gain (at least BIG_CHUNK_GAP chunks)
780 if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 &&
781 zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >=
783 mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS);
784 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
786 } else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 &&
787 TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle
788 + zhdr->middle_chunks) >=
790 unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks -
792 mchunk_memmove(zhdr, new_start);
793 zhdr->start_middle = new_start;
800 static void do_compact_page(struct z3fold_header *zhdr, bool locked)
802 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
805 page = virt_to_page(zhdr);
807 WARN_ON(z3fold_page_trylock(zhdr));
809 z3fold_page_lock(zhdr);
810 if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) {
811 z3fold_page_unlock(zhdr);
814 spin_lock(&pool->lock);
815 list_del_init(&zhdr->buddy);
816 spin_unlock(&pool->lock);
818 if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) {
819 atomic64_dec(&pool->pages_nr);
823 if (test_bit(PAGE_STALE, &page->private) ||
824 test_and_set_bit(PAGE_CLAIMED, &page->private)) {
825 z3fold_page_unlock(zhdr);
829 if (!zhdr->foreign_handles && buddy_single(zhdr) &&
830 zhdr->mapped_count == 0 && compact_single_buddy(zhdr)) {
831 if (kref_put(&zhdr->refcount, release_z3fold_page_locked))
832 atomic64_dec(&pool->pages_nr);
834 clear_bit(PAGE_CLAIMED, &page->private);
835 z3fold_page_unlock(zhdr);
840 z3fold_compact_page(zhdr);
841 add_to_unbuddied(pool, zhdr);
842 clear_bit(PAGE_CLAIMED, &page->private);
843 z3fold_page_unlock(zhdr);
846 static void compact_page_work(struct work_struct *w)
848 struct z3fold_header *zhdr = container_of(w, struct z3fold_header,
851 do_compact_page(zhdr, false);
854 /* returns _locked_ z3fold page header or NULL */
855 static inline struct z3fold_header *__z3fold_alloc(struct z3fold_pool *pool,
856 size_t size, bool can_sleep)
858 struct z3fold_header *zhdr = NULL;
860 struct list_head *unbuddied;
861 int chunks = size_to_chunks(size), i;
865 /* First, try to find an unbuddied z3fold page. */
866 unbuddied = this_cpu_ptr(pool->unbuddied);
867 for_each_unbuddied_list(i, chunks) {
868 struct list_head *l = &unbuddied[i];
870 zhdr = list_first_entry_or_null(READ_ONCE(l),
871 struct z3fold_header, buddy);
876 /* Re-check under lock. */
877 spin_lock(&pool->lock);
879 if (unlikely(zhdr != list_first_entry(READ_ONCE(l),
880 struct z3fold_header, buddy)) ||
881 !z3fold_page_trylock(zhdr)) {
882 spin_unlock(&pool->lock);
889 list_del_init(&zhdr->buddy);
891 spin_unlock(&pool->lock);
893 page = virt_to_page(zhdr);
894 if (test_bit(NEEDS_COMPACTING, &page->private) ||
895 test_bit(PAGE_CLAIMED, &page->private)) {
896 z3fold_page_unlock(zhdr);
905 * this page could not be removed from its unbuddied
906 * list while pool lock was held, and then we've taken
907 * page lock so kref_put could not be called before
908 * we got here, so it's safe to just call kref_get()
910 kref_get(&zhdr->refcount);
918 /* look for _exact_ match on other cpus' lists */
919 for_each_online_cpu(cpu) {
922 unbuddied = per_cpu_ptr(pool->unbuddied, cpu);
923 spin_lock(&pool->lock);
924 l = &unbuddied[chunks];
926 zhdr = list_first_entry_or_null(READ_ONCE(l),
927 struct z3fold_header, buddy);
929 if (!zhdr || !z3fold_page_trylock(zhdr)) {
930 spin_unlock(&pool->lock);
934 list_del_init(&zhdr->buddy);
936 spin_unlock(&pool->lock);
938 page = virt_to_page(zhdr);
939 if (test_bit(NEEDS_COMPACTING, &page->private) ||
940 test_bit(PAGE_CLAIMED, &page->private)) {
941 z3fold_page_unlock(zhdr);
947 kref_get(&zhdr->refcount);
952 if (zhdr && !zhdr->slots)
953 zhdr->slots = alloc_slots(pool,
954 can_sleep ? GFP_NOIO : GFP_ATOMIC);
963 * z3fold_create_pool() - create a new z3fold pool
965 * @gfp: gfp flags when allocating the z3fold pool structure
966 * @ops: user-defined operations for the z3fold pool
968 * Return: pointer to the new z3fold pool or NULL if the metadata allocation
971 static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp,
972 const struct z3fold_ops *ops)
974 struct z3fold_pool *pool = NULL;
977 pool = kzalloc(sizeof(struct z3fold_pool), gfp);
980 pool->c_handle = kmem_cache_create("z3fold_handle",
981 sizeof(struct z3fold_buddy_slots),
982 SLOTS_ALIGN, 0, NULL);
985 spin_lock_init(&pool->lock);
986 spin_lock_init(&pool->stale_lock);
987 pool->unbuddied = __alloc_percpu(sizeof(struct list_head) * NCHUNKS,
988 __alignof__(struct list_head));
989 if (!pool->unbuddied)
991 for_each_possible_cpu(cpu) {
992 struct list_head *unbuddied =
993 per_cpu_ptr(pool->unbuddied, cpu);
994 for_each_unbuddied_list(i, 0)
995 INIT_LIST_HEAD(&unbuddied[i]);
997 INIT_LIST_HEAD(&pool->lru);
998 INIT_LIST_HEAD(&pool->stale);
999 atomic64_set(&pool->pages_nr, 0);
1001 pool->compact_wq = create_singlethread_workqueue(pool->name);
1002 if (!pool->compact_wq)
1004 pool->release_wq = create_singlethread_workqueue(pool->name);
1005 if (!pool->release_wq)
1007 if (z3fold_register_migration(pool))
1009 INIT_WORK(&pool->work, free_pages_work);
1014 destroy_workqueue(pool->release_wq);
1016 destroy_workqueue(pool->compact_wq);
1018 free_percpu(pool->unbuddied);
1020 kmem_cache_destroy(pool->c_handle);
1028 * z3fold_destroy_pool() - destroys an existing z3fold pool
1029 * @pool: the z3fold pool to be destroyed
1031 * The pool should be emptied before this function is called.
1033 static void z3fold_destroy_pool(struct z3fold_pool *pool)
1035 kmem_cache_destroy(pool->c_handle);
1038 * We need to destroy pool->compact_wq before pool->release_wq,
1039 * as any pending work on pool->compact_wq will call
1040 * queue_work(pool->release_wq, &pool->work).
1042 * There are still outstanding pages until both workqueues are drained,
1043 * so we cannot unregister migration until then.
1046 destroy_workqueue(pool->compact_wq);
1047 destroy_workqueue(pool->release_wq);
1048 z3fold_unregister_migration(pool);
1053 * z3fold_alloc() - allocates a region of a given size
1054 * @pool: z3fold pool from which to allocate
1055 * @size: size in bytes of the desired allocation
1056 * @gfp: gfp flags used if the pool needs to grow
1057 * @handle: handle of the new allocation
1059 * This function will attempt to find a free region in the pool large enough to
1060 * satisfy the allocation request. A search of the unbuddied lists is
1061 * performed first. If no suitable free region is found, then a new page is
1062 * allocated and added to the pool to satisfy the request.
1064 * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used
1065 * as z3fold pool pages.
1067 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
1068 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
1071 static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
1072 unsigned long *handle)
1074 int chunks = size_to_chunks(size);
1075 struct z3fold_header *zhdr = NULL;
1076 struct page *page = NULL;
1078 bool can_sleep = gfpflags_allow_blocking(gfp);
1083 if (size > PAGE_SIZE)
1086 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
1090 zhdr = __z3fold_alloc(pool, size, can_sleep);
1092 bud = get_free_buddy(zhdr, chunks);
1093 if (bud == HEADLESS) {
1094 if (kref_put(&zhdr->refcount,
1095 release_z3fold_page_locked))
1096 atomic64_dec(&pool->pages_nr);
1098 z3fold_page_unlock(zhdr);
1099 pr_err("No free chunks in unbuddied\n");
1103 page = virt_to_page(zhdr);
1111 spin_lock(&pool->stale_lock);
1112 zhdr = list_first_entry_or_null(&pool->stale,
1113 struct z3fold_header, buddy);
1115 * Before allocating a page, let's see if we can take one from
1116 * the stale pages list. cancel_work_sync() can sleep so we
1117 * limit this case to the contexts where we can sleep
1120 list_del(&zhdr->buddy);
1121 spin_unlock(&pool->stale_lock);
1122 cancel_work_sync(&zhdr->work);
1123 page = virt_to_page(zhdr);
1125 spin_unlock(&pool->stale_lock);
1129 page = alloc_page(gfp);
1134 zhdr = init_z3fold_page(page, bud == HEADLESS, pool, gfp);
1139 atomic64_inc(&pool->pages_nr);
1141 if (bud == HEADLESS) {
1142 set_bit(PAGE_HEADLESS, &page->private);
1147 __SetPageMovable(page, pool->inode->i_mapping);
1150 if (trylock_page(page)) {
1151 __SetPageMovable(page, pool->inode->i_mapping);
1155 z3fold_page_lock(zhdr);
1159 zhdr->first_chunks = chunks;
1160 else if (bud == LAST)
1161 zhdr->last_chunks = chunks;
1163 zhdr->middle_chunks = chunks;
1164 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
1166 add_to_unbuddied(pool, zhdr);
1169 spin_lock(&pool->lock);
1170 /* Add/move z3fold page to beginning of LRU */
1171 if (!list_empty(&page->lru))
1172 list_del(&page->lru);
1174 list_add(&page->lru, &pool->lru);
1176 *handle = encode_handle(zhdr, bud);
1177 spin_unlock(&pool->lock);
1178 if (bud != HEADLESS)
1179 z3fold_page_unlock(zhdr);
1185 * z3fold_free() - frees the allocation associated with the given handle
1186 * @pool: pool in which the allocation resided
1187 * @handle: handle associated with the allocation returned by z3fold_alloc()
1189 * In the case that the z3fold page in which the allocation resides is under
1190 * reclaim, as indicated by the PG_reclaim flag being set, this function
1191 * only sets the first|last_chunks to 0. The page is actually freed
1192 * once both buddies are evicted (see z3fold_reclaim_page() below).
1194 static void z3fold_free(struct z3fold_pool *pool, unsigned long handle)
1196 struct z3fold_header *zhdr;
1201 zhdr = get_z3fold_header(handle);
1202 page = virt_to_page(zhdr);
1203 page_claimed = test_and_set_bit(PAGE_CLAIMED, &page->private);
1205 if (test_bit(PAGE_HEADLESS, &page->private)) {
1206 /* if a headless page is under reclaim, just leave.
1207 * NB: we use test_and_set_bit for a reason: if the bit
1208 * has not been set before, we release this page
1209 * immediately so we don't care about its value any more.
1211 if (!page_claimed) {
1212 spin_lock(&pool->lock);
1213 list_del(&page->lru);
1214 spin_unlock(&pool->lock);
1215 put_z3fold_header(zhdr);
1216 free_z3fold_page(page, true);
1217 atomic64_dec(&pool->pages_nr);
1222 /* Non-headless case */
1223 bud = handle_to_buddy(handle);
1227 zhdr->first_chunks = 0;
1230 zhdr->middle_chunks = 0;
1233 zhdr->last_chunks = 0;
1236 pr_err("%s: unknown bud %d\n", __func__, bud);
1238 put_z3fold_header(zhdr);
1243 free_handle(handle, zhdr);
1244 if (kref_put(&zhdr->refcount, release_z3fold_page_locked_list)) {
1245 atomic64_dec(&pool->pages_nr);
1249 /* the page has not been claimed by us */
1250 z3fold_page_unlock(zhdr);
1253 if (test_and_set_bit(NEEDS_COMPACTING, &page->private)) {
1254 put_z3fold_header(zhdr);
1255 clear_bit(PAGE_CLAIMED, &page->private);
1258 if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) {
1259 spin_lock(&pool->lock);
1260 list_del_init(&zhdr->buddy);
1261 spin_unlock(&pool->lock);
1263 kref_get(&zhdr->refcount);
1264 clear_bit(PAGE_CLAIMED, &page->private);
1265 do_compact_page(zhdr, true);
1268 kref_get(&zhdr->refcount);
1269 clear_bit(PAGE_CLAIMED, &page->private);
1270 queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work);
1271 put_z3fold_header(zhdr);
1275 * z3fold_reclaim_page() - evicts allocations from a pool page and frees it
1276 * @pool: pool from which a page will attempt to be evicted
1277 * @retries: number of pages on the LRU list for which eviction will
1278 * be attempted before failing
1280 * z3fold reclaim is different from normal system reclaim in that it is done
1281 * from the bottom, up. This is because only the bottom layer, z3fold, has
1282 * information on how the allocations are organized within each z3fold page.
1283 * This has the potential to create interesting locking situations between
1284 * z3fold and the user, however.
1286 * To avoid these, this is how z3fold_reclaim_page() should be called:
1288 * The user detects a page should be reclaimed and calls z3fold_reclaim_page().
1289 * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and
1290 * call the user-defined eviction handler with the pool and handle as
1293 * If the handle can not be evicted, the eviction handler should return
1294 * non-zero. z3fold_reclaim_page() will add the z3fold page back to the
1295 * appropriate list and try the next z3fold page on the LRU up to
1296 * a user defined number of retries.
1298 * If the handle is successfully evicted, the eviction handler should
1299 * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free()
1300 * contains logic to delay freeing the page if the page is under reclaim,
1301 * as indicated by the setting of the PG_reclaim flag on the underlying page.
1303 * If all buddies in the z3fold page are successfully evicted, then the
1304 * z3fold page can be freed.
1306 * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are
1307 * no pages to evict or an eviction handler is not registered, -EAGAIN if
1308 * the retry limit was hit.
1310 static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries)
1313 struct z3fold_header *zhdr = NULL;
1314 struct page *page = NULL;
1315 struct list_head *pos;
1316 unsigned long first_handle = 0, middle_handle = 0, last_handle = 0;
1317 struct z3fold_buddy_slots slots __attribute__((aligned(SLOTS_ALIGN)));
1319 rwlock_init(&slots.lock);
1320 slots.pool = (unsigned long)pool | (1 << HANDLES_NOFREE);
1322 spin_lock(&pool->lock);
1323 if (!pool->ops || !pool->ops->evict || retries == 0) {
1324 spin_unlock(&pool->lock);
1327 for (i = 0; i < retries; i++) {
1328 if (list_empty(&pool->lru)) {
1329 spin_unlock(&pool->lock);
1332 list_for_each_prev(pos, &pool->lru) {
1333 page = list_entry(pos, struct page, lru);
1335 zhdr = page_address(page);
1336 if (test_bit(PAGE_HEADLESS, &page->private)) {
1338 * For non-headless pages, we wait to do this
1339 * until we have the page lock to avoid racing
1340 * with __z3fold_alloc(). Headless pages don't
1341 * have a lock (and __z3fold_alloc() will never
1342 * see them), but we still need to test and set
1343 * PAGE_CLAIMED to avoid racing with
1344 * z3fold_free(), so just do it now before
1347 if (test_and_set_bit(PAGE_CLAIMED, &page->private))
1353 if (kref_get_unless_zero(&zhdr->refcount) == 0) {
1357 if (!z3fold_page_trylock(zhdr)) {
1358 if (kref_put(&zhdr->refcount,
1359 release_z3fold_page))
1360 atomic64_dec(&pool->pages_nr);
1362 continue; /* can't evict at this point */
1365 /* test_and_set_bit is of course atomic, but we still
1366 * need to do it under page lock, otherwise checking
1367 * that bit in __z3fold_alloc wouldn't make sense
1369 if (zhdr->foreign_handles ||
1370 test_and_set_bit(PAGE_CLAIMED, &page->private)) {
1371 if (kref_put(&zhdr->refcount,
1372 release_z3fold_page))
1373 atomic64_dec(&pool->pages_nr);
1375 z3fold_page_unlock(zhdr);
1377 continue; /* can't evict such page */
1379 list_del_init(&zhdr->buddy);
1387 list_del_init(&page->lru);
1388 spin_unlock(&pool->lock);
1390 if (!test_bit(PAGE_HEADLESS, &page->private)) {
1392 * We need encode the handles before unlocking, and
1393 * use our local slots structure because z3fold_free
1394 * can zero out zhdr->slots and we can't do much
1400 memset(slots.slot, 0, sizeof(slots.slot));
1401 if (zhdr->first_chunks)
1402 first_handle = __encode_handle(zhdr, &slots,
1404 if (zhdr->middle_chunks)
1405 middle_handle = __encode_handle(zhdr, &slots,
1407 if (zhdr->last_chunks)
1408 last_handle = __encode_handle(zhdr, &slots,
1411 * it's safe to unlock here because we hold a
1412 * reference to this page
1414 z3fold_page_unlock(zhdr);
1416 first_handle = encode_handle(zhdr, HEADLESS);
1417 last_handle = middle_handle = 0;
1419 /* Issue the eviction callback(s) */
1420 if (middle_handle) {
1421 ret = pool->ops->evict(pool, middle_handle);
1426 ret = pool->ops->evict(pool, first_handle);
1431 ret = pool->ops->evict(pool, last_handle);
1436 if (test_bit(PAGE_HEADLESS, &page->private)) {
1438 free_z3fold_page(page, true);
1439 atomic64_dec(&pool->pages_nr);
1442 spin_lock(&pool->lock);
1443 list_add(&page->lru, &pool->lru);
1444 spin_unlock(&pool->lock);
1445 clear_bit(PAGE_CLAIMED, &page->private);
1447 struct z3fold_buddy_slots *slots = zhdr->slots;
1448 z3fold_page_lock(zhdr);
1449 if (kref_put(&zhdr->refcount,
1450 release_z3fold_page_locked)) {
1451 kmem_cache_free(pool->c_handle, slots);
1452 atomic64_dec(&pool->pages_nr);
1456 * if we are here, the page is still not completely
1457 * free. Take the global pool lock then to be able
1458 * to add it back to the lru list
1460 spin_lock(&pool->lock);
1461 list_add(&page->lru, &pool->lru);
1462 spin_unlock(&pool->lock);
1463 z3fold_page_unlock(zhdr);
1464 clear_bit(PAGE_CLAIMED, &page->private);
1467 /* We started off locked to we need to lock the pool back */
1468 spin_lock(&pool->lock);
1470 spin_unlock(&pool->lock);
1475 * z3fold_map() - maps the allocation associated with the given handle
1476 * @pool: pool in which the allocation resides
1477 * @handle: handle associated with the allocation to be mapped
1479 * Extracts the buddy number from handle and constructs the pointer to the
1480 * correct starting chunk within the page.
1482 * Returns: a pointer to the mapped allocation
1484 static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle)
1486 struct z3fold_header *zhdr;
1491 zhdr = get_z3fold_header(handle);
1493 page = virt_to_page(zhdr);
1495 if (test_bit(PAGE_HEADLESS, &page->private))
1498 buddy = handle_to_buddy(handle);
1501 addr += ZHDR_SIZE_ALIGNED;
1504 addr += zhdr->start_middle << CHUNK_SHIFT;
1505 set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1508 addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT);
1511 pr_err("unknown buddy id %d\n", buddy);
1518 zhdr->mapped_count++;
1520 put_z3fold_header(zhdr);
1525 * z3fold_unmap() - unmaps the allocation associated with the given handle
1526 * @pool: pool in which the allocation resides
1527 * @handle: handle associated with the allocation to be unmapped
1529 static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle)
1531 struct z3fold_header *zhdr;
1535 zhdr = get_z3fold_header(handle);
1536 page = virt_to_page(zhdr);
1538 if (test_bit(PAGE_HEADLESS, &page->private))
1541 buddy = handle_to_buddy(handle);
1542 if (buddy == MIDDLE)
1543 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1544 zhdr->mapped_count--;
1545 put_z3fold_header(zhdr);
1549 * z3fold_get_pool_size() - gets the z3fold pool size in pages
1550 * @pool: pool whose size is being queried
1552 * Returns: size in pages of the given pool.
1554 static u64 z3fold_get_pool_size(struct z3fold_pool *pool)
1556 return atomic64_read(&pool->pages_nr);
1559 static bool z3fold_page_isolate(struct page *page, isolate_mode_t mode)
1561 struct z3fold_header *zhdr;
1562 struct z3fold_pool *pool;
1564 VM_BUG_ON_PAGE(!PageMovable(page), page);
1565 VM_BUG_ON_PAGE(PageIsolated(page), page);
1567 if (test_bit(PAGE_HEADLESS, &page->private))
1570 zhdr = page_address(page);
1571 z3fold_page_lock(zhdr);
1572 if (test_bit(NEEDS_COMPACTING, &page->private) ||
1573 test_bit(PAGE_STALE, &page->private))
1576 if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0)
1579 if (test_and_set_bit(PAGE_CLAIMED, &page->private))
1581 pool = zhdr_to_pool(zhdr);
1582 spin_lock(&pool->lock);
1583 if (!list_empty(&zhdr->buddy))
1584 list_del_init(&zhdr->buddy);
1585 if (!list_empty(&page->lru))
1586 list_del_init(&page->lru);
1587 spin_unlock(&pool->lock);
1589 kref_get(&zhdr->refcount);
1590 z3fold_page_unlock(zhdr);
1594 z3fold_page_unlock(zhdr);
1598 static int z3fold_page_migrate(struct address_space *mapping, struct page *newpage,
1599 struct page *page, enum migrate_mode mode)
1601 struct z3fold_header *zhdr, *new_zhdr;
1602 struct z3fold_pool *pool;
1603 struct address_space *new_mapping;
1605 VM_BUG_ON_PAGE(!PageMovable(page), page);
1606 VM_BUG_ON_PAGE(!PageIsolated(page), page);
1607 VM_BUG_ON_PAGE(!test_bit(PAGE_CLAIMED, &page->private), page);
1608 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
1610 zhdr = page_address(page);
1611 pool = zhdr_to_pool(zhdr);
1613 if (!z3fold_page_trylock(zhdr))
1615 if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0) {
1616 z3fold_page_unlock(zhdr);
1617 clear_bit(PAGE_CLAIMED, &page->private);
1620 if (work_pending(&zhdr->work)) {
1621 z3fold_page_unlock(zhdr);
1624 new_zhdr = page_address(newpage);
1625 memcpy(new_zhdr, zhdr, PAGE_SIZE);
1626 newpage->private = page->private;
1628 z3fold_page_unlock(zhdr);
1629 spin_lock_init(&new_zhdr->page_lock);
1630 INIT_WORK(&new_zhdr->work, compact_page_work);
1632 * z3fold_page_isolate() ensures that new_zhdr->buddy is empty,
1633 * so we only have to reinitialize it.
1635 INIT_LIST_HEAD(&new_zhdr->buddy);
1636 new_mapping = page_mapping(page);
1637 __ClearPageMovable(page);
1638 ClearPagePrivate(page);
1641 z3fold_page_lock(new_zhdr);
1642 if (new_zhdr->first_chunks)
1643 encode_handle(new_zhdr, FIRST);
1644 if (new_zhdr->last_chunks)
1645 encode_handle(new_zhdr, LAST);
1646 if (new_zhdr->middle_chunks)
1647 encode_handle(new_zhdr, MIDDLE);
1648 set_bit(NEEDS_COMPACTING, &newpage->private);
1649 new_zhdr->cpu = smp_processor_id();
1650 spin_lock(&pool->lock);
1651 list_add(&newpage->lru, &pool->lru);
1652 spin_unlock(&pool->lock);
1653 __SetPageMovable(newpage, new_mapping);
1654 z3fold_page_unlock(new_zhdr);
1656 queue_work_on(new_zhdr->cpu, pool->compact_wq, &new_zhdr->work);
1658 page_mapcount_reset(page);
1659 clear_bit(PAGE_CLAIMED, &page->private);
1664 static void z3fold_page_putback(struct page *page)
1666 struct z3fold_header *zhdr;
1667 struct z3fold_pool *pool;
1669 zhdr = page_address(page);
1670 pool = zhdr_to_pool(zhdr);
1672 z3fold_page_lock(zhdr);
1673 if (!list_empty(&zhdr->buddy))
1674 list_del_init(&zhdr->buddy);
1675 INIT_LIST_HEAD(&page->lru);
1676 if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) {
1677 atomic64_dec(&pool->pages_nr);
1680 spin_lock(&pool->lock);
1681 list_add(&page->lru, &pool->lru);
1682 spin_unlock(&pool->lock);
1683 clear_bit(PAGE_CLAIMED, &page->private);
1684 z3fold_page_unlock(zhdr);
1687 static const struct address_space_operations z3fold_aops = {
1688 .isolate_page = z3fold_page_isolate,
1689 .migratepage = z3fold_page_migrate,
1690 .putback_page = z3fold_page_putback,
1697 static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle)
1699 if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict)
1700 return pool->zpool_ops->evict(pool->zpool, handle);
1705 static const struct z3fold_ops z3fold_zpool_ops = {
1706 .evict = z3fold_zpool_evict
1709 static void *z3fold_zpool_create(const char *name, gfp_t gfp,
1710 const struct zpool_ops *zpool_ops,
1711 struct zpool *zpool)
1713 struct z3fold_pool *pool;
1715 pool = z3fold_create_pool(name, gfp,
1716 zpool_ops ? &z3fold_zpool_ops : NULL);
1718 pool->zpool = zpool;
1719 pool->zpool_ops = zpool_ops;
1724 static void z3fold_zpool_destroy(void *pool)
1726 z3fold_destroy_pool(pool);
1729 static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp,
1730 unsigned long *handle)
1732 return z3fold_alloc(pool, size, gfp, handle);
1734 static void z3fold_zpool_free(void *pool, unsigned long handle)
1736 z3fold_free(pool, handle);
1739 static int z3fold_zpool_shrink(void *pool, unsigned int pages,
1740 unsigned int *reclaimed)
1742 unsigned int total = 0;
1745 while (total < pages) {
1746 ret = z3fold_reclaim_page(pool, 8);
1758 static void *z3fold_zpool_map(void *pool, unsigned long handle,
1759 enum zpool_mapmode mm)
1761 return z3fold_map(pool, handle);
1763 static void z3fold_zpool_unmap(void *pool, unsigned long handle)
1765 z3fold_unmap(pool, handle);
1768 static u64 z3fold_zpool_total_size(void *pool)
1770 return z3fold_get_pool_size(pool) * PAGE_SIZE;
1773 static struct zpool_driver z3fold_zpool_driver = {
1775 .sleep_mapped = true,
1776 .owner = THIS_MODULE,
1777 .create = z3fold_zpool_create,
1778 .destroy = z3fold_zpool_destroy,
1779 .malloc = z3fold_zpool_malloc,
1780 .free = z3fold_zpool_free,
1781 .shrink = z3fold_zpool_shrink,
1782 .map = z3fold_zpool_map,
1783 .unmap = z3fold_zpool_unmap,
1784 .total_size = z3fold_zpool_total_size,
1787 MODULE_ALIAS("zpool-z3fold");
1789 static int __init init_z3fold(void)
1794 * Make sure the z3fold header is not larger than the page size and
1795 * there has remaining spaces for its buddy.
1797 BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE - CHUNK_SIZE);
1798 ret = z3fold_mount();
1802 zpool_register_driver(&z3fold_zpool_driver);
1807 static void __exit exit_z3fold(void)
1810 zpool_unregister_driver(&z3fold_zpool_driver);
1813 module_init(init_z3fold);
1814 module_exit(exit_z3fold);
1816 MODULE_LICENSE("GPL");
1817 MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>");
1818 MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages");