4 * Author: Vitaly Wool <vitaly.wool@konsulko.com>
5 * Copyright (C) 2016, Sony Mobile Communications Inc.
7 * This implementation is based on zbud written by Seth Jennings.
9 * z3fold is an special purpose allocator for storing compressed pages. It
10 * can store up to three compressed pages per page which improves the
11 * compression ratio of zbud while retaining its main concepts (e. g. always
12 * storing an integral number of objects per page) and simplicity.
13 * It still has simple and deterministic reclaim properties that make it
14 * preferable to a higher density approach (with no requirement on integral
15 * number of object per page) when reclaim is used.
17 * As in zbud, pages are divided into "chunks". The size of the chunks is
18 * fixed at compile time and is determined by NCHUNKS_ORDER below.
20 * z3fold doesn't export any API and is meant to be used via zpool API.
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25 #include <linux/atomic.h>
26 #include <linux/list.h>
28 #include <linux/module.h>
29 #include <linux/preempt.h>
30 #include <linux/slab.h>
31 #include <linux/spinlock.h>
32 #include <linux/zpool.h>
38 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
39 * adjusting internal fragmentation. It also determines the number of
40 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
41 * allocation granularity will be in chunks of size PAGE_SIZE/64. As one chunk
42 * in allocated page is occupied by z3fold header, NCHUNKS will be calculated
43 * to 63 which shows the max number of free chunks in z3fold page, also there
44 * will be 63 freelists per pool.
46 #define NCHUNKS_ORDER 6
48 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER)
49 #define CHUNK_SIZE (1 << CHUNK_SHIFT)
50 #define ZHDR_SIZE_ALIGNED CHUNK_SIZE
51 #define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT)
53 #define BUDDY_MASK ((1 << NCHUNKS_ORDER) - 1)
57 int (*evict)(struct z3fold_pool *pool, unsigned long handle);
61 * struct z3fold_pool - stores metadata for each z3fold pool
62 * @lock: protects all pool fields and first|last_chunk fields of any
63 * z3fold page in the pool
64 * @unbuddied: array of lists tracking z3fold pages that contain 2- buddies;
65 * the lists each z3fold page is added to depends on the size of
67 * @buddied: list tracking the z3fold pages that contain 3 buddies;
68 * these z3fold pages are full
69 * @lru: list tracking the z3fold pages in LRU order by most recently
71 * @pages_nr: number of z3fold pages in the pool.
72 * @ops: pointer to a structure of user defined operations specified at
75 * This structure is allocated at pool creation time and maintains metadata
76 * pertaining to a particular z3fold pool.
80 struct list_head unbuddied[NCHUNKS];
81 struct list_head buddied;
84 const struct z3fold_ops *ops;
86 const struct zpool_ops *zpool_ops;
98 * struct z3fold_header - z3fold page metadata occupying the first chunk of each
99 * z3fold page, except for HEADLESS pages
100 * @buddy: links the z3fold page into the relevant list in the pool
101 * @first_chunks: the size of the first buddy in chunks, 0 if free
102 * @middle_chunks: the size of the middle buddy in chunks, 0 if free
103 * @last_chunks: the size of the last buddy in chunks, 0 if free
104 * @first_num: the starting number (for the first handle)
106 struct z3fold_header {
107 struct list_head buddy;
108 unsigned short first_chunks;
109 unsigned short middle_chunks;
110 unsigned short last_chunks;
111 unsigned short start_middle;
112 unsigned short first_num:NCHUNKS_ORDER;
116 * Internal z3fold page flags
118 enum z3fold_page_flags {
128 /* Converts an allocation size in bytes to size in z3fold chunks */
129 static int size_to_chunks(size_t size)
131 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
134 #define for_each_unbuddied_list(_iter, _begin) \
135 for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
137 /* Initializes the z3fold header of a newly allocated z3fold page */
138 static struct z3fold_header *init_z3fold_page(struct page *page)
140 struct z3fold_header *zhdr = page_address(page);
142 INIT_LIST_HEAD(&page->lru);
143 clear_bit(UNDER_RECLAIM, &page->private);
144 clear_bit(PAGE_HEADLESS, &page->private);
145 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
147 zhdr->first_chunks = 0;
148 zhdr->middle_chunks = 0;
149 zhdr->last_chunks = 0;
151 zhdr->start_middle = 0;
152 INIT_LIST_HEAD(&zhdr->buddy);
156 /* Resets the struct page fields and frees the page */
157 static void free_z3fold_page(struct z3fold_header *zhdr)
159 __free_page(virt_to_page(zhdr));
163 * Encodes the handle of a particular buddy within a z3fold page
164 * Pool lock should be held as this function accesses first_num
166 static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud)
168 unsigned long handle;
170 handle = (unsigned long)zhdr;
172 handle += (bud + zhdr->first_num) & BUDDY_MASK;
176 /* Returns the z3fold page where a given handle is stored */
177 static struct z3fold_header *handle_to_z3fold_header(unsigned long handle)
179 return (struct z3fold_header *)(handle & PAGE_MASK);
182 /* Returns buddy number */
183 static enum buddy handle_to_buddy(unsigned long handle)
185 struct z3fold_header *zhdr = handle_to_z3fold_header(handle);
186 return (handle - zhdr->first_num) & BUDDY_MASK;
190 * Returns the number of free chunks in a z3fold page.
191 * NB: can't be used with HEADLESS pages.
193 static int num_free_chunks(struct z3fold_header *zhdr)
197 * If there is a middle object, pick up the bigger free space
198 * either before or after it. Otherwise just subtract the number
199 * of chunks occupied by the first and the last objects.
201 if (zhdr->middle_chunks != 0) {
202 int nfree_before = zhdr->first_chunks ?
203 0 : zhdr->start_middle - 1;
204 int nfree_after = zhdr->last_chunks ?
205 0 : NCHUNKS - zhdr->start_middle - zhdr->middle_chunks;
206 nfree = max(nfree_before, nfree_after);
208 nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
216 * z3fold_create_pool() - create a new z3fold pool
217 * @gfp: gfp flags when allocating the z3fold pool structure
218 * @ops: user-defined operations for the z3fold pool
220 * Return: pointer to the new z3fold pool or NULL if the metadata allocation
223 static struct z3fold_pool *z3fold_create_pool(gfp_t gfp,
224 const struct z3fold_ops *ops)
226 struct z3fold_pool *pool;
229 pool = kzalloc(sizeof(struct z3fold_pool), gfp);
232 spin_lock_init(&pool->lock);
233 for_each_unbuddied_list(i, 0)
234 INIT_LIST_HEAD(&pool->unbuddied[i]);
235 INIT_LIST_HEAD(&pool->buddied);
236 INIT_LIST_HEAD(&pool->lru);
243 * z3fold_destroy_pool() - destroys an existing z3fold pool
244 * @pool: the z3fold pool to be destroyed
246 * The pool should be emptied before this function is called.
248 static void z3fold_destroy_pool(struct z3fold_pool *pool)
253 /* Has to be called with lock held */
254 static int z3fold_compact_page(struct z3fold_header *zhdr)
256 struct page *page = virt_to_page(zhdr);
260 if (!test_bit(MIDDLE_CHUNK_MAPPED, &page->private) &&
261 zhdr->middle_chunks != 0 &&
262 zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
263 memmove(beg + ZHDR_SIZE_ALIGNED,
264 beg + (zhdr->start_middle << CHUNK_SHIFT),
265 zhdr->middle_chunks << CHUNK_SHIFT);
266 zhdr->first_chunks = zhdr->middle_chunks;
267 zhdr->middle_chunks = 0;
268 zhdr->start_middle = 0;
276 * z3fold_alloc() - allocates a region of a given size
277 * @pool: z3fold pool from which to allocate
278 * @size: size in bytes of the desired allocation
279 * @gfp: gfp flags used if the pool needs to grow
280 * @handle: handle of the new allocation
282 * This function will attempt to find a free region in the pool large enough to
283 * satisfy the allocation request. A search of the unbuddied lists is
284 * performed first. If no suitable free region is found, then a new page is
285 * allocated and added to the pool to satisfy the request.
287 * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used
288 * as z3fold pool pages.
290 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
291 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
294 static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
295 unsigned long *handle)
297 int chunks = 0, i, freechunks;
298 struct z3fold_header *zhdr = NULL;
302 if (!size || (gfp & __GFP_HIGHMEM))
305 if (size > PAGE_SIZE)
308 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
311 chunks = size_to_chunks(size);
312 spin_lock(&pool->lock);
314 /* First, try to find an unbuddied z3fold page. */
316 for_each_unbuddied_list(i, chunks) {
317 if (!list_empty(&pool->unbuddied[i])) {
318 zhdr = list_first_entry(&pool->unbuddied[i],
319 struct z3fold_header, buddy);
320 page = virt_to_page(zhdr);
321 if (zhdr->first_chunks == 0) {
322 if (zhdr->middle_chunks != 0 &&
323 chunks >= zhdr->start_middle)
327 } else if (zhdr->last_chunks == 0)
329 else if (zhdr->middle_chunks == 0)
332 pr_err("No free chunks in unbuddied\n");
336 list_del(&zhdr->buddy);
341 spin_unlock(&pool->lock);
344 /* Couldn't find unbuddied z3fold page, create new one */
345 page = alloc_page(gfp);
348 spin_lock(&pool->lock);
350 zhdr = init_z3fold_page(page);
352 if (bud == HEADLESS) {
353 set_bit(PAGE_HEADLESS, &page->private);
359 zhdr->first_chunks = chunks;
360 else if (bud == LAST)
361 zhdr->last_chunks = chunks;
363 zhdr->middle_chunks = chunks;
364 zhdr->start_middle = zhdr->first_chunks + 1;
367 if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 ||
368 zhdr->middle_chunks == 0) {
369 /* Add to unbuddied list */
370 freechunks = num_free_chunks(zhdr);
371 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
373 /* Add to buddied list */
374 list_add(&zhdr->buddy, &pool->buddied);
378 /* Add/move z3fold page to beginning of LRU */
379 if (!list_empty(&page->lru))
380 list_del(&page->lru);
382 list_add(&page->lru, &pool->lru);
384 *handle = encode_handle(zhdr, bud);
385 spin_unlock(&pool->lock);
391 * z3fold_free() - frees the allocation associated with the given handle
392 * @pool: pool in which the allocation resided
393 * @handle: handle associated with the allocation returned by z3fold_alloc()
395 * In the case that the z3fold page in which the allocation resides is under
396 * reclaim, as indicated by the PG_reclaim flag being set, this function
397 * only sets the first|last_chunks to 0. The page is actually freed
398 * once both buddies are evicted (see z3fold_reclaim_page() below).
400 static void z3fold_free(struct z3fold_pool *pool, unsigned long handle)
402 struct z3fold_header *zhdr;
407 spin_lock(&pool->lock);
408 zhdr = handle_to_z3fold_header(handle);
409 page = virt_to_page(zhdr);
411 if (test_bit(PAGE_HEADLESS, &page->private)) {
412 /* HEADLESS page stored */
415 bud = handle_to_buddy(handle);
419 zhdr->first_chunks = 0;
422 zhdr->middle_chunks = 0;
423 zhdr->start_middle = 0;
426 zhdr->last_chunks = 0;
429 pr_err("%s: unknown bud %d\n", __func__, bud);
431 spin_unlock(&pool->lock);
436 if (test_bit(UNDER_RECLAIM, &page->private)) {
437 /* z3fold page is under reclaim, reclaim will free */
438 spin_unlock(&pool->lock);
442 if (bud != HEADLESS) {
443 /* Remove from existing buddy list */
444 list_del(&zhdr->buddy);
447 if (bud == HEADLESS ||
448 (zhdr->first_chunks == 0 && zhdr->middle_chunks == 0 &&
449 zhdr->last_chunks == 0)) {
450 /* z3fold page is empty, free */
451 list_del(&page->lru);
452 clear_bit(PAGE_HEADLESS, &page->private);
453 free_z3fold_page(zhdr);
456 z3fold_compact_page(zhdr);
457 /* Add to the unbuddied list */
458 freechunks = num_free_chunks(zhdr);
459 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
462 spin_unlock(&pool->lock);
466 * z3fold_reclaim_page() - evicts allocations from a pool page and frees it
467 * @pool: pool from which a page will attempt to be evicted
468 * @retires: number of pages on the LRU list for which eviction will
469 * be attempted before failing
471 * z3fold reclaim is different from normal system reclaim in that it is done
472 * from the bottom, up. This is because only the bottom layer, z3fold, has
473 * information on how the allocations are organized within each z3fold page.
474 * This has the potential to create interesting locking situations between
475 * z3fold and the user, however.
477 * To avoid these, this is how z3fold_reclaim_page() should be called:
479 * The user detects a page should be reclaimed and calls z3fold_reclaim_page().
480 * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and
481 * call the user-defined eviction handler with the pool and handle as
484 * If the handle can not be evicted, the eviction handler should return
485 * non-zero. z3fold_reclaim_page() will add the z3fold page back to the
486 * appropriate list and try the next z3fold page on the LRU up to
487 * a user defined number of retries.
489 * If the handle is successfully evicted, the eviction handler should
490 * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free()
491 * contains logic to delay freeing the page if the page is under reclaim,
492 * as indicated by the setting of the PG_reclaim flag on the underlying page.
494 * If all buddies in the z3fold page are successfully evicted, then the
495 * z3fold page can be freed.
497 * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are
498 * no pages to evict or an eviction handler is not registered, -EAGAIN if
499 * the retry limit was hit.
501 static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries)
503 int i, ret = 0, freechunks;
504 struct z3fold_header *zhdr;
506 unsigned long first_handle = 0, middle_handle = 0, last_handle = 0;
508 spin_lock(&pool->lock);
509 if (!pool->ops || !pool->ops->evict || list_empty(&pool->lru) ||
511 spin_unlock(&pool->lock);
514 for (i = 0; i < retries; i++) {
515 page = list_last_entry(&pool->lru, struct page, lru);
516 list_del(&page->lru);
518 /* Protect z3fold page against free */
519 set_bit(UNDER_RECLAIM, &page->private);
520 zhdr = page_address(page);
521 if (!test_bit(PAGE_HEADLESS, &page->private)) {
522 list_del(&zhdr->buddy);
524 * We need encode the handles before unlocking, since
525 * we can race with free that will set
526 * (first|last)_chunks to 0
531 if (zhdr->first_chunks)
532 first_handle = encode_handle(zhdr, FIRST);
533 if (zhdr->middle_chunks)
534 middle_handle = encode_handle(zhdr, MIDDLE);
535 if (zhdr->last_chunks)
536 last_handle = encode_handle(zhdr, LAST);
538 first_handle = encode_handle(zhdr, HEADLESS);
539 last_handle = middle_handle = 0;
542 spin_unlock(&pool->lock);
544 /* Issue the eviction callback(s) */
546 ret = pool->ops->evict(pool, middle_handle);
551 ret = pool->ops->evict(pool, first_handle);
556 ret = pool->ops->evict(pool, last_handle);
561 spin_lock(&pool->lock);
562 clear_bit(UNDER_RECLAIM, &page->private);
563 if ((test_bit(PAGE_HEADLESS, &page->private) && ret == 0) ||
564 (zhdr->first_chunks == 0 && zhdr->last_chunks == 0 &&
565 zhdr->middle_chunks == 0)) {
567 * All buddies are now free, free the z3fold page and
570 clear_bit(PAGE_HEADLESS, &page->private);
571 free_z3fold_page(zhdr);
573 spin_unlock(&pool->lock);
575 } else if (!test_bit(PAGE_HEADLESS, &page->private)) {
576 if (zhdr->first_chunks != 0 &&
577 zhdr->last_chunks != 0 &&
578 zhdr->middle_chunks != 0) {
579 /* Full, add to buddied list */
580 list_add(&zhdr->buddy, &pool->buddied);
582 z3fold_compact_page(zhdr);
583 /* add to unbuddied list */
584 freechunks = num_free_chunks(zhdr);
585 list_add(&zhdr->buddy,
586 &pool->unbuddied[freechunks]);
590 /* add to beginning of LRU */
591 list_add(&page->lru, &pool->lru);
593 spin_unlock(&pool->lock);
598 * z3fold_map() - maps the allocation associated with the given handle
599 * @pool: pool in which the allocation resides
600 * @handle: handle associated with the allocation to be mapped
602 * Extracts the buddy number from handle and constructs the pointer to the
603 * correct starting chunk within the page.
605 * Returns: a pointer to the mapped allocation
607 static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle)
609 struct z3fold_header *zhdr;
614 spin_lock(&pool->lock);
615 zhdr = handle_to_z3fold_header(handle);
617 page = virt_to_page(zhdr);
619 if (test_bit(PAGE_HEADLESS, &page->private))
622 buddy = handle_to_buddy(handle);
625 addr += ZHDR_SIZE_ALIGNED;
628 addr += zhdr->start_middle << CHUNK_SHIFT;
629 set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
632 addr += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
635 pr_err("unknown buddy id %d\n", buddy);
641 spin_unlock(&pool->lock);
646 * z3fold_unmap() - unmaps the allocation associated with the given handle
647 * @pool: pool in which the allocation resides
648 * @handle: handle associated with the allocation to be unmapped
650 static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle)
652 struct z3fold_header *zhdr;
656 spin_lock(&pool->lock);
657 zhdr = handle_to_z3fold_header(handle);
658 page = virt_to_page(zhdr);
660 if (test_bit(PAGE_HEADLESS, &page->private)) {
661 spin_unlock(&pool->lock);
665 buddy = handle_to_buddy(handle);
667 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
668 spin_unlock(&pool->lock);
672 * z3fold_get_pool_size() - gets the z3fold pool size in pages
673 * @pool: pool whose size is being queried
675 * Returns: size in pages of the given pool. The pool lock need not be
676 * taken to access pages_nr.
678 static u64 z3fold_get_pool_size(struct z3fold_pool *pool)
680 return pool->pages_nr;
687 static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle)
689 if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict)
690 return pool->zpool_ops->evict(pool->zpool, handle);
695 static const struct z3fold_ops z3fold_zpool_ops = {
696 .evict = z3fold_zpool_evict
699 static void *z3fold_zpool_create(const char *name, gfp_t gfp,
700 const struct zpool_ops *zpool_ops,
703 struct z3fold_pool *pool;
705 pool = z3fold_create_pool(gfp, zpool_ops ? &z3fold_zpool_ops : NULL);
708 pool->zpool_ops = zpool_ops;
713 static void z3fold_zpool_destroy(void *pool)
715 z3fold_destroy_pool(pool);
718 static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp,
719 unsigned long *handle)
721 return z3fold_alloc(pool, size, gfp, handle);
723 static void z3fold_zpool_free(void *pool, unsigned long handle)
725 z3fold_free(pool, handle);
728 static int z3fold_zpool_shrink(void *pool, unsigned int pages,
729 unsigned int *reclaimed)
731 unsigned int total = 0;
734 while (total < pages) {
735 ret = z3fold_reclaim_page(pool, 8);
747 static void *z3fold_zpool_map(void *pool, unsigned long handle,
748 enum zpool_mapmode mm)
750 return z3fold_map(pool, handle);
752 static void z3fold_zpool_unmap(void *pool, unsigned long handle)
754 z3fold_unmap(pool, handle);
757 static u64 z3fold_zpool_total_size(void *pool)
759 return z3fold_get_pool_size(pool) * PAGE_SIZE;
762 static struct zpool_driver z3fold_zpool_driver = {
764 .owner = THIS_MODULE,
765 .create = z3fold_zpool_create,
766 .destroy = z3fold_zpool_destroy,
767 .malloc = z3fold_zpool_malloc,
768 .free = z3fold_zpool_free,
769 .shrink = z3fold_zpool_shrink,
770 .map = z3fold_zpool_map,
771 .unmap = z3fold_zpool_unmap,
772 .total_size = z3fold_zpool_total_size,
775 MODULE_ALIAS("zpool-z3fold");
777 static int __init init_z3fold(void)
779 /* Make sure the z3fold header will fit in one chunk */
780 BUILD_BUG_ON(sizeof(struct z3fold_header) > ZHDR_SIZE_ALIGNED);
781 zpool_register_driver(&z3fold_zpool_driver);
786 static void __exit exit_z3fold(void)
788 zpool_unregister_driver(&z3fold_zpool_driver);
791 module_init(init_z3fold);
792 module_exit(exit_z3fold);
794 MODULE_LICENSE("GPL");
795 MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>");
796 MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages");