1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 2013, Seth Jennings, IBM
7 * Concepts based on zcache internal zbud allocator by Dan Magenheimer.
9 * zbud is an special purpose allocator for storing compressed pages. Contrary
10 * to what its name may suggest, zbud is not a buddy allocator, but rather an
11 * allocator that "buddies" two compressed pages together in a single memory
14 * While this design limits storage density, it has simple and deterministic
15 * reclaim properties that make it preferable to a higher density approach when
16 * reclaim will be used.
18 * zbud works by storing compressed pages, or "zpages", together in pairs in a
19 * single memory page called a "zbud page". The first buddy is "left
20 * justified" at the beginning of the zbud page, and the last buddy is "right
21 * justified" at the end of the zbud page. The benefit is that if either
22 * buddy is freed, the freed buddy space, coalesced with whatever slack space
23 * that existed between the buddies, results in the largest possible free region
24 * within the zbud page.
26 * zbud also provides an attractive lower bound on density. The ratio of zpages
27 * to zbud pages can not be less than 1. This ensures that zbud can never "do
28 * harm" by using more pages to store zpages than the uncompressed zpages would
29 * have used on their own.
31 * zbud pages are divided into "chunks". The size of the chunks is fixed at
32 * compile time and determined by NCHUNKS_ORDER below. Dividing zbud pages
33 * into chunks allows organizing unbuddied zbud pages into a manageable number
34 * of unbuddied lists according to the number of free chunks available in the
37 * The zbud API differs from that of conventional allocators in that the
38 * allocation function, zbud_alloc(), returns an opaque handle to the user,
39 * not a dereferenceable pointer. The user must map the handle using
40 * zbud_map() in order to get a usable pointer by which to access the
41 * allocation data and unmap the handle with zbud_unmap() when operations
42 * on the allocation data are complete.
45 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
47 #include <linux/atomic.h>
48 #include <linux/list.h>
50 #include <linux/module.h>
51 #include <linux/preempt.h>
52 #include <linux/slab.h>
53 #include <linux/spinlock.h>
54 #include <linux/zpool.h>
60 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
61 * adjusting internal fragmentation. It also determines the number of
62 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
63 * allocation granularity will be in chunks of size PAGE_SIZE/64. As one chunk
64 * in allocated page is occupied by zbud header, NCHUNKS will be calculated to
65 * 63 which shows the max number of free chunks in zbud page, also there will be
66 * 63 freelists per pool.
68 #define NCHUNKS_ORDER 6
70 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER)
71 #define CHUNK_SIZE (1 << CHUNK_SHIFT)
72 #define ZHDR_SIZE_ALIGNED CHUNK_SIZE
73 #define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT)
78 * struct zbud_pool - stores metadata for each zbud pool
79 * @lock: protects all pool fields and first|last_chunk fields of any
80 * zbud page in the pool
81 * @unbuddied: array of lists tracking zbud pages that only contain one buddy;
82 * the lists each zbud page is added to depends on the size of
84 * @buddied: list tracking the zbud pages that contain two buddies;
85 * these zbud pages are full
86 * @lru: list tracking the zbud pages in LRU order by most recently
88 * @pages_nr: number of zbud pages in the pool.
89 * @zpool: zpool driver
90 * @zpool_ops: zpool operations structure with an evict callback
92 * This structure is allocated at pool creation time and maintains metadata
93 * pertaining to a particular zbud pool.
99 * Reuse unbuddied[0] as buddied on the ground that
100 * unbuddied[0] is unused.
102 struct list_head buddied
;
103 struct list_head unbuddied
[NCHUNKS
];
105 struct list_head lru
;
108 const struct zpool_ops
*zpool_ops
;
112 * struct zbud_header - zbud page metadata occupying the first chunk of each
114 * @buddy: links the zbud page into the unbuddied/buddied lists in the pool
115 * @lru: links the zbud page into the lru list in the pool
116 * @first_chunks: the size of the first buddy in chunks, 0 if free
117 * @last_chunks: the size of the last buddy in chunks, 0 if free
120 struct list_head buddy
;
121 struct list_head lru
;
122 unsigned int first_chunks
;
123 unsigned int last_chunks
;
130 /* Just to make the code easier to read */
136 /* Converts an allocation size in bytes to size in zbud chunks */
137 static int size_to_chunks(size_t size
)
139 return (size
+ CHUNK_SIZE
- 1) >> CHUNK_SHIFT
;
142 #define for_each_unbuddied_list(_iter, _begin) \
143 for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
145 /* Initializes the zbud header of a newly allocated zbud page */
146 static struct zbud_header
*init_zbud_page(struct page
*page
)
148 struct zbud_header
*zhdr
= page_address(page
);
149 zhdr
->first_chunks
= 0;
150 zhdr
->last_chunks
= 0;
151 INIT_LIST_HEAD(&zhdr
->buddy
);
152 INIT_LIST_HEAD(&zhdr
->lru
);
153 zhdr
->under_reclaim
= false;
157 /* Resets the struct page fields and frees the page */
158 static void free_zbud_page(struct zbud_header
*zhdr
)
160 __free_page(virt_to_page(zhdr
));
164 * Encodes the handle of a particular buddy within a zbud page
165 * Pool lock should be held as this function accesses first|last_chunks
167 static unsigned long encode_handle(struct zbud_header
*zhdr
, enum buddy bud
)
169 unsigned long handle
;
172 * For now, the encoded handle is actually just the pointer to the data
173 * but this might not always be the case. A little information hiding.
174 * Add CHUNK_SIZE to the handle if it is the first allocation to jump
175 * over the zbud header in the first chunk.
177 handle
= (unsigned long)zhdr
;
179 /* skip over zbud header */
180 handle
+= ZHDR_SIZE_ALIGNED
;
181 else /* bud == LAST */
182 handle
+= PAGE_SIZE
- (zhdr
->last_chunks
<< CHUNK_SHIFT
);
186 /* Returns the zbud page where a given handle is stored */
187 static struct zbud_header
*handle_to_zbud_header(unsigned long handle
)
189 return (struct zbud_header
*)(handle
& PAGE_MASK
);
192 /* Returns the number of free chunks in a zbud page */
193 static int num_free_chunks(struct zbud_header
*zhdr
)
196 * Rather than branch for different situations, just use the fact that
197 * free buddies have a length of zero to simplify everything.
199 return NCHUNKS
- zhdr
->first_chunks
- zhdr
->last_chunks
;
206 * zbud_create_pool() - create a new zbud pool
207 * @gfp: gfp flags when allocating the zbud pool structure
209 * Return: pointer to the new zbud pool or NULL if the metadata allocation
212 static struct zbud_pool
*zbud_create_pool(gfp_t gfp
)
214 struct zbud_pool
*pool
;
217 pool
= kzalloc(sizeof(struct zbud_pool
), gfp
);
220 spin_lock_init(&pool
->lock
);
221 for_each_unbuddied_list(i
, 0)
222 INIT_LIST_HEAD(&pool
->unbuddied
[i
]);
223 INIT_LIST_HEAD(&pool
->buddied
);
224 INIT_LIST_HEAD(&pool
->lru
);
230 * zbud_destroy_pool() - destroys an existing zbud pool
231 * @pool: the zbud pool to be destroyed
233 * The pool should be emptied before this function is called.
235 static void zbud_destroy_pool(struct zbud_pool
*pool
)
241 * zbud_alloc() - allocates a region of a given size
242 * @pool: zbud pool from which to allocate
243 * @size: size in bytes of the desired allocation
244 * @gfp: gfp flags used if the pool needs to grow
245 * @handle: handle of the new allocation
247 * This function will attempt to find a free region in the pool large enough to
248 * satisfy the allocation request. A search of the unbuddied lists is
249 * performed first. If no suitable free region is found, then a new page is
250 * allocated and added to the pool to satisfy the request.
252 * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used
253 * as zbud pool pages.
255 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
256 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
259 static int zbud_alloc(struct zbud_pool
*pool
, size_t size
, gfp_t gfp
,
260 unsigned long *handle
)
262 int chunks
, i
, freechunks
;
263 struct zbud_header
*zhdr
= NULL
;
267 if (!size
|| (gfp
& __GFP_HIGHMEM
))
269 if (size
> PAGE_SIZE
- ZHDR_SIZE_ALIGNED
- CHUNK_SIZE
)
271 chunks
= size_to_chunks(size
);
272 spin_lock(&pool
->lock
);
274 /* First, try to find an unbuddied zbud page. */
275 for_each_unbuddied_list(i
, chunks
) {
276 if (!list_empty(&pool
->unbuddied
[i
])) {
277 zhdr
= list_first_entry(&pool
->unbuddied
[i
],
278 struct zbud_header
, buddy
);
279 list_del(&zhdr
->buddy
);
280 if (zhdr
->first_chunks
== 0)
288 /* Couldn't find unbuddied zbud page, create new one */
289 spin_unlock(&pool
->lock
);
290 page
= alloc_page(gfp
);
293 spin_lock(&pool
->lock
);
295 zhdr
= init_zbud_page(page
);
300 zhdr
->first_chunks
= chunks
;
302 zhdr
->last_chunks
= chunks
;
304 if (zhdr
->first_chunks
== 0 || zhdr
->last_chunks
== 0) {
305 /* Add to unbuddied list */
306 freechunks
= num_free_chunks(zhdr
);
307 list_add(&zhdr
->buddy
, &pool
->unbuddied
[freechunks
]);
309 /* Add to buddied list */
310 list_add(&zhdr
->buddy
, &pool
->buddied
);
313 /* Add/move zbud page to beginning of LRU */
314 if (!list_empty(&zhdr
->lru
))
315 list_del(&zhdr
->lru
);
316 list_add(&zhdr
->lru
, &pool
->lru
);
318 *handle
= encode_handle(zhdr
, bud
);
319 spin_unlock(&pool
->lock
);
325 * zbud_free() - frees the allocation associated with the given handle
326 * @pool: pool in which the allocation resided
327 * @handle: handle associated with the allocation returned by zbud_alloc()
329 * In the case that the zbud page in which the allocation resides is under
330 * reclaim, as indicated by the PG_reclaim flag being set, this function
331 * only sets the first|last_chunks to 0. The page is actually freed
332 * once both buddies are evicted (see zbud_reclaim_page() below).
334 static void zbud_free(struct zbud_pool
*pool
, unsigned long handle
)
336 struct zbud_header
*zhdr
;
339 spin_lock(&pool
->lock
);
340 zhdr
= handle_to_zbud_header(handle
);
342 /* If first buddy, handle will be page aligned */
343 if ((handle
- ZHDR_SIZE_ALIGNED
) & ~PAGE_MASK
)
344 zhdr
->last_chunks
= 0;
346 zhdr
->first_chunks
= 0;
348 if (zhdr
->under_reclaim
) {
349 /* zbud page is under reclaim, reclaim will free */
350 spin_unlock(&pool
->lock
);
354 /* Remove from existing buddy list */
355 list_del(&zhdr
->buddy
);
357 if (zhdr
->first_chunks
== 0 && zhdr
->last_chunks
== 0) {
358 /* zbud page is empty, free */
359 list_del(&zhdr
->lru
);
360 free_zbud_page(zhdr
);
363 /* Add to unbuddied list */
364 freechunks
= num_free_chunks(zhdr
);
365 list_add(&zhdr
->buddy
, &pool
->unbuddied
[freechunks
]);
368 spin_unlock(&pool
->lock
);
372 * zbud_reclaim_page() - evicts allocations from a pool page and frees it
373 * @pool: pool from which a page will attempt to be evicted
374 * @retries: number of pages on the LRU list for which eviction will
375 * be attempted before failing
377 * zbud reclaim is different from normal system reclaim in that the reclaim is
378 * done from the bottom, up. This is because only the bottom layer, zbud, has
379 * information on how the allocations are organized within each zbud page. This
380 * has the potential to create interesting locking situations between zbud and
383 * To avoid these, this is how zbud_reclaim_page() should be called:
385 * The user detects a page should be reclaimed and calls zbud_reclaim_page().
386 * zbud_reclaim_page() will remove a zbud page from the pool LRU list and call
387 * the user-defined eviction handler with the pool and handle as arguments.
389 * If the handle can not be evicted, the eviction handler should return
390 * non-zero. zbud_reclaim_page() will add the zbud page back to the
391 * appropriate list and try the next zbud page on the LRU up to
392 * a user defined number of retries.
394 * If the handle is successfully evicted, the eviction handler should
395 * return 0 _and_ should have called zbud_free() on the handle. zbud_free()
396 * contains logic to delay freeing the page if the page is under reclaim,
397 * as indicated by the setting of the PG_reclaim flag on the underlying page.
399 * If all buddies in the zbud page are successfully evicted, then the
400 * zbud page can be freed.
402 * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are
403 * no pages to evict or an eviction handler is not registered, -EAGAIN if
404 * the retry limit was hit.
406 static int zbud_reclaim_page(struct zbud_pool
*pool
, unsigned int retries
)
408 int i
, ret
, freechunks
;
409 struct zbud_header
*zhdr
;
410 unsigned long first_handle
= 0, last_handle
= 0;
412 spin_lock(&pool
->lock
);
413 if (list_empty(&pool
->lru
)) {
414 spin_unlock(&pool
->lock
);
417 for (i
= 0; i
< retries
; i
++) {
418 zhdr
= list_last_entry(&pool
->lru
, struct zbud_header
, lru
);
419 list_del(&zhdr
->lru
);
420 list_del(&zhdr
->buddy
);
421 /* Protect zbud page against free */
422 zhdr
->under_reclaim
= true;
424 * We need encode the handles before unlocking, since we can
425 * race with free that will set (first|last)_chunks to 0
429 if (zhdr
->first_chunks
)
430 first_handle
= encode_handle(zhdr
, FIRST
);
431 if (zhdr
->last_chunks
)
432 last_handle
= encode_handle(zhdr
, LAST
);
433 spin_unlock(&pool
->lock
);
435 /* Issue the eviction callback(s) */
437 ret
= pool
->zpool_ops
->evict(pool
->zpool
, first_handle
);
442 ret
= pool
->zpool_ops
->evict(pool
->zpool
, last_handle
);
447 spin_lock(&pool
->lock
);
448 zhdr
->under_reclaim
= false;
449 if (zhdr
->first_chunks
== 0 && zhdr
->last_chunks
== 0) {
451 * Both buddies are now free, free the zbud page and
454 free_zbud_page(zhdr
);
456 spin_unlock(&pool
->lock
);
458 } else if (zhdr
->first_chunks
== 0 ||
459 zhdr
->last_chunks
== 0) {
460 /* add to unbuddied list */
461 freechunks
= num_free_chunks(zhdr
);
462 list_add(&zhdr
->buddy
, &pool
->unbuddied
[freechunks
]);
464 /* add to buddied list */
465 list_add(&zhdr
->buddy
, &pool
->buddied
);
468 /* add to beginning of LRU */
469 list_add(&zhdr
->lru
, &pool
->lru
);
471 spin_unlock(&pool
->lock
);
476 * zbud_map() - maps the allocation associated with the given handle
477 * @pool: pool in which the allocation resides
478 * @handle: handle associated with the allocation to be mapped
480 * While trivial for zbud, the mapping functions for others allocators
481 * implementing this allocation API could have more complex information encoded
482 * in the handle and could create temporary mappings to make the data
483 * accessible to the user.
485 * Returns: a pointer to the mapped allocation
487 static void *zbud_map(struct zbud_pool
*pool
, unsigned long handle
)
489 return (void *)(handle
);
493 * zbud_unmap() - maps the allocation associated with the given handle
494 * @pool: pool in which the allocation resides
495 * @handle: handle associated with the allocation to be unmapped
497 static void zbud_unmap(struct zbud_pool
*pool
, unsigned long handle
)
502 * zbud_get_pool_size() - gets the zbud pool size in pages
503 * @pool: pool whose size is being queried
505 * Returns: size in pages of the given pool. The pool lock need not be
506 * taken to access pages_nr.
508 static u64
zbud_get_pool_size(struct zbud_pool
*pool
)
510 return pool
->pages_nr
;
517 static void *zbud_zpool_create(const char *name
, gfp_t gfp
,
518 const struct zpool_ops
*zpool_ops
,
521 struct zbud_pool
*pool
;
523 pool
= zbud_create_pool(gfp
);
526 pool
->zpool_ops
= zpool_ops
;
531 static void zbud_zpool_destroy(void *pool
)
533 zbud_destroy_pool(pool
);
536 static int zbud_zpool_malloc(void *pool
, size_t size
, gfp_t gfp
,
537 unsigned long *handle
)
539 return zbud_alloc(pool
, size
, gfp
, handle
);
541 static void zbud_zpool_free(void *pool
, unsigned long handle
)
543 zbud_free(pool
, handle
);
546 static int zbud_zpool_shrink(void *pool
, unsigned int pages
,
547 unsigned int *reclaimed
)
549 unsigned int total
= 0;
552 while (total
< pages
) {
553 ret
= zbud_reclaim_page(pool
, 8);
565 static void *zbud_zpool_map(void *pool
, unsigned long handle
,
566 enum zpool_mapmode mm
)
568 return zbud_map(pool
, handle
);
570 static void zbud_zpool_unmap(void *pool
, unsigned long handle
)
572 zbud_unmap(pool
, handle
);
575 static u64
zbud_zpool_total_size(void *pool
)
577 return zbud_get_pool_size(pool
) * PAGE_SIZE
;
580 static struct zpool_driver zbud_zpool_driver
= {
582 .sleep_mapped
= true,
583 .owner
= THIS_MODULE
,
584 .create
= zbud_zpool_create
,
585 .destroy
= zbud_zpool_destroy
,
586 .malloc
= zbud_zpool_malloc
,
587 .free
= zbud_zpool_free
,
588 .shrink
= zbud_zpool_shrink
,
589 .map
= zbud_zpool_map
,
590 .unmap
= zbud_zpool_unmap
,
591 .total_size
= zbud_zpool_total_size
,
594 MODULE_ALIAS("zpool-zbud");
596 static int __init
init_zbud(void)
598 /* Make sure the zbud header will fit in one chunk */
599 BUILD_BUG_ON(sizeof(struct zbud_header
) > ZHDR_SIZE_ALIGNED
);
602 zpool_register_driver(&zbud_zpool_driver
);
607 static void __exit
exit_zbud(void)
609 zpool_unregister_driver(&zbud_zpool_driver
);
610 pr_info("unloaded\n");
613 module_init(init_zbud
);
614 module_exit(exit_zbud
);
616 MODULE_LICENSE("GPL");
617 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
618 MODULE_DESCRIPTION("Buddy Allocator for Compressed Pages");