xref: /openbmc/linux/mm/zbud.c (revision ff7b437f)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * zbud.c
4  *
5  * Copyright (C) 2013, Seth Jennings, IBM
6  *
7  * Concepts based on zcache internal zbud allocator by Dan Magenheimer.
8  *
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
12  * page.
13  *
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.
17  *
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.
25  *
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.
30  *
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
35  * zbud page.
36  *
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.
43  */
44 
45 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
46 
47 #include <linux/atomic.h>
48 #include <linux/list.h>
49 #include <linux/mm.h>
50 #include <linux/module.h>
51 #include <linux/preempt.h>
52 #include <linux/slab.h>
53 #include <linux/spinlock.h>
54 #include <linux/zbud.h>
55 #include <linux/zpool.h>
56 
57 /*****************
58  * Structures
59 *****************/
60 /*
61  * NCHUNKS_ORDER determines the internal allocation granularity, effectively
62  * adjusting internal fragmentation.  It also determines the number of
63  * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
64  * allocation granularity will be in chunks of size PAGE_SIZE/64. As one chunk
65  * in allocated page is occupied by zbud header, NCHUNKS will be calculated to
66  * 63 which shows the max number of free chunks in zbud page, also there will be
67  * 63 freelists per pool.
68  */
69 #define NCHUNKS_ORDER	6
70 
71 #define CHUNK_SHIFT	(PAGE_SHIFT - NCHUNKS_ORDER)
72 #define CHUNK_SIZE	(1 << CHUNK_SHIFT)
73 #define ZHDR_SIZE_ALIGNED CHUNK_SIZE
74 #define NCHUNKS		((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT)
75 
76 /**
77  * struct zbud_pool - stores metadata for each zbud pool
78  * @lock:	protects all pool fields and first|last_chunk fields of any
79  *		zbud page in the pool
80  * @unbuddied:	array of lists tracking zbud pages that only contain one buddy;
81  *		the lists each zbud page is added to depends on the size of
82  *		its free region.
83  * @buddied:	list tracking the zbud pages that contain two buddies;
84  *		these zbud pages are full
85  * @lru:	list tracking the zbud pages in LRU order by most recently
86  *		added buddy.
87  * @pages_nr:	number of zbud pages in the pool.
88  * @ops:	pointer to a structure of user defined operations specified at
89  *		pool creation time.
90  *
91  * This structure is allocated at pool creation time and maintains metadata
92  * pertaining to a particular zbud pool.
93  */
94 struct zbud_pool {
95 	spinlock_t lock;
96 	struct list_head unbuddied[NCHUNKS];
97 	struct list_head buddied;
98 	struct list_head lru;
99 	u64 pages_nr;
100 	const struct zbud_ops *ops;
101 #ifdef CONFIG_ZPOOL
102 	struct zpool *zpool;
103 	const struct zpool_ops *zpool_ops;
104 #endif
105 };
106 
107 /*
108  * struct zbud_header - zbud page metadata occupying the first chunk of each
109  *			zbud page.
110  * @buddy:	links the zbud page into the unbuddied/buddied lists in the pool
111  * @lru:	links the zbud page into the lru list in the pool
112  * @first_chunks:	the size of the first buddy in chunks, 0 if free
113  * @last_chunks:	the size of the last buddy in chunks, 0 if free
114  */
115 struct zbud_header {
116 	struct list_head buddy;
117 	struct list_head lru;
118 	unsigned int first_chunks;
119 	unsigned int last_chunks;
120 	bool under_reclaim;
121 };
122 
123 /*****************
124  * zpool
125  ****************/
126 
127 #ifdef CONFIG_ZPOOL
128 
129 static int zbud_zpool_evict(struct zbud_pool *pool, unsigned long handle)
130 {
131 	if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict)
132 		return pool->zpool_ops->evict(pool->zpool, handle);
133 	else
134 		return -ENOENT;
135 }
136 
137 static const struct zbud_ops zbud_zpool_ops = {
138 	.evict =	zbud_zpool_evict
139 };
140 
141 static void *zbud_zpool_create(const char *name, gfp_t gfp,
142 			       const struct zpool_ops *zpool_ops,
143 			       struct zpool *zpool)
144 {
145 	struct zbud_pool *pool;
146 
147 	pool = zbud_create_pool(gfp, zpool_ops ? &zbud_zpool_ops : NULL);
148 	if (pool) {
149 		pool->zpool = zpool;
150 		pool->zpool_ops = zpool_ops;
151 	}
152 	return pool;
153 }
154 
155 static void zbud_zpool_destroy(void *pool)
156 {
157 	zbud_destroy_pool(pool);
158 }
159 
160 static int zbud_zpool_malloc(void *pool, size_t size, gfp_t gfp,
161 			unsigned long *handle)
162 {
163 	return zbud_alloc(pool, size, gfp, handle);
164 }
165 static void zbud_zpool_free(void *pool, unsigned long handle)
166 {
167 	zbud_free(pool, handle);
168 }
169 
170 static int zbud_zpool_shrink(void *pool, unsigned int pages,
171 			unsigned int *reclaimed)
172 {
173 	unsigned int total = 0;
174 	int ret = -EINVAL;
175 
176 	while (total < pages) {
177 		ret = zbud_reclaim_page(pool, 8);
178 		if (ret < 0)
179 			break;
180 		total++;
181 	}
182 
183 	if (reclaimed)
184 		*reclaimed = total;
185 
186 	return ret;
187 }
188 
189 static void *zbud_zpool_map(void *pool, unsigned long handle,
190 			enum zpool_mapmode mm)
191 {
192 	return zbud_map(pool, handle);
193 }
194 static void zbud_zpool_unmap(void *pool, unsigned long handle)
195 {
196 	zbud_unmap(pool, handle);
197 }
198 
199 static u64 zbud_zpool_total_size(void *pool)
200 {
201 	return zbud_get_pool_size(pool) * PAGE_SIZE;
202 }
203 
204 static struct zpool_driver zbud_zpool_driver = {
205 	.type =		"zbud",
206 	.owner =	THIS_MODULE,
207 	.create =	zbud_zpool_create,
208 	.destroy =	zbud_zpool_destroy,
209 	.malloc =	zbud_zpool_malloc,
210 	.free =		zbud_zpool_free,
211 	.shrink =	zbud_zpool_shrink,
212 	.map =		zbud_zpool_map,
213 	.unmap =	zbud_zpool_unmap,
214 	.total_size =	zbud_zpool_total_size,
215 };
216 
217 MODULE_ALIAS("zpool-zbud");
218 #endif /* CONFIG_ZPOOL */
219 
220 /*****************
221  * Helpers
222 *****************/
223 /* Just to make the code easier to read */
224 enum buddy {
225 	FIRST,
226 	LAST
227 };
228 
229 /* Converts an allocation size in bytes to size in zbud chunks */
230 static int size_to_chunks(size_t size)
231 {
232 	return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
233 }
234 
235 #define for_each_unbuddied_list(_iter, _begin) \
236 	for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
237 
238 /* Initializes the zbud header of a newly allocated zbud page */
239 static struct zbud_header *init_zbud_page(struct page *page)
240 {
241 	struct zbud_header *zhdr = page_address(page);
242 	zhdr->first_chunks = 0;
243 	zhdr->last_chunks = 0;
244 	INIT_LIST_HEAD(&zhdr->buddy);
245 	INIT_LIST_HEAD(&zhdr->lru);
246 	zhdr->under_reclaim = 0;
247 	return zhdr;
248 }
249 
250 /* Resets the struct page fields and frees the page */
251 static void free_zbud_page(struct zbud_header *zhdr)
252 {
253 	__free_page(virt_to_page(zhdr));
254 }
255 
256 /*
257  * Encodes the handle of a particular buddy within a zbud page
258  * Pool lock should be held as this function accesses first|last_chunks
259  */
260 static unsigned long encode_handle(struct zbud_header *zhdr, enum buddy bud)
261 {
262 	unsigned long handle;
263 
264 	/*
265 	 * For now, the encoded handle is actually just the pointer to the data
266 	 * but this might not always be the case.  A little information hiding.
267 	 * Add CHUNK_SIZE to the handle if it is the first allocation to jump
268 	 * over the zbud header in the first chunk.
269 	 */
270 	handle = (unsigned long)zhdr;
271 	if (bud == FIRST)
272 		/* skip over zbud header */
273 		handle += ZHDR_SIZE_ALIGNED;
274 	else /* bud == LAST */
275 		handle += PAGE_SIZE - (zhdr->last_chunks  << CHUNK_SHIFT);
276 	return handle;
277 }
278 
279 /* Returns the zbud page where a given handle is stored */
280 static struct zbud_header *handle_to_zbud_header(unsigned long handle)
281 {
282 	return (struct zbud_header *)(handle & PAGE_MASK);
283 }
284 
285 /* Returns the number of free chunks in a zbud page */
286 static int num_free_chunks(struct zbud_header *zhdr)
287 {
288 	/*
289 	 * Rather than branch for different situations, just use the fact that
290 	 * free buddies have a length of zero to simplify everything.
291 	 */
292 	return NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
293 }
294 
295 /*****************
296  * API Functions
297 *****************/
298 /**
299  * zbud_create_pool() - create a new zbud pool
300  * @gfp:	gfp flags when allocating the zbud pool structure
301  * @ops:	user-defined operations for the zbud pool
302  *
303  * Return: pointer to the new zbud pool or NULL if the metadata allocation
304  * failed.
305  */
306 struct zbud_pool *zbud_create_pool(gfp_t gfp, const struct zbud_ops *ops)
307 {
308 	struct zbud_pool *pool;
309 	int i;
310 
311 	pool = kzalloc(sizeof(struct zbud_pool), gfp);
312 	if (!pool)
313 		return NULL;
314 	spin_lock_init(&pool->lock);
315 	for_each_unbuddied_list(i, 0)
316 		INIT_LIST_HEAD(&pool->unbuddied[i]);
317 	INIT_LIST_HEAD(&pool->buddied);
318 	INIT_LIST_HEAD(&pool->lru);
319 	pool->pages_nr = 0;
320 	pool->ops = ops;
321 	return pool;
322 }
323 
324 /**
325  * zbud_destroy_pool() - destroys an existing zbud pool
326  * @pool:	the zbud pool to be destroyed
327  *
328  * The pool should be emptied before this function is called.
329  */
330 void zbud_destroy_pool(struct zbud_pool *pool)
331 {
332 	kfree(pool);
333 }
334 
335 /**
336  * zbud_alloc() - allocates a region of a given size
337  * @pool:	zbud pool from which to allocate
338  * @size:	size in bytes of the desired allocation
339  * @gfp:	gfp flags used if the pool needs to grow
340  * @handle:	handle of the new allocation
341  *
342  * This function will attempt to find a free region in the pool large enough to
343  * satisfy the allocation request.  A search of the unbuddied lists is
344  * performed first. If no suitable free region is found, then a new page is
345  * allocated and added to the pool to satisfy the request.
346  *
347  * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used
348  * as zbud pool pages.
349  *
350  * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
351  * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
352  * a new page.
353  */
354 int zbud_alloc(struct zbud_pool *pool, size_t size, gfp_t gfp,
355 			unsigned long *handle)
356 {
357 	int chunks, i, freechunks;
358 	struct zbud_header *zhdr = NULL;
359 	enum buddy bud;
360 	struct page *page;
361 
362 	if (!size || (gfp & __GFP_HIGHMEM))
363 		return -EINVAL;
364 	if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
365 		return -ENOSPC;
366 	chunks = size_to_chunks(size);
367 	spin_lock(&pool->lock);
368 
369 	/* First, try to find an unbuddied zbud page. */
370 	zhdr = NULL;
371 	for_each_unbuddied_list(i, chunks) {
372 		if (!list_empty(&pool->unbuddied[i])) {
373 			zhdr = list_first_entry(&pool->unbuddied[i],
374 					struct zbud_header, buddy);
375 			list_del(&zhdr->buddy);
376 			if (zhdr->first_chunks == 0)
377 				bud = FIRST;
378 			else
379 				bud = LAST;
380 			goto found;
381 		}
382 	}
383 
384 	/* Couldn't find unbuddied zbud page, create new one */
385 	spin_unlock(&pool->lock);
386 	page = alloc_page(gfp);
387 	if (!page)
388 		return -ENOMEM;
389 	spin_lock(&pool->lock);
390 	pool->pages_nr++;
391 	zhdr = init_zbud_page(page);
392 	bud = FIRST;
393 
394 found:
395 	if (bud == FIRST)
396 		zhdr->first_chunks = chunks;
397 	else
398 		zhdr->last_chunks = chunks;
399 
400 	if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0) {
401 		/* Add to unbuddied list */
402 		freechunks = num_free_chunks(zhdr);
403 		list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
404 	} else {
405 		/* Add to buddied list */
406 		list_add(&zhdr->buddy, &pool->buddied);
407 	}
408 
409 	/* Add/move zbud page to beginning of LRU */
410 	if (!list_empty(&zhdr->lru))
411 		list_del(&zhdr->lru);
412 	list_add(&zhdr->lru, &pool->lru);
413 
414 	*handle = encode_handle(zhdr, bud);
415 	spin_unlock(&pool->lock);
416 
417 	return 0;
418 }
419 
420 /**
421  * zbud_free() - frees the allocation associated with the given handle
422  * @pool:	pool in which the allocation resided
423  * @handle:	handle associated with the allocation returned by zbud_alloc()
424  *
425  * In the case that the zbud page in which the allocation resides is under
426  * reclaim, as indicated by the PG_reclaim flag being set, this function
427  * only sets the first|last_chunks to 0.  The page is actually freed
428  * once both buddies are evicted (see zbud_reclaim_page() below).
429  */
430 void zbud_free(struct zbud_pool *pool, unsigned long handle)
431 {
432 	struct zbud_header *zhdr;
433 	int freechunks;
434 
435 	spin_lock(&pool->lock);
436 	zhdr = handle_to_zbud_header(handle);
437 
438 	/* If first buddy, handle will be page aligned */
439 	if ((handle - ZHDR_SIZE_ALIGNED) & ~PAGE_MASK)
440 		zhdr->last_chunks = 0;
441 	else
442 		zhdr->first_chunks = 0;
443 
444 	if (zhdr->under_reclaim) {
445 		/* zbud page is under reclaim, reclaim will free */
446 		spin_unlock(&pool->lock);
447 		return;
448 	}
449 
450 	/* Remove from existing buddy list */
451 	list_del(&zhdr->buddy);
452 
453 	if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
454 		/* zbud page is empty, free */
455 		list_del(&zhdr->lru);
456 		free_zbud_page(zhdr);
457 		pool->pages_nr--;
458 	} else {
459 		/* Add to unbuddied list */
460 		freechunks = num_free_chunks(zhdr);
461 		list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
462 	}
463 
464 	spin_unlock(&pool->lock);
465 }
466 
467 /**
468  * zbud_reclaim_page() - evicts allocations from a pool page and frees it
469  * @pool:	pool from which a page will attempt to be evicted
470  * @retries:	number of pages on the LRU list for which eviction will
471  *		be attempted before failing
472  *
473  * zbud reclaim is different from normal system reclaim in that the reclaim is
474  * done from the bottom, up.  This is because only the bottom layer, zbud, has
475  * information on how the allocations are organized within each zbud page. This
476  * has the potential to create interesting locking situations between zbud and
477  * the user, however.
478  *
479  * To avoid these, this is how zbud_reclaim_page() should be called:
480  *
481  * The user detects a page should be reclaimed and calls zbud_reclaim_page().
482  * zbud_reclaim_page() will remove a zbud page from the pool LRU list and call
483  * the user-defined eviction handler with the pool and handle as arguments.
484  *
485  * If the handle can not be evicted, the eviction handler should return
486  * non-zero. zbud_reclaim_page() will add the zbud page back to the
487  * appropriate list and try the next zbud page on the LRU up to
488  * a user defined number of retries.
489  *
490  * If the handle is successfully evicted, the eviction handler should
491  * return 0 _and_ should have called zbud_free() on the handle. zbud_free()
492  * contains logic to delay freeing the page if the page is under reclaim,
493  * as indicated by the setting of the PG_reclaim flag on the underlying page.
494  *
495  * If all buddies in the zbud page are successfully evicted, then the
496  * zbud page can be freed.
497  *
498  * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are
499  * no pages to evict or an eviction handler is not registered, -EAGAIN if
500  * the retry limit was hit.
501  */
502 int zbud_reclaim_page(struct zbud_pool *pool, unsigned int retries)
503 {
504 	int i, ret, freechunks;
505 	struct zbud_header *zhdr;
506 	unsigned long first_handle = 0, last_handle = 0;
507 
508 	spin_lock(&pool->lock);
509 	if (!pool->ops || !pool->ops->evict || list_empty(&pool->lru) ||
510 			retries == 0) {
511 		spin_unlock(&pool->lock);
512 		return -EINVAL;
513 	}
514 	for (i = 0; i < retries; i++) {
515 		zhdr = list_last_entry(&pool->lru, struct zbud_header, lru);
516 		list_del(&zhdr->lru);
517 		list_del(&zhdr->buddy);
518 		/* Protect zbud page against free */
519 		zhdr->under_reclaim = true;
520 		/*
521 		 * We need encode the handles before unlocking, since we can
522 		 * race with free that will set (first|last)_chunks to 0
523 		 */
524 		first_handle = 0;
525 		last_handle = 0;
526 		if (zhdr->first_chunks)
527 			first_handle = encode_handle(zhdr, FIRST);
528 		if (zhdr->last_chunks)
529 			last_handle = encode_handle(zhdr, LAST);
530 		spin_unlock(&pool->lock);
531 
532 		/* Issue the eviction callback(s) */
533 		if (first_handle) {
534 			ret = pool->ops->evict(pool, first_handle);
535 			if (ret)
536 				goto next;
537 		}
538 		if (last_handle) {
539 			ret = pool->ops->evict(pool, last_handle);
540 			if (ret)
541 				goto next;
542 		}
543 next:
544 		spin_lock(&pool->lock);
545 		zhdr->under_reclaim = false;
546 		if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
547 			/*
548 			 * Both buddies are now free, free the zbud page and
549 			 * return success.
550 			 */
551 			free_zbud_page(zhdr);
552 			pool->pages_nr--;
553 			spin_unlock(&pool->lock);
554 			return 0;
555 		} else if (zhdr->first_chunks == 0 ||
556 				zhdr->last_chunks == 0) {
557 			/* add to unbuddied list */
558 			freechunks = num_free_chunks(zhdr);
559 			list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
560 		} else {
561 			/* add to buddied list */
562 			list_add(&zhdr->buddy, &pool->buddied);
563 		}
564 
565 		/* add to beginning of LRU */
566 		list_add(&zhdr->lru, &pool->lru);
567 	}
568 	spin_unlock(&pool->lock);
569 	return -EAGAIN;
570 }
571 
572 /**
573  * zbud_map() - maps the allocation associated with the given handle
574  * @pool:	pool in which the allocation resides
575  * @handle:	handle associated with the allocation to be mapped
576  *
577  * While trivial for zbud, the mapping functions for others allocators
578  * implementing this allocation API could have more complex information encoded
579  * in the handle and could create temporary mappings to make the data
580  * accessible to the user.
581  *
582  * Returns: a pointer to the mapped allocation
583  */
584 void *zbud_map(struct zbud_pool *pool, unsigned long handle)
585 {
586 	return (void *)(handle);
587 }
588 
589 /**
590  * zbud_unmap() - maps the allocation associated with the given handle
591  * @pool:	pool in which the allocation resides
592  * @handle:	handle associated with the allocation to be unmapped
593  */
594 void zbud_unmap(struct zbud_pool *pool, unsigned long handle)
595 {
596 }
597 
598 /**
599  * zbud_get_pool_size() - gets the zbud pool size in pages
600  * @pool:	pool whose size is being queried
601  *
602  * Returns: size in pages of the given pool.  The pool lock need not be
603  * taken to access pages_nr.
604  */
605 u64 zbud_get_pool_size(struct zbud_pool *pool)
606 {
607 	return pool->pages_nr;
608 }
609 
610 static int __init init_zbud(void)
611 {
612 	/* Make sure the zbud header will fit in one chunk */
613 	BUILD_BUG_ON(sizeof(struct zbud_header) > ZHDR_SIZE_ALIGNED);
614 	pr_info("loaded\n");
615 
616 #ifdef CONFIG_ZPOOL
617 	zpool_register_driver(&zbud_zpool_driver);
618 #endif
619 
620 	return 0;
621 }
622 
623 static void __exit exit_zbud(void)
624 {
625 #ifdef CONFIG_ZPOOL
626 	zpool_unregister_driver(&zbud_zpool_driver);
627 #endif
628 
629 	pr_info("unloaded\n");
630 }
631 
632 module_init(init_zbud);
633 module_exit(exit_zbud);
634 
635 MODULE_LICENSE("GPL");
636 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
637 MODULE_DESCRIPTION("Buddy Allocator for Compressed Pages");
638