blob: 98ab01f910bc27caa9d15a81e32d132f5d12d372 [file] [log] [blame]
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001/*
2 * z3fold.c
3 *
4 * Author: Vitaly Wool <vitaly.wool@konsulko.com>
5 * Copyright (C) 2016, Sony Mobile Communications Inc.
6 *
7 * This implementation is based on zbud written by Seth Jennings.
8 *
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.
16 *
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.
19 *
20 * z3fold doesn't export any API and is meant to be used via zpool API.
21 */
22
23#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24
25#include <linux/atomic.h>
26#include <linux/list.h>
27#include <linux/mm.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>
33
34/*****************
35 * Structures
36*****************/
Vitaly Woolede93212017-02-24 14:57:17 -080037struct z3fold_pool;
38struct z3fold_ops {
39 int (*evict)(struct z3fold_pool *pool, unsigned long handle);
40};
41
42enum buddy {
43 HEADLESS = 0,
44 FIRST,
45 MIDDLE,
46 LAST,
47 BUDDIES_MAX
48};
49
50/*
51 * struct z3fold_header - z3fold page metadata occupying the first chunk of each
52 * z3fold page, except for HEADLESS pages
53 * @buddy: links the z3fold page into the relevant list in the pool
54 * @first_chunks: the size of the first buddy in chunks, 0 if free
55 * @middle_chunks: the size of the middle buddy in chunks, 0 if free
56 * @last_chunks: the size of the last buddy in chunks, 0 if free
57 * @first_num: the starting number (for the first handle)
58 */
59struct z3fold_header {
60 struct list_head buddy;
61 unsigned short first_chunks;
62 unsigned short middle_chunks;
63 unsigned short last_chunks;
64 unsigned short start_middle;
65 unsigned short first_num:2;
66};
67
Vitaly Wool9a001fc2016-05-20 16:58:30 -070068/*
69 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
70 * adjusting internal fragmentation. It also determines the number of
71 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
Vitaly Woolede93212017-02-24 14:57:17 -080072 * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks
73 * in the beginning of an allocated page are occupied by z3fold header, so
74 * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y),
75 * which shows the max number of free chunks in z3fold page, also there will
76 * be 63, or 62, respectively, freelists per pool.
Vitaly Wool9a001fc2016-05-20 16:58:30 -070077 */
78#define NCHUNKS_ORDER 6
79
80#define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER)
81#define CHUNK_SIZE (1 << CHUNK_SHIFT)
Vitaly Woolede93212017-02-24 14:57:17 -080082#define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE)
83#define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT)
84#define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SHIFT)
Vitaly Wool9a001fc2016-05-20 16:58:30 -070085#define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT)
86
zhong jiangf201ebd2017-02-22 15:46:51 -080087#define BUDDY_MASK (0x3)
Vitaly Wool9a001fc2016-05-20 16:58:30 -070088
Vitaly Wool9a001fc2016-05-20 16:58:30 -070089/**
90 * struct z3fold_pool - stores metadata for each z3fold pool
91 * @lock: protects all pool fields and first|last_chunk fields of any
92 * z3fold page in the pool
93 * @unbuddied: array of lists tracking z3fold pages that contain 2- buddies;
94 * the lists each z3fold page is added to depends on the size of
95 * its free region.
96 * @buddied: list tracking the z3fold pages that contain 3 buddies;
97 * these z3fold pages are full
98 * @lru: list tracking the z3fold pages in LRU order by most recently
99 * added buddy.
100 * @pages_nr: number of z3fold pages in the pool.
101 * @ops: pointer to a structure of user defined operations specified at
102 * pool creation time.
103 *
104 * This structure is allocated at pool creation time and maintains metadata
105 * pertaining to a particular z3fold pool.
106 */
107struct z3fold_pool {
108 spinlock_t lock;
109 struct list_head unbuddied[NCHUNKS];
110 struct list_head buddied;
111 struct list_head lru;
Vitaly Wool12d59ae2017-02-24 14:57:15 -0800112 atomic64_t pages_nr;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700113 const struct z3fold_ops *ops;
114 struct zpool *zpool;
115 const struct zpool_ops *zpool_ops;
116};
117
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700118/*
119 * Internal z3fold page flags
120 */
121enum z3fold_page_flags {
122 UNDER_RECLAIM = 0,
123 PAGE_HEADLESS,
124 MIDDLE_CHUNK_MAPPED,
125};
126
Vitaly Woolede93212017-02-24 14:57:17 -0800127
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700128/*****************
129 * Helpers
130*****************/
131
132/* Converts an allocation size in bytes to size in z3fold chunks */
133static int size_to_chunks(size_t size)
134{
135 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
136}
137
138#define for_each_unbuddied_list(_iter, _begin) \
139 for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
140
141/* Initializes the z3fold header of a newly allocated z3fold page */
142static struct z3fold_header *init_z3fold_page(struct page *page)
143{
144 struct z3fold_header *zhdr = page_address(page);
145
146 INIT_LIST_HEAD(&page->lru);
147 clear_bit(UNDER_RECLAIM, &page->private);
148 clear_bit(PAGE_HEADLESS, &page->private);
149 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
150
151 zhdr->first_chunks = 0;
152 zhdr->middle_chunks = 0;
153 zhdr->last_chunks = 0;
154 zhdr->first_num = 0;
155 zhdr->start_middle = 0;
156 INIT_LIST_HEAD(&zhdr->buddy);
157 return zhdr;
158}
159
160/* Resets the struct page fields and frees the page */
161static void free_z3fold_page(struct z3fold_header *zhdr)
162{
163 __free_page(virt_to_page(zhdr));
164}
165
166/*
167 * Encodes the handle of a particular buddy within a z3fold page
168 * Pool lock should be held as this function accesses first_num
169 */
170static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud)
171{
172 unsigned long handle;
173
174 handle = (unsigned long)zhdr;
175 if (bud != HEADLESS)
176 handle += (bud + zhdr->first_num) & BUDDY_MASK;
177 return handle;
178}
179
180/* Returns the z3fold page where a given handle is stored */
181static struct z3fold_header *handle_to_z3fold_header(unsigned long handle)
182{
183 return (struct z3fold_header *)(handle & PAGE_MASK);
184}
185
zhong jiangf201ebd2017-02-22 15:46:51 -0800186/*
187 * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle
188 * but that doesn't matter. because the masking will result in the
189 * correct buddy number.
190 */
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700191static enum buddy handle_to_buddy(unsigned long handle)
192{
193 struct z3fold_header *zhdr = handle_to_z3fold_header(handle);
194 return (handle - zhdr->first_num) & BUDDY_MASK;
195}
196
197/*
198 * Returns the number of free chunks in a z3fold page.
199 * NB: can't be used with HEADLESS pages.
200 */
201static int num_free_chunks(struct z3fold_header *zhdr)
202{
203 int nfree;
204 /*
205 * If there is a middle object, pick up the bigger free space
206 * either before or after it. Otherwise just subtract the number
207 * of chunks occupied by the first and the last objects.
208 */
209 if (zhdr->middle_chunks != 0) {
210 int nfree_before = zhdr->first_chunks ?
Vitaly Woolede93212017-02-24 14:57:17 -0800211 0 : zhdr->start_middle - ZHDR_CHUNKS;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700212 int nfree_after = zhdr->last_chunks ?
Vitaly Woolede93212017-02-24 14:57:17 -0800213 0 : TOTAL_CHUNKS -
214 (zhdr->start_middle + zhdr->middle_chunks);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700215 nfree = max(nfree_before, nfree_after);
216 } else
217 nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
218 return nfree;
219}
220
221/*****************
222 * API Functions
223*****************/
224/**
225 * z3fold_create_pool() - create a new z3fold pool
226 * @gfp: gfp flags when allocating the z3fold pool structure
227 * @ops: user-defined operations for the z3fold pool
228 *
229 * Return: pointer to the new z3fold pool or NULL if the metadata allocation
230 * failed.
231 */
232static struct z3fold_pool *z3fold_create_pool(gfp_t gfp,
233 const struct z3fold_ops *ops)
234{
235 struct z3fold_pool *pool;
236 int i;
237
238 pool = kzalloc(sizeof(struct z3fold_pool), gfp);
239 if (!pool)
240 return NULL;
241 spin_lock_init(&pool->lock);
242 for_each_unbuddied_list(i, 0)
243 INIT_LIST_HEAD(&pool->unbuddied[i]);
244 INIT_LIST_HEAD(&pool->buddied);
245 INIT_LIST_HEAD(&pool->lru);
Vitaly Wool12d59ae2017-02-24 14:57:15 -0800246 atomic64_set(&pool->pages_nr, 0);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700247 pool->ops = ops;
248 return pool;
249}
250
251/**
252 * z3fold_destroy_pool() - destroys an existing z3fold pool
253 * @pool: the z3fold pool to be destroyed
254 *
255 * The pool should be emptied before this function is called.
256 */
257static void z3fold_destroy_pool(struct z3fold_pool *pool)
258{
259 kfree(pool);
260}
261
Vitaly Woolede93212017-02-24 14:57:17 -0800262static inline void *mchunk_memmove(struct z3fold_header *zhdr,
263 unsigned short dst_chunk)
264{
265 void *beg = zhdr;
266 return memmove(beg + (dst_chunk << CHUNK_SHIFT),
267 beg + (zhdr->start_middle << CHUNK_SHIFT),
268 zhdr->middle_chunks << CHUNK_SHIFT);
269}
270
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700271/* Has to be called with lock held */
272static int z3fold_compact_page(struct z3fold_header *zhdr)
273{
274 struct page *page = virt_to_page(zhdr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700275
Vitaly Woolede93212017-02-24 14:57:17 -0800276 if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private))
277 return 0; /* can't move middle chunk, it's used */
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700278
Vitaly Woolede93212017-02-24 14:57:17 -0800279 if (zhdr->middle_chunks == 0)
280 return 0; /* nothing to compact */
281
282 if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
283 /* move to the beginning */
284 mchunk_memmove(zhdr, ZHDR_CHUNKS);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700285 zhdr->first_chunks = zhdr->middle_chunks;
286 zhdr->middle_chunks = 0;
287 zhdr->start_middle = 0;
288 zhdr->first_num++;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700289 }
Vitaly Woolede93212017-02-24 14:57:17 -0800290 return 1;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700291}
292
293/**
294 * z3fold_alloc() - allocates a region of a given size
295 * @pool: z3fold pool from which to allocate
296 * @size: size in bytes of the desired allocation
297 * @gfp: gfp flags used if the pool needs to grow
298 * @handle: handle of the new allocation
299 *
300 * This function will attempt to find a free region in the pool large enough to
301 * satisfy the allocation request. A search of the unbuddied lists is
302 * performed first. If no suitable free region is found, then a new page is
303 * allocated and added to the pool to satisfy the request.
304 *
305 * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used
306 * as z3fold pool pages.
307 *
308 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
309 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
310 * a new page.
311 */
312static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
313 unsigned long *handle)
314{
315 int chunks = 0, i, freechunks;
316 struct z3fold_header *zhdr = NULL;
317 enum buddy bud;
318 struct page *page;
319
320 if (!size || (gfp & __GFP_HIGHMEM))
321 return -EINVAL;
322
323 if (size > PAGE_SIZE)
324 return -ENOSPC;
325
326 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
327 bud = HEADLESS;
328 else {
329 chunks = size_to_chunks(size);
330 spin_lock(&pool->lock);
331
332 /* First, try to find an unbuddied z3fold page. */
333 zhdr = NULL;
334 for_each_unbuddied_list(i, chunks) {
335 if (!list_empty(&pool->unbuddied[i])) {
336 zhdr = list_first_entry(&pool->unbuddied[i],
337 struct z3fold_header, buddy);
338 page = virt_to_page(zhdr);
339 if (zhdr->first_chunks == 0) {
340 if (zhdr->middle_chunks != 0 &&
341 chunks >= zhdr->start_middle)
342 bud = LAST;
343 else
344 bud = FIRST;
345 } else if (zhdr->last_chunks == 0)
346 bud = LAST;
347 else if (zhdr->middle_chunks == 0)
348 bud = MIDDLE;
349 else {
350 pr_err("No free chunks in unbuddied\n");
351 WARN_ON(1);
352 continue;
353 }
354 list_del(&zhdr->buddy);
355 goto found;
356 }
357 }
358 bud = FIRST;
359 spin_unlock(&pool->lock);
360 }
361
362 /* Couldn't find unbuddied z3fold page, create new one */
363 page = alloc_page(gfp);
364 if (!page)
365 return -ENOMEM;
366 spin_lock(&pool->lock);
Vitaly Wool12d59ae2017-02-24 14:57:15 -0800367 atomic64_inc(&pool->pages_nr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700368 zhdr = init_z3fold_page(page);
369
370 if (bud == HEADLESS) {
371 set_bit(PAGE_HEADLESS, &page->private);
372 goto headless;
373 }
374
375found:
376 if (bud == FIRST)
377 zhdr->first_chunks = chunks;
378 else if (bud == LAST)
379 zhdr->last_chunks = chunks;
380 else {
381 zhdr->middle_chunks = chunks;
Vitaly Woolede93212017-02-24 14:57:17 -0800382 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700383 }
384
385 if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 ||
386 zhdr->middle_chunks == 0) {
387 /* Add to unbuddied list */
388 freechunks = num_free_chunks(zhdr);
389 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
390 } else {
391 /* Add to buddied list */
392 list_add(&zhdr->buddy, &pool->buddied);
393 }
394
395headless:
396 /* Add/move z3fold page to beginning of LRU */
397 if (!list_empty(&page->lru))
398 list_del(&page->lru);
399
400 list_add(&page->lru, &pool->lru);
401
402 *handle = encode_handle(zhdr, bud);
403 spin_unlock(&pool->lock);
404
405 return 0;
406}
407
408/**
409 * z3fold_free() - frees the allocation associated with the given handle
410 * @pool: pool in which the allocation resided
411 * @handle: handle associated with the allocation returned by z3fold_alloc()
412 *
413 * In the case that the z3fold page in which the allocation resides is under
414 * reclaim, as indicated by the PG_reclaim flag being set, this function
415 * only sets the first|last_chunks to 0. The page is actually freed
416 * once both buddies are evicted (see z3fold_reclaim_page() below).
417 */
418static void z3fold_free(struct z3fold_pool *pool, unsigned long handle)
419{
420 struct z3fold_header *zhdr;
421 int freechunks;
422 struct page *page;
423 enum buddy bud;
424
425 spin_lock(&pool->lock);
426 zhdr = handle_to_z3fold_header(handle);
427 page = virt_to_page(zhdr);
428
429 if (test_bit(PAGE_HEADLESS, &page->private)) {
430 /* HEADLESS page stored */
431 bud = HEADLESS;
432 } else {
Vitaly Wool43afc192016-06-03 14:55:47 -0700433 bud = handle_to_buddy(handle);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700434
435 switch (bud) {
436 case FIRST:
437 zhdr->first_chunks = 0;
438 break;
439 case MIDDLE:
440 zhdr->middle_chunks = 0;
441 zhdr->start_middle = 0;
442 break;
443 case LAST:
444 zhdr->last_chunks = 0;
445 break;
446 default:
447 pr_err("%s: unknown bud %d\n", __func__, bud);
448 WARN_ON(1);
449 spin_unlock(&pool->lock);
450 return;
451 }
452 }
453
454 if (test_bit(UNDER_RECLAIM, &page->private)) {
455 /* z3fold page is under reclaim, reclaim will free */
456 spin_unlock(&pool->lock);
457 return;
458 }
459
Vitaly Wool12d59ae2017-02-24 14:57:15 -0800460 /* Remove from existing buddy list */
461 if (bud != HEADLESS)
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700462 list_del(&zhdr->buddy);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700463
464 if (bud == HEADLESS ||
465 (zhdr->first_chunks == 0 && zhdr->middle_chunks == 0 &&
466 zhdr->last_chunks == 0)) {
467 /* z3fold page is empty, free */
468 list_del(&page->lru);
469 clear_bit(PAGE_HEADLESS, &page->private);
470 free_z3fold_page(zhdr);
Vitaly Wool12d59ae2017-02-24 14:57:15 -0800471 atomic64_dec(&pool->pages_nr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700472 } else {
473 z3fold_compact_page(zhdr);
474 /* Add to the unbuddied list */
475 freechunks = num_free_chunks(zhdr);
476 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
477 }
478
479 spin_unlock(&pool->lock);
480}
481
482/**
483 * z3fold_reclaim_page() - evicts allocations from a pool page and frees it
484 * @pool: pool from which a page will attempt to be evicted
485 * @retires: number of pages on the LRU list for which eviction will
486 * be attempted before failing
487 *
488 * z3fold reclaim is different from normal system reclaim in that it is done
489 * from the bottom, up. This is because only the bottom layer, z3fold, has
490 * information on how the allocations are organized within each z3fold page.
491 * This has the potential to create interesting locking situations between
492 * z3fold and the user, however.
493 *
494 * To avoid these, this is how z3fold_reclaim_page() should be called:
495
496 * The user detects a page should be reclaimed and calls z3fold_reclaim_page().
497 * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and
498 * call the user-defined eviction handler with the pool and handle as
499 * arguments.
500 *
501 * If the handle can not be evicted, the eviction handler should return
502 * non-zero. z3fold_reclaim_page() will add the z3fold page back to the
503 * appropriate list and try the next z3fold page on the LRU up to
504 * a user defined number of retries.
505 *
506 * If the handle is successfully evicted, the eviction handler should
507 * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free()
508 * contains logic to delay freeing the page if the page is under reclaim,
509 * as indicated by the setting of the PG_reclaim flag on the underlying page.
510 *
511 * If all buddies in the z3fold page are successfully evicted, then the
512 * z3fold page can be freed.
513 *
514 * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are
515 * no pages to evict or an eviction handler is not registered, -EAGAIN if
516 * the retry limit was hit.
517 */
518static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries)
519{
520 int i, ret = 0, freechunks;
521 struct z3fold_header *zhdr;
522 struct page *page;
523 unsigned long first_handle = 0, middle_handle = 0, last_handle = 0;
524
525 spin_lock(&pool->lock);
526 if (!pool->ops || !pool->ops->evict || list_empty(&pool->lru) ||
527 retries == 0) {
528 spin_unlock(&pool->lock);
529 return -EINVAL;
530 }
531 for (i = 0; i < retries; i++) {
532 page = list_last_entry(&pool->lru, struct page, lru);
533 list_del(&page->lru);
534
535 /* Protect z3fold page against free */
536 set_bit(UNDER_RECLAIM, &page->private);
537 zhdr = page_address(page);
538 if (!test_bit(PAGE_HEADLESS, &page->private)) {
539 list_del(&zhdr->buddy);
540 /*
541 * We need encode the handles before unlocking, since
542 * we can race with free that will set
543 * (first|last)_chunks to 0
544 */
545 first_handle = 0;
546 last_handle = 0;
547 middle_handle = 0;
548 if (zhdr->first_chunks)
549 first_handle = encode_handle(zhdr, FIRST);
550 if (zhdr->middle_chunks)
551 middle_handle = encode_handle(zhdr, MIDDLE);
552 if (zhdr->last_chunks)
553 last_handle = encode_handle(zhdr, LAST);
554 } else {
555 first_handle = encode_handle(zhdr, HEADLESS);
556 last_handle = middle_handle = 0;
557 }
558
559 spin_unlock(&pool->lock);
560
561 /* Issue the eviction callback(s) */
562 if (middle_handle) {
563 ret = pool->ops->evict(pool, middle_handle);
564 if (ret)
565 goto next;
566 }
567 if (first_handle) {
568 ret = pool->ops->evict(pool, first_handle);
569 if (ret)
570 goto next;
571 }
572 if (last_handle) {
573 ret = pool->ops->evict(pool, last_handle);
574 if (ret)
575 goto next;
576 }
577next:
578 spin_lock(&pool->lock);
579 clear_bit(UNDER_RECLAIM, &page->private);
580 if ((test_bit(PAGE_HEADLESS, &page->private) && ret == 0) ||
581 (zhdr->first_chunks == 0 && zhdr->last_chunks == 0 &&
582 zhdr->middle_chunks == 0)) {
583 /*
584 * All buddies are now free, free the z3fold page and
585 * return success.
586 */
587 clear_bit(PAGE_HEADLESS, &page->private);
588 free_z3fold_page(zhdr);
Vitaly Wool12d59ae2017-02-24 14:57:15 -0800589 atomic64_dec(&pool->pages_nr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700590 spin_unlock(&pool->lock);
591 return 0;
Vitaly Wool43afc192016-06-03 14:55:47 -0700592 } else if (!test_bit(PAGE_HEADLESS, &page->private)) {
593 if (zhdr->first_chunks != 0 &&
594 zhdr->last_chunks != 0 &&
595 zhdr->middle_chunks != 0) {
596 /* Full, add to buddied list */
597 list_add(&zhdr->buddy, &pool->buddied);
598 } else {
599 z3fold_compact_page(zhdr);
600 /* add to unbuddied list */
601 freechunks = num_free_chunks(zhdr);
602 list_add(&zhdr->buddy,
603 &pool->unbuddied[freechunks]);
604 }
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700605 }
606
607 /* add to beginning of LRU */
608 list_add(&page->lru, &pool->lru);
609 }
610 spin_unlock(&pool->lock);
611 return -EAGAIN;
612}
613
614/**
615 * z3fold_map() - maps the allocation associated with the given handle
616 * @pool: pool in which the allocation resides
617 * @handle: handle associated with the allocation to be mapped
618 *
619 * Extracts the buddy number from handle and constructs the pointer to the
620 * correct starting chunk within the page.
621 *
622 * Returns: a pointer to the mapped allocation
623 */
624static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle)
625{
626 struct z3fold_header *zhdr;
627 struct page *page;
628 void *addr;
629 enum buddy buddy;
630
631 spin_lock(&pool->lock);
632 zhdr = handle_to_z3fold_header(handle);
633 addr = zhdr;
634 page = virt_to_page(zhdr);
635
636 if (test_bit(PAGE_HEADLESS, &page->private))
637 goto out;
638
639 buddy = handle_to_buddy(handle);
640 switch (buddy) {
641 case FIRST:
642 addr += ZHDR_SIZE_ALIGNED;
643 break;
644 case MIDDLE:
645 addr += zhdr->start_middle << CHUNK_SHIFT;
646 set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
647 break;
648 case LAST:
649 addr += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
650 break;
651 default:
652 pr_err("unknown buddy id %d\n", buddy);
653 WARN_ON(1);
654 addr = NULL;
655 break;
656 }
657out:
658 spin_unlock(&pool->lock);
659 return addr;
660}
661
662/**
663 * z3fold_unmap() - unmaps the allocation associated with the given handle
664 * @pool: pool in which the allocation resides
665 * @handle: handle associated with the allocation to be unmapped
666 */
667static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle)
668{
669 struct z3fold_header *zhdr;
670 struct page *page;
671 enum buddy buddy;
672
673 spin_lock(&pool->lock);
674 zhdr = handle_to_z3fold_header(handle);
675 page = virt_to_page(zhdr);
676
677 if (test_bit(PAGE_HEADLESS, &page->private)) {
678 spin_unlock(&pool->lock);
679 return;
680 }
681
682 buddy = handle_to_buddy(handle);
683 if (buddy == MIDDLE)
684 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
685 spin_unlock(&pool->lock);
686}
687
688/**
689 * z3fold_get_pool_size() - gets the z3fold pool size in pages
690 * @pool: pool whose size is being queried
691 *
Vitaly Wool12d59ae2017-02-24 14:57:15 -0800692 * Returns: size in pages of the given pool.
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700693 */
694static u64 z3fold_get_pool_size(struct z3fold_pool *pool)
695{
Vitaly Wool12d59ae2017-02-24 14:57:15 -0800696 return atomic64_read(&pool->pages_nr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700697}
698
699/*****************
700 * zpool
701 ****************/
702
703static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle)
704{
705 if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict)
706 return pool->zpool_ops->evict(pool->zpool, handle);
707 else
708 return -ENOENT;
709}
710
711static const struct z3fold_ops z3fold_zpool_ops = {
712 .evict = z3fold_zpool_evict
713};
714
715static void *z3fold_zpool_create(const char *name, gfp_t gfp,
716 const struct zpool_ops *zpool_ops,
717 struct zpool *zpool)
718{
719 struct z3fold_pool *pool;
720
721 pool = z3fold_create_pool(gfp, zpool_ops ? &z3fold_zpool_ops : NULL);
722 if (pool) {
723 pool->zpool = zpool;
724 pool->zpool_ops = zpool_ops;
725 }
726 return pool;
727}
728
729static void z3fold_zpool_destroy(void *pool)
730{
731 z3fold_destroy_pool(pool);
732}
733
734static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp,
735 unsigned long *handle)
736{
737 return z3fold_alloc(pool, size, gfp, handle);
738}
739static void z3fold_zpool_free(void *pool, unsigned long handle)
740{
741 z3fold_free(pool, handle);
742}
743
744static int z3fold_zpool_shrink(void *pool, unsigned int pages,
745 unsigned int *reclaimed)
746{
747 unsigned int total = 0;
748 int ret = -EINVAL;
749
750 while (total < pages) {
751 ret = z3fold_reclaim_page(pool, 8);
752 if (ret < 0)
753 break;
754 total++;
755 }
756
757 if (reclaimed)
758 *reclaimed = total;
759
760 return ret;
761}
762
763static void *z3fold_zpool_map(void *pool, unsigned long handle,
764 enum zpool_mapmode mm)
765{
766 return z3fold_map(pool, handle);
767}
768static void z3fold_zpool_unmap(void *pool, unsigned long handle)
769{
770 z3fold_unmap(pool, handle);
771}
772
773static u64 z3fold_zpool_total_size(void *pool)
774{
775 return z3fold_get_pool_size(pool) * PAGE_SIZE;
776}
777
778static struct zpool_driver z3fold_zpool_driver = {
779 .type = "z3fold",
780 .owner = THIS_MODULE,
781 .create = z3fold_zpool_create,
782 .destroy = z3fold_zpool_destroy,
783 .malloc = z3fold_zpool_malloc,
784 .free = z3fold_zpool_free,
785 .shrink = z3fold_zpool_shrink,
786 .map = z3fold_zpool_map,
787 .unmap = z3fold_zpool_unmap,
788 .total_size = z3fold_zpool_total_size,
789};
790
791MODULE_ALIAS("zpool-z3fold");
792
793static int __init init_z3fold(void)
794{
Vitaly Woolede93212017-02-24 14:57:17 -0800795 /* Make sure the z3fold header is not larger than the page size */
796 BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700797 zpool_register_driver(&z3fold_zpool_driver);
798
799 return 0;
800}
801
802static void __exit exit_z3fold(void)
803{
804 zpool_unregister_driver(&z3fold_zpool_driver);
805}
806
807module_init(init_z3fold);
808module_exit(exit_z3fold);
809
810MODULE_LICENSE("GPL");
811MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>");
812MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages");