Commit | Line | Data |
---|---|---|
9a001fc1 VW |
1 | /* |
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 | *****************/ | |
ede93213 VW |
37 | struct z3fold_pool; |
38 | struct z3fold_ops { | |
39 | int (*evict)(struct z3fold_pool *pool, unsigned long handle); | |
40 | }; | |
41 | ||
42 | enum 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 | |
2f1e5e4d | 54 | * @page_lock: per-page lock |
5a27aa82 | 55 | * @refcount: reference cound for the z3fold page |
ede93213 VW |
56 | * @first_chunks: the size of the first buddy in chunks, 0 if free |
57 | * @middle_chunks: the size of the middle buddy in chunks, 0 if free | |
58 | * @last_chunks: the size of the last buddy in chunks, 0 if free | |
59 | * @first_num: the starting number (for the first handle) | |
60 | */ | |
61 | struct z3fold_header { | |
62 | struct list_head buddy; | |
2f1e5e4d | 63 | spinlock_t page_lock; |
5a27aa82 | 64 | struct kref refcount; |
ede93213 VW |
65 | unsigned short first_chunks; |
66 | unsigned short middle_chunks; | |
67 | unsigned short last_chunks; | |
68 | unsigned short start_middle; | |
69 | unsigned short first_num:2; | |
70 | }; | |
71 | ||
9a001fc1 VW |
72 | /* |
73 | * NCHUNKS_ORDER determines the internal allocation granularity, effectively | |
74 | * adjusting internal fragmentation. It also determines the number of | |
75 | * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the | |
ede93213 VW |
76 | * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks |
77 | * in the beginning of an allocated page are occupied by z3fold header, so | |
78 | * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y), | |
79 | * which shows the max number of free chunks in z3fold page, also there will | |
80 | * be 63, or 62, respectively, freelists per pool. | |
9a001fc1 VW |
81 | */ |
82 | #define NCHUNKS_ORDER 6 | |
83 | ||
84 | #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER) | |
85 | #define CHUNK_SIZE (1 << CHUNK_SHIFT) | |
ede93213 VW |
86 | #define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE) |
87 | #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT) | |
88 | #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SHIFT) | |
9a001fc1 VW |
89 | #define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT) |
90 | ||
f201ebd8 | 91 | #define BUDDY_MASK (0x3) |
9a001fc1 | 92 | |
9a001fc1 VW |
93 | /** |
94 | * struct z3fold_pool - stores metadata for each z3fold pool | |
95 | * @lock: protects all pool fields and first|last_chunk fields of any | |
96 | * z3fold page in the pool | |
97 | * @unbuddied: array of lists tracking z3fold pages that contain 2- buddies; | |
98 | * the lists each z3fold page is added to depends on the size of | |
99 | * its free region. | |
9a001fc1 VW |
100 | * @lru: list tracking the z3fold pages in LRU order by most recently |
101 | * added buddy. | |
102 | * @pages_nr: number of z3fold pages in the pool. | |
103 | * @ops: pointer to a structure of user defined operations specified at | |
104 | * pool creation time. | |
105 | * | |
106 | * This structure is allocated at pool creation time and maintains metadata | |
107 | * pertaining to a particular z3fold pool. | |
108 | */ | |
109 | struct z3fold_pool { | |
110 | spinlock_t lock; | |
111 | struct list_head unbuddied[NCHUNKS]; | |
9a001fc1 | 112 | struct list_head lru; |
12d59ae6 | 113 | atomic64_t pages_nr; |
9a001fc1 VW |
114 | const struct z3fold_ops *ops; |
115 | struct zpool *zpool; | |
116 | const struct zpool_ops *zpool_ops; | |
117 | }; | |
118 | ||
9a001fc1 VW |
119 | /* |
120 | * Internal z3fold page flags | |
121 | */ | |
122 | enum z3fold_page_flags { | |
5a27aa82 | 123 | PAGE_HEADLESS = 0, |
9a001fc1 VW |
124 | MIDDLE_CHUNK_MAPPED, |
125 | }; | |
126 | ||
ede93213 | 127 | |
9a001fc1 VW |
128 | /***************** |
129 | * Helpers | |
130 | *****************/ | |
131 | ||
132 | /* Converts an allocation size in bytes to size in z3fold chunks */ | |
133 | static 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 */ | |
142 | static 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); | |
9a001fc1 VW |
147 | clear_bit(PAGE_HEADLESS, &page->private); |
148 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
149 | ||
2f1e5e4d | 150 | spin_lock_init(&zhdr->page_lock); |
5a27aa82 | 151 | kref_init(&zhdr->refcount); |
9a001fc1 VW |
152 | zhdr->first_chunks = 0; |
153 | zhdr->middle_chunks = 0; | |
154 | zhdr->last_chunks = 0; | |
155 | zhdr->first_num = 0; | |
156 | zhdr->start_middle = 0; | |
157 | INIT_LIST_HEAD(&zhdr->buddy); | |
158 | return zhdr; | |
159 | } | |
160 | ||
161 | /* Resets the struct page fields and frees the page */ | |
5a27aa82 | 162 | static void free_z3fold_page(struct page *page) |
9a001fc1 | 163 | { |
5a27aa82 VW |
164 | __free_page(page); |
165 | } | |
166 | ||
167 | static void release_z3fold_page(struct kref *ref) | |
168 | { | |
169 | struct z3fold_header *zhdr; | |
170 | struct page *page; | |
171 | ||
172 | zhdr = container_of(ref, struct z3fold_header, refcount); | |
173 | page = virt_to_page(zhdr); | |
174 | ||
175 | if (!list_empty(&zhdr->buddy)) | |
176 | list_del(&zhdr->buddy); | |
177 | if (!list_empty(&page->lru)) | |
178 | list_del(&page->lru); | |
179 | free_z3fold_page(page); | |
9a001fc1 VW |
180 | } |
181 | ||
2f1e5e4d VW |
182 | /* Lock a z3fold page */ |
183 | static inline void z3fold_page_lock(struct z3fold_header *zhdr) | |
184 | { | |
185 | spin_lock(&zhdr->page_lock); | |
186 | } | |
187 | ||
188 | /* Unlock a z3fold page */ | |
189 | static inline void z3fold_page_unlock(struct z3fold_header *zhdr) | |
190 | { | |
191 | spin_unlock(&zhdr->page_lock); | |
192 | } | |
193 | ||
9a001fc1 VW |
194 | /* |
195 | * Encodes the handle of a particular buddy within a z3fold page | |
196 | * Pool lock should be held as this function accesses first_num | |
197 | */ | |
198 | static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud) | |
199 | { | |
200 | unsigned long handle; | |
201 | ||
202 | handle = (unsigned long)zhdr; | |
203 | if (bud != HEADLESS) | |
204 | handle += (bud + zhdr->first_num) & BUDDY_MASK; | |
205 | return handle; | |
206 | } | |
207 | ||
208 | /* Returns the z3fold page where a given handle is stored */ | |
209 | static struct z3fold_header *handle_to_z3fold_header(unsigned long handle) | |
210 | { | |
211 | return (struct z3fold_header *)(handle & PAGE_MASK); | |
212 | } | |
213 | ||
f201ebd8 | 214 | /* |
215 | * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle | |
216 | * but that doesn't matter. because the masking will result in the | |
217 | * correct buddy number. | |
218 | */ | |
9a001fc1 VW |
219 | static enum buddy handle_to_buddy(unsigned long handle) |
220 | { | |
221 | struct z3fold_header *zhdr = handle_to_z3fold_header(handle); | |
222 | return (handle - zhdr->first_num) & BUDDY_MASK; | |
223 | } | |
224 | ||
225 | /* | |
226 | * Returns the number of free chunks in a z3fold page. | |
227 | * NB: can't be used with HEADLESS pages. | |
228 | */ | |
229 | static int num_free_chunks(struct z3fold_header *zhdr) | |
230 | { | |
231 | int nfree; | |
232 | /* | |
233 | * If there is a middle object, pick up the bigger free space | |
234 | * either before or after it. Otherwise just subtract the number | |
235 | * of chunks occupied by the first and the last objects. | |
236 | */ | |
237 | if (zhdr->middle_chunks != 0) { | |
238 | int nfree_before = zhdr->first_chunks ? | |
ede93213 | 239 | 0 : zhdr->start_middle - ZHDR_CHUNKS; |
9a001fc1 | 240 | int nfree_after = zhdr->last_chunks ? |
ede93213 VW |
241 | 0 : TOTAL_CHUNKS - |
242 | (zhdr->start_middle + zhdr->middle_chunks); | |
9a001fc1 VW |
243 | nfree = max(nfree_before, nfree_after); |
244 | } else | |
245 | nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks; | |
246 | return nfree; | |
247 | } | |
248 | ||
249 | /***************** | |
250 | * API Functions | |
251 | *****************/ | |
252 | /** | |
253 | * z3fold_create_pool() - create a new z3fold pool | |
254 | * @gfp: gfp flags when allocating the z3fold pool structure | |
255 | * @ops: user-defined operations for the z3fold pool | |
256 | * | |
257 | * Return: pointer to the new z3fold pool or NULL if the metadata allocation | |
258 | * failed. | |
259 | */ | |
260 | static struct z3fold_pool *z3fold_create_pool(gfp_t gfp, | |
261 | const struct z3fold_ops *ops) | |
262 | { | |
263 | struct z3fold_pool *pool; | |
264 | int i; | |
265 | ||
266 | pool = kzalloc(sizeof(struct z3fold_pool), gfp); | |
267 | if (!pool) | |
268 | return NULL; | |
269 | spin_lock_init(&pool->lock); | |
270 | for_each_unbuddied_list(i, 0) | |
271 | INIT_LIST_HEAD(&pool->unbuddied[i]); | |
9a001fc1 | 272 | INIT_LIST_HEAD(&pool->lru); |
12d59ae6 | 273 | atomic64_set(&pool->pages_nr, 0); |
9a001fc1 VW |
274 | pool->ops = ops; |
275 | return pool; | |
276 | } | |
277 | ||
278 | /** | |
279 | * z3fold_destroy_pool() - destroys an existing z3fold pool | |
280 | * @pool: the z3fold pool to be destroyed | |
281 | * | |
282 | * The pool should be emptied before this function is called. | |
283 | */ | |
284 | static void z3fold_destroy_pool(struct z3fold_pool *pool) | |
285 | { | |
286 | kfree(pool); | |
287 | } | |
288 | ||
ede93213 VW |
289 | static inline void *mchunk_memmove(struct z3fold_header *zhdr, |
290 | unsigned short dst_chunk) | |
291 | { | |
292 | void *beg = zhdr; | |
293 | return memmove(beg + (dst_chunk << CHUNK_SHIFT), | |
294 | beg + (zhdr->start_middle << CHUNK_SHIFT), | |
295 | zhdr->middle_chunks << CHUNK_SHIFT); | |
296 | } | |
297 | ||
1b096e5a | 298 | #define BIG_CHUNK_GAP 3 |
9a001fc1 VW |
299 | /* Has to be called with lock held */ |
300 | static int z3fold_compact_page(struct z3fold_header *zhdr) | |
301 | { | |
302 | struct page *page = virt_to_page(zhdr); | |
9a001fc1 | 303 | |
ede93213 VW |
304 | if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private)) |
305 | return 0; /* can't move middle chunk, it's used */ | |
9a001fc1 | 306 | |
ede93213 VW |
307 | if (zhdr->middle_chunks == 0) |
308 | return 0; /* nothing to compact */ | |
309 | ||
310 | if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) { | |
311 | /* move to the beginning */ | |
312 | mchunk_memmove(zhdr, ZHDR_CHUNKS); | |
9a001fc1 VW |
313 | zhdr->first_chunks = zhdr->middle_chunks; |
314 | zhdr->middle_chunks = 0; | |
315 | zhdr->start_middle = 0; | |
316 | zhdr->first_num++; | |
1b096e5a | 317 | return 1; |
9a001fc1 | 318 | } |
1b096e5a VW |
319 | |
320 | /* | |
321 | * moving data is expensive, so let's only do that if | |
322 | * there's substantial gain (at least BIG_CHUNK_GAP chunks) | |
323 | */ | |
324 | if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 && | |
325 | zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >= | |
326 | BIG_CHUNK_GAP) { | |
327 | mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS); | |
328 | zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS; | |
329 | return 1; | |
330 | } else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 && | |
331 | TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle | |
332 | + zhdr->middle_chunks) >= | |
333 | BIG_CHUNK_GAP) { | |
334 | unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks - | |
335 | zhdr->middle_chunks; | |
336 | mchunk_memmove(zhdr, new_start); | |
337 | zhdr->start_middle = new_start; | |
338 | return 1; | |
339 | } | |
340 | ||
341 | return 0; | |
9a001fc1 VW |
342 | } |
343 | ||
344 | /** | |
345 | * z3fold_alloc() - allocates a region of a given size | |
346 | * @pool: z3fold pool from which to allocate | |
347 | * @size: size in bytes of the desired allocation | |
348 | * @gfp: gfp flags used if the pool needs to grow | |
349 | * @handle: handle of the new allocation | |
350 | * | |
351 | * This function will attempt to find a free region in the pool large enough to | |
352 | * satisfy the allocation request. A search of the unbuddied lists is | |
353 | * performed first. If no suitable free region is found, then a new page is | |
354 | * allocated and added to the pool to satisfy the request. | |
355 | * | |
356 | * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used | |
357 | * as z3fold pool pages. | |
358 | * | |
359 | * Return: 0 if success and handle is set, otherwise -EINVAL if the size or | |
360 | * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate | |
361 | * a new page. | |
362 | */ | |
363 | static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp, | |
364 | unsigned long *handle) | |
365 | { | |
366 | int chunks = 0, i, freechunks; | |
367 | struct z3fold_header *zhdr = NULL; | |
368 | enum buddy bud; | |
369 | struct page *page; | |
370 | ||
371 | if (!size || (gfp & __GFP_HIGHMEM)) | |
372 | return -EINVAL; | |
373 | ||
374 | if (size > PAGE_SIZE) | |
375 | return -ENOSPC; | |
376 | ||
377 | if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE) | |
378 | bud = HEADLESS; | |
379 | else { | |
380 | chunks = size_to_chunks(size); | |
9a001fc1 VW |
381 | |
382 | /* First, try to find an unbuddied z3fold page. */ | |
383 | zhdr = NULL; | |
384 | for_each_unbuddied_list(i, chunks) { | |
2f1e5e4d VW |
385 | spin_lock(&pool->lock); |
386 | zhdr = list_first_entry_or_null(&pool->unbuddied[i], | |
9a001fc1 | 387 | struct z3fold_header, buddy); |
2f1e5e4d VW |
388 | if (!zhdr) { |
389 | spin_unlock(&pool->lock); | |
390 | continue; | |
391 | } | |
5a27aa82 | 392 | kref_get(&zhdr->refcount); |
2f1e5e4d VW |
393 | list_del_init(&zhdr->buddy); |
394 | spin_unlock(&pool->lock); | |
395 | ||
396 | page = virt_to_page(zhdr); | |
397 | z3fold_page_lock(zhdr); | |
398 | if (zhdr->first_chunks == 0) { | |
399 | if (zhdr->middle_chunks != 0 && | |
400 | chunks >= zhdr->start_middle) | |
9a001fc1 | 401 | bud = LAST; |
2f1e5e4d VW |
402 | else |
403 | bud = FIRST; | |
404 | } else if (zhdr->last_chunks == 0) | |
405 | bud = LAST; | |
406 | else if (zhdr->middle_chunks == 0) | |
407 | bud = MIDDLE; | |
408 | else { | |
5a27aa82 | 409 | z3fold_page_unlock(zhdr); |
2f1e5e4d | 410 | spin_lock(&pool->lock); |
5a27aa82 VW |
411 | if (kref_put(&zhdr->refcount, |
412 | release_z3fold_page)) | |
413 | atomic64_dec(&pool->pages_nr); | |
2f1e5e4d | 414 | spin_unlock(&pool->lock); |
2f1e5e4d VW |
415 | pr_err("No free chunks in unbuddied\n"); |
416 | WARN_ON(1); | |
417 | continue; | |
9a001fc1 | 418 | } |
2f1e5e4d | 419 | goto found; |
9a001fc1 VW |
420 | } |
421 | bud = FIRST; | |
9a001fc1 VW |
422 | } |
423 | ||
424 | /* Couldn't find unbuddied z3fold page, create new one */ | |
425 | page = alloc_page(gfp); | |
426 | if (!page) | |
427 | return -ENOMEM; | |
2f1e5e4d | 428 | |
12d59ae6 | 429 | atomic64_inc(&pool->pages_nr); |
9a001fc1 VW |
430 | zhdr = init_z3fold_page(page); |
431 | ||
432 | if (bud == HEADLESS) { | |
433 | set_bit(PAGE_HEADLESS, &page->private); | |
2f1e5e4d | 434 | spin_lock(&pool->lock); |
9a001fc1 VW |
435 | goto headless; |
436 | } | |
2f1e5e4d | 437 | z3fold_page_lock(zhdr); |
9a001fc1 VW |
438 | |
439 | found: | |
440 | if (bud == FIRST) | |
441 | zhdr->first_chunks = chunks; | |
442 | else if (bud == LAST) | |
443 | zhdr->last_chunks = chunks; | |
444 | else { | |
445 | zhdr->middle_chunks = chunks; | |
ede93213 | 446 | zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS; |
9a001fc1 VW |
447 | } |
448 | ||
2f1e5e4d | 449 | spin_lock(&pool->lock); |
9a001fc1 VW |
450 | if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 || |
451 | zhdr->middle_chunks == 0) { | |
452 | /* Add to unbuddied list */ | |
453 | freechunks = num_free_chunks(zhdr); | |
454 | list_add(&zhdr->buddy, &pool->unbuddied[freechunks]); | |
9a001fc1 VW |
455 | } |
456 | ||
457 | headless: | |
458 | /* Add/move z3fold page to beginning of LRU */ | |
459 | if (!list_empty(&page->lru)) | |
460 | list_del(&page->lru); | |
461 | ||
462 | list_add(&page->lru, &pool->lru); | |
463 | ||
464 | *handle = encode_handle(zhdr, bud); | |
465 | spin_unlock(&pool->lock); | |
2f1e5e4d VW |
466 | if (bud != HEADLESS) |
467 | z3fold_page_unlock(zhdr); | |
9a001fc1 VW |
468 | |
469 | return 0; | |
470 | } | |
471 | ||
472 | /** | |
473 | * z3fold_free() - frees the allocation associated with the given handle | |
474 | * @pool: pool in which the allocation resided | |
475 | * @handle: handle associated with the allocation returned by z3fold_alloc() | |
476 | * | |
477 | * In the case that the z3fold page in which the allocation resides is under | |
478 | * reclaim, as indicated by the PG_reclaim flag being set, this function | |
479 | * only sets the first|last_chunks to 0. The page is actually freed | |
480 | * once both buddies are evicted (see z3fold_reclaim_page() below). | |
481 | */ | |
482 | static void z3fold_free(struct z3fold_pool *pool, unsigned long handle) | |
483 | { | |
484 | struct z3fold_header *zhdr; | |
485 | int freechunks; | |
486 | struct page *page; | |
487 | enum buddy bud; | |
488 | ||
9a001fc1 VW |
489 | zhdr = handle_to_z3fold_header(handle); |
490 | page = virt_to_page(zhdr); | |
491 | ||
492 | if (test_bit(PAGE_HEADLESS, &page->private)) { | |
493 | /* HEADLESS page stored */ | |
494 | bud = HEADLESS; | |
495 | } else { | |
2f1e5e4d | 496 | z3fold_page_lock(zhdr); |
43afc194 | 497 | bud = handle_to_buddy(handle); |
9a001fc1 VW |
498 | |
499 | switch (bud) { | |
500 | case FIRST: | |
501 | zhdr->first_chunks = 0; | |
502 | break; | |
503 | case MIDDLE: | |
504 | zhdr->middle_chunks = 0; | |
505 | zhdr->start_middle = 0; | |
506 | break; | |
507 | case LAST: | |
508 | zhdr->last_chunks = 0; | |
509 | break; | |
510 | default: | |
511 | pr_err("%s: unknown bud %d\n", __func__, bud); | |
512 | WARN_ON(1); | |
2f1e5e4d | 513 | z3fold_page_unlock(zhdr); |
9a001fc1 VW |
514 | return; |
515 | } | |
516 | } | |
517 | ||
5a27aa82 | 518 | if (bud == HEADLESS) { |
2f1e5e4d | 519 | spin_lock(&pool->lock); |
9a001fc1 | 520 | list_del(&page->lru); |
2f1e5e4d | 521 | spin_unlock(&pool->lock); |
5a27aa82 | 522 | free_z3fold_page(page); |
12d59ae6 | 523 | atomic64_dec(&pool->pages_nr); |
9a001fc1 | 524 | } else { |
5a27aa82 VW |
525 | if (zhdr->first_chunks != 0 || zhdr->middle_chunks != 0 || |
526 | zhdr->last_chunks != 0) { | |
527 | z3fold_compact_page(zhdr); | |
528 | /* Add to the unbuddied list */ | |
529 | spin_lock(&pool->lock); | |
530 | if (!list_empty(&zhdr->buddy)) | |
531 | list_del(&zhdr->buddy); | |
532 | freechunks = num_free_chunks(zhdr); | |
533 | list_add(&zhdr->buddy, &pool->unbuddied[freechunks]); | |
534 | spin_unlock(&pool->lock); | |
535 | } | |
536 | z3fold_page_unlock(zhdr); | |
2f1e5e4d | 537 | spin_lock(&pool->lock); |
5a27aa82 VW |
538 | if (kref_put(&zhdr->refcount, release_z3fold_page)) |
539 | atomic64_dec(&pool->pages_nr); | |
2f1e5e4d | 540 | spin_unlock(&pool->lock); |
9a001fc1 VW |
541 | } |
542 | ||
9a001fc1 VW |
543 | } |
544 | ||
545 | /** | |
546 | * z3fold_reclaim_page() - evicts allocations from a pool page and frees it | |
547 | * @pool: pool from which a page will attempt to be evicted | |
548 | * @retires: number of pages on the LRU list for which eviction will | |
549 | * be attempted before failing | |
550 | * | |
551 | * z3fold reclaim is different from normal system reclaim in that it is done | |
552 | * from the bottom, up. This is because only the bottom layer, z3fold, has | |
553 | * information on how the allocations are organized within each z3fold page. | |
554 | * This has the potential to create interesting locking situations between | |
555 | * z3fold and the user, however. | |
556 | * | |
557 | * To avoid these, this is how z3fold_reclaim_page() should be called: | |
558 | ||
559 | * The user detects a page should be reclaimed and calls z3fold_reclaim_page(). | |
560 | * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and | |
561 | * call the user-defined eviction handler with the pool and handle as | |
562 | * arguments. | |
563 | * | |
564 | * If the handle can not be evicted, the eviction handler should return | |
565 | * non-zero. z3fold_reclaim_page() will add the z3fold page back to the | |
566 | * appropriate list and try the next z3fold page on the LRU up to | |
567 | * a user defined number of retries. | |
568 | * | |
569 | * If the handle is successfully evicted, the eviction handler should | |
570 | * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free() | |
571 | * contains logic to delay freeing the page if the page is under reclaim, | |
572 | * as indicated by the setting of the PG_reclaim flag on the underlying page. | |
573 | * | |
574 | * If all buddies in the z3fold page are successfully evicted, then the | |
575 | * z3fold page can be freed. | |
576 | * | |
577 | * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are | |
578 | * no pages to evict or an eviction handler is not registered, -EAGAIN if | |
579 | * the retry limit was hit. | |
580 | */ | |
581 | static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries) | |
582 | { | |
583 | int i, ret = 0, freechunks; | |
584 | struct z3fold_header *zhdr; | |
585 | struct page *page; | |
586 | unsigned long first_handle = 0, middle_handle = 0, last_handle = 0; | |
587 | ||
588 | spin_lock(&pool->lock); | |
2f1e5e4d | 589 | if (!pool->ops || !pool->ops->evict || retries == 0) { |
9a001fc1 VW |
590 | spin_unlock(&pool->lock); |
591 | return -EINVAL; | |
592 | } | |
593 | for (i = 0; i < retries; i++) { | |
2f1e5e4d VW |
594 | if (list_empty(&pool->lru)) { |
595 | spin_unlock(&pool->lock); | |
596 | return -EINVAL; | |
597 | } | |
9a001fc1 | 598 | page = list_last_entry(&pool->lru, struct page, lru); |
5a27aa82 | 599 | list_del_init(&page->lru); |
9a001fc1 | 600 | |
9a001fc1 VW |
601 | zhdr = page_address(page); |
602 | if (!test_bit(PAGE_HEADLESS, &page->private)) { | |
5a27aa82 VW |
603 | if (!list_empty(&zhdr->buddy)) |
604 | list_del_init(&zhdr->buddy); | |
605 | kref_get(&zhdr->refcount); | |
2f1e5e4d VW |
606 | spin_unlock(&pool->lock); |
607 | z3fold_page_lock(zhdr); | |
9a001fc1 VW |
608 | /* |
609 | * We need encode the handles before unlocking, since | |
610 | * we can race with free that will set | |
611 | * (first|last)_chunks to 0 | |
612 | */ | |
613 | first_handle = 0; | |
614 | last_handle = 0; | |
615 | middle_handle = 0; | |
616 | if (zhdr->first_chunks) | |
617 | first_handle = encode_handle(zhdr, FIRST); | |
618 | if (zhdr->middle_chunks) | |
619 | middle_handle = encode_handle(zhdr, MIDDLE); | |
620 | if (zhdr->last_chunks) | |
621 | last_handle = encode_handle(zhdr, LAST); | |
2f1e5e4d | 622 | z3fold_page_unlock(zhdr); |
9a001fc1 VW |
623 | } else { |
624 | first_handle = encode_handle(zhdr, HEADLESS); | |
625 | last_handle = middle_handle = 0; | |
2f1e5e4d | 626 | spin_unlock(&pool->lock); |
9a001fc1 VW |
627 | } |
628 | ||
9a001fc1 VW |
629 | /* Issue the eviction callback(s) */ |
630 | if (middle_handle) { | |
631 | ret = pool->ops->evict(pool, middle_handle); | |
632 | if (ret) | |
633 | goto next; | |
634 | } | |
635 | if (first_handle) { | |
636 | ret = pool->ops->evict(pool, first_handle); | |
637 | if (ret) | |
638 | goto next; | |
639 | } | |
640 | if (last_handle) { | |
641 | ret = pool->ops->evict(pool, last_handle); | |
642 | if (ret) | |
643 | goto next; | |
644 | } | |
645 | next: | |
5a27aa82 VW |
646 | if (test_bit(PAGE_HEADLESS, &page->private)) { |
647 | if (ret == 0) { | |
648 | free_z3fold_page(page); | |
649 | return 0; | |
43afc194 | 650 | } else { |
5a27aa82 VW |
651 | spin_lock(&pool->lock); |
652 | } | |
653 | } else { | |
654 | z3fold_page_lock(zhdr); | |
655 | if ((zhdr->first_chunks || zhdr->last_chunks || | |
656 | zhdr->middle_chunks) && | |
657 | !(zhdr->first_chunks && zhdr->last_chunks && | |
658 | zhdr->middle_chunks)) { | |
43afc194 VW |
659 | z3fold_compact_page(zhdr); |
660 | /* add to unbuddied list */ | |
2f1e5e4d | 661 | spin_lock(&pool->lock); |
43afc194 VW |
662 | freechunks = num_free_chunks(zhdr); |
663 | list_add(&zhdr->buddy, | |
664 | &pool->unbuddied[freechunks]); | |
2f1e5e4d | 665 | spin_unlock(&pool->lock); |
43afc194 | 666 | } |
2f1e5e4d | 667 | z3fold_page_unlock(zhdr); |
5a27aa82 VW |
668 | spin_lock(&pool->lock); |
669 | if (kref_put(&zhdr->refcount, release_z3fold_page)) { | |
670 | atomic64_dec(&pool->pages_nr); | |
671 | return 0; | |
672 | } | |
673 | } | |
2f1e5e4d | 674 | |
5a27aa82 VW |
675 | /* |
676 | * Add to the beginning of LRU. | |
677 | * Pool lock has to be kept here to ensure the page has | |
678 | * not already been released | |
679 | */ | |
9a001fc1 VW |
680 | list_add(&page->lru, &pool->lru); |
681 | } | |
682 | spin_unlock(&pool->lock); | |
683 | return -EAGAIN; | |
684 | } | |
685 | ||
686 | /** | |
687 | * z3fold_map() - maps the allocation associated with the given handle | |
688 | * @pool: pool in which the allocation resides | |
689 | * @handle: handle associated with the allocation to be mapped | |
690 | * | |
691 | * Extracts the buddy number from handle and constructs the pointer to the | |
692 | * correct starting chunk within the page. | |
693 | * | |
694 | * Returns: a pointer to the mapped allocation | |
695 | */ | |
696 | static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle) | |
697 | { | |
698 | struct z3fold_header *zhdr; | |
699 | struct page *page; | |
700 | void *addr; | |
701 | enum buddy buddy; | |
702 | ||
9a001fc1 VW |
703 | zhdr = handle_to_z3fold_header(handle); |
704 | addr = zhdr; | |
705 | page = virt_to_page(zhdr); | |
706 | ||
707 | if (test_bit(PAGE_HEADLESS, &page->private)) | |
708 | goto out; | |
709 | ||
2f1e5e4d | 710 | z3fold_page_lock(zhdr); |
9a001fc1 VW |
711 | buddy = handle_to_buddy(handle); |
712 | switch (buddy) { | |
713 | case FIRST: | |
714 | addr += ZHDR_SIZE_ALIGNED; | |
715 | break; | |
716 | case MIDDLE: | |
717 | addr += zhdr->start_middle << CHUNK_SHIFT; | |
718 | set_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
719 | break; | |
720 | case LAST: | |
721 | addr += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT); | |
722 | break; | |
723 | default: | |
724 | pr_err("unknown buddy id %d\n", buddy); | |
725 | WARN_ON(1); | |
726 | addr = NULL; | |
727 | break; | |
728 | } | |
2f1e5e4d VW |
729 | |
730 | z3fold_page_unlock(zhdr); | |
9a001fc1 | 731 | out: |
9a001fc1 VW |
732 | return addr; |
733 | } | |
734 | ||
735 | /** | |
736 | * z3fold_unmap() - unmaps the allocation associated with the given handle | |
737 | * @pool: pool in which the allocation resides | |
738 | * @handle: handle associated with the allocation to be unmapped | |
739 | */ | |
740 | static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle) | |
741 | { | |
742 | struct z3fold_header *zhdr; | |
743 | struct page *page; | |
744 | enum buddy buddy; | |
745 | ||
9a001fc1 VW |
746 | zhdr = handle_to_z3fold_header(handle); |
747 | page = virt_to_page(zhdr); | |
748 | ||
2f1e5e4d | 749 | if (test_bit(PAGE_HEADLESS, &page->private)) |
9a001fc1 | 750 | return; |
9a001fc1 | 751 | |
2f1e5e4d | 752 | z3fold_page_lock(zhdr); |
9a001fc1 VW |
753 | buddy = handle_to_buddy(handle); |
754 | if (buddy == MIDDLE) | |
755 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
2f1e5e4d | 756 | z3fold_page_unlock(zhdr); |
9a001fc1 VW |
757 | } |
758 | ||
759 | /** | |
760 | * z3fold_get_pool_size() - gets the z3fold pool size in pages | |
761 | * @pool: pool whose size is being queried | |
762 | * | |
12d59ae6 | 763 | * Returns: size in pages of the given pool. |
9a001fc1 VW |
764 | */ |
765 | static u64 z3fold_get_pool_size(struct z3fold_pool *pool) | |
766 | { | |
12d59ae6 | 767 | return atomic64_read(&pool->pages_nr); |
9a001fc1 VW |
768 | } |
769 | ||
770 | /***************** | |
771 | * zpool | |
772 | ****************/ | |
773 | ||
774 | static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle) | |
775 | { | |
776 | if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict) | |
777 | return pool->zpool_ops->evict(pool->zpool, handle); | |
778 | else | |
779 | return -ENOENT; | |
780 | } | |
781 | ||
782 | static const struct z3fold_ops z3fold_zpool_ops = { | |
783 | .evict = z3fold_zpool_evict | |
784 | }; | |
785 | ||
786 | static void *z3fold_zpool_create(const char *name, gfp_t gfp, | |
787 | const struct zpool_ops *zpool_ops, | |
788 | struct zpool *zpool) | |
789 | { | |
790 | struct z3fold_pool *pool; | |
791 | ||
792 | pool = z3fold_create_pool(gfp, zpool_ops ? &z3fold_zpool_ops : NULL); | |
793 | if (pool) { | |
794 | pool->zpool = zpool; | |
795 | pool->zpool_ops = zpool_ops; | |
796 | } | |
797 | return pool; | |
798 | } | |
799 | ||
800 | static void z3fold_zpool_destroy(void *pool) | |
801 | { | |
802 | z3fold_destroy_pool(pool); | |
803 | } | |
804 | ||
805 | static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp, | |
806 | unsigned long *handle) | |
807 | { | |
808 | return z3fold_alloc(pool, size, gfp, handle); | |
809 | } | |
810 | static void z3fold_zpool_free(void *pool, unsigned long handle) | |
811 | { | |
812 | z3fold_free(pool, handle); | |
813 | } | |
814 | ||
815 | static int z3fold_zpool_shrink(void *pool, unsigned int pages, | |
816 | unsigned int *reclaimed) | |
817 | { | |
818 | unsigned int total = 0; | |
819 | int ret = -EINVAL; | |
820 | ||
821 | while (total < pages) { | |
822 | ret = z3fold_reclaim_page(pool, 8); | |
823 | if (ret < 0) | |
824 | break; | |
825 | total++; | |
826 | } | |
827 | ||
828 | if (reclaimed) | |
829 | *reclaimed = total; | |
830 | ||
831 | return ret; | |
832 | } | |
833 | ||
834 | static void *z3fold_zpool_map(void *pool, unsigned long handle, | |
835 | enum zpool_mapmode mm) | |
836 | { | |
837 | return z3fold_map(pool, handle); | |
838 | } | |
839 | static void z3fold_zpool_unmap(void *pool, unsigned long handle) | |
840 | { | |
841 | z3fold_unmap(pool, handle); | |
842 | } | |
843 | ||
844 | static u64 z3fold_zpool_total_size(void *pool) | |
845 | { | |
846 | return z3fold_get_pool_size(pool) * PAGE_SIZE; | |
847 | } | |
848 | ||
849 | static struct zpool_driver z3fold_zpool_driver = { | |
850 | .type = "z3fold", | |
851 | .owner = THIS_MODULE, | |
852 | .create = z3fold_zpool_create, | |
853 | .destroy = z3fold_zpool_destroy, | |
854 | .malloc = z3fold_zpool_malloc, | |
855 | .free = z3fold_zpool_free, | |
856 | .shrink = z3fold_zpool_shrink, | |
857 | .map = z3fold_zpool_map, | |
858 | .unmap = z3fold_zpool_unmap, | |
859 | .total_size = z3fold_zpool_total_size, | |
860 | }; | |
861 | ||
862 | MODULE_ALIAS("zpool-z3fold"); | |
863 | ||
864 | static int __init init_z3fold(void) | |
865 | { | |
ede93213 VW |
866 | /* Make sure the z3fold header is not larger than the page size */ |
867 | BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE); | |
9a001fc1 VW |
868 | zpool_register_driver(&z3fold_zpool_driver); |
869 | ||
870 | return 0; | |
871 | } | |
872 | ||
873 | static void __exit exit_z3fold(void) | |
874 | { | |
875 | zpool_unregister_driver(&z3fold_zpool_driver); | |
876 | } | |
877 | ||
878 | module_init(init_z3fold); | |
879 | module_exit(exit_z3fold); | |
880 | ||
881 | MODULE_LICENSE("GPL"); | |
882 | MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>"); | |
883 | MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages"); |