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 | *****************/ | |
37 | /* | |
38 | * NCHUNKS_ORDER determines the internal allocation granularity, effectively | |
39 | * adjusting internal fragmentation. It also determines the number of | |
40 | * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the | |
41 | * allocation granularity will be in chunks of size PAGE_SIZE/64. As one chunk | |
42 | * in allocated page is occupied by z3fold header, NCHUNKS will be calculated | |
43 | * to 63 which shows the max number of free chunks in z3fold page, also there | |
44 | * will be 63 freelists per pool. | |
45 | */ | |
46 | #define NCHUNKS_ORDER 6 | |
47 | ||
48 | #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER) | |
49 | #define CHUNK_SIZE (1 << CHUNK_SHIFT) | |
50 | #define ZHDR_SIZE_ALIGNED CHUNK_SIZE | |
51 | #define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT) | |
52 | ||
f201ebd8 | 53 | #define BUDDY_MASK (0x3) |
9a001fc1 VW |
54 | |
55 | struct z3fold_pool; | |
56 | struct z3fold_ops { | |
57 | int (*evict)(struct z3fold_pool *pool, unsigned long handle); | |
58 | }; | |
59 | ||
60 | /** | |
61 | * struct z3fold_pool - stores metadata for each z3fold pool | |
62 | * @lock: protects all pool fields and first|last_chunk fields of any | |
63 | * z3fold page in the pool | |
64 | * @unbuddied: array of lists tracking z3fold pages that contain 2- buddies; | |
65 | * the lists each z3fold page is added to depends on the size of | |
66 | * its free region. | |
67 | * @buddied: list tracking the z3fold pages that contain 3 buddies; | |
68 | * these z3fold pages are full | |
69 | * @lru: list tracking the z3fold pages in LRU order by most recently | |
70 | * added buddy. | |
71 | * @pages_nr: number of z3fold pages in the pool. | |
72 | * @ops: pointer to a structure of user defined operations specified at | |
73 | * pool creation time. | |
74 | * | |
75 | * This structure is allocated at pool creation time and maintains metadata | |
76 | * pertaining to a particular z3fold pool. | |
77 | */ | |
78 | struct z3fold_pool { | |
79 | spinlock_t lock; | |
80 | struct list_head unbuddied[NCHUNKS]; | |
81 | struct list_head buddied; | |
82 | struct list_head lru; | |
83 | u64 pages_nr; | |
84 | const struct z3fold_ops *ops; | |
85 | struct zpool *zpool; | |
86 | const struct zpool_ops *zpool_ops; | |
87 | }; | |
88 | ||
89 | enum buddy { | |
90 | HEADLESS = 0, | |
91 | FIRST, | |
92 | MIDDLE, | |
93 | LAST, | |
94 | BUDDIES_MAX | |
95 | }; | |
96 | ||
97 | /* | |
98 | * struct z3fold_header - z3fold page metadata occupying the first chunk of each | |
99 | * z3fold page, except for HEADLESS pages | |
100 | * @buddy: links the z3fold page into the relevant list in the pool | |
101 | * @first_chunks: the size of the first buddy in chunks, 0 if free | |
102 | * @middle_chunks: the size of the middle buddy in chunks, 0 if free | |
103 | * @last_chunks: the size of the last buddy in chunks, 0 if free | |
104 | * @first_num: the starting number (for the first handle) | |
105 | */ | |
106 | struct z3fold_header { | |
107 | struct list_head buddy; | |
108 | unsigned short first_chunks; | |
109 | unsigned short middle_chunks; | |
110 | unsigned short last_chunks; | |
111 | unsigned short start_middle; | |
f201ebd8 | 112 | unsigned short first_num:2; |
9a001fc1 VW |
113 | }; |
114 | ||
115 | /* | |
116 | * Internal z3fold page flags | |
117 | */ | |
118 | enum z3fold_page_flags { | |
119 | UNDER_RECLAIM = 0, | |
120 | PAGE_HEADLESS, | |
121 | MIDDLE_CHUNK_MAPPED, | |
122 | }; | |
123 | ||
124 | /***************** | |
125 | * Helpers | |
126 | *****************/ | |
127 | ||
128 | /* Converts an allocation size in bytes to size in z3fold chunks */ | |
129 | static int size_to_chunks(size_t size) | |
130 | { | |
131 | return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT; | |
132 | } | |
133 | ||
134 | #define for_each_unbuddied_list(_iter, _begin) \ | |
135 | for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++) | |
136 | ||
137 | /* Initializes the z3fold header of a newly allocated z3fold page */ | |
138 | static struct z3fold_header *init_z3fold_page(struct page *page) | |
139 | { | |
140 | struct z3fold_header *zhdr = page_address(page); | |
141 | ||
142 | INIT_LIST_HEAD(&page->lru); | |
143 | clear_bit(UNDER_RECLAIM, &page->private); | |
144 | clear_bit(PAGE_HEADLESS, &page->private); | |
145 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
146 | ||
147 | zhdr->first_chunks = 0; | |
148 | zhdr->middle_chunks = 0; | |
149 | zhdr->last_chunks = 0; | |
150 | zhdr->first_num = 0; | |
151 | zhdr->start_middle = 0; | |
152 | INIT_LIST_HEAD(&zhdr->buddy); | |
153 | return zhdr; | |
154 | } | |
155 | ||
156 | /* Resets the struct page fields and frees the page */ | |
157 | static void free_z3fold_page(struct z3fold_header *zhdr) | |
158 | { | |
159 | __free_page(virt_to_page(zhdr)); | |
160 | } | |
161 | ||
162 | /* | |
163 | * Encodes the handle of a particular buddy within a z3fold page | |
164 | * Pool lock should be held as this function accesses first_num | |
165 | */ | |
166 | static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud) | |
167 | { | |
168 | unsigned long handle; | |
169 | ||
170 | handle = (unsigned long)zhdr; | |
171 | if (bud != HEADLESS) | |
172 | handle += (bud + zhdr->first_num) & BUDDY_MASK; | |
173 | return handle; | |
174 | } | |
175 | ||
176 | /* Returns the z3fold page where a given handle is stored */ | |
177 | static struct z3fold_header *handle_to_z3fold_header(unsigned long handle) | |
178 | { | |
179 | return (struct z3fold_header *)(handle & PAGE_MASK); | |
180 | } | |
181 | ||
f201ebd8 | 182 | /* |
183 | * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle | |
184 | * but that doesn't matter. because the masking will result in the | |
185 | * correct buddy number. | |
186 | */ | |
9a001fc1 VW |
187 | static enum buddy handle_to_buddy(unsigned long handle) |
188 | { | |
189 | struct z3fold_header *zhdr = handle_to_z3fold_header(handle); | |
190 | return (handle - zhdr->first_num) & BUDDY_MASK; | |
191 | } | |
192 | ||
193 | /* | |
194 | * Returns the number of free chunks in a z3fold page. | |
195 | * NB: can't be used with HEADLESS pages. | |
196 | */ | |
197 | static int num_free_chunks(struct z3fold_header *zhdr) | |
198 | { | |
199 | int nfree; | |
200 | /* | |
201 | * If there is a middle object, pick up the bigger free space | |
202 | * either before or after it. Otherwise just subtract the number | |
203 | * of chunks occupied by the first and the last objects. | |
204 | */ | |
205 | if (zhdr->middle_chunks != 0) { | |
206 | int nfree_before = zhdr->first_chunks ? | |
207 | 0 : zhdr->start_middle - 1; | |
208 | int nfree_after = zhdr->last_chunks ? | |
209 | 0 : NCHUNKS - zhdr->start_middle - zhdr->middle_chunks; | |
210 | nfree = max(nfree_before, nfree_after); | |
211 | } else | |
212 | nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks; | |
213 | return nfree; | |
214 | } | |
215 | ||
216 | /***************** | |
217 | * API Functions | |
218 | *****************/ | |
219 | /** | |
220 | * z3fold_create_pool() - create a new z3fold pool | |
221 | * @gfp: gfp flags when allocating the z3fold pool structure | |
222 | * @ops: user-defined operations for the z3fold pool | |
223 | * | |
224 | * Return: pointer to the new z3fold pool or NULL if the metadata allocation | |
225 | * failed. | |
226 | */ | |
227 | static struct z3fold_pool *z3fold_create_pool(gfp_t gfp, | |
228 | const struct z3fold_ops *ops) | |
229 | { | |
230 | struct z3fold_pool *pool; | |
231 | int i; | |
232 | ||
233 | pool = kzalloc(sizeof(struct z3fold_pool), gfp); | |
234 | if (!pool) | |
235 | return NULL; | |
236 | spin_lock_init(&pool->lock); | |
237 | for_each_unbuddied_list(i, 0) | |
238 | INIT_LIST_HEAD(&pool->unbuddied[i]); | |
239 | INIT_LIST_HEAD(&pool->buddied); | |
240 | INIT_LIST_HEAD(&pool->lru); | |
241 | pool->pages_nr = 0; | |
242 | pool->ops = ops; | |
243 | return pool; | |
244 | } | |
245 | ||
246 | /** | |
247 | * z3fold_destroy_pool() - destroys an existing z3fold pool | |
248 | * @pool: the z3fold pool to be destroyed | |
249 | * | |
250 | * The pool should be emptied before this function is called. | |
251 | */ | |
252 | static void z3fold_destroy_pool(struct z3fold_pool *pool) | |
253 | { | |
254 | kfree(pool); | |
255 | } | |
256 | ||
257 | /* Has to be called with lock held */ | |
258 | static int z3fold_compact_page(struct z3fold_header *zhdr) | |
259 | { | |
260 | struct page *page = virt_to_page(zhdr); | |
261 | void *beg = zhdr; | |
262 | ||
263 | ||
264 | if (!test_bit(MIDDLE_CHUNK_MAPPED, &page->private) && | |
265 | zhdr->middle_chunks != 0 && | |
266 | zhdr->first_chunks == 0 && zhdr->last_chunks == 0) { | |
267 | memmove(beg + ZHDR_SIZE_ALIGNED, | |
268 | beg + (zhdr->start_middle << CHUNK_SHIFT), | |
269 | zhdr->middle_chunks << CHUNK_SHIFT); | |
270 | zhdr->first_chunks = zhdr->middle_chunks; | |
271 | zhdr->middle_chunks = 0; | |
272 | zhdr->start_middle = 0; | |
273 | zhdr->first_num++; | |
274 | return 1; | |
275 | } | |
276 | return 0; | |
277 | } | |
278 | ||
279 | /** | |
280 | * z3fold_alloc() - allocates a region of a given size | |
281 | * @pool: z3fold pool from which to allocate | |
282 | * @size: size in bytes of the desired allocation | |
283 | * @gfp: gfp flags used if the pool needs to grow | |
284 | * @handle: handle of the new allocation | |
285 | * | |
286 | * This function will attempt to find a free region in the pool large enough to | |
287 | * satisfy the allocation request. A search of the unbuddied lists is | |
288 | * performed first. If no suitable free region is found, then a new page is | |
289 | * allocated and added to the pool to satisfy the request. | |
290 | * | |
291 | * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used | |
292 | * as z3fold pool pages. | |
293 | * | |
294 | * Return: 0 if success and handle is set, otherwise -EINVAL if the size or | |
295 | * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate | |
296 | * a new page. | |
297 | */ | |
298 | static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp, | |
299 | unsigned long *handle) | |
300 | { | |
301 | int chunks = 0, i, freechunks; | |
302 | struct z3fold_header *zhdr = NULL; | |
303 | enum buddy bud; | |
304 | struct page *page; | |
305 | ||
306 | if (!size || (gfp & __GFP_HIGHMEM)) | |
307 | return -EINVAL; | |
308 | ||
309 | if (size > PAGE_SIZE) | |
310 | return -ENOSPC; | |
311 | ||
312 | if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE) | |
313 | bud = HEADLESS; | |
314 | else { | |
315 | chunks = size_to_chunks(size); | |
316 | spin_lock(&pool->lock); | |
317 | ||
318 | /* First, try to find an unbuddied z3fold page. */ | |
319 | zhdr = NULL; | |
320 | for_each_unbuddied_list(i, chunks) { | |
321 | if (!list_empty(&pool->unbuddied[i])) { | |
322 | zhdr = list_first_entry(&pool->unbuddied[i], | |
323 | struct z3fold_header, buddy); | |
324 | page = virt_to_page(zhdr); | |
325 | if (zhdr->first_chunks == 0) { | |
326 | if (zhdr->middle_chunks != 0 && | |
327 | chunks >= zhdr->start_middle) | |
328 | bud = LAST; | |
329 | else | |
330 | bud = FIRST; | |
331 | } else if (zhdr->last_chunks == 0) | |
332 | bud = LAST; | |
333 | else if (zhdr->middle_chunks == 0) | |
334 | bud = MIDDLE; | |
335 | else { | |
336 | pr_err("No free chunks in unbuddied\n"); | |
337 | WARN_ON(1); | |
338 | continue; | |
339 | } | |
340 | list_del(&zhdr->buddy); | |
341 | goto found; | |
342 | } | |
343 | } | |
344 | bud = FIRST; | |
345 | spin_unlock(&pool->lock); | |
346 | } | |
347 | ||
348 | /* Couldn't find unbuddied z3fold page, create new one */ | |
349 | page = alloc_page(gfp); | |
350 | if (!page) | |
351 | return -ENOMEM; | |
352 | spin_lock(&pool->lock); | |
353 | pool->pages_nr++; | |
354 | zhdr = init_z3fold_page(page); | |
355 | ||
356 | if (bud == HEADLESS) { | |
357 | set_bit(PAGE_HEADLESS, &page->private); | |
358 | goto headless; | |
359 | } | |
360 | ||
361 | found: | |
362 | if (bud == FIRST) | |
363 | zhdr->first_chunks = chunks; | |
364 | else if (bud == LAST) | |
365 | zhdr->last_chunks = chunks; | |
366 | else { | |
367 | zhdr->middle_chunks = chunks; | |
368 | zhdr->start_middle = zhdr->first_chunks + 1; | |
369 | } | |
370 | ||
371 | if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 || | |
372 | zhdr->middle_chunks == 0) { | |
373 | /* Add to unbuddied list */ | |
374 | freechunks = num_free_chunks(zhdr); | |
375 | list_add(&zhdr->buddy, &pool->unbuddied[freechunks]); | |
376 | } else { | |
377 | /* Add to buddied list */ | |
378 | list_add(&zhdr->buddy, &pool->buddied); | |
379 | } | |
380 | ||
381 | headless: | |
382 | /* Add/move z3fold page to beginning of LRU */ | |
383 | if (!list_empty(&page->lru)) | |
384 | list_del(&page->lru); | |
385 | ||
386 | list_add(&page->lru, &pool->lru); | |
387 | ||
388 | *handle = encode_handle(zhdr, bud); | |
389 | spin_unlock(&pool->lock); | |
390 | ||
391 | return 0; | |
392 | } | |
393 | ||
394 | /** | |
395 | * z3fold_free() - frees the allocation associated with the given handle | |
396 | * @pool: pool in which the allocation resided | |
397 | * @handle: handle associated with the allocation returned by z3fold_alloc() | |
398 | * | |
399 | * In the case that the z3fold page in which the allocation resides is under | |
400 | * reclaim, as indicated by the PG_reclaim flag being set, this function | |
401 | * only sets the first|last_chunks to 0. The page is actually freed | |
402 | * once both buddies are evicted (see z3fold_reclaim_page() below). | |
403 | */ | |
404 | static void z3fold_free(struct z3fold_pool *pool, unsigned long handle) | |
405 | { | |
406 | struct z3fold_header *zhdr; | |
407 | int freechunks; | |
408 | struct page *page; | |
409 | enum buddy bud; | |
410 | ||
411 | spin_lock(&pool->lock); | |
412 | zhdr = handle_to_z3fold_header(handle); | |
413 | page = virt_to_page(zhdr); | |
414 | ||
415 | if (test_bit(PAGE_HEADLESS, &page->private)) { | |
416 | /* HEADLESS page stored */ | |
417 | bud = HEADLESS; | |
418 | } else { | |
43afc194 | 419 | bud = handle_to_buddy(handle); |
9a001fc1 VW |
420 | |
421 | switch (bud) { | |
422 | case FIRST: | |
423 | zhdr->first_chunks = 0; | |
424 | break; | |
425 | case MIDDLE: | |
426 | zhdr->middle_chunks = 0; | |
427 | zhdr->start_middle = 0; | |
428 | break; | |
429 | case LAST: | |
430 | zhdr->last_chunks = 0; | |
431 | break; | |
432 | default: | |
433 | pr_err("%s: unknown bud %d\n", __func__, bud); | |
434 | WARN_ON(1); | |
435 | spin_unlock(&pool->lock); | |
436 | return; | |
437 | } | |
438 | } | |
439 | ||
440 | if (test_bit(UNDER_RECLAIM, &page->private)) { | |
441 | /* z3fold page is under reclaim, reclaim will free */ | |
442 | spin_unlock(&pool->lock); | |
443 | return; | |
444 | } | |
445 | ||
446 | if (bud != HEADLESS) { | |
447 | /* Remove from existing buddy list */ | |
448 | list_del(&zhdr->buddy); | |
449 | } | |
450 | ||
451 | if (bud == HEADLESS || | |
452 | (zhdr->first_chunks == 0 && zhdr->middle_chunks == 0 && | |
453 | zhdr->last_chunks == 0)) { | |
454 | /* z3fold page is empty, free */ | |
455 | list_del(&page->lru); | |
456 | clear_bit(PAGE_HEADLESS, &page->private); | |
457 | free_z3fold_page(zhdr); | |
458 | pool->pages_nr--; | |
459 | } else { | |
460 | z3fold_compact_page(zhdr); | |
461 | /* Add to the unbuddied list */ | |
462 | freechunks = num_free_chunks(zhdr); | |
463 | list_add(&zhdr->buddy, &pool->unbuddied[freechunks]); | |
464 | } | |
465 | ||
466 | spin_unlock(&pool->lock); | |
467 | } | |
468 | ||
469 | /** | |
470 | * z3fold_reclaim_page() - evicts allocations from a pool page and frees it | |
471 | * @pool: pool from which a page will attempt to be evicted | |
472 | * @retires: number of pages on the LRU list for which eviction will | |
473 | * be attempted before failing | |
474 | * | |
475 | * z3fold reclaim is different from normal system reclaim in that it is done | |
476 | * from the bottom, up. This is because only the bottom layer, z3fold, has | |
477 | * information on how the allocations are organized within each z3fold page. | |
478 | * This has the potential to create interesting locking situations between | |
479 | * z3fold and the user, however. | |
480 | * | |
481 | * To avoid these, this is how z3fold_reclaim_page() should be called: | |
482 | ||
483 | * The user detects a page should be reclaimed and calls z3fold_reclaim_page(). | |
484 | * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and | |
485 | * call the user-defined eviction handler with the pool and handle as | |
486 | * arguments. | |
487 | * | |
488 | * If the handle can not be evicted, the eviction handler should return | |
489 | * non-zero. z3fold_reclaim_page() will add the z3fold page back to the | |
490 | * appropriate list and try the next z3fold page on the LRU up to | |
491 | * a user defined number of retries. | |
492 | * | |
493 | * If the handle is successfully evicted, the eviction handler should | |
494 | * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free() | |
495 | * contains logic to delay freeing the page if the page is under reclaim, | |
496 | * as indicated by the setting of the PG_reclaim flag on the underlying page. | |
497 | * | |
498 | * If all buddies in the z3fold page are successfully evicted, then the | |
499 | * z3fold page can be freed. | |
500 | * | |
501 | * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are | |
502 | * no pages to evict or an eviction handler is not registered, -EAGAIN if | |
503 | * the retry limit was hit. | |
504 | */ | |
505 | static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries) | |
506 | { | |
507 | int i, ret = 0, freechunks; | |
508 | struct z3fold_header *zhdr; | |
509 | struct page *page; | |
510 | unsigned long first_handle = 0, middle_handle = 0, last_handle = 0; | |
511 | ||
512 | spin_lock(&pool->lock); | |
513 | if (!pool->ops || !pool->ops->evict || list_empty(&pool->lru) || | |
514 | retries == 0) { | |
515 | spin_unlock(&pool->lock); | |
516 | return -EINVAL; | |
517 | } | |
518 | for (i = 0; i < retries; i++) { | |
519 | page = list_last_entry(&pool->lru, struct page, lru); | |
520 | list_del(&page->lru); | |
521 | ||
522 | /* Protect z3fold page against free */ | |
523 | set_bit(UNDER_RECLAIM, &page->private); | |
524 | zhdr = page_address(page); | |
525 | if (!test_bit(PAGE_HEADLESS, &page->private)) { | |
526 | list_del(&zhdr->buddy); | |
527 | /* | |
528 | * We need encode the handles before unlocking, since | |
529 | * we can race with free that will set | |
530 | * (first|last)_chunks to 0 | |
531 | */ | |
532 | first_handle = 0; | |
533 | last_handle = 0; | |
534 | middle_handle = 0; | |
535 | if (zhdr->first_chunks) | |
536 | first_handle = encode_handle(zhdr, FIRST); | |
537 | if (zhdr->middle_chunks) | |
538 | middle_handle = encode_handle(zhdr, MIDDLE); | |
539 | if (zhdr->last_chunks) | |
540 | last_handle = encode_handle(zhdr, LAST); | |
541 | } else { | |
542 | first_handle = encode_handle(zhdr, HEADLESS); | |
543 | last_handle = middle_handle = 0; | |
544 | } | |
545 | ||
546 | spin_unlock(&pool->lock); | |
547 | ||
548 | /* Issue the eviction callback(s) */ | |
549 | if (middle_handle) { | |
550 | ret = pool->ops->evict(pool, middle_handle); | |
551 | if (ret) | |
552 | goto next; | |
553 | } | |
554 | if (first_handle) { | |
555 | ret = pool->ops->evict(pool, first_handle); | |
556 | if (ret) | |
557 | goto next; | |
558 | } | |
559 | if (last_handle) { | |
560 | ret = pool->ops->evict(pool, last_handle); | |
561 | if (ret) | |
562 | goto next; | |
563 | } | |
564 | next: | |
565 | spin_lock(&pool->lock); | |
566 | clear_bit(UNDER_RECLAIM, &page->private); | |
567 | if ((test_bit(PAGE_HEADLESS, &page->private) && ret == 0) || | |
568 | (zhdr->first_chunks == 0 && zhdr->last_chunks == 0 && | |
569 | zhdr->middle_chunks == 0)) { | |
570 | /* | |
571 | * All buddies are now free, free the z3fold page and | |
572 | * return success. | |
573 | */ | |
574 | clear_bit(PAGE_HEADLESS, &page->private); | |
575 | free_z3fold_page(zhdr); | |
576 | pool->pages_nr--; | |
577 | spin_unlock(&pool->lock); | |
578 | return 0; | |
43afc194 VW |
579 | } else if (!test_bit(PAGE_HEADLESS, &page->private)) { |
580 | if (zhdr->first_chunks != 0 && | |
581 | zhdr->last_chunks != 0 && | |
582 | zhdr->middle_chunks != 0) { | |
583 | /* Full, add to buddied list */ | |
584 | list_add(&zhdr->buddy, &pool->buddied); | |
585 | } else { | |
586 | z3fold_compact_page(zhdr); | |
587 | /* add to unbuddied list */ | |
588 | freechunks = num_free_chunks(zhdr); | |
589 | list_add(&zhdr->buddy, | |
590 | &pool->unbuddied[freechunks]); | |
591 | } | |
9a001fc1 VW |
592 | } |
593 | ||
594 | /* add to beginning of LRU */ | |
595 | list_add(&page->lru, &pool->lru); | |
596 | } | |
597 | spin_unlock(&pool->lock); | |
598 | return -EAGAIN; | |
599 | } | |
600 | ||
601 | /** | |
602 | * z3fold_map() - maps the allocation associated with the given handle | |
603 | * @pool: pool in which the allocation resides | |
604 | * @handle: handle associated with the allocation to be mapped | |
605 | * | |
606 | * Extracts the buddy number from handle and constructs the pointer to the | |
607 | * correct starting chunk within the page. | |
608 | * | |
609 | * Returns: a pointer to the mapped allocation | |
610 | */ | |
611 | static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle) | |
612 | { | |
613 | struct z3fold_header *zhdr; | |
614 | struct page *page; | |
615 | void *addr; | |
616 | enum buddy buddy; | |
617 | ||
618 | spin_lock(&pool->lock); | |
619 | zhdr = handle_to_z3fold_header(handle); | |
620 | addr = zhdr; | |
621 | page = virt_to_page(zhdr); | |
622 | ||
623 | if (test_bit(PAGE_HEADLESS, &page->private)) | |
624 | goto out; | |
625 | ||
626 | buddy = handle_to_buddy(handle); | |
627 | switch (buddy) { | |
628 | case FIRST: | |
629 | addr += ZHDR_SIZE_ALIGNED; | |
630 | break; | |
631 | case MIDDLE: | |
632 | addr += zhdr->start_middle << CHUNK_SHIFT; | |
633 | set_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
634 | break; | |
635 | case LAST: | |
636 | addr += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT); | |
637 | break; | |
638 | default: | |
639 | pr_err("unknown buddy id %d\n", buddy); | |
640 | WARN_ON(1); | |
641 | addr = NULL; | |
642 | break; | |
643 | } | |
644 | out: | |
645 | spin_unlock(&pool->lock); | |
646 | return addr; | |
647 | } | |
648 | ||
649 | /** | |
650 | * z3fold_unmap() - unmaps the allocation associated with the given handle | |
651 | * @pool: pool in which the allocation resides | |
652 | * @handle: handle associated with the allocation to be unmapped | |
653 | */ | |
654 | static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle) | |
655 | { | |
656 | struct z3fold_header *zhdr; | |
657 | struct page *page; | |
658 | enum buddy buddy; | |
659 | ||
660 | spin_lock(&pool->lock); | |
661 | zhdr = handle_to_z3fold_header(handle); | |
662 | page = virt_to_page(zhdr); | |
663 | ||
664 | if (test_bit(PAGE_HEADLESS, &page->private)) { | |
665 | spin_unlock(&pool->lock); | |
666 | return; | |
667 | } | |
668 | ||
669 | buddy = handle_to_buddy(handle); | |
670 | if (buddy == MIDDLE) | |
671 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
672 | spin_unlock(&pool->lock); | |
673 | } | |
674 | ||
675 | /** | |
676 | * z3fold_get_pool_size() - gets the z3fold pool size in pages | |
677 | * @pool: pool whose size is being queried | |
678 | * | |
679 | * Returns: size in pages of the given pool. The pool lock need not be | |
680 | * taken to access pages_nr. | |
681 | */ | |
682 | static u64 z3fold_get_pool_size(struct z3fold_pool *pool) | |
683 | { | |
684 | return pool->pages_nr; | |
685 | } | |
686 | ||
687 | /***************** | |
688 | * zpool | |
689 | ****************/ | |
690 | ||
691 | static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle) | |
692 | { | |
693 | if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict) | |
694 | return pool->zpool_ops->evict(pool->zpool, handle); | |
695 | else | |
696 | return -ENOENT; | |
697 | } | |
698 | ||
699 | static const struct z3fold_ops z3fold_zpool_ops = { | |
700 | .evict = z3fold_zpool_evict | |
701 | }; | |
702 | ||
703 | static void *z3fold_zpool_create(const char *name, gfp_t gfp, | |
704 | const struct zpool_ops *zpool_ops, | |
705 | struct zpool *zpool) | |
706 | { | |
707 | struct z3fold_pool *pool; | |
708 | ||
709 | pool = z3fold_create_pool(gfp, zpool_ops ? &z3fold_zpool_ops : NULL); | |
710 | if (pool) { | |
711 | pool->zpool = zpool; | |
712 | pool->zpool_ops = zpool_ops; | |
713 | } | |
714 | return pool; | |
715 | } | |
716 | ||
717 | static void z3fold_zpool_destroy(void *pool) | |
718 | { | |
719 | z3fold_destroy_pool(pool); | |
720 | } | |
721 | ||
722 | static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp, | |
723 | unsigned long *handle) | |
724 | { | |
725 | return z3fold_alloc(pool, size, gfp, handle); | |
726 | } | |
727 | static void z3fold_zpool_free(void *pool, unsigned long handle) | |
728 | { | |
729 | z3fold_free(pool, handle); | |
730 | } | |
731 | ||
732 | static int z3fold_zpool_shrink(void *pool, unsigned int pages, | |
733 | unsigned int *reclaimed) | |
734 | { | |
735 | unsigned int total = 0; | |
736 | int ret = -EINVAL; | |
737 | ||
738 | while (total < pages) { | |
739 | ret = z3fold_reclaim_page(pool, 8); | |
740 | if (ret < 0) | |
741 | break; | |
742 | total++; | |
743 | } | |
744 | ||
745 | if (reclaimed) | |
746 | *reclaimed = total; | |
747 | ||
748 | return ret; | |
749 | } | |
750 | ||
751 | static void *z3fold_zpool_map(void *pool, unsigned long handle, | |
752 | enum zpool_mapmode mm) | |
753 | { | |
754 | return z3fold_map(pool, handle); | |
755 | } | |
756 | static void z3fold_zpool_unmap(void *pool, unsigned long handle) | |
757 | { | |
758 | z3fold_unmap(pool, handle); | |
759 | } | |
760 | ||
761 | static u64 z3fold_zpool_total_size(void *pool) | |
762 | { | |
763 | return z3fold_get_pool_size(pool) * PAGE_SIZE; | |
764 | } | |
765 | ||
766 | static struct zpool_driver z3fold_zpool_driver = { | |
767 | .type = "z3fold", | |
768 | .owner = THIS_MODULE, | |
769 | .create = z3fold_zpool_create, | |
770 | .destroy = z3fold_zpool_destroy, | |
771 | .malloc = z3fold_zpool_malloc, | |
772 | .free = z3fold_zpool_free, | |
773 | .shrink = z3fold_zpool_shrink, | |
774 | .map = z3fold_zpool_map, | |
775 | .unmap = z3fold_zpool_unmap, | |
776 | .total_size = z3fold_zpool_total_size, | |
777 | }; | |
778 | ||
779 | MODULE_ALIAS("zpool-z3fold"); | |
780 | ||
781 | static int __init init_z3fold(void) | |
782 | { | |
783 | /* Make sure the z3fold header will fit in one chunk */ | |
784 | BUILD_BUG_ON(sizeof(struct z3fold_header) > ZHDR_SIZE_ALIGNED); | |
785 | zpool_register_driver(&z3fold_zpool_driver); | |
786 | ||
787 | return 0; | |
788 | } | |
789 | ||
790 | static void __exit exit_z3fold(void) | |
791 | { | |
792 | zpool_unregister_driver(&z3fold_zpool_driver); | |
793 | } | |
794 | ||
795 | module_init(init_z3fold); | |
796 | module_exit(exit_z3fold); | |
797 | ||
798 | MODULE_LICENSE("GPL"); | |
799 | MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>"); | |
800 | MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages"); |