]>
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> | |
d30561c5 | 26 | #include <linux/sched.h> |
9a001fc1 VW |
27 | #include <linux/list.h> |
28 | #include <linux/mm.h> | |
29 | #include <linux/module.h> | |
d30561c5 | 30 | #include <linux/percpu.h> |
9a001fc1 | 31 | #include <linux/preempt.h> |
d30561c5 | 32 | #include <linux/workqueue.h> |
9a001fc1 VW |
33 | #include <linux/slab.h> |
34 | #include <linux/spinlock.h> | |
35 | #include <linux/zpool.h> | |
36 | ||
37 | /***************** | |
38 | * Structures | |
39 | *****************/ | |
ede93213 VW |
40 | struct z3fold_pool; |
41 | struct z3fold_ops { | |
42 | int (*evict)(struct z3fold_pool *pool, unsigned long handle); | |
43 | }; | |
44 | ||
45 | enum buddy { | |
46 | HEADLESS = 0, | |
47 | FIRST, | |
48 | MIDDLE, | |
49 | LAST, | |
50 | BUDDIES_MAX | |
51 | }; | |
52 | ||
53 | /* | |
d30561c5 | 54 | * struct z3fold_header - z3fold page metadata occupying first chunks of each |
ede93213 | 55 | * z3fold page, except for HEADLESS pages |
d30561c5 VW |
56 | * @buddy: links the z3fold page into the relevant list in the |
57 | * pool | |
2f1e5e4d | 58 | * @page_lock: per-page lock |
d30561c5 VW |
59 | * @refcount: reference count for the z3fold page |
60 | * @work: work_struct for page layout optimization | |
61 | * @pool: pointer to the pool which this page belongs to | |
62 | * @cpu: CPU which this page "belongs" to | |
ede93213 VW |
63 | * @first_chunks: the size of the first buddy in chunks, 0 if free |
64 | * @middle_chunks: the size of the middle buddy in chunks, 0 if free | |
65 | * @last_chunks: the size of the last buddy in chunks, 0 if free | |
66 | * @first_num: the starting number (for the first handle) | |
67 | */ | |
68 | struct z3fold_header { | |
69 | struct list_head buddy; | |
2f1e5e4d | 70 | spinlock_t page_lock; |
5a27aa82 | 71 | struct kref refcount; |
d30561c5 VW |
72 | struct work_struct work; |
73 | struct z3fold_pool *pool; | |
74 | short cpu; | |
ede93213 VW |
75 | unsigned short first_chunks; |
76 | unsigned short middle_chunks; | |
77 | unsigned short last_chunks; | |
78 | unsigned short start_middle; | |
79 | unsigned short first_num:2; | |
80 | }; | |
81 | ||
9a001fc1 VW |
82 | /* |
83 | * NCHUNKS_ORDER determines the internal allocation granularity, effectively | |
84 | * adjusting internal fragmentation. It also determines the number of | |
85 | * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the | |
ede93213 VW |
86 | * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks |
87 | * in the beginning of an allocated page are occupied by z3fold header, so | |
88 | * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y), | |
89 | * which shows the max number of free chunks in z3fold page, also there will | |
90 | * be 63, or 62, respectively, freelists per pool. | |
9a001fc1 VW |
91 | */ |
92 | #define NCHUNKS_ORDER 6 | |
93 | ||
94 | #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER) | |
95 | #define CHUNK_SIZE (1 << CHUNK_SHIFT) | |
ede93213 VW |
96 | #define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE) |
97 | #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT) | |
98 | #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SHIFT) | |
9a001fc1 VW |
99 | #define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT) |
100 | ||
f201ebd8 | 101 | #define BUDDY_MASK (0x3) |
9a001fc1 | 102 | |
9a001fc1 VW |
103 | /** |
104 | * struct z3fold_pool - stores metadata for each z3fold pool | |
d30561c5 VW |
105 | * @name: pool name |
106 | * @lock: protects pool unbuddied/lru lists | |
107 | * @stale_lock: protects pool stale page list | |
108 | * @unbuddied: per-cpu array of lists tracking z3fold pages that contain 2- | |
109 | * buddies; the list each z3fold page is added to depends on | |
110 | * the size of its free region. | |
9a001fc1 VW |
111 | * @lru: list tracking the z3fold pages in LRU order by most recently |
112 | * added buddy. | |
d30561c5 | 113 | * @stale: list of pages marked for freeing |
9a001fc1 VW |
114 | * @pages_nr: number of z3fold pages in the pool. |
115 | * @ops: pointer to a structure of user defined operations specified at | |
116 | * pool creation time. | |
d30561c5 VW |
117 | * @compact_wq: workqueue for page layout background optimization |
118 | * @release_wq: workqueue for safe page release | |
119 | * @work: work_struct for safe page release | |
9a001fc1 VW |
120 | * |
121 | * This structure is allocated at pool creation time and maintains metadata | |
122 | * pertaining to a particular z3fold pool. | |
123 | */ | |
124 | struct z3fold_pool { | |
d30561c5 | 125 | const char *name; |
9a001fc1 | 126 | spinlock_t lock; |
d30561c5 VW |
127 | spinlock_t stale_lock; |
128 | struct list_head *unbuddied; | |
9a001fc1 | 129 | struct list_head lru; |
d30561c5 | 130 | struct list_head stale; |
12d59ae6 | 131 | atomic64_t pages_nr; |
9a001fc1 VW |
132 | const struct z3fold_ops *ops; |
133 | struct zpool *zpool; | |
134 | const struct zpool_ops *zpool_ops; | |
d30561c5 VW |
135 | struct workqueue_struct *compact_wq; |
136 | struct workqueue_struct *release_wq; | |
137 | struct work_struct work; | |
9a001fc1 VW |
138 | }; |
139 | ||
9a001fc1 VW |
140 | /* |
141 | * Internal z3fold page flags | |
142 | */ | |
143 | enum z3fold_page_flags { | |
5a27aa82 | 144 | PAGE_HEADLESS = 0, |
9a001fc1 | 145 | MIDDLE_CHUNK_MAPPED, |
d30561c5 | 146 | NEEDS_COMPACTING, |
9fc5adb0 VW |
147 | PAGE_STALE, |
148 | UNDER_RECLAIM | |
9a001fc1 VW |
149 | }; |
150 | ||
151 | /***************** | |
152 | * Helpers | |
153 | *****************/ | |
154 | ||
155 | /* Converts an allocation size in bytes to size in z3fold chunks */ | |
156 | static int size_to_chunks(size_t size) | |
157 | { | |
158 | return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT; | |
159 | } | |
160 | ||
161 | #define for_each_unbuddied_list(_iter, _begin) \ | |
162 | for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++) | |
163 | ||
d30561c5 VW |
164 | static void compact_page_work(struct work_struct *w); |
165 | ||
9a001fc1 | 166 | /* Initializes the z3fold header of a newly allocated z3fold page */ |
d30561c5 VW |
167 | static struct z3fold_header *init_z3fold_page(struct page *page, |
168 | struct z3fold_pool *pool) | |
9a001fc1 VW |
169 | { |
170 | struct z3fold_header *zhdr = page_address(page); | |
171 | ||
172 | INIT_LIST_HEAD(&page->lru); | |
9a001fc1 VW |
173 | clear_bit(PAGE_HEADLESS, &page->private); |
174 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
d30561c5 VW |
175 | clear_bit(NEEDS_COMPACTING, &page->private); |
176 | clear_bit(PAGE_STALE, &page->private); | |
9fc5adb0 | 177 | clear_bit(UNDER_RECLAIM, &page->private); |
9a001fc1 | 178 | |
2f1e5e4d | 179 | spin_lock_init(&zhdr->page_lock); |
5a27aa82 | 180 | kref_init(&zhdr->refcount); |
9a001fc1 VW |
181 | zhdr->first_chunks = 0; |
182 | zhdr->middle_chunks = 0; | |
183 | zhdr->last_chunks = 0; | |
184 | zhdr->first_num = 0; | |
185 | zhdr->start_middle = 0; | |
d30561c5 VW |
186 | zhdr->cpu = -1; |
187 | zhdr->pool = pool; | |
9a001fc1 | 188 | INIT_LIST_HEAD(&zhdr->buddy); |
d30561c5 | 189 | INIT_WORK(&zhdr->work, compact_page_work); |
9a001fc1 VW |
190 | return zhdr; |
191 | } | |
192 | ||
193 | /* Resets the struct page fields and frees the page */ | |
5a27aa82 | 194 | static void free_z3fold_page(struct page *page) |
9a001fc1 | 195 | { |
5a27aa82 VW |
196 | __free_page(page); |
197 | } | |
198 | ||
2f1e5e4d VW |
199 | /* Lock a z3fold page */ |
200 | static inline void z3fold_page_lock(struct z3fold_header *zhdr) | |
201 | { | |
202 | spin_lock(&zhdr->page_lock); | |
203 | } | |
204 | ||
76e32a2a VW |
205 | /* Try to lock a z3fold page */ |
206 | static inline int z3fold_page_trylock(struct z3fold_header *zhdr) | |
207 | { | |
208 | return spin_trylock(&zhdr->page_lock); | |
209 | } | |
210 | ||
2f1e5e4d VW |
211 | /* Unlock a z3fold page */ |
212 | static inline void z3fold_page_unlock(struct z3fold_header *zhdr) | |
213 | { | |
214 | spin_unlock(&zhdr->page_lock); | |
215 | } | |
216 | ||
9a001fc1 VW |
217 | /* |
218 | * Encodes the handle of a particular buddy within a z3fold page | |
219 | * Pool lock should be held as this function accesses first_num | |
220 | */ | |
221 | static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud) | |
222 | { | |
223 | unsigned long handle; | |
224 | ||
225 | handle = (unsigned long)zhdr; | |
226 | if (bud != HEADLESS) | |
227 | handle += (bud + zhdr->first_num) & BUDDY_MASK; | |
228 | return handle; | |
229 | } | |
230 | ||
231 | /* Returns the z3fold page where a given handle is stored */ | |
232 | static struct z3fold_header *handle_to_z3fold_header(unsigned long handle) | |
233 | { | |
234 | return (struct z3fold_header *)(handle & PAGE_MASK); | |
235 | } | |
236 | ||
f201ebd8 | 237 | /* |
238 | * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle | |
239 | * but that doesn't matter. because the masking will result in the | |
240 | * correct buddy number. | |
241 | */ | |
9a001fc1 VW |
242 | static enum buddy handle_to_buddy(unsigned long handle) |
243 | { | |
244 | struct z3fold_header *zhdr = handle_to_z3fold_header(handle); | |
245 | return (handle - zhdr->first_num) & BUDDY_MASK; | |
246 | } | |
247 | ||
d30561c5 VW |
248 | static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked) |
249 | { | |
250 | struct page *page = virt_to_page(zhdr); | |
251 | struct z3fold_pool *pool = zhdr->pool; | |
252 | ||
253 | WARN_ON(!list_empty(&zhdr->buddy)); | |
254 | set_bit(PAGE_STALE, &page->private); | |
35529357 | 255 | clear_bit(NEEDS_COMPACTING, &page->private); |
d30561c5 VW |
256 | spin_lock(&pool->lock); |
257 | if (!list_empty(&page->lru)) | |
258 | list_del(&page->lru); | |
259 | spin_unlock(&pool->lock); | |
260 | if (locked) | |
261 | z3fold_page_unlock(zhdr); | |
262 | spin_lock(&pool->stale_lock); | |
263 | list_add(&zhdr->buddy, &pool->stale); | |
264 | queue_work(pool->release_wq, &pool->work); | |
265 | spin_unlock(&pool->stale_lock); | |
266 | } | |
267 | ||
268 | static void __attribute__((__unused__)) | |
269 | release_z3fold_page(struct kref *ref) | |
270 | { | |
271 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, | |
272 | refcount); | |
273 | __release_z3fold_page(zhdr, false); | |
274 | } | |
275 | ||
276 | static void release_z3fold_page_locked(struct kref *ref) | |
277 | { | |
278 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, | |
279 | refcount); | |
280 | WARN_ON(z3fold_page_trylock(zhdr)); | |
281 | __release_z3fold_page(zhdr, true); | |
282 | } | |
283 | ||
284 | static void release_z3fold_page_locked_list(struct kref *ref) | |
285 | { | |
286 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, | |
287 | refcount); | |
288 | spin_lock(&zhdr->pool->lock); | |
289 | list_del_init(&zhdr->buddy); | |
290 | spin_unlock(&zhdr->pool->lock); | |
291 | ||
292 | WARN_ON(z3fold_page_trylock(zhdr)); | |
293 | __release_z3fold_page(zhdr, true); | |
294 | } | |
295 | ||
296 | static void free_pages_work(struct work_struct *w) | |
297 | { | |
298 | struct z3fold_pool *pool = container_of(w, struct z3fold_pool, work); | |
299 | ||
300 | spin_lock(&pool->stale_lock); | |
301 | while (!list_empty(&pool->stale)) { | |
302 | struct z3fold_header *zhdr = list_first_entry(&pool->stale, | |
303 | struct z3fold_header, buddy); | |
304 | struct page *page = virt_to_page(zhdr); | |
305 | ||
306 | list_del(&zhdr->buddy); | |
307 | if (WARN_ON(!test_bit(PAGE_STALE, &page->private))) | |
308 | continue; | |
d30561c5 VW |
309 | spin_unlock(&pool->stale_lock); |
310 | cancel_work_sync(&zhdr->work); | |
311 | free_z3fold_page(page); | |
312 | cond_resched(); | |
313 | spin_lock(&pool->stale_lock); | |
314 | } | |
315 | spin_unlock(&pool->stale_lock); | |
316 | } | |
317 | ||
9a001fc1 VW |
318 | /* |
319 | * Returns the number of free chunks in a z3fold page. | |
320 | * NB: can't be used with HEADLESS pages. | |
321 | */ | |
322 | static int num_free_chunks(struct z3fold_header *zhdr) | |
323 | { | |
324 | int nfree; | |
325 | /* | |
326 | * If there is a middle object, pick up the bigger free space | |
327 | * either before or after it. Otherwise just subtract the number | |
328 | * of chunks occupied by the first and the last objects. | |
329 | */ | |
330 | if (zhdr->middle_chunks != 0) { | |
331 | int nfree_before = zhdr->first_chunks ? | |
ede93213 | 332 | 0 : zhdr->start_middle - ZHDR_CHUNKS; |
9a001fc1 | 333 | int nfree_after = zhdr->last_chunks ? |
ede93213 VW |
334 | 0 : TOTAL_CHUNKS - |
335 | (zhdr->start_middle + zhdr->middle_chunks); | |
9a001fc1 VW |
336 | nfree = max(nfree_before, nfree_after); |
337 | } else | |
338 | nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks; | |
339 | return nfree; | |
340 | } | |
341 | ||
ede93213 VW |
342 | static inline void *mchunk_memmove(struct z3fold_header *zhdr, |
343 | unsigned short dst_chunk) | |
344 | { | |
345 | void *beg = zhdr; | |
346 | return memmove(beg + (dst_chunk << CHUNK_SHIFT), | |
347 | beg + (zhdr->start_middle << CHUNK_SHIFT), | |
348 | zhdr->middle_chunks << CHUNK_SHIFT); | |
349 | } | |
350 | ||
1b096e5a | 351 | #define BIG_CHUNK_GAP 3 |
9a001fc1 VW |
352 | /* Has to be called with lock held */ |
353 | static int z3fold_compact_page(struct z3fold_header *zhdr) | |
354 | { | |
355 | struct page *page = virt_to_page(zhdr); | |
9a001fc1 | 356 | |
ede93213 VW |
357 | if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private)) |
358 | return 0; /* can't move middle chunk, it's used */ | |
9a001fc1 | 359 | |
ede93213 VW |
360 | if (zhdr->middle_chunks == 0) |
361 | return 0; /* nothing to compact */ | |
362 | ||
363 | if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) { | |
364 | /* move to the beginning */ | |
365 | mchunk_memmove(zhdr, ZHDR_CHUNKS); | |
9a001fc1 VW |
366 | zhdr->first_chunks = zhdr->middle_chunks; |
367 | zhdr->middle_chunks = 0; | |
368 | zhdr->start_middle = 0; | |
369 | zhdr->first_num++; | |
1b096e5a | 370 | return 1; |
9a001fc1 | 371 | } |
1b096e5a VW |
372 | |
373 | /* | |
374 | * moving data is expensive, so let's only do that if | |
375 | * there's substantial gain (at least BIG_CHUNK_GAP chunks) | |
376 | */ | |
377 | if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 && | |
378 | zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >= | |
379 | BIG_CHUNK_GAP) { | |
380 | mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS); | |
381 | zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS; | |
382 | return 1; | |
383 | } else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 && | |
384 | TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle | |
385 | + zhdr->middle_chunks) >= | |
386 | BIG_CHUNK_GAP) { | |
387 | unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks - | |
388 | zhdr->middle_chunks; | |
389 | mchunk_memmove(zhdr, new_start); | |
390 | zhdr->start_middle = new_start; | |
391 | return 1; | |
392 | } | |
393 | ||
394 | return 0; | |
9a001fc1 VW |
395 | } |
396 | ||
d30561c5 VW |
397 | static void do_compact_page(struct z3fold_header *zhdr, bool locked) |
398 | { | |
399 | struct z3fold_pool *pool = zhdr->pool; | |
400 | struct page *page; | |
401 | struct list_head *unbuddied; | |
402 | int fchunks; | |
403 | ||
404 | page = virt_to_page(zhdr); | |
405 | if (locked) | |
406 | WARN_ON(z3fold_page_trylock(zhdr)); | |
407 | else | |
408 | z3fold_page_lock(zhdr); | |
5d03a661 | 409 | if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) { |
d30561c5 VW |
410 | z3fold_page_unlock(zhdr); |
411 | return; | |
412 | } | |
413 | spin_lock(&pool->lock); | |
414 | list_del_init(&zhdr->buddy); | |
415 | spin_unlock(&pool->lock); | |
416 | ||
5d03a661 VW |
417 | if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) { |
418 | atomic64_dec(&pool->pages_nr); | |
419 | return; | |
420 | } | |
421 | ||
d30561c5 VW |
422 | z3fold_compact_page(zhdr); |
423 | unbuddied = get_cpu_ptr(pool->unbuddied); | |
424 | fchunks = num_free_chunks(zhdr); | |
425 | if (fchunks < NCHUNKS && | |
426 | (!zhdr->first_chunks || !zhdr->middle_chunks || | |
427 | !zhdr->last_chunks)) { | |
428 | /* the page's not completely free and it's unbuddied */ | |
429 | spin_lock(&pool->lock); | |
430 | list_add(&zhdr->buddy, &unbuddied[fchunks]); | |
431 | spin_unlock(&pool->lock); | |
432 | zhdr->cpu = smp_processor_id(); | |
433 | } | |
434 | put_cpu_ptr(pool->unbuddied); | |
435 | z3fold_page_unlock(zhdr); | |
436 | } | |
437 | ||
438 | static void compact_page_work(struct work_struct *w) | |
439 | { | |
440 | struct z3fold_header *zhdr = container_of(w, struct z3fold_header, | |
441 | work); | |
442 | ||
443 | do_compact_page(zhdr, false); | |
444 | } | |
445 | ||
446 | ||
447 | /* | |
448 | * API Functions | |
449 | */ | |
450 | ||
451 | /** | |
452 | * z3fold_create_pool() - create a new z3fold pool | |
453 | * @name: pool name | |
454 | * @gfp: gfp flags when allocating the z3fold pool structure | |
455 | * @ops: user-defined operations for the z3fold pool | |
456 | * | |
457 | * Return: pointer to the new z3fold pool or NULL if the metadata allocation | |
458 | * failed. | |
459 | */ | |
460 | static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp, | |
461 | const struct z3fold_ops *ops) | |
462 | { | |
463 | struct z3fold_pool *pool = NULL; | |
464 | int i, cpu; | |
465 | ||
466 | pool = kzalloc(sizeof(struct z3fold_pool), gfp); | |
467 | if (!pool) | |
468 | goto out; | |
469 | spin_lock_init(&pool->lock); | |
470 | spin_lock_init(&pool->stale_lock); | |
471 | pool->unbuddied = __alloc_percpu(sizeof(struct list_head)*NCHUNKS, 2); | |
472 | for_each_possible_cpu(cpu) { | |
473 | struct list_head *unbuddied = | |
474 | per_cpu_ptr(pool->unbuddied, cpu); | |
475 | for_each_unbuddied_list(i, 0) | |
476 | INIT_LIST_HEAD(&unbuddied[i]); | |
477 | } | |
478 | INIT_LIST_HEAD(&pool->lru); | |
479 | INIT_LIST_HEAD(&pool->stale); | |
480 | atomic64_set(&pool->pages_nr, 0); | |
481 | pool->name = name; | |
482 | pool->compact_wq = create_singlethread_workqueue(pool->name); | |
483 | if (!pool->compact_wq) | |
484 | goto out; | |
485 | pool->release_wq = create_singlethread_workqueue(pool->name); | |
486 | if (!pool->release_wq) | |
487 | goto out_wq; | |
488 | INIT_WORK(&pool->work, free_pages_work); | |
489 | pool->ops = ops; | |
490 | return pool; | |
491 | ||
492 | out_wq: | |
493 | destroy_workqueue(pool->compact_wq); | |
494 | out: | |
495 | kfree(pool); | |
496 | return NULL; | |
497 | } | |
498 | ||
499 | /** | |
500 | * z3fold_destroy_pool() - destroys an existing z3fold pool | |
501 | * @pool: the z3fold pool to be destroyed | |
502 | * | |
503 | * The pool should be emptied before this function is called. | |
504 | */ | |
505 | static void z3fold_destroy_pool(struct z3fold_pool *pool) | |
506 | { | |
507 | destroy_workqueue(pool->release_wq); | |
508 | destroy_workqueue(pool->compact_wq); | |
509 | kfree(pool); | |
510 | } | |
511 | ||
9a001fc1 VW |
512 | /** |
513 | * z3fold_alloc() - allocates a region of a given size | |
514 | * @pool: z3fold pool from which to allocate | |
515 | * @size: size in bytes of the desired allocation | |
516 | * @gfp: gfp flags used if the pool needs to grow | |
517 | * @handle: handle of the new allocation | |
518 | * | |
519 | * This function will attempt to find a free region in the pool large enough to | |
520 | * satisfy the allocation request. A search of the unbuddied lists is | |
521 | * performed first. If no suitable free region is found, then a new page is | |
522 | * allocated and added to the pool to satisfy the request. | |
523 | * | |
524 | * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used | |
525 | * as z3fold pool pages. | |
526 | * | |
527 | * Return: 0 if success and handle is set, otherwise -EINVAL if the size or | |
528 | * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate | |
529 | * a new page. | |
530 | */ | |
531 | static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp, | |
532 | unsigned long *handle) | |
533 | { | |
534 | int chunks = 0, i, freechunks; | |
535 | struct z3fold_header *zhdr = NULL; | |
d30561c5 | 536 | struct page *page = NULL; |
9a001fc1 | 537 | enum buddy bud; |
d30561c5 | 538 | bool can_sleep = (gfp & __GFP_RECLAIM) == __GFP_RECLAIM; |
9a001fc1 VW |
539 | |
540 | if (!size || (gfp & __GFP_HIGHMEM)) | |
541 | return -EINVAL; | |
542 | ||
543 | if (size > PAGE_SIZE) | |
544 | return -ENOSPC; | |
545 | ||
546 | if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE) | |
547 | bud = HEADLESS; | |
548 | else { | |
d30561c5 | 549 | struct list_head *unbuddied; |
9a001fc1 | 550 | chunks = size_to_chunks(size); |
9a001fc1 | 551 | |
d30561c5 | 552 | lookup: |
9a001fc1 | 553 | /* First, try to find an unbuddied z3fold page. */ |
d30561c5 | 554 | unbuddied = get_cpu_ptr(pool->unbuddied); |
9a001fc1 | 555 | for_each_unbuddied_list(i, chunks) { |
d30561c5 VW |
556 | struct list_head *l = &unbuddied[i]; |
557 | ||
558 | zhdr = list_first_entry_or_null(READ_ONCE(l), | |
9a001fc1 | 559 | struct z3fold_header, buddy); |
d30561c5 VW |
560 | |
561 | if (!zhdr) | |
2f1e5e4d | 562 | continue; |
d30561c5 VW |
563 | |
564 | /* Re-check under lock. */ | |
565 | spin_lock(&pool->lock); | |
566 | l = &unbuddied[i]; | |
567 | if (unlikely(zhdr != list_first_entry(READ_ONCE(l), | |
568 | struct z3fold_header, buddy)) || | |
569 | !z3fold_page_trylock(zhdr)) { | |
570 | spin_unlock(&pool->lock); | |
571 | put_cpu_ptr(pool->unbuddied); | |
572 | goto lookup; | |
2f1e5e4d VW |
573 | } |
574 | list_del_init(&zhdr->buddy); | |
d30561c5 | 575 | zhdr->cpu = -1; |
2f1e5e4d VW |
576 | spin_unlock(&pool->lock); |
577 | ||
578 | page = virt_to_page(zhdr); | |
d30561c5 VW |
579 | if (test_bit(NEEDS_COMPACTING, &page->private)) { |
580 | z3fold_page_unlock(zhdr); | |
581 | zhdr = NULL; | |
582 | put_cpu_ptr(pool->unbuddied); | |
583 | if (can_sleep) | |
584 | cond_resched(); | |
585 | goto lookup; | |
586 | } | |
587 | ||
588 | /* | |
589 | * this page could not be removed from its unbuddied | |
590 | * list while pool lock was held, and then we've taken | |
591 | * page lock so kref_put could not be called before | |
592 | * we got here, so it's safe to just call kref_get() | |
593 | */ | |
594 | kref_get(&zhdr->refcount); | |
595 | break; | |
596 | } | |
597 | put_cpu_ptr(pool->unbuddied); | |
598 | ||
599 | if (zhdr) { | |
2f1e5e4d VW |
600 | if (zhdr->first_chunks == 0) { |
601 | if (zhdr->middle_chunks != 0 && | |
602 | chunks >= zhdr->start_middle) | |
9a001fc1 | 603 | bud = LAST; |
2f1e5e4d VW |
604 | else |
605 | bud = FIRST; | |
606 | } else if (zhdr->last_chunks == 0) | |
607 | bud = LAST; | |
608 | else if (zhdr->middle_chunks == 0) | |
609 | bud = MIDDLE; | |
610 | else { | |
5a27aa82 | 611 | if (kref_put(&zhdr->refcount, |
d30561c5 | 612 | release_z3fold_page_locked)) |
5a27aa82 | 613 | atomic64_dec(&pool->pages_nr); |
d30561c5 VW |
614 | else |
615 | z3fold_page_unlock(zhdr); | |
2f1e5e4d VW |
616 | pr_err("No free chunks in unbuddied\n"); |
617 | WARN_ON(1); | |
d30561c5 | 618 | goto lookup; |
9a001fc1 | 619 | } |
2f1e5e4d | 620 | goto found; |
9a001fc1 VW |
621 | } |
622 | bud = FIRST; | |
9a001fc1 VW |
623 | } |
624 | ||
d30561c5 VW |
625 | spin_lock(&pool->stale_lock); |
626 | zhdr = list_first_entry_or_null(&pool->stale, | |
627 | struct z3fold_header, buddy); | |
628 | /* | |
629 | * Before allocating a page, let's see if we can take one from the | |
630 | * stale pages list. cancel_work_sync() can sleep so we must make | |
631 | * sure it won't be called in case we're in atomic context. | |
632 | */ | |
35529357 | 633 | if (zhdr && (can_sleep || !work_pending(&zhdr->work))) { |
d30561c5 | 634 | list_del(&zhdr->buddy); |
d30561c5 VW |
635 | spin_unlock(&pool->stale_lock); |
636 | if (can_sleep) | |
637 | cancel_work_sync(&zhdr->work); | |
638 | page = virt_to_page(zhdr); | |
639 | } else { | |
640 | spin_unlock(&pool->stale_lock); | |
641 | page = alloc_page(gfp); | |
642 | } | |
643 | ||
9a001fc1 VW |
644 | if (!page) |
645 | return -ENOMEM; | |
2f1e5e4d | 646 | |
12d59ae6 | 647 | atomic64_inc(&pool->pages_nr); |
d30561c5 | 648 | zhdr = init_z3fold_page(page, pool); |
9a001fc1 VW |
649 | |
650 | if (bud == HEADLESS) { | |
651 | set_bit(PAGE_HEADLESS, &page->private); | |
652 | goto headless; | |
653 | } | |
2f1e5e4d | 654 | z3fold_page_lock(zhdr); |
9a001fc1 VW |
655 | |
656 | found: | |
657 | if (bud == FIRST) | |
658 | zhdr->first_chunks = chunks; | |
659 | else if (bud == LAST) | |
660 | zhdr->last_chunks = chunks; | |
661 | else { | |
662 | zhdr->middle_chunks = chunks; | |
ede93213 | 663 | zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS; |
9a001fc1 VW |
664 | } |
665 | ||
666 | if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 || | |
667 | zhdr->middle_chunks == 0) { | |
d30561c5 VW |
668 | struct list_head *unbuddied = get_cpu_ptr(pool->unbuddied); |
669 | ||
9a001fc1 VW |
670 | /* Add to unbuddied list */ |
671 | freechunks = num_free_chunks(zhdr); | |
d30561c5 VW |
672 | spin_lock(&pool->lock); |
673 | list_add(&zhdr->buddy, &unbuddied[freechunks]); | |
674 | spin_unlock(&pool->lock); | |
675 | zhdr->cpu = smp_processor_id(); | |
676 | put_cpu_ptr(pool->unbuddied); | |
9a001fc1 VW |
677 | } |
678 | ||
679 | headless: | |
d30561c5 | 680 | spin_lock(&pool->lock); |
9a001fc1 VW |
681 | /* Add/move z3fold page to beginning of LRU */ |
682 | if (!list_empty(&page->lru)) | |
683 | list_del(&page->lru); | |
684 | ||
685 | list_add(&page->lru, &pool->lru); | |
686 | ||
687 | *handle = encode_handle(zhdr, bud); | |
688 | spin_unlock(&pool->lock); | |
2f1e5e4d VW |
689 | if (bud != HEADLESS) |
690 | z3fold_page_unlock(zhdr); | |
9a001fc1 VW |
691 | |
692 | return 0; | |
693 | } | |
694 | ||
695 | /** | |
696 | * z3fold_free() - frees the allocation associated with the given handle | |
697 | * @pool: pool in which the allocation resided | |
698 | * @handle: handle associated with the allocation returned by z3fold_alloc() | |
699 | * | |
700 | * In the case that the z3fold page in which the allocation resides is under | |
701 | * reclaim, as indicated by the PG_reclaim flag being set, this function | |
702 | * only sets the first|last_chunks to 0. The page is actually freed | |
703 | * once both buddies are evicted (see z3fold_reclaim_page() below). | |
704 | */ | |
705 | static void z3fold_free(struct z3fold_pool *pool, unsigned long handle) | |
706 | { | |
707 | struct z3fold_header *zhdr; | |
9a001fc1 VW |
708 | struct page *page; |
709 | enum buddy bud; | |
710 | ||
9a001fc1 VW |
711 | zhdr = handle_to_z3fold_header(handle); |
712 | page = virt_to_page(zhdr); | |
713 | ||
714 | if (test_bit(PAGE_HEADLESS, &page->private)) { | |
715 | /* HEADLESS page stored */ | |
716 | bud = HEADLESS; | |
717 | } else { | |
2f1e5e4d | 718 | z3fold_page_lock(zhdr); |
43afc194 | 719 | bud = handle_to_buddy(handle); |
9a001fc1 VW |
720 | |
721 | switch (bud) { | |
722 | case FIRST: | |
723 | zhdr->first_chunks = 0; | |
724 | break; | |
725 | case MIDDLE: | |
726 | zhdr->middle_chunks = 0; | |
727 | zhdr->start_middle = 0; | |
728 | break; | |
729 | case LAST: | |
730 | zhdr->last_chunks = 0; | |
731 | break; | |
732 | default: | |
733 | pr_err("%s: unknown bud %d\n", __func__, bud); | |
734 | WARN_ON(1); | |
2f1e5e4d | 735 | z3fold_page_unlock(zhdr); |
9a001fc1 VW |
736 | return; |
737 | } | |
738 | } | |
739 | ||
5a27aa82 | 740 | if (bud == HEADLESS) { |
2f1e5e4d | 741 | spin_lock(&pool->lock); |
9a001fc1 | 742 | list_del(&page->lru); |
2f1e5e4d | 743 | spin_unlock(&pool->lock); |
5a27aa82 | 744 | free_z3fold_page(page); |
12d59ae6 | 745 | atomic64_dec(&pool->pages_nr); |
d30561c5 VW |
746 | return; |
747 | } | |
748 | ||
749 | if (kref_put(&zhdr->refcount, release_z3fold_page_locked_list)) { | |
750 | atomic64_dec(&pool->pages_nr); | |
751 | return; | |
752 | } | |
9fc5adb0 VW |
753 | if (test_bit(UNDER_RECLAIM, &page->private)) { |
754 | z3fold_page_unlock(zhdr); | |
755 | return; | |
756 | } | |
d30561c5 | 757 | if (test_and_set_bit(NEEDS_COMPACTING, &page->private)) { |
5a27aa82 | 758 | z3fold_page_unlock(zhdr); |
d30561c5 VW |
759 | return; |
760 | } | |
761 | if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) { | |
2f1e5e4d | 762 | spin_lock(&pool->lock); |
d30561c5 | 763 | list_del_init(&zhdr->buddy); |
2f1e5e4d | 764 | spin_unlock(&pool->lock); |
d30561c5 | 765 | zhdr->cpu = -1; |
5d03a661 | 766 | kref_get(&zhdr->refcount); |
d30561c5 VW |
767 | do_compact_page(zhdr, true); |
768 | return; | |
9a001fc1 | 769 | } |
5d03a661 | 770 | kref_get(&zhdr->refcount); |
d30561c5 VW |
771 | queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work); |
772 | z3fold_page_unlock(zhdr); | |
9a001fc1 VW |
773 | } |
774 | ||
775 | /** | |
776 | * z3fold_reclaim_page() - evicts allocations from a pool page and frees it | |
777 | * @pool: pool from which a page will attempt to be evicted | |
778 | * @retires: number of pages on the LRU list for which eviction will | |
779 | * be attempted before failing | |
780 | * | |
781 | * z3fold reclaim is different from normal system reclaim in that it is done | |
782 | * from the bottom, up. This is because only the bottom layer, z3fold, has | |
783 | * information on how the allocations are organized within each z3fold page. | |
784 | * This has the potential to create interesting locking situations between | |
785 | * z3fold and the user, however. | |
786 | * | |
787 | * To avoid these, this is how z3fold_reclaim_page() should be called: | |
788 | ||
789 | * The user detects a page should be reclaimed and calls z3fold_reclaim_page(). | |
790 | * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and | |
791 | * call the user-defined eviction handler with the pool and handle as | |
792 | * arguments. | |
793 | * | |
794 | * If the handle can not be evicted, the eviction handler should return | |
795 | * non-zero. z3fold_reclaim_page() will add the z3fold page back to the | |
796 | * appropriate list and try the next z3fold page on the LRU up to | |
797 | * a user defined number of retries. | |
798 | * | |
799 | * If the handle is successfully evicted, the eviction handler should | |
800 | * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free() | |
801 | * contains logic to delay freeing the page if the page is under reclaim, | |
802 | * as indicated by the setting of the PG_reclaim flag on the underlying page. | |
803 | * | |
804 | * If all buddies in the z3fold page are successfully evicted, then the | |
805 | * z3fold page can be freed. | |
806 | * | |
807 | * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are | |
808 | * no pages to evict or an eviction handler is not registered, -EAGAIN if | |
809 | * the retry limit was hit. | |
810 | */ | |
811 | static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries) | |
812 | { | |
d30561c5 VW |
813 | int i, ret = 0; |
814 | struct z3fold_header *zhdr = NULL; | |
815 | struct page *page = NULL; | |
816 | struct list_head *pos; | |
9a001fc1 VW |
817 | unsigned long first_handle = 0, middle_handle = 0, last_handle = 0; |
818 | ||
819 | spin_lock(&pool->lock); | |
2f1e5e4d | 820 | if (!pool->ops || !pool->ops->evict || retries == 0) { |
9a001fc1 VW |
821 | spin_unlock(&pool->lock); |
822 | return -EINVAL; | |
823 | } | |
824 | for (i = 0; i < retries; i++) { | |
2f1e5e4d VW |
825 | if (list_empty(&pool->lru)) { |
826 | spin_unlock(&pool->lock); | |
827 | return -EINVAL; | |
828 | } | |
d30561c5 VW |
829 | list_for_each_prev(pos, &pool->lru) { |
830 | page = list_entry(pos, struct page, lru); | |
831 | if (test_bit(PAGE_HEADLESS, &page->private)) | |
832 | /* candidate found */ | |
833 | break; | |
834 | ||
835 | zhdr = page_address(page); | |
836 | if (!z3fold_page_trylock(zhdr)) | |
837 | continue; /* can't evict at this point */ | |
838 | kref_get(&zhdr->refcount); | |
839 | list_del_init(&zhdr->buddy); | |
840 | zhdr->cpu = -1; | |
9fc5adb0 VW |
841 | set_bit(UNDER_RECLAIM, &page->private); |
842 | break; | |
d30561c5 VW |
843 | } |
844 | ||
5a27aa82 | 845 | list_del_init(&page->lru); |
d30561c5 | 846 | spin_unlock(&pool->lock); |
9a001fc1 | 847 | |
9a001fc1 | 848 | if (!test_bit(PAGE_HEADLESS, &page->private)) { |
9a001fc1 VW |
849 | /* |
850 | * We need encode the handles before unlocking, since | |
851 | * we can race with free that will set | |
852 | * (first|last)_chunks to 0 | |
853 | */ | |
854 | first_handle = 0; | |
855 | last_handle = 0; | |
856 | middle_handle = 0; | |
857 | if (zhdr->first_chunks) | |
858 | first_handle = encode_handle(zhdr, FIRST); | |
859 | if (zhdr->middle_chunks) | |
860 | middle_handle = encode_handle(zhdr, MIDDLE); | |
861 | if (zhdr->last_chunks) | |
862 | last_handle = encode_handle(zhdr, LAST); | |
d30561c5 VW |
863 | /* |
864 | * it's safe to unlock here because we hold a | |
865 | * reference to this page | |
866 | */ | |
2f1e5e4d | 867 | z3fold_page_unlock(zhdr); |
9a001fc1 VW |
868 | } else { |
869 | first_handle = encode_handle(zhdr, HEADLESS); | |
870 | last_handle = middle_handle = 0; | |
871 | } | |
872 | ||
9a001fc1 VW |
873 | /* Issue the eviction callback(s) */ |
874 | if (middle_handle) { | |
875 | ret = pool->ops->evict(pool, middle_handle); | |
876 | if (ret) | |
877 | goto next; | |
878 | } | |
879 | if (first_handle) { | |
880 | ret = pool->ops->evict(pool, first_handle); | |
881 | if (ret) | |
882 | goto next; | |
883 | } | |
884 | if (last_handle) { | |
885 | ret = pool->ops->evict(pool, last_handle); | |
886 | if (ret) | |
887 | goto next; | |
888 | } | |
889 | next: | |
5a27aa82 VW |
890 | if (test_bit(PAGE_HEADLESS, &page->private)) { |
891 | if (ret == 0) { | |
892 | free_z3fold_page(page); | |
893 | return 0; | |
5a27aa82 | 894 | } |
9fc5adb0 VW |
895 | spin_lock(&pool->lock); |
896 | list_add(&page->lru, &pool->lru); | |
897 | spin_unlock(&pool->lock); | |
898 | } else { | |
899 | z3fold_page_lock(zhdr); | |
900 | clear_bit(UNDER_RECLAIM, &page->private); | |
901 | if (kref_put(&zhdr->refcount, | |
902 | release_z3fold_page_locked)) { | |
903 | atomic64_dec(&pool->pages_nr); | |
904 | return 0; | |
905 | } | |
906 | /* | |
907 | * if we are here, the page is still not completely | |
908 | * free. Take the global pool lock then to be able | |
909 | * to add it back to the lru list | |
910 | */ | |
911 | spin_lock(&pool->lock); | |
912 | list_add(&page->lru, &pool->lru); | |
d5567c9d | 913 | spin_unlock(&pool->lock); |
9fc5adb0 | 914 | z3fold_page_unlock(zhdr); |
5a27aa82 | 915 | } |
2f1e5e4d | 916 | |
9fc5adb0 VW |
917 | /* We started off locked to we need to lock the pool back */ |
918 | spin_lock(&pool->lock); | |
9a001fc1 VW |
919 | } |
920 | spin_unlock(&pool->lock); | |
921 | return -EAGAIN; | |
922 | } | |
923 | ||
924 | /** | |
925 | * z3fold_map() - maps the allocation associated with the given handle | |
926 | * @pool: pool in which the allocation resides | |
927 | * @handle: handle associated with the allocation to be mapped | |
928 | * | |
929 | * Extracts the buddy number from handle and constructs the pointer to the | |
930 | * correct starting chunk within the page. | |
931 | * | |
932 | * Returns: a pointer to the mapped allocation | |
933 | */ | |
934 | static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle) | |
935 | { | |
936 | struct z3fold_header *zhdr; | |
937 | struct page *page; | |
938 | void *addr; | |
939 | enum buddy buddy; | |
940 | ||
9a001fc1 VW |
941 | zhdr = handle_to_z3fold_header(handle); |
942 | addr = zhdr; | |
943 | page = virt_to_page(zhdr); | |
944 | ||
945 | if (test_bit(PAGE_HEADLESS, &page->private)) | |
946 | goto out; | |
947 | ||
2f1e5e4d | 948 | z3fold_page_lock(zhdr); |
9a001fc1 VW |
949 | buddy = handle_to_buddy(handle); |
950 | switch (buddy) { | |
951 | case FIRST: | |
952 | addr += ZHDR_SIZE_ALIGNED; | |
953 | break; | |
954 | case MIDDLE: | |
955 | addr += zhdr->start_middle << CHUNK_SHIFT; | |
956 | set_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
957 | break; | |
958 | case LAST: | |
959 | addr += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT); | |
960 | break; | |
961 | default: | |
962 | pr_err("unknown buddy id %d\n", buddy); | |
963 | WARN_ON(1); | |
964 | addr = NULL; | |
965 | break; | |
966 | } | |
2f1e5e4d VW |
967 | |
968 | z3fold_page_unlock(zhdr); | |
9a001fc1 | 969 | out: |
9a001fc1 VW |
970 | return addr; |
971 | } | |
972 | ||
973 | /** | |
974 | * z3fold_unmap() - unmaps the allocation associated with the given handle | |
975 | * @pool: pool in which the allocation resides | |
976 | * @handle: handle associated with the allocation to be unmapped | |
977 | */ | |
978 | static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle) | |
979 | { | |
980 | struct z3fold_header *zhdr; | |
981 | struct page *page; | |
982 | enum buddy buddy; | |
983 | ||
9a001fc1 VW |
984 | zhdr = handle_to_z3fold_header(handle); |
985 | page = virt_to_page(zhdr); | |
986 | ||
2f1e5e4d | 987 | if (test_bit(PAGE_HEADLESS, &page->private)) |
9a001fc1 | 988 | return; |
9a001fc1 | 989 | |
2f1e5e4d | 990 | z3fold_page_lock(zhdr); |
9a001fc1 VW |
991 | buddy = handle_to_buddy(handle); |
992 | if (buddy == MIDDLE) | |
993 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
2f1e5e4d | 994 | z3fold_page_unlock(zhdr); |
9a001fc1 VW |
995 | } |
996 | ||
997 | /** | |
998 | * z3fold_get_pool_size() - gets the z3fold pool size in pages | |
999 | * @pool: pool whose size is being queried | |
1000 | * | |
12d59ae6 | 1001 | * Returns: size in pages of the given pool. |
9a001fc1 VW |
1002 | */ |
1003 | static u64 z3fold_get_pool_size(struct z3fold_pool *pool) | |
1004 | { | |
12d59ae6 | 1005 | return atomic64_read(&pool->pages_nr); |
9a001fc1 VW |
1006 | } |
1007 | ||
1008 | /***************** | |
1009 | * zpool | |
1010 | ****************/ | |
1011 | ||
1012 | static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle) | |
1013 | { | |
1014 | if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict) | |
1015 | return pool->zpool_ops->evict(pool->zpool, handle); | |
1016 | else | |
1017 | return -ENOENT; | |
1018 | } | |
1019 | ||
1020 | static const struct z3fold_ops z3fold_zpool_ops = { | |
1021 | .evict = z3fold_zpool_evict | |
1022 | }; | |
1023 | ||
1024 | static void *z3fold_zpool_create(const char *name, gfp_t gfp, | |
1025 | const struct zpool_ops *zpool_ops, | |
1026 | struct zpool *zpool) | |
1027 | { | |
1028 | struct z3fold_pool *pool; | |
1029 | ||
d30561c5 VW |
1030 | pool = z3fold_create_pool(name, gfp, |
1031 | zpool_ops ? &z3fold_zpool_ops : NULL); | |
9a001fc1 VW |
1032 | if (pool) { |
1033 | pool->zpool = zpool; | |
1034 | pool->zpool_ops = zpool_ops; | |
1035 | } | |
1036 | return pool; | |
1037 | } | |
1038 | ||
1039 | static void z3fold_zpool_destroy(void *pool) | |
1040 | { | |
1041 | z3fold_destroy_pool(pool); | |
1042 | } | |
1043 | ||
1044 | static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp, | |
1045 | unsigned long *handle) | |
1046 | { | |
1047 | return z3fold_alloc(pool, size, gfp, handle); | |
1048 | } | |
1049 | static void z3fold_zpool_free(void *pool, unsigned long handle) | |
1050 | { | |
1051 | z3fold_free(pool, handle); | |
1052 | } | |
1053 | ||
1054 | static int z3fold_zpool_shrink(void *pool, unsigned int pages, | |
1055 | unsigned int *reclaimed) | |
1056 | { | |
1057 | unsigned int total = 0; | |
1058 | int ret = -EINVAL; | |
1059 | ||
1060 | while (total < pages) { | |
1061 | ret = z3fold_reclaim_page(pool, 8); | |
1062 | if (ret < 0) | |
1063 | break; | |
1064 | total++; | |
1065 | } | |
1066 | ||
1067 | if (reclaimed) | |
1068 | *reclaimed = total; | |
1069 | ||
1070 | return ret; | |
1071 | } | |
1072 | ||
1073 | static void *z3fold_zpool_map(void *pool, unsigned long handle, | |
1074 | enum zpool_mapmode mm) | |
1075 | { | |
1076 | return z3fold_map(pool, handle); | |
1077 | } | |
1078 | static void z3fold_zpool_unmap(void *pool, unsigned long handle) | |
1079 | { | |
1080 | z3fold_unmap(pool, handle); | |
1081 | } | |
1082 | ||
1083 | static u64 z3fold_zpool_total_size(void *pool) | |
1084 | { | |
1085 | return z3fold_get_pool_size(pool) * PAGE_SIZE; | |
1086 | } | |
1087 | ||
1088 | static struct zpool_driver z3fold_zpool_driver = { | |
1089 | .type = "z3fold", | |
1090 | .owner = THIS_MODULE, | |
1091 | .create = z3fold_zpool_create, | |
1092 | .destroy = z3fold_zpool_destroy, | |
1093 | .malloc = z3fold_zpool_malloc, | |
1094 | .free = z3fold_zpool_free, | |
1095 | .shrink = z3fold_zpool_shrink, | |
1096 | .map = z3fold_zpool_map, | |
1097 | .unmap = z3fold_zpool_unmap, | |
1098 | .total_size = z3fold_zpool_total_size, | |
1099 | }; | |
1100 | ||
1101 | MODULE_ALIAS("zpool-z3fold"); | |
1102 | ||
1103 | static int __init init_z3fold(void) | |
1104 | { | |
ede93213 VW |
1105 | /* Make sure the z3fold header is not larger than the page size */ |
1106 | BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE); | |
9a001fc1 VW |
1107 | zpool_register_driver(&z3fold_zpool_driver); |
1108 | ||
1109 | return 0; | |
1110 | } | |
1111 | ||
1112 | static void __exit exit_z3fold(void) | |
1113 | { | |
1114 | zpool_unregister_driver(&z3fold_zpool_driver); | |
1115 | } | |
1116 | ||
1117 | module_init(init_z3fold); | |
1118 | module_exit(exit_z3fold); | |
1119 | ||
1120 | MODULE_LICENSE("GPL"); | |
1121 | MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>"); | |
1122 | MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages"); |