]>
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) |
ca0246bb | 102 | #define BUDDY_SHIFT 2 |
9a001fc1 | 103 | |
9a001fc1 VW |
104 | /** |
105 | * struct z3fold_pool - stores metadata for each z3fold pool | |
d30561c5 VW |
106 | * @name: pool name |
107 | * @lock: protects pool unbuddied/lru lists | |
108 | * @stale_lock: protects pool stale page list | |
109 | * @unbuddied: per-cpu array of lists tracking z3fold pages that contain 2- | |
110 | * buddies; the list each z3fold page is added to depends on | |
111 | * the size of its free region. | |
9a001fc1 VW |
112 | * @lru: list tracking the z3fold pages in LRU order by most recently |
113 | * added buddy. | |
d30561c5 | 114 | * @stale: list of pages marked for freeing |
9a001fc1 VW |
115 | * @pages_nr: number of z3fold pages in the pool. |
116 | * @ops: pointer to a structure of user defined operations specified at | |
117 | * pool creation time. | |
d30561c5 VW |
118 | * @compact_wq: workqueue for page layout background optimization |
119 | * @release_wq: workqueue for safe page release | |
120 | * @work: work_struct for safe page release | |
9a001fc1 VW |
121 | * |
122 | * This structure is allocated at pool creation time and maintains metadata | |
123 | * pertaining to a particular z3fold pool. | |
124 | */ | |
125 | struct z3fold_pool { | |
d30561c5 | 126 | const char *name; |
9a001fc1 | 127 | spinlock_t lock; |
d30561c5 VW |
128 | spinlock_t stale_lock; |
129 | struct list_head *unbuddied; | |
9a001fc1 | 130 | struct list_head lru; |
d30561c5 | 131 | struct list_head stale; |
12d59ae6 | 132 | atomic64_t pages_nr; |
9a001fc1 VW |
133 | const struct z3fold_ops *ops; |
134 | struct zpool *zpool; | |
135 | const struct zpool_ops *zpool_ops; | |
d30561c5 VW |
136 | struct workqueue_struct *compact_wq; |
137 | struct workqueue_struct *release_wq; | |
138 | struct work_struct work; | |
9a001fc1 VW |
139 | }; |
140 | ||
9a001fc1 VW |
141 | /* |
142 | * Internal z3fold page flags | |
143 | */ | |
144 | enum z3fold_page_flags { | |
5a27aa82 | 145 | PAGE_HEADLESS = 0, |
9a001fc1 | 146 | MIDDLE_CHUNK_MAPPED, |
d30561c5 | 147 | NEEDS_COMPACTING, |
6098d7e1 | 148 | PAGE_STALE, |
ca0246bb | 149 | PAGE_CLAIMED, /* by either reclaim or free */ |
9a001fc1 VW |
150 | }; |
151 | ||
152 | /***************** | |
153 | * Helpers | |
154 | *****************/ | |
155 | ||
156 | /* Converts an allocation size in bytes to size in z3fold chunks */ | |
157 | static int size_to_chunks(size_t size) | |
158 | { | |
159 | return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT; | |
160 | } | |
161 | ||
162 | #define for_each_unbuddied_list(_iter, _begin) \ | |
163 | for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++) | |
164 | ||
d30561c5 VW |
165 | static void compact_page_work(struct work_struct *w); |
166 | ||
9a001fc1 | 167 | /* Initializes the z3fold header of a newly allocated z3fold page */ |
d30561c5 VW |
168 | static struct z3fold_header *init_z3fold_page(struct page *page, |
169 | struct z3fold_pool *pool) | |
9a001fc1 VW |
170 | { |
171 | struct z3fold_header *zhdr = page_address(page); | |
172 | ||
173 | INIT_LIST_HEAD(&page->lru); | |
9a001fc1 VW |
174 | clear_bit(PAGE_HEADLESS, &page->private); |
175 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
d30561c5 VW |
176 | clear_bit(NEEDS_COMPACTING, &page->private); |
177 | clear_bit(PAGE_STALE, &page->private); | |
ca0246bb | 178 | clear_bit(PAGE_CLAIMED, &page->private); |
9a001fc1 | 179 | |
2f1e5e4d | 180 | spin_lock_init(&zhdr->page_lock); |
5a27aa82 | 181 | kref_init(&zhdr->refcount); |
9a001fc1 VW |
182 | zhdr->first_chunks = 0; |
183 | zhdr->middle_chunks = 0; | |
184 | zhdr->last_chunks = 0; | |
185 | zhdr->first_num = 0; | |
186 | zhdr->start_middle = 0; | |
d30561c5 VW |
187 | zhdr->cpu = -1; |
188 | zhdr->pool = pool; | |
9a001fc1 | 189 | INIT_LIST_HEAD(&zhdr->buddy); |
d30561c5 | 190 | INIT_WORK(&zhdr->work, compact_page_work); |
9a001fc1 VW |
191 | return zhdr; |
192 | } | |
193 | ||
194 | /* Resets the struct page fields and frees the page */ | |
5a27aa82 | 195 | static void free_z3fold_page(struct page *page) |
9a001fc1 | 196 | { |
5a27aa82 VW |
197 | __free_page(page); |
198 | } | |
199 | ||
2f1e5e4d VW |
200 | /* Lock a z3fold page */ |
201 | static inline void z3fold_page_lock(struct z3fold_header *zhdr) | |
202 | { | |
203 | spin_lock(&zhdr->page_lock); | |
204 | } | |
205 | ||
76e32a2a VW |
206 | /* Try to lock a z3fold page */ |
207 | static inline int z3fold_page_trylock(struct z3fold_header *zhdr) | |
208 | { | |
209 | return spin_trylock(&zhdr->page_lock); | |
210 | } | |
211 | ||
2f1e5e4d VW |
212 | /* Unlock a z3fold page */ |
213 | static inline void z3fold_page_unlock(struct z3fold_header *zhdr) | |
214 | { | |
215 | spin_unlock(&zhdr->page_lock); | |
216 | } | |
217 | ||
9a001fc1 VW |
218 | /* |
219 | * Encodes the handle of a particular buddy within a z3fold page | |
220 | * Pool lock should be held as this function accesses first_num | |
221 | */ | |
222 | static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud) | |
223 | { | |
224 | unsigned long handle; | |
225 | ||
226 | handle = (unsigned long)zhdr; | |
ca0246bb VW |
227 | if (bud != HEADLESS) { |
228 | handle |= (bud + zhdr->first_num) & BUDDY_MASK; | |
229 | if (bud == LAST) | |
230 | handle |= (zhdr->last_chunks << BUDDY_SHIFT); | |
231 | } | |
9a001fc1 VW |
232 | return handle; |
233 | } | |
234 | ||
235 | /* Returns the z3fold page where a given handle is stored */ | |
236 | static struct z3fold_header *handle_to_z3fold_header(unsigned long handle) | |
237 | { | |
238 | return (struct z3fold_header *)(handle & PAGE_MASK); | |
239 | } | |
240 | ||
ca0246bb VW |
241 | /* only for LAST bud, returns zero otherwise */ |
242 | static unsigned short handle_to_chunks(unsigned long handle) | |
243 | { | |
244 | return (handle & ~PAGE_MASK) >> BUDDY_SHIFT; | |
245 | } | |
246 | ||
f201ebd8 | 247 | /* |
248 | * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle | |
249 | * but that doesn't matter. because the masking will result in the | |
250 | * correct buddy number. | |
251 | */ | |
9a001fc1 VW |
252 | static enum buddy handle_to_buddy(unsigned long handle) |
253 | { | |
254 | struct z3fold_header *zhdr = handle_to_z3fold_header(handle); | |
255 | return (handle - zhdr->first_num) & BUDDY_MASK; | |
256 | } | |
257 | ||
9050cce1 VW |
258 | static inline struct z3fold_pool *zhdr_to_pool(struct z3fold_header *zhdr) |
259 | { | |
260 | return zhdr->pool; | |
261 | } | |
262 | ||
d30561c5 VW |
263 | static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked) |
264 | { | |
265 | struct page *page = virt_to_page(zhdr); | |
9050cce1 | 266 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); |
d30561c5 VW |
267 | |
268 | WARN_ON(!list_empty(&zhdr->buddy)); | |
269 | set_bit(PAGE_STALE, &page->private); | |
35529357 | 270 | clear_bit(NEEDS_COMPACTING, &page->private); |
d30561c5 VW |
271 | spin_lock(&pool->lock); |
272 | if (!list_empty(&page->lru)) | |
273 | list_del(&page->lru); | |
274 | spin_unlock(&pool->lock); | |
275 | if (locked) | |
276 | z3fold_page_unlock(zhdr); | |
277 | spin_lock(&pool->stale_lock); | |
278 | list_add(&zhdr->buddy, &pool->stale); | |
279 | queue_work(pool->release_wq, &pool->work); | |
280 | spin_unlock(&pool->stale_lock); | |
281 | } | |
282 | ||
283 | static void __attribute__((__unused__)) | |
284 | release_z3fold_page(struct kref *ref) | |
285 | { | |
286 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, | |
287 | refcount); | |
288 | __release_z3fold_page(zhdr, false); | |
289 | } | |
290 | ||
291 | static void release_z3fold_page_locked(struct kref *ref) | |
292 | { | |
293 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, | |
294 | refcount); | |
295 | WARN_ON(z3fold_page_trylock(zhdr)); | |
296 | __release_z3fold_page(zhdr, true); | |
297 | } | |
298 | ||
299 | static void release_z3fold_page_locked_list(struct kref *ref) | |
300 | { | |
301 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, | |
302 | refcount); | |
9050cce1 VW |
303 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); |
304 | spin_lock(&pool->lock); | |
d30561c5 | 305 | list_del_init(&zhdr->buddy); |
9050cce1 | 306 | spin_unlock(&pool->lock); |
d30561c5 VW |
307 | |
308 | WARN_ON(z3fold_page_trylock(zhdr)); | |
309 | __release_z3fold_page(zhdr, true); | |
310 | } | |
311 | ||
312 | static void free_pages_work(struct work_struct *w) | |
313 | { | |
314 | struct z3fold_pool *pool = container_of(w, struct z3fold_pool, work); | |
315 | ||
316 | spin_lock(&pool->stale_lock); | |
317 | while (!list_empty(&pool->stale)) { | |
318 | struct z3fold_header *zhdr = list_first_entry(&pool->stale, | |
319 | struct z3fold_header, buddy); | |
320 | struct page *page = virt_to_page(zhdr); | |
321 | ||
322 | list_del(&zhdr->buddy); | |
323 | if (WARN_ON(!test_bit(PAGE_STALE, &page->private))) | |
324 | continue; | |
d30561c5 VW |
325 | spin_unlock(&pool->stale_lock); |
326 | cancel_work_sync(&zhdr->work); | |
327 | free_z3fold_page(page); | |
328 | cond_resched(); | |
329 | spin_lock(&pool->stale_lock); | |
330 | } | |
331 | spin_unlock(&pool->stale_lock); | |
332 | } | |
333 | ||
9a001fc1 VW |
334 | /* |
335 | * Returns the number of free chunks in a z3fold page. | |
336 | * NB: can't be used with HEADLESS pages. | |
337 | */ | |
338 | static int num_free_chunks(struct z3fold_header *zhdr) | |
339 | { | |
340 | int nfree; | |
341 | /* | |
342 | * If there is a middle object, pick up the bigger free space | |
343 | * either before or after it. Otherwise just subtract the number | |
344 | * of chunks occupied by the first and the last objects. | |
345 | */ | |
346 | if (zhdr->middle_chunks != 0) { | |
347 | int nfree_before = zhdr->first_chunks ? | |
ede93213 | 348 | 0 : zhdr->start_middle - ZHDR_CHUNKS; |
9a001fc1 | 349 | int nfree_after = zhdr->last_chunks ? |
ede93213 VW |
350 | 0 : TOTAL_CHUNKS - |
351 | (zhdr->start_middle + zhdr->middle_chunks); | |
9a001fc1 VW |
352 | nfree = max(nfree_before, nfree_after); |
353 | } else | |
354 | nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks; | |
355 | return nfree; | |
356 | } | |
357 | ||
9050cce1 VW |
358 | /* Add to the appropriate unbuddied list */ |
359 | static inline void add_to_unbuddied(struct z3fold_pool *pool, | |
360 | struct z3fold_header *zhdr) | |
361 | { | |
362 | if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 || | |
363 | zhdr->middle_chunks == 0) { | |
364 | struct list_head *unbuddied = get_cpu_ptr(pool->unbuddied); | |
365 | ||
366 | int freechunks = num_free_chunks(zhdr); | |
367 | spin_lock(&pool->lock); | |
368 | list_add(&zhdr->buddy, &unbuddied[freechunks]); | |
369 | spin_unlock(&pool->lock); | |
370 | zhdr->cpu = smp_processor_id(); | |
371 | put_cpu_ptr(pool->unbuddied); | |
372 | } | |
373 | } | |
374 | ||
ede93213 VW |
375 | static inline void *mchunk_memmove(struct z3fold_header *zhdr, |
376 | unsigned short dst_chunk) | |
377 | { | |
378 | void *beg = zhdr; | |
379 | return memmove(beg + (dst_chunk << CHUNK_SHIFT), | |
380 | beg + (zhdr->start_middle << CHUNK_SHIFT), | |
381 | zhdr->middle_chunks << CHUNK_SHIFT); | |
382 | } | |
383 | ||
1b096e5a | 384 | #define BIG_CHUNK_GAP 3 |
9a001fc1 VW |
385 | /* Has to be called with lock held */ |
386 | static int z3fold_compact_page(struct z3fold_header *zhdr) | |
387 | { | |
388 | struct page *page = virt_to_page(zhdr); | |
9a001fc1 | 389 | |
ede93213 VW |
390 | if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private)) |
391 | return 0; /* can't move middle chunk, it's used */ | |
9a001fc1 | 392 | |
ede93213 VW |
393 | if (zhdr->middle_chunks == 0) |
394 | return 0; /* nothing to compact */ | |
395 | ||
396 | if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) { | |
397 | /* move to the beginning */ | |
398 | mchunk_memmove(zhdr, ZHDR_CHUNKS); | |
9a001fc1 VW |
399 | zhdr->first_chunks = zhdr->middle_chunks; |
400 | zhdr->middle_chunks = 0; | |
401 | zhdr->start_middle = 0; | |
402 | zhdr->first_num++; | |
1b096e5a | 403 | return 1; |
9a001fc1 | 404 | } |
1b096e5a VW |
405 | |
406 | /* | |
407 | * moving data is expensive, so let's only do that if | |
408 | * there's substantial gain (at least BIG_CHUNK_GAP chunks) | |
409 | */ | |
410 | if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 && | |
411 | zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >= | |
412 | BIG_CHUNK_GAP) { | |
413 | mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS); | |
414 | zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS; | |
415 | return 1; | |
416 | } else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 && | |
417 | TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle | |
418 | + zhdr->middle_chunks) >= | |
419 | BIG_CHUNK_GAP) { | |
420 | unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks - | |
421 | zhdr->middle_chunks; | |
422 | mchunk_memmove(zhdr, new_start); | |
423 | zhdr->start_middle = new_start; | |
424 | return 1; | |
425 | } | |
426 | ||
427 | return 0; | |
9a001fc1 VW |
428 | } |
429 | ||
d30561c5 VW |
430 | static void do_compact_page(struct z3fold_header *zhdr, bool locked) |
431 | { | |
9050cce1 | 432 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); |
d30561c5 | 433 | struct page *page; |
d30561c5 VW |
434 | |
435 | page = virt_to_page(zhdr); | |
436 | if (locked) | |
437 | WARN_ON(z3fold_page_trylock(zhdr)); | |
438 | else | |
439 | z3fold_page_lock(zhdr); | |
5d03a661 | 440 | if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) { |
d30561c5 VW |
441 | z3fold_page_unlock(zhdr); |
442 | return; | |
443 | } | |
444 | spin_lock(&pool->lock); | |
445 | list_del_init(&zhdr->buddy); | |
446 | spin_unlock(&pool->lock); | |
447 | ||
5d03a661 VW |
448 | if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) { |
449 | atomic64_dec(&pool->pages_nr); | |
450 | return; | |
451 | } | |
452 | ||
d30561c5 | 453 | z3fold_compact_page(zhdr); |
9050cce1 | 454 | add_to_unbuddied(pool, zhdr); |
d30561c5 VW |
455 | z3fold_page_unlock(zhdr); |
456 | } | |
457 | ||
458 | static void compact_page_work(struct work_struct *w) | |
459 | { | |
460 | struct z3fold_header *zhdr = container_of(w, struct z3fold_header, | |
461 | work); | |
462 | ||
463 | do_compact_page(zhdr, false); | |
464 | } | |
465 | ||
9050cce1 VW |
466 | /* returns _locked_ z3fold page header or NULL */ |
467 | static inline struct z3fold_header *__z3fold_alloc(struct z3fold_pool *pool, | |
468 | size_t size, bool can_sleep) | |
469 | { | |
470 | struct z3fold_header *zhdr = NULL; | |
471 | struct page *page; | |
472 | struct list_head *unbuddied; | |
473 | int chunks = size_to_chunks(size), i; | |
474 | ||
475 | lookup: | |
476 | /* First, try to find an unbuddied z3fold page. */ | |
477 | unbuddied = get_cpu_ptr(pool->unbuddied); | |
478 | for_each_unbuddied_list(i, chunks) { | |
479 | struct list_head *l = &unbuddied[i]; | |
480 | ||
481 | zhdr = list_first_entry_or_null(READ_ONCE(l), | |
482 | struct z3fold_header, buddy); | |
483 | ||
484 | if (!zhdr) | |
485 | continue; | |
486 | ||
487 | /* Re-check under lock. */ | |
488 | spin_lock(&pool->lock); | |
489 | l = &unbuddied[i]; | |
490 | if (unlikely(zhdr != list_first_entry(READ_ONCE(l), | |
491 | struct z3fold_header, buddy)) || | |
492 | !z3fold_page_trylock(zhdr)) { | |
493 | spin_unlock(&pool->lock); | |
494 | zhdr = NULL; | |
495 | put_cpu_ptr(pool->unbuddied); | |
496 | if (can_sleep) | |
497 | cond_resched(); | |
498 | goto lookup; | |
499 | } | |
500 | list_del_init(&zhdr->buddy); | |
501 | zhdr->cpu = -1; | |
502 | spin_unlock(&pool->lock); | |
503 | ||
504 | page = virt_to_page(zhdr); | |
505 | if (test_bit(NEEDS_COMPACTING, &page->private)) { | |
506 | z3fold_page_unlock(zhdr); | |
507 | zhdr = NULL; | |
508 | put_cpu_ptr(pool->unbuddied); | |
509 | if (can_sleep) | |
510 | cond_resched(); | |
511 | goto lookup; | |
512 | } | |
513 | ||
514 | /* | |
515 | * this page could not be removed from its unbuddied | |
516 | * list while pool lock was held, and then we've taken | |
517 | * page lock so kref_put could not be called before | |
518 | * we got here, so it's safe to just call kref_get() | |
519 | */ | |
520 | kref_get(&zhdr->refcount); | |
521 | break; | |
522 | } | |
523 | put_cpu_ptr(pool->unbuddied); | |
524 | ||
525 | return zhdr; | |
526 | } | |
d30561c5 VW |
527 | |
528 | /* | |
529 | * API Functions | |
530 | */ | |
531 | ||
532 | /** | |
533 | * z3fold_create_pool() - create a new z3fold pool | |
534 | * @name: pool name | |
535 | * @gfp: gfp flags when allocating the z3fold pool structure | |
536 | * @ops: user-defined operations for the z3fold pool | |
537 | * | |
538 | * Return: pointer to the new z3fold pool or NULL if the metadata allocation | |
539 | * failed. | |
540 | */ | |
541 | static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp, | |
542 | const struct z3fold_ops *ops) | |
543 | { | |
544 | struct z3fold_pool *pool = NULL; | |
545 | int i, cpu; | |
546 | ||
547 | pool = kzalloc(sizeof(struct z3fold_pool), gfp); | |
548 | if (!pool) | |
549 | goto out; | |
550 | spin_lock_init(&pool->lock); | |
551 | spin_lock_init(&pool->stale_lock); | |
552 | pool->unbuddied = __alloc_percpu(sizeof(struct list_head)*NCHUNKS, 2); | |
1ec6995d XW |
553 | if (!pool->unbuddied) |
554 | goto out_pool; | |
d30561c5 VW |
555 | for_each_possible_cpu(cpu) { |
556 | struct list_head *unbuddied = | |
557 | per_cpu_ptr(pool->unbuddied, cpu); | |
558 | for_each_unbuddied_list(i, 0) | |
559 | INIT_LIST_HEAD(&unbuddied[i]); | |
560 | } | |
561 | INIT_LIST_HEAD(&pool->lru); | |
562 | INIT_LIST_HEAD(&pool->stale); | |
563 | atomic64_set(&pool->pages_nr, 0); | |
564 | pool->name = name; | |
565 | pool->compact_wq = create_singlethread_workqueue(pool->name); | |
566 | if (!pool->compact_wq) | |
1ec6995d | 567 | goto out_unbuddied; |
d30561c5 VW |
568 | pool->release_wq = create_singlethread_workqueue(pool->name); |
569 | if (!pool->release_wq) | |
570 | goto out_wq; | |
571 | INIT_WORK(&pool->work, free_pages_work); | |
572 | pool->ops = ops; | |
573 | return pool; | |
574 | ||
575 | out_wq: | |
576 | destroy_workqueue(pool->compact_wq); | |
1ec6995d XW |
577 | out_unbuddied: |
578 | free_percpu(pool->unbuddied); | |
579 | out_pool: | |
d30561c5 | 580 | kfree(pool); |
1ec6995d | 581 | out: |
d30561c5 VW |
582 | return NULL; |
583 | } | |
584 | ||
585 | /** | |
586 | * z3fold_destroy_pool() - destroys an existing z3fold pool | |
587 | * @pool: the z3fold pool to be destroyed | |
588 | * | |
589 | * The pool should be emptied before this function is called. | |
590 | */ | |
591 | static void z3fold_destroy_pool(struct z3fold_pool *pool) | |
592 | { | |
593 | destroy_workqueue(pool->release_wq); | |
594 | destroy_workqueue(pool->compact_wq); | |
595 | kfree(pool); | |
596 | } | |
597 | ||
9a001fc1 VW |
598 | /** |
599 | * z3fold_alloc() - allocates a region of a given size | |
600 | * @pool: z3fold pool from which to allocate | |
601 | * @size: size in bytes of the desired allocation | |
602 | * @gfp: gfp flags used if the pool needs to grow | |
603 | * @handle: handle of the new allocation | |
604 | * | |
605 | * This function will attempt to find a free region in the pool large enough to | |
606 | * satisfy the allocation request. A search of the unbuddied lists is | |
607 | * performed first. If no suitable free region is found, then a new page is | |
608 | * allocated and added to the pool to satisfy the request. | |
609 | * | |
610 | * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used | |
611 | * as z3fold pool pages. | |
612 | * | |
613 | * Return: 0 if success and handle is set, otherwise -EINVAL if the size or | |
614 | * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate | |
615 | * a new page. | |
616 | */ | |
617 | static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp, | |
618 | unsigned long *handle) | |
619 | { | |
9050cce1 | 620 | int chunks = size_to_chunks(size); |
9a001fc1 | 621 | struct z3fold_header *zhdr = NULL; |
d30561c5 | 622 | struct page *page = NULL; |
9a001fc1 | 623 | enum buddy bud; |
8a97ea54 | 624 | bool can_sleep = gfpflags_allow_blocking(gfp); |
9a001fc1 VW |
625 | |
626 | if (!size || (gfp & __GFP_HIGHMEM)) | |
627 | return -EINVAL; | |
628 | ||
629 | if (size > PAGE_SIZE) | |
630 | return -ENOSPC; | |
631 | ||
632 | if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE) | |
633 | bud = HEADLESS; | |
634 | else { | |
9050cce1 VW |
635 | retry: |
636 | zhdr = __z3fold_alloc(pool, size, can_sleep); | |
d30561c5 | 637 | if (zhdr) { |
2f1e5e4d VW |
638 | if (zhdr->first_chunks == 0) { |
639 | if (zhdr->middle_chunks != 0 && | |
640 | chunks >= zhdr->start_middle) | |
9a001fc1 | 641 | bud = LAST; |
2f1e5e4d VW |
642 | else |
643 | bud = FIRST; | |
644 | } else if (zhdr->last_chunks == 0) | |
645 | bud = LAST; | |
646 | else if (zhdr->middle_chunks == 0) | |
647 | bud = MIDDLE; | |
648 | else { | |
5a27aa82 | 649 | if (kref_put(&zhdr->refcount, |
d30561c5 | 650 | release_z3fold_page_locked)) |
5a27aa82 | 651 | atomic64_dec(&pool->pages_nr); |
d30561c5 VW |
652 | else |
653 | z3fold_page_unlock(zhdr); | |
2f1e5e4d VW |
654 | pr_err("No free chunks in unbuddied\n"); |
655 | WARN_ON(1); | |
9050cce1 | 656 | goto retry; |
9a001fc1 | 657 | } |
9050cce1 | 658 | page = virt_to_page(zhdr); |
2f1e5e4d | 659 | goto found; |
9a001fc1 VW |
660 | } |
661 | bud = FIRST; | |
9a001fc1 VW |
662 | } |
663 | ||
5c9bab59 VW |
664 | page = NULL; |
665 | if (can_sleep) { | |
666 | spin_lock(&pool->stale_lock); | |
667 | zhdr = list_first_entry_or_null(&pool->stale, | |
668 | struct z3fold_header, buddy); | |
669 | /* | |
670 | * Before allocating a page, let's see if we can take one from | |
671 | * the stale pages list. cancel_work_sync() can sleep so we | |
672 | * limit this case to the contexts where we can sleep | |
673 | */ | |
674 | if (zhdr) { | |
675 | list_del(&zhdr->buddy); | |
676 | spin_unlock(&pool->stale_lock); | |
d30561c5 | 677 | cancel_work_sync(&zhdr->work); |
5c9bab59 VW |
678 | page = virt_to_page(zhdr); |
679 | } else { | |
680 | spin_unlock(&pool->stale_lock); | |
681 | } | |
d30561c5 | 682 | } |
5c9bab59 VW |
683 | if (!page) |
684 | page = alloc_page(gfp); | |
d30561c5 | 685 | |
9a001fc1 VW |
686 | if (!page) |
687 | return -ENOMEM; | |
2f1e5e4d | 688 | |
d30561c5 | 689 | zhdr = init_z3fold_page(page, pool); |
9050cce1 VW |
690 | if (!zhdr) { |
691 | __free_page(page); | |
692 | return -ENOMEM; | |
693 | } | |
694 | atomic64_inc(&pool->pages_nr); | |
9a001fc1 VW |
695 | |
696 | if (bud == HEADLESS) { | |
697 | set_bit(PAGE_HEADLESS, &page->private); | |
698 | goto headless; | |
699 | } | |
2f1e5e4d | 700 | z3fold_page_lock(zhdr); |
9a001fc1 VW |
701 | |
702 | found: | |
703 | if (bud == FIRST) | |
704 | zhdr->first_chunks = chunks; | |
705 | else if (bud == LAST) | |
706 | zhdr->last_chunks = chunks; | |
707 | else { | |
708 | zhdr->middle_chunks = chunks; | |
ede93213 | 709 | zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS; |
9a001fc1 | 710 | } |
9050cce1 | 711 | add_to_unbuddied(pool, zhdr); |
9a001fc1 VW |
712 | |
713 | headless: | |
d30561c5 | 714 | spin_lock(&pool->lock); |
9a001fc1 VW |
715 | /* Add/move z3fold page to beginning of LRU */ |
716 | if (!list_empty(&page->lru)) | |
717 | list_del(&page->lru); | |
718 | ||
719 | list_add(&page->lru, &pool->lru); | |
720 | ||
721 | *handle = encode_handle(zhdr, bud); | |
722 | spin_unlock(&pool->lock); | |
2f1e5e4d VW |
723 | if (bud != HEADLESS) |
724 | z3fold_page_unlock(zhdr); | |
9a001fc1 VW |
725 | |
726 | return 0; | |
727 | } | |
728 | ||
729 | /** | |
730 | * z3fold_free() - frees the allocation associated with the given handle | |
731 | * @pool: pool in which the allocation resided | |
732 | * @handle: handle associated with the allocation returned by z3fold_alloc() | |
733 | * | |
734 | * In the case that the z3fold page in which the allocation resides is under | |
735 | * reclaim, as indicated by the PG_reclaim flag being set, this function | |
736 | * only sets the first|last_chunks to 0. The page is actually freed | |
737 | * once both buddies are evicted (see z3fold_reclaim_page() below). | |
738 | */ | |
739 | static void z3fold_free(struct z3fold_pool *pool, unsigned long handle) | |
740 | { | |
741 | struct z3fold_header *zhdr; | |
9a001fc1 VW |
742 | struct page *page; |
743 | enum buddy bud; | |
744 | ||
9a001fc1 VW |
745 | zhdr = handle_to_z3fold_header(handle); |
746 | page = virt_to_page(zhdr); | |
747 | ||
748 | if (test_bit(PAGE_HEADLESS, &page->private)) { | |
ca0246bb VW |
749 | /* if a headless page is under reclaim, just leave. |
750 | * NB: we use test_and_set_bit for a reason: if the bit | |
751 | * has not been set before, we release this page | |
752 | * immediately so we don't care about its value any more. | |
753 | */ | |
754 | if (!test_and_set_bit(PAGE_CLAIMED, &page->private)) { | |
755 | spin_lock(&pool->lock); | |
756 | list_del(&page->lru); | |
757 | spin_unlock(&pool->lock); | |
758 | free_z3fold_page(page); | |
759 | atomic64_dec(&pool->pages_nr); | |
9a001fc1 | 760 | } |
ca0246bb | 761 | return; |
9a001fc1 VW |
762 | } |
763 | ||
ca0246bb VW |
764 | /* Non-headless case */ |
765 | z3fold_page_lock(zhdr); | |
766 | bud = handle_to_buddy(handle); | |
767 | ||
768 | switch (bud) { | |
769 | case FIRST: | |
770 | zhdr->first_chunks = 0; | |
771 | break; | |
772 | case MIDDLE: | |
773 | zhdr->middle_chunks = 0; | |
774 | break; | |
775 | case LAST: | |
776 | zhdr->last_chunks = 0; | |
777 | break; | |
778 | default: | |
779 | pr_err("%s: unknown bud %d\n", __func__, bud); | |
780 | WARN_ON(1); | |
781 | z3fold_page_unlock(zhdr); | |
d30561c5 VW |
782 | return; |
783 | } | |
784 | ||
785 | if (kref_put(&zhdr->refcount, release_z3fold_page_locked_list)) { | |
786 | atomic64_dec(&pool->pages_nr); | |
787 | return; | |
788 | } | |
ca0246bb | 789 | if (test_bit(PAGE_CLAIMED, &page->private)) { |
6098d7e1 VW |
790 | z3fold_page_unlock(zhdr); |
791 | return; | |
792 | } | |
d30561c5 | 793 | if (test_and_set_bit(NEEDS_COMPACTING, &page->private)) { |
5a27aa82 | 794 | z3fold_page_unlock(zhdr); |
d30561c5 VW |
795 | return; |
796 | } | |
797 | if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) { | |
2f1e5e4d | 798 | spin_lock(&pool->lock); |
d30561c5 | 799 | list_del_init(&zhdr->buddy); |
2f1e5e4d | 800 | spin_unlock(&pool->lock); |
d30561c5 | 801 | zhdr->cpu = -1; |
5d03a661 | 802 | kref_get(&zhdr->refcount); |
d30561c5 VW |
803 | do_compact_page(zhdr, true); |
804 | return; | |
9a001fc1 | 805 | } |
5d03a661 | 806 | kref_get(&zhdr->refcount); |
d30561c5 VW |
807 | queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work); |
808 | z3fold_page_unlock(zhdr); | |
9a001fc1 VW |
809 | } |
810 | ||
811 | /** | |
812 | * z3fold_reclaim_page() - evicts allocations from a pool page and frees it | |
813 | * @pool: pool from which a page will attempt to be evicted | |
f144c390 | 814 | * @retries: number of pages on the LRU list for which eviction will |
9a001fc1 VW |
815 | * be attempted before failing |
816 | * | |
817 | * z3fold reclaim is different from normal system reclaim in that it is done | |
818 | * from the bottom, up. This is because only the bottom layer, z3fold, has | |
819 | * information on how the allocations are organized within each z3fold page. | |
820 | * This has the potential to create interesting locking situations between | |
821 | * z3fold and the user, however. | |
822 | * | |
823 | * To avoid these, this is how z3fold_reclaim_page() should be called: | |
f144c390 | 824 | * |
9a001fc1 VW |
825 | * The user detects a page should be reclaimed and calls z3fold_reclaim_page(). |
826 | * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and | |
827 | * call the user-defined eviction handler with the pool and handle as | |
828 | * arguments. | |
829 | * | |
830 | * If the handle can not be evicted, the eviction handler should return | |
831 | * non-zero. z3fold_reclaim_page() will add the z3fold page back to the | |
832 | * appropriate list and try the next z3fold page on the LRU up to | |
833 | * a user defined number of retries. | |
834 | * | |
835 | * If the handle is successfully evicted, the eviction handler should | |
836 | * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free() | |
837 | * contains logic to delay freeing the page if the page is under reclaim, | |
838 | * as indicated by the setting of the PG_reclaim flag on the underlying page. | |
839 | * | |
840 | * If all buddies in the z3fold page are successfully evicted, then the | |
841 | * z3fold page can be freed. | |
842 | * | |
843 | * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are | |
844 | * no pages to evict or an eviction handler is not registered, -EAGAIN if | |
845 | * the retry limit was hit. | |
846 | */ | |
847 | static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries) | |
848 | { | |
d30561c5 VW |
849 | int i, ret = 0; |
850 | struct z3fold_header *zhdr = NULL; | |
851 | struct page *page = NULL; | |
852 | struct list_head *pos; | |
9a001fc1 VW |
853 | unsigned long first_handle = 0, middle_handle = 0, last_handle = 0; |
854 | ||
855 | spin_lock(&pool->lock); | |
2f1e5e4d | 856 | if (!pool->ops || !pool->ops->evict || retries == 0) { |
9a001fc1 VW |
857 | spin_unlock(&pool->lock); |
858 | return -EINVAL; | |
859 | } | |
860 | for (i = 0; i < retries; i++) { | |
2f1e5e4d VW |
861 | if (list_empty(&pool->lru)) { |
862 | spin_unlock(&pool->lock); | |
863 | return -EINVAL; | |
864 | } | |
d30561c5 VW |
865 | list_for_each_prev(pos, &pool->lru) { |
866 | page = list_entry(pos, struct page, lru); | |
ca0246bb VW |
867 | |
868 | /* this bit could have been set by free, in which case | |
869 | * we pass over to the next page in the pool. | |
870 | */ | |
871 | if (test_and_set_bit(PAGE_CLAIMED, &page->private)) | |
872 | continue; | |
873 | ||
874 | zhdr = page_address(page); | |
d30561c5 | 875 | if (test_bit(PAGE_HEADLESS, &page->private)) |
d30561c5 VW |
876 | break; |
877 | ||
ca0246bb VW |
878 | if (!z3fold_page_trylock(zhdr)) { |
879 | zhdr = NULL; | |
d30561c5 | 880 | continue; /* can't evict at this point */ |
ca0246bb | 881 | } |
d30561c5 VW |
882 | kref_get(&zhdr->refcount); |
883 | list_del_init(&zhdr->buddy); | |
884 | zhdr->cpu = -1; | |
6098d7e1 | 885 | break; |
d30561c5 VW |
886 | } |
887 | ||
ca0246bb VW |
888 | if (!zhdr) |
889 | break; | |
890 | ||
5a27aa82 | 891 | list_del_init(&page->lru); |
d30561c5 | 892 | spin_unlock(&pool->lock); |
9a001fc1 | 893 | |
9a001fc1 | 894 | if (!test_bit(PAGE_HEADLESS, &page->private)) { |
9a001fc1 VW |
895 | /* |
896 | * We need encode the handles before unlocking, since | |
897 | * we can race with free that will set | |
898 | * (first|last)_chunks to 0 | |
899 | */ | |
900 | first_handle = 0; | |
901 | last_handle = 0; | |
902 | middle_handle = 0; | |
903 | if (zhdr->first_chunks) | |
904 | first_handle = encode_handle(zhdr, FIRST); | |
905 | if (zhdr->middle_chunks) | |
906 | middle_handle = encode_handle(zhdr, MIDDLE); | |
907 | if (zhdr->last_chunks) | |
908 | last_handle = encode_handle(zhdr, LAST); | |
d30561c5 VW |
909 | /* |
910 | * it's safe to unlock here because we hold a | |
911 | * reference to this page | |
912 | */ | |
2f1e5e4d | 913 | z3fold_page_unlock(zhdr); |
9a001fc1 VW |
914 | } else { |
915 | first_handle = encode_handle(zhdr, HEADLESS); | |
916 | last_handle = middle_handle = 0; | |
917 | } | |
918 | ||
9a001fc1 VW |
919 | /* Issue the eviction callback(s) */ |
920 | if (middle_handle) { | |
921 | ret = pool->ops->evict(pool, middle_handle); | |
922 | if (ret) | |
923 | goto next; | |
924 | } | |
925 | if (first_handle) { | |
926 | ret = pool->ops->evict(pool, first_handle); | |
927 | if (ret) | |
928 | goto next; | |
929 | } | |
930 | if (last_handle) { | |
931 | ret = pool->ops->evict(pool, last_handle); | |
932 | if (ret) | |
933 | goto next; | |
934 | } | |
935 | next: | |
5a27aa82 VW |
936 | if (test_bit(PAGE_HEADLESS, &page->private)) { |
937 | if (ret == 0) { | |
938 | free_z3fold_page(page); | |
ca0246bb | 939 | atomic64_dec(&pool->pages_nr); |
5a27aa82 | 940 | return 0; |
5a27aa82 | 941 | } |
6098d7e1 VW |
942 | spin_lock(&pool->lock); |
943 | list_add(&page->lru, &pool->lru); | |
944 | spin_unlock(&pool->lock); | |
945 | } else { | |
946 | z3fold_page_lock(zhdr); | |
ca0246bb | 947 | clear_bit(PAGE_CLAIMED, &page->private); |
6098d7e1 VW |
948 | if (kref_put(&zhdr->refcount, |
949 | release_z3fold_page_locked)) { | |
950 | atomic64_dec(&pool->pages_nr); | |
951 | return 0; | |
952 | } | |
953 | /* | |
954 | * if we are here, the page is still not completely | |
955 | * free. Take the global pool lock then to be able | |
956 | * to add it back to the lru list | |
957 | */ | |
958 | spin_lock(&pool->lock); | |
959 | list_add(&page->lru, &pool->lru); | |
d5567c9d | 960 | spin_unlock(&pool->lock); |
6098d7e1 | 961 | z3fold_page_unlock(zhdr); |
5a27aa82 | 962 | } |
2f1e5e4d | 963 | |
6098d7e1 VW |
964 | /* We started off locked to we need to lock the pool back */ |
965 | spin_lock(&pool->lock); | |
9a001fc1 VW |
966 | } |
967 | spin_unlock(&pool->lock); | |
968 | return -EAGAIN; | |
969 | } | |
970 | ||
971 | /** | |
972 | * z3fold_map() - maps the allocation associated with the given handle | |
973 | * @pool: pool in which the allocation resides | |
974 | * @handle: handle associated with the allocation to be mapped | |
975 | * | |
976 | * Extracts the buddy number from handle and constructs the pointer to the | |
977 | * correct starting chunk within the page. | |
978 | * | |
979 | * Returns: a pointer to the mapped allocation | |
980 | */ | |
981 | static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle) | |
982 | { | |
983 | struct z3fold_header *zhdr; | |
984 | struct page *page; | |
985 | void *addr; | |
986 | enum buddy buddy; | |
987 | ||
9a001fc1 VW |
988 | zhdr = handle_to_z3fold_header(handle); |
989 | addr = zhdr; | |
990 | page = virt_to_page(zhdr); | |
991 | ||
992 | if (test_bit(PAGE_HEADLESS, &page->private)) | |
993 | goto out; | |
994 | ||
2f1e5e4d | 995 | z3fold_page_lock(zhdr); |
9a001fc1 VW |
996 | buddy = handle_to_buddy(handle); |
997 | switch (buddy) { | |
998 | case FIRST: | |
999 | addr += ZHDR_SIZE_ALIGNED; | |
1000 | break; | |
1001 | case MIDDLE: | |
1002 | addr += zhdr->start_middle << CHUNK_SHIFT; | |
1003 | set_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
1004 | break; | |
1005 | case LAST: | |
ca0246bb | 1006 | addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT); |
9a001fc1 VW |
1007 | break; |
1008 | default: | |
1009 | pr_err("unknown buddy id %d\n", buddy); | |
1010 | WARN_ON(1); | |
1011 | addr = NULL; | |
1012 | break; | |
1013 | } | |
2f1e5e4d VW |
1014 | |
1015 | z3fold_page_unlock(zhdr); | |
9a001fc1 | 1016 | out: |
9a001fc1 VW |
1017 | return addr; |
1018 | } | |
1019 | ||
1020 | /** | |
1021 | * z3fold_unmap() - unmaps the allocation associated with the given handle | |
1022 | * @pool: pool in which the allocation resides | |
1023 | * @handle: handle associated with the allocation to be unmapped | |
1024 | */ | |
1025 | static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle) | |
1026 | { | |
1027 | struct z3fold_header *zhdr; | |
1028 | struct page *page; | |
1029 | enum buddy buddy; | |
1030 | ||
9a001fc1 VW |
1031 | zhdr = handle_to_z3fold_header(handle); |
1032 | page = virt_to_page(zhdr); | |
1033 | ||
2f1e5e4d | 1034 | if (test_bit(PAGE_HEADLESS, &page->private)) |
9a001fc1 | 1035 | return; |
9a001fc1 | 1036 | |
2f1e5e4d | 1037 | z3fold_page_lock(zhdr); |
9a001fc1 VW |
1038 | buddy = handle_to_buddy(handle); |
1039 | if (buddy == MIDDLE) | |
1040 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
2f1e5e4d | 1041 | z3fold_page_unlock(zhdr); |
9a001fc1 VW |
1042 | } |
1043 | ||
1044 | /** | |
1045 | * z3fold_get_pool_size() - gets the z3fold pool size in pages | |
1046 | * @pool: pool whose size is being queried | |
1047 | * | |
12d59ae6 | 1048 | * Returns: size in pages of the given pool. |
9a001fc1 VW |
1049 | */ |
1050 | static u64 z3fold_get_pool_size(struct z3fold_pool *pool) | |
1051 | { | |
12d59ae6 | 1052 | return atomic64_read(&pool->pages_nr); |
9a001fc1 VW |
1053 | } |
1054 | ||
1055 | /***************** | |
1056 | * zpool | |
1057 | ****************/ | |
1058 | ||
1059 | static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle) | |
1060 | { | |
1061 | if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict) | |
1062 | return pool->zpool_ops->evict(pool->zpool, handle); | |
1063 | else | |
1064 | return -ENOENT; | |
1065 | } | |
1066 | ||
1067 | static const struct z3fold_ops z3fold_zpool_ops = { | |
1068 | .evict = z3fold_zpool_evict | |
1069 | }; | |
1070 | ||
1071 | static void *z3fold_zpool_create(const char *name, gfp_t gfp, | |
1072 | const struct zpool_ops *zpool_ops, | |
1073 | struct zpool *zpool) | |
1074 | { | |
1075 | struct z3fold_pool *pool; | |
1076 | ||
d30561c5 VW |
1077 | pool = z3fold_create_pool(name, gfp, |
1078 | zpool_ops ? &z3fold_zpool_ops : NULL); | |
9a001fc1 VW |
1079 | if (pool) { |
1080 | pool->zpool = zpool; | |
1081 | pool->zpool_ops = zpool_ops; | |
1082 | } | |
1083 | return pool; | |
1084 | } | |
1085 | ||
1086 | static void z3fold_zpool_destroy(void *pool) | |
1087 | { | |
1088 | z3fold_destroy_pool(pool); | |
1089 | } | |
1090 | ||
1091 | static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp, | |
1092 | unsigned long *handle) | |
1093 | { | |
1094 | return z3fold_alloc(pool, size, gfp, handle); | |
1095 | } | |
1096 | static void z3fold_zpool_free(void *pool, unsigned long handle) | |
1097 | { | |
1098 | z3fold_free(pool, handle); | |
1099 | } | |
1100 | ||
1101 | static int z3fold_zpool_shrink(void *pool, unsigned int pages, | |
1102 | unsigned int *reclaimed) | |
1103 | { | |
1104 | unsigned int total = 0; | |
1105 | int ret = -EINVAL; | |
1106 | ||
1107 | while (total < pages) { | |
1108 | ret = z3fold_reclaim_page(pool, 8); | |
1109 | if (ret < 0) | |
1110 | break; | |
1111 | total++; | |
1112 | } | |
1113 | ||
1114 | if (reclaimed) | |
1115 | *reclaimed = total; | |
1116 | ||
1117 | return ret; | |
1118 | } | |
1119 | ||
1120 | static void *z3fold_zpool_map(void *pool, unsigned long handle, | |
1121 | enum zpool_mapmode mm) | |
1122 | { | |
1123 | return z3fold_map(pool, handle); | |
1124 | } | |
1125 | static void z3fold_zpool_unmap(void *pool, unsigned long handle) | |
1126 | { | |
1127 | z3fold_unmap(pool, handle); | |
1128 | } | |
1129 | ||
1130 | static u64 z3fold_zpool_total_size(void *pool) | |
1131 | { | |
1132 | return z3fold_get_pool_size(pool) * PAGE_SIZE; | |
1133 | } | |
1134 | ||
1135 | static struct zpool_driver z3fold_zpool_driver = { | |
1136 | .type = "z3fold", | |
1137 | .owner = THIS_MODULE, | |
1138 | .create = z3fold_zpool_create, | |
1139 | .destroy = z3fold_zpool_destroy, | |
1140 | .malloc = z3fold_zpool_malloc, | |
1141 | .free = z3fold_zpool_free, | |
1142 | .shrink = z3fold_zpool_shrink, | |
1143 | .map = z3fold_zpool_map, | |
1144 | .unmap = z3fold_zpool_unmap, | |
1145 | .total_size = z3fold_zpool_total_size, | |
1146 | }; | |
1147 | ||
1148 | MODULE_ALIAS("zpool-z3fold"); | |
1149 | ||
1150 | static int __init init_z3fold(void) | |
1151 | { | |
ede93213 VW |
1152 | /* Make sure the z3fold header is not larger than the page size */ |
1153 | BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE); | |
9a001fc1 VW |
1154 | zpool_register_driver(&z3fold_zpool_driver); |
1155 | ||
1156 | return 0; | |
1157 | } | |
1158 | ||
1159 | static void __exit exit_z3fold(void) | |
1160 | { | |
1161 | zpool_unregister_driver(&z3fold_zpool_driver); | |
1162 | } | |
1163 | ||
1164 | module_init(init_z3fold); | |
1165 | module_exit(exit_z3fold); | |
1166 | ||
1167 | MODULE_LICENSE("GPL"); | |
1168 | MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>"); | |
1169 | MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages"); |