Commit | Line | Data |
---|---|---|
09c434b8 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
9a001fc1 VW |
2 | /* |
3 | * z3fold.c | |
4 | * | |
5 | * Author: Vitaly Wool <vitaly.wool@konsulko.com> | |
6 | * Copyright (C) 2016, Sony Mobile Communications Inc. | |
7 | * | |
8 | * This implementation is based on zbud written by Seth Jennings. | |
9 | * | |
10 | * z3fold is an special purpose allocator for storing compressed pages. It | |
11 | * can store up to three compressed pages per page which improves the | |
12 | * compression ratio of zbud while retaining its main concepts (e. g. always | |
13 | * storing an integral number of objects per page) and simplicity. | |
14 | * It still has simple and deterministic reclaim properties that make it | |
15 | * preferable to a higher density approach (with no requirement on integral | |
16 | * number of object per page) when reclaim is used. | |
17 | * | |
18 | * As in zbud, pages are divided into "chunks". The size of the chunks is | |
19 | * fixed at compile time and is determined by NCHUNKS_ORDER below. | |
20 | * | |
21 | * z3fold doesn't export any API and is meant to be used via zpool API. | |
22 | */ | |
23 | ||
24 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
25 | ||
26 | #include <linux/atomic.h> | |
d30561c5 | 27 | #include <linux/sched.h> |
1f862989 | 28 | #include <linux/cpumask.h> |
9a001fc1 VW |
29 | #include <linux/list.h> |
30 | #include <linux/mm.h> | |
31 | #include <linux/module.h> | |
1f862989 VW |
32 | #include <linux/page-flags.h> |
33 | #include <linux/migrate.h> | |
34 | #include <linux/node.h> | |
35 | #include <linux/compaction.h> | |
d30561c5 | 36 | #include <linux/percpu.h> |
1f862989 | 37 | #include <linux/mount.h> |
ea8157ab | 38 | #include <linux/pseudo_fs.h> |
1f862989 | 39 | #include <linux/fs.h> |
9a001fc1 | 40 | #include <linux/preempt.h> |
d30561c5 | 41 | #include <linux/workqueue.h> |
9a001fc1 VW |
42 | #include <linux/slab.h> |
43 | #include <linux/spinlock.h> | |
44 | #include <linux/zpool.h> | |
ea8157ab | 45 | #include <linux/magic.h> |
af4798a5 | 46 | #include <linux/kmemleak.h> |
9a001fc1 | 47 | |
7c2b8baa VW |
48 | /* |
49 | * NCHUNKS_ORDER determines the internal allocation granularity, effectively | |
50 | * adjusting internal fragmentation. It also determines the number of | |
51 | * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the | |
52 | * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks | |
53 | * in the beginning of an allocated page are occupied by z3fold header, so | |
54 | * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y), | |
55 | * which shows the max number of free chunks in z3fold page, also there will | |
56 | * be 63, or 62, respectively, freelists per pool. | |
57 | */ | |
58 | #define NCHUNKS_ORDER 6 | |
59 | ||
60 | #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER) | |
61 | #define CHUNK_SIZE (1 << CHUNK_SHIFT) | |
62 | #define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE) | |
63 | #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT) | |
64 | #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SHIFT) | |
e3c0db4f | 65 | #define NCHUNKS (TOTAL_CHUNKS - ZHDR_CHUNKS) |
7c2b8baa VW |
66 | |
67 | #define BUDDY_MASK (0x3) | |
68 | #define BUDDY_SHIFT 2 | |
69 | #define SLOTS_ALIGN (0x40) | |
70 | ||
9a001fc1 VW |
71 | /***************** |
72 | * Structures | |
73 | *****************/ | |
ede93213 VW |
74 | struct z3fold_pool; |
75 | struct z3fold_ops { | |
76 | int (*evict)(struct z3fold_pool *pool, unsigned long handle); | |
77 | }; | |
78 | ||
79 | enum buddy { | |
80 | HEADLESS = 0, | |
81 | FIRST, | |
82 | MIDDLE, | |
83 | LAST, | |
7c2b8baa VW |
84 | BUDDIES_MAX = LAST |
85 | }; | |
86 | ||
87 | struct z3fold_buddy_slots { | |
88 | /* | |
89 | * we are using BUDDY_MASK in handle_to_buddy etc. so there should | |
90 | * be enough slots to hold all possible variants | |
91 | */ | |
92 | unsigned long slot[BUDDY_MASK + 1]; | |
fc548865 | 93 | unsigned long pool; /* back link */ |
4a3ac931 | 94 | rwlock_t lock; |
ede93213 | 95 | }; |
7c2b8baa | 96 | #define HANDLE_FLAG_MASK (0x03) |
ede93213 VW |
97 | |
98 | /* | |
d30561c5 | 99 | * struct z3fold_header - z3fold page metadata occupying first chunks of each |
ede93213 | 100 | * z3fold page, except for HEADLESS pages |
d30561c5 VW |
101 | * @buddy: links the z3fold page into the relevant list in the |
102 | * pool | |
2f1e5e4d | 103 | * @page_lock: per-page lock |
d30561c5 VW |
104 | * @refcount: reference count for the z3fold page |
105 | * @work: work_struct for page layout optimization | |
7c2b8baa | 106 | * @slots: pointer to the structure holding buddy slots |
bb9a374d | 107 | * @pool: pointer to the containing pool |
d30561c5 | 108 | * @cpu: CPU which this page "belongs" to |
ede93213 VW |
109 | * @first_chunks: the size of the first buddy in chunks, 0 if free |
110 | * @middle_chunks: the size of the middle buddy in chunks, 0 if free | |
111 | * @last_chunks: the size of the last buddy in chunks, 0 if free | |
112 | * @first_num: the starting number (for the first handle) | |
1f862989 | 113 | * @mapped_count: the number of objects currently mapped |
ede93213 VW |
114 | */ |
115 | struct z3fold_header { | |
116 | struct list_head buddy; | |
2f1e5e4d | 117 | spinlock_t page_lock; |
5a27aa82 | 118 | struct kref refcount; |
d30561c5 | 119 | struct work_struct work; |
7c2b8baa | 120 | struct z3fold_buddy_slots *slots; |
bb9a374d | 121 | struct z3fold_pool *pool; |
d30561c5 | 122 | short cpu; |
ede93213 VW |
123 | unsigned short first_chunks; |
124 | unsigned short middle_chunks; | |
125 | unsigned short last_chunks; | |
126 | unsigned short start_middle; | |
127 | unsigned short first_num:2; | |
1f862989 | 128 | unsigned short mapped_count:2; |
4a3ac931 | 129 | unsigned short foreign_handles:2; |
ede93213 VW |
130 | }; |
131 | ||
9a001fc1 VW |
132 | /** |
133 | * struct z3fold_pool - stores metadata for each z3fold pool | |
d30561c5 VW |
134 | * @name: pool name |
135 | * @lock: protects pool unbuddied/lru lists | |
136 | * @stale_lock: protects pool stale page list | |
137 | * @unbuddied: per-cpu array of lists tracking z3fold pages that contain 2- | |
138 | * buddies; the list each z3fold page is added to depends on | |
139 | * the size of its free region. | |
9a001fc1 VW |
140 | * @lru: list tracking the z3fold pages in LRU order by most recently |
141 | * added buddy. | |
d30561c5 | 142 | * @stale: list of pages marked for freeing |
9a001fc1 | 143 | * @pages_nr: number of z3fold pages in the pool. |
7c2b8baa | 144 | * @c_handle: cache for z3fold_buddy_slots allocation |
9a001fc1 VW |
145 | * @ops: pointer to a structure of user defined operations specified at |
146 | * pool creation time. | |
30522175 MG |
147 | * @zpool: zpool driver |
148 | * @zpool_ops: zpool operations structure with an evict callback | |
d30561c5 VW |
149 | * @compact_wq: workqueue for page layout background optimization |
150 | * @release_wq: workqueue for safe page release | |
151 | * @work: work_struct for safe page release | |
1f862989 | 152 | * @inode: inode for z3fold pseudo filesystem |
9a001fc1 VW |
153 | * |
154 | * This structure is allocated at pool creation time and maintains metadata | |
155 | * pertaining to a particular z3fold pool. | |
156 | */ | |
157 | struct z3fold_pool { | |
d30561c5 | 158 | const char *name; |
9a001fc1 | 159 | spinlock_t lock; |
d30561c5 VW |
160 | spinlock_t stale_lock; |
161 | struct list_head *unbuddied; | |
9a001fc1 | 162 | struct list_head lru; |
d30561c5 | 163 | struct list_head stale; |
12d59ae6 | 164 | atomic64_t pages_nr; |
7c2b8baa | 165 | struct kmem_cache *c_handle; |
9a001fc1 VW |
166 | const struct z3fold_ops *ops; |
167 | struct zpool *zpool; | |
168 | const struct zpool_ops *zpool_ops; | |
d30561c5 VW |
169 | struct workqueue_struct *compact_wq; |
170 | struct workqueue_struct *release_wq; | |
171 | struct work_struct work; | |
1f862989 | 172 | struct inode *inode; |
9a001fc1 VW |
173 | }; |
174 | ||
9a001fc1 VW |
175 | /* |
176 | * Internal z3fold page flags | |
177 | */ | |
178 | enum z3fold_page_flags { | |
5a27aa82 | 179 | PAGE_HEADLESS = 0, |
9a001fc1 | 180 | MIDDLE_CHUNK_MAPPED, |
d30561c5 | 181 | NEEDS_COMPACTING, |
6098d7e1 | 182 | PAGE_STALE, |
ca0246bb | 183 | PAGE_CLAIMED, /* by either reclaim or free */ |
9a001fc1 VW |
184 | }; |
185 | ||
dcf5aedb VW |
186 | /* |
187 | * handle flags, go under HANDLE_FLAG_MASK | |
188 | */ | |
189 | enum z3fold_handle_flags { | |
190 | HANDLES_NOFREE = 0, | |
191 | }; | |
192 | ||
4a3ac931 VW |
193 | /* |
194 | * Forward declarations | |
195 | */ | |
196 | static struct z3fold_header *__z3fold_alloc(struct z3fold_pool *, size_t, bool); | |
197 | static void compact_page_work(struct work_struct *w); | |
198 | ||
9a001fc1 VW |
199 | /***************** |
200 | * Helpers | |
201 | *****************/ | |
202 | ||
203 | /* Converts an allocation size in bytes to size in z3fold chunks */ | |
204 | static int size_to_chunks(size_t size) | |
205 | { | |
206 | return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT; | |
207 | } | |
208 | ||
209 | #define for_each_unbuddied_list(_iter, _begin) \ | |
210 | for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++) | |
211 | ||
bb9f6f63 VW |
212 | static inline struct z3fold_buddy_slots *alloc_slots(struct z3fold_pool *pool, |
213 | gfp_t gfp) | |
7c2b8baa | 214 | { |
f1549cb5 HB |
215 | struct z3fold_buddy_slots *slots; |
216 | ||
f94afee9 | 217 | slots = kmem_cache_zalloc(pool->c_handle, |
f1549cb5 | 218 | (gfp & ~(__GFP_HIGHMEM | __GFP_MOVABLE))); |
7c2b8baa VW |
219 | |
220 | if (slots) { | |
af4798a5 QC |
221 | /* It will be freed separately in free_handle(). */ |
222 | kmemleak_not_leak(slots); | |
7c2b8baa | 223 | slots->pool = (unsigned long)pool; |
4a3ac931 | 224 | rwlock_init(&slots->lock); |
7c2b8baa VW |
225 | } |
226 | ||
227 | return slots; | |
228 | } | |
229 | ||
230 | static inline struct z3fold_pool *slots_to_pool(struct z3fold_buddy_slots *s) | |
231 | { | |
232 | return (struct z3fold_pool *)(s->pool & ~HANDLE_FLAG_MASK); | |
233 | } | |
234 | ||
235 | static inline struct z3fold_buddy_slots *handle_to_slots(unsigned long handle) | |
236 | { | |
237 | return (struct z3fold_buddy_slots *)(handle & ~(SLOTS_ALIGN - 1)); | |
238 | } | |
239 | ||
4a3ac931 VW |
240 | /* Lock a z3fold page */ |
241 | static inline void z3fold_page_lock(struct z3fold_header *zhdr) | |
242 | { | |
243 | spin_lock(&zhdr->page_lock); | |
244 | } | |
245 | ||
246 | /* Try to lock a z3fold page */ | |
247 | static inline int z3fold_page_trylock(struct z3fold_header *zhdr) | |
248 | { | |
249 | return spin_trylock(&zhdr->page_lock); | |
250 | } | |
251 | ||
252 | /* Unlock a z3fold page */ | |
253 | static inline void z3fold_page_unlock(struct z3fold_header *zhdr) | |
254 | { | |
255 | spin_unlock(&zhdr->page_lock); | |
256 | } | |
257 | ||
767cc6c5 ML |
258 | /* return locked z3fold page if it's not headless */ |
259 | static inline struct z3fold_header *get_z3fold_header(unsigned long handle) | |
4a3ac931 VW |
260 | { |
261 | struct z3fold_buddy_slots *slots; | |
262 | struct z3fold_header *zhdr; | |
263 | int locked = 0; | |
264 | ||
265 | if (!(handle & (1 << PAGE_HEADLESS))) { | |
266 | slots = handle_to_slots(handle); | |
267 | do { | |
268 | unsigned long addr; | |
269 | ||
270 | read_lock(&slots->lock); | |
271 | addr = *(unsigned long *)handle; | |
272 | zhdr = (struct z3fold_header *)(addr & PAGE_MASK); | |
767cc6c5 | 273 | locked = z3fold_page_trylock(zhdr); |
4a3ac931 VW |
274 | read_unlock(&slots->lock); |
275 | if (locked) | |
276 | break; | |
277 | cpu_relax(); | |
767cc6c5 | 278 | } while (true); |
4a3ac931 VW |
279 | } else { |
280 | zhdr = (struct z3fold_header *)(handle & PAGE_MASK); | |
281 | } | |
282 | ||
283 | return zhdr; | |
284 | } | |
285 | ||
4a3ac931 VW |
286 | static inline void put_z3fold_header(struct z3fold_header *zhdr) |
287 | { | |
288 | struct page *page = virt_to_page(zhdr); | |
289 | ||
290 | if (!test_bit(PAGE_HEADLESS, &page->private)) | |
291 | z3fold_page_unlock(zhdr); | |
292 | } | |
293 | ||
fc548865 | 294 | static inline void free_handle(unsigned long handle, struct z3fold_header *zhdr) |
7c2b8baa VW |
295 | { |
296 | struct z3fold_buddy_slots *slots; | |
297 | int i; | |
298 | bool is_free; | |
299 | ||
4a3ac931 VW |
300 | if (WARN_ON(*(unsigned long *)handle == 0)) |
301 | return; | |
302 | ||
7c2b8baa | 303 | slots = handle_to_slots(handle); |
4a3ac931 VW |
304 | write_lock(&slots->lock); |
305 | *(unsigned long *)handle = 0; | |
dcf5aedb VW |
306 | |
307 | if (test_bit(HANDLES_NOFREE, &slots->pool)) { | |
308 | write_unlock(&slots->lock); | |
309 | return; /* simple case, nothing else to do */ | |
310 | } | |
311 | ||
fc548865 VW |
312 | if (zhdr->slots != slots) |
313 | zhdr->foreign_handles--; | |
4a3ac931 | 314 | |
7c2b8baa VW |
315 | is_free = true; |
316 | for (i = 0; i <= BUDDY_MASK; i++) { | |
317 | if (slots->slot[i]) { | |
318 | is_free = false; | |
319 | break; | |
320 | } | |
321 | } | |
d8f117ab | 322 | write_unlock(&slots->lock); |
7c2b8baa VW |
323 | |
324 | if (is_free) { | |
325 | struct z3fold_pool *pool = slots_to_pool(slots); | |
326 | ||
fc548865 VW |
327 | if (zhdr->slots == slots) |
328 | zhdr->slots = NULL; | |
7c2b8baa VW |
329 | kmem_cache_free(pool->c_handle, slots); |
330 | } | |
331 | } | |
332 | ||
ea8157ab | 333 | static int z3fold_init_fs_context(struct fs_context *fc) |
1f862989 | 334 | { |
ea8157ab | 335 | return init_pseudo(fc, Z3FOLD_MAGIC) ? 0 : -ENOMEM; |
1f862989 VW |
336 | } |
337 | ||
338 | static struct file_system_type z3fold_fs = { | |
339 | .name = "z3fold", | |
ea8157ab | 340 | .init_fs_context = z3fold_init_fs_context, |
1f862989 VW |
341 | .kill_sb = kill_anon_super, |
342 | }; | |
343 | ||
344 | static struct vfsmount *z3fold_mnt; | |
dc3a1f30 | 345 | static int __init z3fold_mount(void) |
1f862989 VW |
346 | { |
347 | int ret = 0; | |
348 | ||
349 | z3fold_mnt = kern_mount(&z3fold_fs); | |
350 | if (IS_ERR(z3fold_mnt)) | |
351 | ret = PTR_ERR(z3fold_mnt); | |
352 | ||
353 | return ret; | |
354 | } | |
355 | ||
356 | static void z3fold_unmount(void) | |
357 | { | |
358 | kern_unmount(z3fold_mnt); | |
359 | } | |
360 | ||
361 | static const struct address_space_operations z3fold_aops; | |
362 | static int z3fold_register_migration(struct z3fold_pool *pool) | |
363 | { | |
364 | pool->inode = alloc_anon_inode(z3fold_mnt->mnt_sb); | |
365 | if (IS_ERR(pool->inode)) { | |
366 | pool->inode = NULL; | |
367 | return 1; | |
368 | } | |
369 | ||
370 | pool->inode->i_mapping->private_data = pool; | |
371 | pool->inode->i_mapping->a_ops = &z3fold_aops; | |
372 | return 0; | |
373 | } | |
374 | ||
375 | static void z3fold_unregister_migration(struct z3fold_pool *pool) | |
376 | { | |
377 | if (pool->inode) | |
378 | iput(pool->inode); | |
cb152a1a | 379 | } |
1f862989 | 380 | |
9a001fc1 | 381 | /* Initializes the z3fold header of a newly allocated z3fold page */ |
63398413 | 382 | static struct z3fold_header *init_z3fold_page(struct page *page, bool headless, |
bb9f6f63 | 383 | struct z3fold_pool *pool, gfp_t gfp) |
9a001fc1 VW |
384 | { |
385 | struct z3fold_header *zhdr = page_address(page); | |
63398413 | 386 | struct z3fold_buddy_slots *slots; |
9a001fc1 VW |
387 | |
388 | INIT_LIST_HEAD(&page->lru); | |
9a001fc1 VW |
389 | clear_bit(PAGE_HEADLESS, &page->private); |
390 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
d30561c5 VW |
391 | clear_bit(NEEDS_COMPACTING, &page->private); |
392 | clear_bit(PAGE_STALE, &page->private); | |
ca0246bb | 393 | clear_bit(PAGE_CLAIMED, &page->private); |
63398413 VW |
394 | if (headless) |
395 | return zhdr; | |
396 | ||
397 | slots = alloc_slots(pool, gfp); | |
398 | if (!slots) | |
399 | return NULL; | |
9a001fc1 | 400 | |
c457cd96 | 401 | memset(zhdr, 0, sizeof(*zhdr)); |
2f1e5e4d | 402 | spin_lock_init(&zhdr->page_lock); |
5a27aa82 | 403 | kref_init(&zhdr->refcount); |
d30561c5 | 404 | zhdr->cpu = -1; |
7c2b8baa | 405 | zhdr->slots = slots; |
bb9a374d | 406 | zhdr->pool = pool; |
9a001fc1 | 407 | INIT_LIST_HEAD(&zhdr->buddy); |
d30561c5 | 408 | INIT_WORK(&zhdr->work, compact_page_work); |
9a001fc1 VW |
409 | return zhdr; |
410 | } | |
411 | ||
412 | /* Resets the struct page fields and frees the page */ | |
1f862989 | 413 | static void free_z3fold_page(struct page *page, bool headless) |
9a001fc1 | 414 | { |
1f862989 VW |
415 | if (!headless) { |
416 | lock_page(page); | |
417 | __ClearPageMovable(page); | |
418 | unlock_page(page); | |
419 | } | |
5a27aa82 VW |
420 | __free_page(page); |
421 | } | |
422 | ||
7c2b8baa VW |
423 | /* Helper function to build the index */ |
424 | static inline int __idx(struct z3fold_header *zhdr, enum buddy bud) | |
425 | { | |
426 | return (bud + zhdr->first_num) & BUDDY_MASK; | |
427 | } | |
428 | ||
9a001fc1 VW |
429 | /* |
430 | * Encodes the handle of a particular buddy within a z3fold page | |
431 | * Pool lock should be held as this function accesses first_num | |
432 | */ | |
3f9d2b57 VW |
433 | static unsigned long __encode_handle(struct z3fold_header *zhdr, |
434 | struct z3fold_buddy_slots *slots, | |
435 | enum buddy bud) | |
9a001fc1 | 436 | { |
7c2b8baa VW |
437 | unsigned long h = (unsigned long)zhdr; |
438 | int idx = 0; | |
9a001fc1 | 439 | |
7c2b8baa VW |
440 | /* |
441 | * For a headless page, its handle is its pointer with the extra | |
442 | * PAGE_HEADLESS bit set | |
443 | */ | |
444 | if (bud == HEADLESS) | |
445 | return h | (1 << PAGE_HEADLESS); | |
446 | ||
447 | /* otherwise, return pointer to encoded handle */ | |
448 | idx = __idx(zhdr, bud); | |
449 | h += idx; | |
450 | if (bud == LAST) | |
451 | h |= (zhdr->last_chunks << BUDDY_SHIFT); | |
452 | ||
4a3ac931 | 453 | write_lock(&slots->lock); |
7c2b8baa | 454 | slots->slot[idx] = h; |
4a3ac931 | 455 | write_unlock(&slots->lock); |
7c2b8baa | 456 | return (unsigned long)&slots->slot[idx]; |
9a001fc1 VW |
457 | } |
458 | ||
3f9d2b57 VW |
459 | static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud) |
460 | { | |
461 | return __encode_handle(zhdr, zhdr->slots, bud); | |
462 | } | |
463 | ||
ca0246bb VW |
464 | /* only for LAST bud, returns zero otherwise */ |
465 | static unsigned short handle_to_chunks(unsigned long handle) | |
466 | { | |
4a3ac931 VW |
467 | struct z3fold_buddy_slots *slots = handle_to_slots(handle); |
468 | unsigned long addr; | |
7c2b8baa | 469 | |
4a3ac931 VW |
470 | read_lock(&slots->lock); |
471 | addr = *(unsigned long *)handle; | |
472 | read_unlock(&slots->lock); | |
7c2b8baa | 473 | return (addr & ~PAGE_MASK) >> BUDDY_SHIFT; |
ca0246bb VW |
474 | } |
475 | ||
f201ebd8 | 476 | /* |
477 | * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle | |
478 | * but that doesn't matter. because the masking will result in the | |
479 | * correct buddy number. | |
480 | */ | |
9a001fc1 VW |
481 | static enum buddy handle_to_buddy(unsigned long handle) |
482 | { | |
7c2b8baa | 483 | struct z3fold_header *zhdr; |
4a3ac931 | 484 | struct z3fold_buddy_slots *slots = handle_to_slots(handle); |
7c2b8baa VW |
485 | unsigned long addr; |
486 | ||
4a3ac931 | 487 | read_lock(&slots->lock); |
7c2b8baa VW |
488 | WARN_ON(handle & (1 << PAGE_HEADLESS)); |
489 | addr = *(unsigned long *)handle; | |
4a3ac931 | 490 | read_unlock(&slots->lock); |
7c2b8baa VW |
491 | zhdr = (struct z3fold_header *)(addr & PAGE_MASK); |
492 | return (addr - zhdr->first_num) & BUDDY_MASK; | |
9a001fc1 VW |
493 | } |
494 | ||
9050cce1 VW |
495 | static inline struct z3fold_pool *zhdr_to_pool(struct z3fold_header *zhdr) |
496 | { | |
bb9a374d | 497 | return zhdr->pool; |
9050cce1 VW |
498 | } |
499 | ||
d30561c5 VW |
500 | static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked) |
501 | { | |
502 | struct page *page = virt_to_page(zhdr); | |
9050cce1 | 503 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); |
d30561c5 VW |
504 | |
505 | WARN_ON(!list_empty(&zhdr->buddy)); | |
506 | set_bit(PAGE_STALE, &page->private); | |
35529357 | 507 | clear_bit(NEEDS_COMPACTING, &page->private); |
d30561c5 VW |
508 | spin_lock(&pool->lock); |
509 | if (!list_empty(&page->lru)) | |
1f862989 | 510 | list_del_init(&page->lru); |
d30561c5 | 511 | spin_unlock(&pool->lock); |
4a3ac931 | 512 | |
d30561c5 VW |
513 | if (locked) |
514 | z3fold_page_unlock(zhdr); | |
4a3ac931 | 515 | |
d30561c5 VW |
516 | spin_lock(&pool->stale_lock); |
517 | list_add(&zhdr->buddy, &pool->stale); | |
518 | queue_work(pool->release_wq, &pool->work); | |
519 | spin_unlock(&pool->stale_lock); | |
5e36c25b ML |
520 | |
521 | atomic64_dec(&pool->pages_nr); | |
d30561c5 VW |
522 | } |
523 | ||
70ad3196 | 524 | static void release_z3fold_page(struct kref *ref) |
d30561c5 VW |
525 | { |
526 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, | |
527 | refcount); | |
528 | __release_z3fold_page(zhdr, false); | |
529 | } | |
530 | ||
531 | static void release_z3fold_page_locked(struct kref *ref) | |
532 | { | |
533 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, | |
534 | refcount); | |
535 | WARN_ON(z3fold_page_trylock(zhdr)); | |
536 | __release_z3fold_page(zhdr, true); | |
537 | } | |
538 | ||
539 | static void release_z3fold_page_locked_list(struct kref *ref) | |
540 | { | |
541 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, | |
542 | refcount); | |
9050cce1 | 543 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); |
4a3ac931 | 544 | |
9050cce1 | 545 | spin_lock(&pool->lock); |
d30561c5 | 546 | list_del_init(&zhdr->buddy); |
9050cce1 | 547 | spin_unlock(&pool->lock); |
d30561c5 VW |
548 | |
549 | WARN_ON(z3fold_page_trylock(zhdr)); | |
550 | __release_z3fold_page(zhdr, true); | |
551 | } | |
552 | ||
553 | static void free_pages_work(struct work_struct *w) | |
554 | { | |
555 | struct z3fold_pool *pool = container_of(w, struct z3fold_pool, work); | |
556 | ||
557 | spin_lock(&pool->stale_lock); | |
558 | while (!list_empty(&pool->stale)) { | |
559 | struct z3fold_header *zhdr = list_first_entry(&pool->stale, | |
560 | struct z3fold_header, buddy); | |
561 | struct page *page = virt_to_page(zhdr); | |
562 | ||
563 | list_del(&zhdr->buddy); | |
564 | if (WARN_ON(!test_bit(PAGE_STALE, &page->private))) | |
565 | continue; | |
d30561c5 VW |
566 | spin_unlock(&pool->stale_lock); |
567 | cancel_work_sync(&zhdr->work); | |
1f862989 | 568 | free_z3fold_page(page, false); |
d30561c5 VW |
569 | cond_resched(); |
570 | spin_lock(&pool->stale_lock); | |
571 | } | |
572 | spin_unlock(&pool->stale_lock); | |
573 | } | |
574 | ||
9a001fc1 VW |
575 | /* |
576 | * Returns the number of free chunks in a z3fold page. | |
577 | * NB: can't be used with HEADLESS pages. | |
578 | */ | |
579 | static int num_free_chunks(struct z3fold_header *zhdr) | |
580 | { | |
581 | int nfree; | |
582 | /* | |
583 | * If there is a middle object, pick up the bigger free space | |
584 | * either before or after it. Otherwise just subtract the number | |
585 | * of chunks occupied by the first and the last objects. | |
586 | */ | |
587 | if (zhdr->middle_chunks != 0) { | |
588 | int nfree_before = zhdr->first_chunks ? | |
ede93213 | 589 | 0 : zhdr->start_middle - ZHDR_CHUNKS; |
9a001fc1 | 590 | int nfree_after = zhdr->last_chunks ? |
ede93213 VW |
591 | 0 : TOTAL_CHUNKS - |
592 | (zhdr->start_middle + zhdr->middle_chunks); | |
9a001fc1 VW |
593 | nfree = max(nfree_before, nfree_after); |
594 | } else | |
595 | nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks; | |
596 | return nfree; | |
597 | } | |
598 | ||
9050cce1 VW |
599 | /* Add to the appropriate unbuddied list */ |
600 | static inline void add_to_unbuddied(struct z3fold_pool *pool, | |
601 | struct z3fold_header *zhdr) | |
602 | { | |
603 | if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 || | |
604 | zhdr->middle_chunks == 0) { | |
135f97fd | 605 | struct list_head *unbuddied; |
9050cce1 | 606 | int freechunks = num_free_chunks(zhdr); |
135f97fd VW |
607 | |
608 | migrate_disable(); | |
609 | unbuddied = this_cpu_ptr(pool->unbuddied); | |
9050cce1 VW |
610 | spin_lock(&pool->lock); |
611 | list_add(&zhdr->buddy, &unbuddied[freechunks]); | |
612 | spin_unlock(&pool->lock); | |
613 | zhdr->cpu = smp_processor_id(); | |
135f97fd | 614 | migrate_enable(); |
9050cce1 VW |
615 | } |
616 | } | |
617 | ||
dcf5aedb VW |
618 | static inline enum buddy get_free_buddy(struct z3fold_header *zhdr, int chunks) |
619 | { | |
620 | enum buddy bud = HEADLESS; | |
621 | ||
622 | if (zhdr->middle_chunks) { | |
623 | if (!zhdr->first_chunks && | |
624 | chunks <= zhdr->start_middle - ZHDR_CHUNKS) | |
625 | bud = FIRST; | |
626 | else if (!zhdr->last_chunks) | |
627 | bud = LAST; | |
628 | } else { | |
629 | if (!zhdr->first_chunks) | |
630 | bud = FIRST; | |
631 | else if (!zhdr->last_chunks) | |
632 | bud = LAST; | |
633 | else | |
634 | bud = MIDDLE; | |
635 | } | |
636 | ||
637 | return bud; | |
638 | } | |
639 | ||
ede93213 VW |
640 | static inline void *mchunk_memmove(struct z3fold_header *zhdr, |
641 | unsigned short dst_chunk) | |
642 | { | |
643 | void *beg = zhdr; | |
644 | return memmove(beg + (dst_chunk << CHUNK_SHIFT), | |
645 | beg + (zhdr->start_middle << CHUNK_SHIFT), | |
646 | zhdr->middle_chunks << CHUNK_SHIFT); | |
647 | } | |
648 | ||
4a3ac931 VW |
649 | static inline bool buddy_single(struct z3fold_header *zhdr) |
650 | { | |
651 | return !((zhdr->first_chunks && zhdr->middle_chunks) || | |
652 | (zhdr->first_chunks && zhdr->last_chunks) || | |
653 | (zhdr->middle_chunks && zhdr->last_chunks)); | |
654 | } | |
655 | ||
656 | static struct z3fold_header *compact_single_buddy(struct z3fold_header *zhdr) | |
657 | { | |
658 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); | |
659 | void *p = zhdr; | |
660 | unsigned long old_handle = 0; | |
661 | size_t sz = 0; | |
662 | struct z3fold_header *new_zhdr = NULL; | |
663 | int first_idx = __idx(zhdr, FIRST); | |
664 | int middle_idx = __idx(zhdr, MIDDLE); | |
665 | int last_idx = __idx(zhdr, LAST); | |
666 | unsigned short *moved_chunks = NULL; | |
667 | ||
668 | /* | |
669 | * No need to protect slots here -- all the slots are "local" and | |
670 | * the page lock is already taken | |
671 | */ | |
672 | if (zhdr->first_chunks && zhdr->slots->slot[first_idx]) { | |
673 | p += ZHDR_SIZE_ALIGNED; | |
674 | sz = zhdr->first_chunks << CHUNK_SHIFT; | |
675 | old_handle = (unsigned long)&zhdr->slots->slot[first_idx]; | |
676 | moved_chunks = &zhdr->first_chunks; | |
677 | } else if (zhdr->middle_chunks && zhdr->slots->slot[middle_idx]) { | |
678 | p += zhdr->start_middle << CHUNK_SHIFT; | |
679 | sz = zhdr->middle_chunks << CHUNK_SHIFT; | |
680 | old_handle = (unsigned long)&zhdr->slots->slot[middle_idx]; | |
681 | moved_chunks = &zhdr->middle_chunks; | |
682 | } else if (zhdr->last_chunks && zhdr->slots->slot[last_idx]) { | |
683 | p += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT); | |
684 | sz = zhdr->last_chunks << CHUNK_SHIFT; | |
685 | old_handle = (unsigned long)&zhdr->slots->slot[last_idx]; | |
686 | moved_chunks = &zhdr->last_chunks; | |
687 | } | |
688 | ||
689 | if (sz > 0) { | |
690 | enum buddy new_bud = HEADLESS; | |
691 | short chunks = size_to_chunks(sz); | |
692 | void *q; | |
693 | ||
694 | new_zhdr = __z3fold_alloc(pool, sz, false); | |
695 | if (!new_zhdr) | |
696 | return NULL; | |
697 | ||
698 | if (WARN_ON(new_zhdr == zhdr)) | |
699 | goto out_fail; | |
700 | ||
dcf5aedb | 701 | new_bud = get_free_buddy(new_zhdr, chunks); |
4a3ac931 VW |
702 | q = new_zhdr; |
703 | switch (new_bud) { | |
704 | case FIRST: | |
705 | new_zhdr->first_chunks = chunks; | |
706 | q += ZHDR_SIZE_ALIGNED; | |
707 | break; | |
708 | case MIDDLE: | |
709 | new_zhdr->middle_chunks = chunks; | |
710 | new_zhdr->start_middle = | |
711 | new_zhdr->first_chunks + ZHDR_CHUNKS; | |
712 | q += new_zhdr->start_middle << CHUNK_SHIFT; | |
713 | break; | |
714 | case LAST: | |
715 | new_zhdr->last_chunks = chunks; | |
716 | q += PAGE_SIZE - (new_zhdr->last_chunks << CHUNK_SHIFT); | |
717 | break; | |
718 | default: | |
719 | goto out_fail; | |
720 | } | |
721 | new_zhdr->foreign_handles++; | |
722 | memcpy(q, p, sz); | |
723 | write_lock(&zhdr->slots->lock); | |
724 | *(unsigned long *)old_handle = (unsigned long)new_zhdr + | |
725 | __idx(new_zhdr, new_bud); | |
726 | if (new_bud == LAST) | |
727 | *(unsigned long *)old_handle |= | |
728 | (new_zhdr->last_chunks << BUDDY_SHIFT); | |
729 | write_unlock(&zhdr->slots->lock); | |
730 | add_to_unbuddied(pool, new_zhdr); | |
731 | z3fold_page_unlock(new_zhdr); | |
732 | ||
733 | *moved_chunks = 0; | |
734 | } | |
735 | ||
736 | return new_zhdr; | |
737 | ||
738 | out_fail: | |
5e36c25b ML |
739 | if (new_zhdr && !kref_put(&new_zhdr->refcount, release_z3fold_page_locked)) { |
740 | add_to_unbuddied(pool, new_zhdr); | |
741 | z3fold_page_unlock(new_zhdr); | |
4a3ac931 VW |
742 | } |
743 | return NULL; | |
744 | ||
745 | } | |
746 | ||
1b096e5a | 747 | #define BIG_CHUNK_GAP 3 |
9a001fc1 VW |
748 | /* Has to be called with lock held */ |
749 | static int z3fold_compact_page(struct z3fold_header *zhdr) | |
750 | { | |
751 | struct page *page = virt_to_page(zhdr); | |
9a001fc1 | 752 | |
ede93213 VW |
753 | if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private)) |
754 | return 0; /* can't move middle chunk, it's used */ | |
9a001fc1 | 755 | |
1f862989 VW |
756 | if (unlikely(PageIsolated(page))) |
757 | return 0; | |
758 | ||
ede93213 VW |
759 | if (zhdr->middle_chunks == 0) |
760 | return 0; /* nothing to compact */ | |
761 | ||
762 | if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) { | |
763 | /* move to the beginning */ | |
764 | mchunk_memmove(zhdr, ZHDR_CHUNKS); | |
9a001fc1 VW |
765 | zhdr->first_chunks = zhdr->middle_chunks; |
766 | zhdr->middle_chunks = 0; | |
767 | zhdr->start_middle = 0; | |
768 | zhdr->first_num++; | |
1b096e5a | 769 | return 1; |
9a001fc1 | 770 | } |
1b096e5a VW |
771 | |
772 | /* | |
773 | * moving data is expensive, so let's only do that if | |
774 | * there's substantial gain (at least BIG_CHUNK_GAP chunks) | |
775 | */ | |
776 | if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 && | |
777 | zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >= | |
778 | BIG_CHUNK_GAP) { | |
779 | mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS); | |
780 | zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS; | |
781 | return 1; | |
782 | } else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 && | |
783 | TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle | |
784 | + zhdr->middle_chunks) >= | |
785 | BIG_CHUNK_GAP) { | |
786 | unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks - | |
787 | zhdr->middle_chunks; | |
788 | mchunk_memmove(zhdr, new_start); | |
789 | zhdr->start_middle = new_start; | |
790 | return 1; | |
791 | } | |
792 | ||
793 | return 0; | |
9a001fc1 VW |
794 | } |
795 | ||
d30561c5 VW |
796 | static void do_compact_page(struct z3fold_header *zhdr, bool locked) |
797 | { | |
9050cce1 | 798 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); |
d30561c5 | 799 | struct page *page; |
d30561c5 VW |
800 | |
801 | page = virt_to_page(zhdr); | |
802 | if (locked) | |
803 | WARN_ON(z3fold_page_trylock(zhdr)); | |
804 | else | |
805 | z3fold_page_lock(zhdr); | |
5d03a661 | 806 | if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) { |
d30561c5 VW |
807 | z3fold_page_unlock(zhdr); |
808 | return; | |
809 | } | |
810 | spin_lock(&pool->lock); | |
811 | list_del_init(&zhdr->buddy); | |
812 | spin_unlock(&pool->lock); | |
813 | ||
5e36c25b | 814 | if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) |
5d03a661 | 815 | return; |
5d03a661 | 816 | |
dcf5aedb VW |
817 | if (test_bit(PAGE_STALE, &page->private) || |
818 | test_and_set_bit(PAGE_CLAIMED, &page->private)) { | |
1f862989 VW |
819 | z3fold_page_unlock(zhdr); |
820 | return; | |
821 | } | |
822 | ||
4a3ac931 VW |
823 | if (!zhdr->foreign_handles && buddy_single(zhdr) && |
824 | zhdr->mapped_count == 0 && compact_single_buddy(zhdr)) { | |
5e36c25b | 825 | if (!kref_put(&zhdr->refcount, release_z3fold_page_locked)) { |
dcf5aedb | 826 | clear_bit(PAGE_CLAIMED, &page->private); |
4a3ac931 | 827 | z3fold_page_unlock(zhdr); |
dcf5aedb | 828 | } |
4a3ac931 VW |
829 | return; |
830 | } | |
831 | ||
d30561c5 | 832 | z3fold_compact_page(zhdr); |
9050cce1 | 833 | add_to_unbuddied(pool, zhdr); |
dcf5aedb | 834 | clear_bit(PAGE_CLAIMED, &page->private); |
d30561c5 VW |
835 | z3fold_page_unlock(zhdr); |
836 | } | |
837 | ||
838 | static void compact_page_work(struct work_struct *w) | |
839 | { | |
840 | struct z3fold_header *zhdr = container_of(w, struct z3fold_header, | |
841 | work); | |
842 | ||
843 | do_compact_page(zhdr, false); | |
844 | } | |
845 | ||
9050cce1 VW |
846 | /* returns _locked_ z3fold page header or NULL */ |
847 | static inline struct z3fold_header *__z3fold_alloc(struct z3fold_pool *pool, | |
848 | size_t size, bool can_sleep) | |
849 | { | |
850 | struct z3fold_header *zhdr = NULL; | |
851 | struct page *page; | |
852 | struct list_head *unbuddied; | |
853 | int chunks = size_to_chunks(size), i; | |
854 | ||
855 | lookup: | |
135f97fd | 856 | migrate_disable(); |
9050cce1 | 857 | /* First, try to find an unbuddied z3fold page. */ |
135f97fd | 858 | unbuddied = this_cpu_ptr(pool->unbuddied); |
9050cce1 VW |
859 | for_each_unbuddied_list(i, chunks) { |
860 | struct list_head *l = &unbuddied[i]; | |
861 | ||
862 | zhdr = list_first_entry_or_null(READ_ONCE(l), | |
863 | struct z3fold_header, buddy); | |
864 | ||
865 | if (!zhdr) | |
866 | continue; | |
867 | ||
868 | /* Re-check under lock. */ | |
869 | spin_lock(&pool->lock); | |
9050cce1 VW |
870 | if (unlikely(zhdr != list_first_entry(READ_ONCE(l), |
871 | struct z3fold_header, buddy)) || | |
872 | !z3fold_page_trylock(zhdr)) { | |
873 | spin_unlock(&pool->lock); | |
874 | zhdr = NULL; | |
135f97fd | 875 | migrate_enable(); |
9050cce1 VW |
876 | if (can_sleep) |
877 | cond_resched(); | |
878 | goto lookup; | |
879 | } | |
880 | list_del_init(&zhdr->buddy); | |
881 | zhdr->cpu = -1; | |
882 | spin_unlock(&pool->lock); | |
883 | ||
884 | page = virt_to_page(zhdr); | |
4a3ac931 VW |
885 | if (test_bit(NEEDS_COMPACTING, &page->private) || |
886 | test_bit(PAGE_CLAIMED, &page->private)) { | |
9050cce1 VW |
887 | z3fold_page_unlock(zhdr); |
888 | zhdr = NULL; | |
135f97fd | 889 | migrate_enable(); |
9050cce1 VW |
890 | if (can_sleep) |
891 | cond_resched(); | |
892 | goto lookup; | |
893 | } | |
894 | ||
895 | /* | |
896 | * this page could not be removed from its unbuddied | |
897 | * list while pool lock was held, and then we've taken | |
898 | * page lock so kref_put could not be called before | |
899 | * we got here, so it's safe to just call kref_get() | |
900 | */ | |
901 | kref_get(&zhdr->refcount); | |
902 | break; | |
903 | } | |
135f97fd | 904 | migrate_enable(); |
9050cce1 | 905 | |
351618b2 VW |
906 | if (!zhdr) { |
907 | int cpu; | |
908 | ||
909 | /* look for _exact_ match on other cpus' lists */ | |
910 | for_each_online_cpu(cpu) { | |
911 | struct list_head *l; | |
912 | ||
913 | unbuddied = per_cpu_ptr(pool->unbuddied, cpu); | |
914 | spin_lock(&pool->lock); | |
915 | l = &unbuddied[chunks]; | |
916 | ||
917 | zhdr = list_first_entry_or_null(READ_ONCE(l), | |
918 | struct z3fold_header, buddy); | |
919 | ||
920 | if (!zhdr || !z3fold_page_trylock(zhdr)) { | |
921 | spin_unlock(&pool->lock); | |
922 | zhdr = NULL; | |
923 | continue; | |
924 | } | |
925 | list_del_init(&zhdr->buddy); | |
926 | zhdr->cpu = -1; | |
927 | spin_unlock(&pool->lock); | |
928 | ||
929 | page = virt_to_page(zhdr); | |
4a3ac931 VW |
930 | if (test_bit(NEEDS_COMPACTING, &page->private) || |
931 | test_bit(PAGE_CLAIMED, &page->private)) { | |
351618b2 VW |
932 | z3fold_page_unlock(zhdr); |
933 | zhdr = NULL; | |
934 | if (can_sleep) | |
935 | cond_resched(); | |
936 | continue; | |
937 | } | |
938 | kref_get(&zhdr->refcount); | |
939 | break; | |
940 | } | |
941 | } | |
942 | ||
fc548865 VW |
943 | if (zhdr && !zhdr->slots) |
944 | zhdr->slots = alloc_slots(pool, | |
945 | can_sleep ? GFP_NOIO : GFP_ATOMIC); | |
9050cce1 VW |
946 | return zhdr; |
947 | } | |
d30561c5 VW |
948 | |
949 | /* | |
950 | * API Functions | |
951 | */ | |
952 | ||
953 | /** | |
954 | * z3fold_create_pool() - create a new z3fold pool | |
955 | * @name: pool name | |
956 | * @gfp: gfp flags when allocating the z3fold pool structure | |
957 | * @ops: user-defined operations for the z3fold pool | |
958 | * | |
959 | * Return: pointer to the new z3fold pool or NULL if the metadata allocation | |
960 | * failed. | |
961 | */ | |
962 | static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp, | |
963 | const struct z3fold_ops *ops) | |
964 | { | |
965 | struct z3fold_pool *pool = NULL; | |
966 | int i, cpu; | |
967 | ||
968 | pool = kzalloc(sizeof(struct z3fold_pool), gfp); | |
969 | if (!pool) | |
970 | goto out; | |
7c2b8baa VW |
971 | pool->c_handle = kmem_cache_create("z3fold_handle", |
972 | sizeof(struct z3fold_buddy_slots), | |
973 | SLOTS_ALIGN, 0, NULL); | |
974 | if (!pool->c_handle) | |
975 | goto out_c; | |
d30561c5 VW |
976 | spin_lock_init(&pool->lock); |
977 | spin_lock_init(&pool->stale_lock); | |
e891f60e ML |
978 | pool->unbuddied = __alloc_percpu(sizeof(struct list_head) * NCHUNKS, |
979 | __alignof__(struct list_head)); | |
1ec6995d XW |
980 | if (!pool->unbuddied) |
981 | goto out_pool; | |
d30561c5 VW |
982 | for_each_possible_cpu(cpu) { |
983 | struct list_head *unbuddied = | |
984 | per_cpu_ptr(pool->unbuddied, cpu); | |
985 | for_each_unbuddied_list(i, 0) | |
986 | INIT_LIST_HEAD(&unbuddied[i]); | |
987 | } | |
988 | INIT_LIST_HEAD(&pool->lru); | |
989 | INIT_LIST_HEAD(&pool->stale); | |
990 | atomic64_set(&pool->pages_nr, 0); | |
991 | pool->name = name; | |
992 | pool->compact_wq = create_singlethread_workqueue(pool->name); | |
993 | if (!pool->compact_wq) | |
1ec6995d | 994 | goto out_unbuddied; |
d30561c5 VW |
995 | pool->release_wq = create_singlethread_workqueue(pool->name); |
996 | if (!pool->release_wq) | |
997 | goto out_wq; | |
1f862989 VW |
998 | if (z3fold_register_migration(pool)) |
999 | goto out_rwq; | |
d30561c5 VW |
1000 | INIT_WORK(&pool->work, free_pages_work); |
1001 | pool->ops = ops; | |
1002 | return pool; | |
1003 | ||
1f862989 VW |
1004 | out_rwq: |
1005 | destroy_workqueue(pool->release_wq); | |
d30561c5 VW |
1006 | out_wq: |
1007 | destroy_workqueue(pool->compact_wq); | |
1ec6995d XW |
1008 | out_unbuddied: |
1009 | free_percpu(pool->unbuddied); | |
1010 | out_pool: | |
7c2b8baa VW |
1011 | kmem_cache_destroy(pool->c_handle); |
1012 | out_c: | |
d30561c5 | 1013 | kfree(pool); |
1ec6995d | 1014 | out: |
d30561c5 VW |
1015 | return NULL; |
1016 | } | |
1017 | ||
1018 | /** | |
1019 | * z3fold_destroy_pool() - destroys an existing z3fold pool | |
1020 | * @pool: the z3fold pool to be destroyed | |
1021 | * | |
1022 | * The pool should be emptied before this function is called. | |
1023 | */ | |
1024 | static void z3fold_destroy_pool(struct z3fold_pool *pool) | |
1025 | { | |
7c2b8baa | 1026 | kmem_cache_destroy(pool->c_handle); |
6051d3bd HB |
1027 | |
1028 | /* | |
1029 | * We need to destroy pool->compact_wq before pool->release_wq, | |
1030 | * as any pending work on pool->compact_wq will call | |
1031 | * queue_work(pool->release_wq, &pool->work). | |
b997052b HB |
1032 | * |
1033 | * There are still outstanding pages until both workqueues are drained, | |
1034 | * so we cannot unregister migration until then. | |
6051d3bd HB |
1035 | */ |
1036 | ||
d30561c5 | 1037 | destroy_workqueue(pool->compact_wq); |
6051d3bd | 1038 | destroy_workqueue(pool->release_wq); |
b997052b | 1039 | z3fold_unregister_migration(pool); |
dac0d1cf | 1040 | free_percpu(pool->unbuddied); |
d30561c5 VW |
1041 | kfree(pool); |
1042 | } | |
1043 | ||
9a001fc1 VW |
1044 | /** |
1045 | * z3fold_alloc() - allocates a region of a given size | |
1046 | * @pool: z3fold pool from which to allocate | |
1047 | * @size: size in bytes of the desired allocation | |
1048 | * @gfp: gfp flags used if the pool needs to grow | |
1049 | * @handle: handle of the new allocation | |
1050 | * | |
1051 | * This function will attempt to find a free region in the pool large enough to | |
1052 | * satisfy the allocation request. A search of the unbuddied lists is | |
1053 | * performed first. If no suitable free region is found, then a new page is | |
1054 | * allocated and added to the pool to satisfy the request. | |
1055 | * | |
9a001fc1 VW |
1056 | * Return: 0 if success and handle is set, otherwise -EINVAL if the size or |
1057 | * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate | |
1058 | * a new page. | |
1059 | */ | |
1060 | static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp, | |
1061 | unsigned long *handle) | |
1062 | { | |
9050cce1 | 1063 | int chunks = size_to_chunks(size); |
9a001fc1 | 1064 | struct z3fold_header *zhdr = NULL; |
d30561c5 | 1065 | struct page *page = NULL; |
9a001fc1 | 1066 | enum buddy bud; |
8a97ea54 | 1067 | bool can_sleep = gfpflags_allow_blocking(gfp); |
9a001fc1 | 1068 | |
f1549cb5 | 1069 | if (!size) |
9a001fc1 VW |
1070 | return -EINVAL; |
1071 | ||
1072 | if (size > PAGE_SIZE) | |
1073 | return -ENOSPC; | |
1074 | ||
1075 | if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE) | |
1076 | bud = HEADLESS; | |
1077 | else { | |
9050cce1 VW |
1078 | retry: |
1079 | zhdr = __z3fold_alloc(pool, size, can_sleep); | |
d30561c5 | 1080 | if (zhdr) { |
dcf5aedb VW |
1081 | bud = get_free_buddy(zhdr, chunks); |
1082 | if (bud == HEADLESS) { | |
5e36c25b | 1083 | if (!kref_put(&zhdr->refcount, |
d30561c5 | 1084 | release_z3fold_page_locked)) |
d30561c5 | 1085 | z3fold_page_unlock(zhdr); |
2f1e5e4d VW |
1086 | pr_err("No free chunks in unbuddied\n"); |
1087 | WARN_ON(1); | |
9050cce1 | 1088 | goto retry; |
9a001fc1 | 1089 | } |
9050cce1 | 1090 | page = virt_to_page(zhdr); |
2f1e5e4d | 1091 | goto found; |
9a001fc1 VW |
1092 | } |
1093 | bud = FIRST; | |
9a001fc1 VW |
1094 | } |
1095 | ||
5c9bab59 VW |
1096 | page = NULL; |
1097 | if (can_sleep) { | |
1098 | spin_lock(&pool->stale_lock); | |
1099 | zhdr = list_first_entry_or_null(&pool->stale, | |
1100 | struct z3fold_header, buddy); | |
1101 | /* | |
1102 | * Before allocating a page, let's see if we can take one from | |
1103 | * the stale pages list. cancel_work_sync() can sleep so we | |
1104 | * limit this case to the contexts where we can sleep | |
1105 | */ | |
1106 | if (zhdr) { | |
1107 | list_del(&zhdr->buddy); | |
1108 | spin_unlock(&pool->stale_lock); | |
d30561c5 | 1109 | cancel_work_sync(&zhdr->work); |
5c9bab59 VW |
1110 | page = virt_to_page(zhdr); |
1111 | } else { | |
1112 | spin_unlock(&pool->stale_lock); | |
1113 | } | |
d30561c5 | 1114 | } |
5c9bab59 VW |
1115 | if (!page) |
1116 | page = alloc_page(gfp); | |
d30561c5 | 1117 | |
9a001fc1 VW |
1118 | if (!page) |
1119 | return -ENOMEM; | |
2f1e5e4d | 1120 | |
63398413 | 1121 | zhdr = init_z3fold_page(page, bud == HEADLESS, pool, gfp); |
9050cce1 VW |
1122 | if (!zhdr) { |
1123 | __free_page(page); | |
1124 | return -ENOMEM; | |
1125 | } | |
1126 | atomic64_inc(&pool->pages_nr); | |
9a001fc1 VW |
1127 | |
1128 | if (bud == HEADLESS) { | |
1129 | set_bit(PAGE_HEADLESS, &page->private); | |
1130 | goto headless; | |
1131 | } | |
810481a2 HB |
1132 | if (can_sleep) { |
1133 | lock_page(page); | |
1134 | __SetPageMovable(page, pool->inode->i_mapping); | |
1135 | unlock_page(page); | |
1136 | } else { | |
1137 | if (trylock_page(page)) { | |
1138 | __SetPageMovable(page, pool->inode->i_mapping); | |
1139 | unlock_page(page); | |
1140 | } | |
1141 | } | |
2f1e5e4d | 1142 | z3fold_page_lock(zhdr); |
9a001fc1 VW |
1143 | |
1144 | found: | |
1145 | if (bud == FIRST) | |
1146 | zhdr->first_chunks = chunks; | |
1147 | else if (bud == LAST) | |
1148 | zhdr->last_chunks = chunks; | |
1149 | else { | |
1150 | zhdr->middle_chunks = chunks; | |
ede93213 | 1151 | zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS; |
9a001fc1 | 1152 | } |
9050cce1 | 1153 | add_to_unbuddied(pool, zhdr); |
9a001fc1 VW |
1154 | |
1155 | headless: | |
d30561c5 | 1156 | spin_lock(&pool->lock); |
9a001fc1 VW |
1157 | /* Add/move z3fold page to beginning of LRU */ |
1158 | if (!list_empty(&page->lru)) | |
1159 | list_del(&page->lru); | |
1160 | ||
1161 | list_add(&page->lru, &pool->lru); | |
1162 | ||
1163 | *handle = encode_handle(zhdr, bud); | |
1164 | spin_unlock(&pool->lock); | |
2f1e5e4d VW |
1165 | if (bud != HEADLESS) |
1166 | z3fold_page_unlock(zhdr); | |
9a001fc1 VW |
1167 | |
1168 | return 0; | |
1169 | } | |
1170 | ||
1171 | /** | |
1172 | * z3fold_free() - frees the allocation associated with the given handle | |
1173 | * @pool: pool in which the allocation resided | |
1174 | * @handle: handle associated with the allocation returned by z3fold_alloc() | |
1175 | * | |
1176 | * In the case that the z3fold page in which the allocation resides is under | |
ed0e5dca ML |
1177 | * reclaim, as indicated by the PAGE_CLAIMED flag being set, this function |
1178 | * only sets the first|middle|last_chunks to 0. The page is actually freed | |
1179 | * once all buddies are evicted (see z3fold_reclaim_page() below). | |
9a001fc1 VW |
1180 | */ |
1181 | static void z3fold_free(struct z3fold_pool *pool, unsigned long handle) | |
1182 | { | |
1183 | struct z3fold_header *zhdr; | |
9a001fc1 VW |
1184 | struct page *page; |
1185 | enum buddy bud; | |
5b6807de | 1186 | bool page_claimed; |
9a001fc1 | 1187 | |
4a3ac931 | 1188 | zhdr = get_z3fold_header(handle); |
9a001fc1 | 1189 | page = virt_to_page(zhdr); |
5b6807de | 1190 | page_claimed = test_and_set_bit(PAGE_CLAIMED, &page->private); |
9a001fc1 VW |
1191 | |
1192 | if (test_bit(PAGE_HEADLESS, &page->private)) { | |
ca0246bb VW |
1193 | /* if a headless page is under reclaim, just leave. |
1194 | * NB: we use test_and_set_bit for a reason: if the bit | |
1195 | * has not been set before, we release this page | |
1196 | * immediately so we don't care about its value any more. | |
1197 | */ | |
5b6807de | 1198 | if (!page_claimed) { |
ca0246bb VW |
1199 | spin_lock(&pool->lock); |
1200 | list_del(&page->lru); | |
1201 | spin_unlock(&pool->lock); | |
4a3ac931 | 1202 | put_z3fold_header(zhdr); |
1f862989 | 1203 | free_z3fold_page(page, true); |
ca0246bb | 1204 | atomic64_dec(&pool->pages_nr); |
9a001fc1 | 1205 | } |
ca0246bb | 1206 | return; |
9a001fc1 VW |
1207 | } |
1208 | ||
ca0246bb | 1209 | /* Non-headless case */ |
ca0246bb VW |
1210 | bud = handle_to_buddy(handle); |
1211 | ||
1212 | switch (bud) { | |
1213 | case FIRST: | |
1214 | zhdr->first_chunks = 0; | |
1215 | break; | |
1216 | case MIDDLE: | |
1217 | zhdr->middle_chunks = 0; | |
1218 | break; | |
1219 | case LAST: | |
1220 | zhdr->last_chunks = 0; | |
1221 | break; | |
1222 | default: | |
1223 | pr_err("%s: unknown bud %d\n", __func__, bud); | |
1224 | WARN_ON(1); | |
4a3ac931 | 1225 | put_z3fold_header(zhdr); |
d30561c5 VW |
1226 | return; |
1227 | } | |
1228 | ||
4a3ac931 | 1229 | if (!page_claimed) |
fc548865 | 1230 | free_handle(handle, zhdr); |
5e36c25b | 1231 | if (kref_put(&zhdr->refcount, release_z3fold_page_locked_list)) |
d30561c5 | 1232 | return; |
5b6807de VW |
1233 | if (page_claimed) { |
1234 | /* the page has not been claimed by us */ | |
ed0e5dca | 1235 | put_z3fold_header(zhdr); |
6098d7e1 VW |
1236 | return; |
1237 | } | |
dcf5aedb | 1238 | if (test_and_set_bit(NEEDS_COMPACTING, &page->private)) { |
4a3ac931 | 1239 | put_z3fold_header(zhdr); |
5b6807de | 1240 | clear_bit(PAGE_CLAIMED, &page->private); |
d30561c5 VW |
1241 | return; |
1242 | } | |
1243 | if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) { | |
d30561c5 | 1244 | zhdr->cpu = -1; |
5d03a661 | 1245 | kref_get(&zhdr->refcount); |
5b6807de | 1246 | clear_bit(PAGE_CLAIMED, &page->private); |
4a3ac931 | 1247 | do_compact_page(zhdr, true); |
d30561c5 | 1248 | return; |
9a001fc1 | 1249 | } |
5d03a661 | 1250 | kref_get(&zhdr->refcount); |
5b6807de | 1251 | clear_bit(PAGE_CLAIMED, &page->private); |
4a3ac931 VW |
1252 | queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work); |
1253 | put_z3fold_header(zhdr); | |
9a001fc1 VW |
1254 | } |
1255 | ||
1256 | /** | |
1257 | * z3fold_reclaim_page() - evicts allocations from a pool page and frees it | |
1258 | * @pool: pool from which a page will attempt to be evicted | |
f144c390 | 1259 | * @retries: number of pages on the LRU list for which eviction will |
9a001fc1 VW |
1260 | * be attempted before failing |
1261 | * | |
1262 | * z3fold reclaim is different from normal system reclaim in that it is done | |
1263 | * from the bottom, up. This is because only the bottom layer, z3fold, has | |
1264 | * information on how the allocations are organized within each z3fold page. | |
1265 | * This has the potential to create interesting locking situations between | |
1266 | * z3fold and the user, however. | |
1267 | * | |
1268 | * To avoid these, this is how z3fold_reclaim_page() should be called: | |
f144c390 | 1269 | * |
9a001fc1 VW |
1270 | * The user detects a page should be reclaimed and calls z3fold_reclaim_page(). |
1271 | * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and | |
1272 | * call the user-defined eviction handler with the pool and handle as | |
1273 | * arguments. | |
1274 | * | |
1275 | * If the handle can not be evicted, the eviction handler should return | |
1276 | * non-zero. z3fold_reclaim_page() will add the z3fold page back to the | |
1277 | * appropriate list and try the next z3fold page on the LRU up to | |
1278 | * a user defined number of retries. | |
1279 | * | |
1280 | * If the handle is successfully evicted, the eviction handler should | |
1281 | * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free() | |
1282 | * contains logic to delay freeing the page if the page is under reclaim, | |
1283 | * as indicated by the setting of the PG_reclaim flag on the underlying page. | |
1284 | * | |
1285 | * If all buddies in the z3fold page are successfully evicted, then the | |
1286 | * z3fold page can be freed. | |
1287 | * | |
1288 | * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are | |
1289 | * no pages to evict or an eviction handler is not registered, -EAGAIN if | |
1290 | * the retry limit was hit. | |
1291 | */ | |
1292 | static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries) | |
1293 | { | |
4a3ac931 | 1294 | int i, ret = -1; |
d30561c5 VW |
1295 | struct z3fold_header *zhdr = NULL; |
1296 | struct page *page = NULL; | |
1297 | struct list_head *pos; | |
9a001fc1 | 1298 | unsigned long first_handle = 0, middle_handle = 0, last_handle = 0; |
dcf5aedb VW |
1299 | struct z3fold_buddy_slots slots __attribute__((aligned(SLOTS_ALIGN))); |
1300 | ||
1301 | rwlock_init(&slots.lock); | |
1302 | slots.pool = (unsigned long)pool | (1 << HANDLES_NOFREE); | |
9a001fc1 VW |
1303 | |
1304 | spin_lock(&pool->lock); | |
2f1e5e4d | 1305 | if (!pool->ops || !pool->ops->evict || retries == 0) { |
9a001fc1 VW |
1306 | spin_unlock(&pool->lock); |
1307 | return -EINVAL; | |
1308 | } | |
1309 | for (i = 0; i < retries; i++) { | |
2f1e5e4d VW |
1310 | if (list_empty(&pool->lru)) { |
1311 | spin_unlock(&pool->lock); | |
1312 | return -EINVAL; | |
1313 | } | |
d30561c5 VW |
1314 | list_for_each_prev(pos, &pool->lru) { |
1315 | page = list_entry(pos, struct page, lru); | |
ca0246bb | 1316 | |
3f9d2b57 | 1317 | zhdr = page_address(page); |
6d679578 TH |
1318 | if (test_bit(PAGE_HEADLESS, &page->private)) { |
1319 | /* | |
1320 | * For non-headless pages, we wait to do this | |
1321 | * until we have the page lock to avoid racing | |
1322 | * with __z3fold_alloc(). Headless pages don't | |
1323 | * have a lock (and __z3fold_alloc() will never | |
1324 | * see them), but we still need to test and set | |
1325 | * PAGE_CLAIMED to avoid racing with | |
1326 | * z3fold_free(), so just do it now before | |
1327 | * leaving the loop. | |
1328 | */ | |
1329 | if (test_and_set_bit(PAGE_CLAIMED, &page->private)) | |
1330 | continue; | |
1331 | ||
d30561c5 | 1332 | break; |
6d679578 | 1333 | } |
d30561c5 | 1334 | |
dcf5aedb VW |
1335 | if (kref_get_unless_zero(&zhdr->refcount) == 0) { |
1336 | zhdr = NULL; | |
1337 | break; | |
1338 | } | |
ca0246bb | 1339 | if (!z3fold_page_trylock(zhdr)) { |
5e36c25b | 1340 | kref_put(&zhdr->refcount, release_z3fold_page); |
ca0246bb | 1341 | zhdr = NULL; |
d30561c5 | 1342 | continue; /* can't evict at this point */ |
ca0246bb | 1343 | } |
dcf5aedb VW |
1344 | |
1345 | /* test_and_set_bit is of course atomic, but we still | |
1346 | * need to do it under page lock, otherwise checking | |
1347 | * that bit in __z3fold_alloc wouldn't make sense | |
1348 | */ | |
1349 | if (zhdr->foreign_handles || | |
1350 | test_and_set_bit(PAGE_CLAIMED, &page->private)) { | |
5e36c25b | 1351 | if (!kref_put(&zhdr->refcount, |
28473d91 | 1352 | release_z3fold_page_locked)) |
dcf5aedb | 1353 | z3fold_page_unlock(zhdr); |
4a3ac931 VW |
1354 | zhdr = NULL; |
1355 | continue; /* can't evict such page */ | |
1356 | } | |
d30561c5 VW |
1357 | list_del_init(&zhdr->buddy); |
1358 | zhdr->cpu = -1; | |
6098d7e1 | 1359 | break; |
d30561c5 VW |
1360 | } |
1361 | ||
ca0246bb VW |
1362 | if (!zhdr) |
1363 | break; | |
1364 | ||
5a27aa82 | 1365 | list_del_init(&page->lru); |
d30561c5 | 1366 | spin_unlock(&pool->lock); |
9a001fc1 | 1367 | |
9a001fc1 | 1368 | if (!test_bit(PAGE_HEADLESS, &page->private)) { |
9a001fc1 | 1369 | /* |
3f9d2b57 VW |
1370 | * We need encode the handles before unlocking, and |
1371 | * use our local slots structure because z3fold_free | |
1372 | * can zero out zhdr->slots and we can't do much | |
1373 | * about that | |
9a001fc1 VW |
1374 | */ |
1375 | first_handle = 0; | |
1376 | last_handle = 0; | |
1377 | middle_handle = 0; | |
dcf5aedb | 1378 | memset(slots.slot, 0, sizeof(slots.slot)); |
9a001fc1 | 1379 | if (zhdr->first_chunks) |
dcf5aedb VW |
1380 | first_handle = __encode_handle(zhdr, &slots, |
1381 | FIRST); | |
9a001fc1 | 1382 | if (zhdr->middle_chunks) |
dcf5aedb VW |
1383 | middle_handle = __encode_handle(zhdr, &slots, |
1384 | MIDDLE); | |
9a001fc1 | 1385 | if (zhdr->last_chunks) |
dcf5aedb VW |
1386 | last_handle = __encode_handle(zhdr, &slots, |
1387 | LAST); | |
d30561c5 VW |
1388 | /* |
1389 | * it's safe to unlock here because we hold a | |
1390 | * reference to this page | |
1391 | */ | |
2f1e5e4d | 1392 | z3fold_page_unlock(zhdr); |
9a001fc1 | 1393 | } else { |
4a3ac931 | 1394 | first_handle = encode_handle(zhdr, HEADLESS); |
9a001fc1 VW |
1395 | last_handle = middle_handle = 0; |
1396 | } | |
9a001fc1 VW |
1397 | /* Issue the eviction callback(s) */ |
1398 | if (middle_handle) { | |
1399 | ret = pool->ops->evict(pool, middle_handle); | |
1400 | if (ret) | |
1401 | goto next; | |
1402 | } | |
1403 | if (first_handle) { | |
1404 | ret = pool->ops->evict(pool, first_handle); | |
1405 | if (ret) | |
1406 | goto next; | |
1407 | } | |
1408 | if (last_handle) { | |
1409 | ret = pool->ops->evict(pool, last_handle); | |
1410 | if (ret) | |
1411 | goto next; | |
1412 | } | |
1413 | next: | |
5a27aa82 VW |
1414 | if (test_bit(PAGE_HEADLESS, &page->private)) { |
1415 | if (ret == 0) { | |
1f862989 | 1416 | free_z3fold_page(page, true); |
ca0246bb | 1417 | atomic64_dec(&pool->pages_nr); |
5a27aa82 | 1418 | return 0; |
5a27aa82 | 1419 | } |
6098d7e1 VW |
1420 | spin_lock(&pool->lock); |
1421 | list_add(&page->lru, &pool->lru); | |
1422 | spin_unlock(&pool->lock); | |
3f9d2b57 | 1423 | clear_bit(PAGE_CLAIMED, &page->private); |
6098d7e1 | 1424 | } else { |
dcf5aedb | 1425 | struct z3fold_buddy_slots *slots = zhdr->slots; |
6098d7e1 | 1426 | z3fold_page_lock(zhdr); |
6098d7e1 VW |
1427 | if (kref_put(&zhdr->refcount, |
1428 | release_z3fold_page_locked)) { | |
dcf5aedb | 1429 | kmem_cache_free(pool->c_handle, slots); |
6098d7e1 VW |
1430 | return 0; |
1431 | } | |
1432 | /* | |
1433 | * if we are here, the page is still not completely | |
1434 | * free. Take the global pool lock then to be able | |
1435 | * to add it back to the lru list | |
1436 | */ | |
1437 | spin_lock(&pool->lock); | |
1438 | list_add(&page->lru, &pool->lru); | |
d5567c9d | 1439 | spin_unlock(&pool->lock); |
6098d7e1 | 1440 | z3fold_page_unlock(zhdr); |
3f9d2b57 | 1441 | clear_bit(PAGE_CLAIMED, &page->private); |
5a27aa82 | 1442 | } |
2f1e5e4d | 1443 | |
6098d7e1 VW |
1444 | /* We started off locked to we need to lock the pool back */ |
1445 | spin_lock(&pool->lock); | |
9a001fc1 VW |
1446 | } |
1447 | spin_unlock(&pool->lock); | |
1448 | return -EAGAIN; | |
1449 | } | |
1450 | ||
1451 | /** | |
1452 | * z3fold_map() - maps the allocation associated with the given handle | |
1453 | * @pool: pool in which the allocation resides | |
1454 | * @handle: handle associated with the allocation to be mapped | |
1455 | * | |
1456 | * Extracts the buddy number from handle and constructs the pointer to the | |
1457 | * correct starting chunk within the page. | |
1458 | * | |
1459 | * Returns: a pointer to the mapped allocation | |
1460 | */ | |
1461 | static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle) | |
1462 | { | |
1463 | struct z3fold_header *zhdr; | |
1464 | struct page *page; | |
1465 | void *addr; | |
1466 | enum buddy buddy; | |
1467 | ||
4a3ac931 | 1468 | zhdr = get_z3fold_header(handle); |
9a001fc1 VW |
1469 | addr = zhdr; |
1470 | page = virt_to_page(zhdr); | |
1471 | ||
1472 | if (test_bit(PAGE_HEADLESS, &page->private)) | |
1473 | goto out; | |
1474 | ||
1475 | buddy = handle_to_buddy(handle); | |
1476 | switch (buddy) { | |
1477 | case FIRST: | |
1478 | addr += ZHDR_SIZE_ALIGNED; | |
1479 | break; | |
1480 | case MIDDLE: | |
1481 | addr += zhdr->start_middle << CHUNK_SHIFT; | |
1482 | set_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
1483 | break; | |
1484 | case LAST: | |
ca0246bb | 1485 | addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT); |
9a001fc1 VW |
1486 | break; |
1487 | default: | |
1488 | pr_err("unknown buddy id %d\n", buddy); | |
1489 | WARN_ON(1); | |
1490 | addr = NULL; | |
1491 | break; | |
1492 | } | |
2f1e5e4d | 1493 | |
1f862989 VW |
1494 | if (addr) |
1495 | zhdr->mapped_count++; | |
9a001fc1 | 1496 | out: |
4a3ac931 | 1497 | put_z3fold_header(zhdr); |
9a001fc1 VW |
1498 | return addr; |
1499 | } | |
1500 | ||
1501 | /** | |
1502 | * z3fold_unmap() - unmaps the allocation associated with the given handle | |
1503 | * @pool: pool in which the allocation resides | |
1504 | * @handle: handle associated with the allocation to be unmapped | |
1505 | */ | |
1506 | static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle) | |
1507 | { | |
1508 | struct z3fold_header *zhdr; | |
1509 | struct page *page; | |
1510 | enum buddy buddy; | |
1511 | ||
4a3ac931 | 1512 | zhdr = get_z3fold_header(handle); |
9a001fc1 VW |
1513 | page = virt_to_page(zhdr); |
1514 | ||
2f1e5e4d | 1515 | if (test_bit(PAGE_HEADLESS, &page->private)) |
9a001fc1 | 1516 | return; |
9a001fc1 VW |
1517 | |
1518 | buddy = handle_to_buddy(handle); | |
1519 | if (buddy == MIDDLE) | |
1520 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); | |
1f862989 | 1521 | zhdr->mapped_count--; |
4a3ac931 | 1522 | put_z3fold_header(zhdr); |
9a001fc1 VW |
1523 | } |
1524 | ||
1525 | /** | |
1526 | * z3fold_get_pool_size() - gets the z3fold pool size in pages | |
1527 | * @pool: pool whose size is being queried | |
1528 | * | |
12d59ae6 | 1529 | * Returns: size in pages of the given pool. |
9a001fc1 VW |
1530 | */ |
1531 | static u64 z3fold_get_pool_size(struct z3fold_pool *pool) | |
1532 | { | |
12d59ae6 | 1533 | return atomic64_read(&pool->pages_nr); |
9a001fc1 VW |
1534 | } |
1535 | ||
1f862989 VW |
1536 | static bool z3fold_page_isolate(struct page *page, isolate_mode_t mode) |
1537 | { | |
1538 | struct z3fold_header *zhdr; | |
1539 | struct z3fold_pool *pool; | |
1540 | ||
1541 | VM_BUG_ON_PAGE(!PageMovable(page), page); | |
1542 | VM_BUG_ON_PAGE(PageIsolated(page), page); | |
1543 | ||
dcf5aedb | 1544 | if (test_bit(PAGE_HEADLESS, &page->private)) |
1f862989 VW |
1545 | return false; |
1546 | ||
1547 | zhdr = page_address(page); | |
1548 | z3fold_page_lock(zhdr); | |
1549 | if (test_bit(NEEDS_COMPACTING, &page->private) || | |
1550 | test_bit(PAGE_STALE, &page->private)) | |
1551 | goto out; | |
1552 | ||
4a3ac931 VW |
1553 | if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0) |
1554 | goto out; | |
1555 | ||
dcf5aedb VW |
1556 | if (test_and_set_bit(PAGE_CLAIMED, &page->private)) |
1557 | goto out; | |
1f862989 | 1558 | pool = zhdr_to_pool(zhdr); |
4a3ac931 VW |
1559 | spin_lock(&pool->lock); |
1560 | if (!list_empty(&zhdr->buddy)) | |
1561 | list_del_init(&zhdr->buddy); | |
1562 | if (!list_empty(&page->lru)) | |
1563 | list_del_init(&page->lru); | |
1564 | spin_unlock(&pool->lock); | |
1565 | ||
1566 | kref_get(&zhdr->refcount); | |
1567 | z3fold_page_unlock(zhdr); | |
1568 | return true; | |
1f862989 | 1569 | |
1f862989 VW |
1570 | out: |
1571 | z3fold_page_unlock(zhdr); | |
1572 | return false; | |
1573 | } | |
1574 | ||
1575 | static int z3fold_page_migrate(struct address_space *mapping, struct page *newpage, | |
1576 | struct page *page, enum migrate_mode mode) | |
1577 | { | |
1578 | struct z3fold_header *zhdr, *new_zhdr; | |
1579 | struct z3fold_pool *pool; | |
1580 | struct address_space *new_mapping; | |
1581 | ||
1582 | VM_BUG_ON_PAGE(!PageMovable(page), page); | |
1583 | VM_BUG_ON_PAGE(!PageIsolated(page), page); | |
dcf5aedb | 1584 | VM_BUG_ON_PAGE(!test_bit(PAGE_CLAIMED, &page->private), page); |
810481a2 | 1585 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); |
1f862989 VW |
1586 | |
1587 | zhdr = page_address(page); | |
1588 | pool = zhdr_to_pool(zhdr); | |
1589 | ||
dcf5aedb | 1590 | if (!z3fold_page_trylock(zhdr)) |
1f862989 | 1591 | return -EAGAIN; |
4a3ac931 | 1592 | if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0) { |
1f862989 | 1593 | z3fold_page_unlock(zhdr); |
dcf5aedb | 1594 | clear_bit(PAGE_CLAIMED, &page->private); |
1f862989 VW |
1595 | return -EBUSY; |
1596 | } | |
c92d2f38 HB |
1597 | if (work_pending(&zhdr->work)) { |
1598 | z3fold_page_unlock(zhdr); | |
1599 | return -EAGAIN; | |
1600 | } | |
1f862989 VW |
1601 | new_zhdr = page_address(newpage); |
1602 | memcpy(new_zhdr, zhdr, PAGE_SIZE); | |
1603 | newpage->private = page->private; | |
1604 | page->private = 0; | |
1605 | z3fold_page_unlock(zhdr); | |
1606 | spin_lock_init(&new_zhdr->page_lock); | |
c92d2f38 HB |
1607 | INIT_WORK(&new_zhdr->work, compact_page_work); |
1608 | /* | |
1609 | * z3fold_page_isolate() ensures that new_zhdr->buddy is empty, | |
1610 | * so we only have to reinitialize it. | |
1611 | */ | |
1612 | INIT_LIST_HEAD(&new_zhdr->buddy); | |
1f862989 VW |
1613 | new_mapping = page_mapping(page); |
1614 | __ClearPageMovable(page); | |
1f862989 VW |
1615 | |
1616 | get_page(newpage); | |
1617 | z3fold_page_lock(new_zhdr); | |
1618 | if (new_zhdr->first_chunks) | |
1619 | encode_handle(new_zhdr, FIRST); | |
1620 | if (new_zhdr->last_chunks) | |
1621 | encode_handle(new_zhdr, LAST); | |
1622 | if (new_zhdr->middle_chunks) | |
1623 | encode_handle(new_zhdr, MIDDLE); | |
1624 | set_bit(NEEDS_COMPACTING, &newpage->private); | |
1625 | new_zhdr->cpu = smp_processor_id(); | |
1626 | spin_lock(&pool->lock); | |
1627 | list_add(&newpage->lru, &pool->lru); | |
1628 | spin_unlock(&pool->lock); | |
1629 | __SetPageMovable(newpage, new_mapping); | |
1630 | z3fold_page_unlock(new_zhdr); | |
1631 | ||
1632 | queue_work_on(new_zhdr->cpu, pool->compact_wq, &new_zhdr->work); | |
1633 | ||
dcf5aedb | 1634 | clear_bit(PAGE_CLAIMED, &page->private); |
1f862989 VW |
1635 | put_page(page); |
1636 | return 0; | |
1637 | } | |
1638 | ||
1639 | static void z3fold_page_putback(struct page *page) | |
1640 | { | |
1641 | struct z3fold_header *zhdr; | |
1642 | struct z3fold_pool *pool; | |
1643 | ||
1644 | zhdr = page_address(page); | |
1645 | pool = zhdr_to_pool(zhdr); | |
1646 | ||
1647 | z3fold_page_lock(zhdr); | |
1648 | if (!list_empty(&zhdr->buddy)) | |
1649 | list_del_init(&zhdr->buddy); | |
1650 | INIT_LIST_HEAD(&page->lru); | |
5e36c25b | 1651 | if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) |
1f862989 | 1652 | return; |
1f862989 VW |
1653 | spin_lock(&pool->lock); |
1654 | list_add(&page->lru, &pool->lru); | |
1655 | spin_unlock(&pool->lock); | |
dcf5aedb | 1656 | clear_bit(PAGE_CLAIMED, &page->private); |
1f862989 VW |
1657 | z3fold_page_unlock(zhdr); |
1658 | } | |
1659 | ||
1660 | static const struct address_space_operations z3fold_aops = { | |
1661 | .isolate_page = z3fold_page_isolate, | |
1662 | .migratepage = z3fold_page_migrate, | |
1663 | .putback_page = z3fold_page_putback, | |
1664 | }; | |
1665 | ||
9a001fc1 VW |
1666 | /***************** |
1667 | * zpool | |
1668 | ****************/ | |
1669 | ||
1670 | static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle) | |
1671 | { | |
1672 | if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict) | |
1673 | return pool->zpool_ops->evict(pool->zpool, handle); | |
1674 | else | |
1675 | return -ENOENT; | |
1676 | } | |
1677 | ||
1678 | static const struct z3fold_ops z3fold_zpool_ops = { | |
1679 | .evict = z3fold_zpool_evict | |
1680 | }; | |
1681 | ||
1682 | static void *z3fold_zpool_create(const char *name, gfp_t gfp, | |
1683 | const struct zpool_ops *zpool_ops, | |
1684 | struct zpool *zpool) | |
1685 | { | |
1686 | struct z3fold_pool *pool; | |
1687 | ||
d30561c5 VW |
1688 | pool = z3fold_create_pool(name, gfp, |
1689 | zpool_ops ? &z3fold_zpool_ops : NULL); | |
9a001fc1 VW |
1690 | if (pool) { |
1691 | pool->zpool = zpool; | |
1692 | pool->zpool_ops = zpool_ops; | |
1693 | } | |
1694 | return pool; | |
1695 | } | |
1696 | ||
1697 | static void z3fold_zpool_destroy(void *pool) | |
1698 | { | |
1699 | z3fold_destroy_pool(pool); | |
1700 | } | |
1701 | ||
1702 | static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp, | |
1703 | unsigned long *handle) | |
1704 | { | |
1705 | return z3fold_alloc(pool, size, gfp, handle); | |
1706 | } | |
1707 | static void z3fold_zpool_free(void *pool, unsigned long handle) | |
1708 | { | |
1709 | z3fold_free(pool, handle); | |
1710 | } | |
1711 | ||
1712 | static int z3fold_zpool_shrink(void *pool, unsigned int pages, | |
1713 | unsigned int *reclaimed) | |
1714 | { | |
1715 | unsigned int total = 0; | |
1716 | int ret = -EINVAL; | |
1717 | ||
1718 | while (total < pages) { | |
1719 | ret = z3fold_reclaim_page(pool, 8); | |
1720 | if (ret < 0) | |
1721 | break; | |
1722 | total++; | |
1723 | } | |
1724 | ||
1725 | if (reclaimed) | |
1726 | *reclaimed = total; | |
1727 | ||
1728 | return ret; | |
1729 | } | |
1730 | ||
1731 | static void *z3fold_zpool_map(void *pool, unsigned long handle, | |
1732 | enum zpool_mapmode mm) | |
1733 | { | |
1734 | return z3fold_map(pool, handle); | |
1735 | } | |
1736 | static void z3fold_zpool_unmap(void *pool, unsigned long handle) | |
1737 | { | |
1738 | z3fold_unmap(pool, handle); | |
1739 | } | |
1740 | ||
1741 | static u64 z3fold_zpool_total_size(void *pool) | |
1742 | { | |
1743 | return z3fold_get_pool_size(pool) * PAGE_SIZE; | |
1744 | } | |
1745 | ||
1746 | static struct zpool_driver z3fold_zpool_driver = { | |
1747 | .type = "z3fold", | |
e818e820 | 1748 | .sleep_mapped = true, |
9a001fc1 VW |
1749 | .owner = THIS_MODULE, |
1750 | .create = z3fold_zpool_create, | |
1751 | .destroy = z3fold_zpool_destroy, | |
1752 | .malloc = z3fold_zpool_malloc, | |
1753 | .free = z3fold_zpool_free, | |
1754 | .shrink = z3fold_zpool_shrink, | |
1755 | .map = z3fold_zpool_map, | |
1756 | .unmap = z3fold_zpool_unmap, | |
1757 | .total_size = z3fold_zpool_total_size, | |
1758 | }; | |
1759 | ||
1760 | MODULE_ALIAS("zpool-z3fold"); | |
1761 | ||
1762 | static int __init init_z3fold(void) | |
1763 | { | |
1f862989 VW |
1764 | int ret; |
1765 | ||
014284a0 ML |
1766 | /* |
1767 | * Make sure the z3fold header is not larger than the page size and | |
1768 | * there has remaining spaces for its buddy. | |
1769 | */ | |
1770 | BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE - CHUNK_SIZE); | |
1f862989 VW |
1771 | ret = z3fold_mount(); |
1772 | if (ret) | |
1773 | return ret; | |
1774 | ||
9a001fc1 VW |
1775 | zpool_register_driver(&z3fold_zpool_driver); |
1776 | ||
1777 | return 0; | |
1778 | } | |
1779 | ||
1780 | static void __exit exit_z3fold(void) | |
1781 | { | |
1f862989 | 1782 | z3fold_unmount(); |
9a001fc1 VW |
1783 | zpool_unregister_driver(&z3fold_zpool_driver); |
1784 | } | |
1785 | ||
1786 | module_init(init_z3fold); | |
1787 | module_exit(exit_z3fold); | |
1788 | ||
1789 | MODULE_LICENSE("GPL"); | |
1790 | MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>"); | |
1791 | MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages"); |