1 // SPDX-License-Identifier: GPL-2.0-only <<
2 /* 1 /*
3 * z3fold.c 2 * z3fold.c
4 * 3 *
5 * Author: Vitaly Wool <vitaly.wool@konsulko.c 4 * Author: Vitaly Wool <vitaly.wool@konsulko.com>
6 * Copyright (C) 2016, Sony Mobile Communicati 5 * Copyright (C) 2016, Sony Mobile Communications Inc.
7 * 6 *
8 * This implementation is based on zbud writte 7 * This implementation is based on zbud written by Seth Jennings.
9 * 8 *
10 * z3fold is an special purpose allocator for 9 * z3fold is an special purpose allocator for storing compressed pages. It
11 * can store up to three compressed pages per 10 * can store up to three compressed pages per page which improves the
12 * compression ratio of zbud while retaining i 11 * compression ratio of zbud while retaining its main concepts (e. g. always
13 * storing an integral number of objects per p 12 * storing an integral number of objects per page) and simplicity.
14 * It still has simple and deterministic recla 13 * It still has simple and deterministic reclaim properties that make it
15 * preferable to a higher density approach (wi 14 * preferable to a higher density approach (with no requirement on integral
16 * number of object per page) when reclaim is 15 * number of object per page) when reclaim is used.
17 * 16 *
18 * As in zbud, pages are divided into "chunks" 17 * As in zbud, pages are divided into "chunks". The size of the chunks is
19 * fixed at compile time and is determined by 18 * fixed at compile time and is determined by NCHUNKS_ORDER below.
20 * 19 *
21 * z3fold doesn't export any API and is meant 20 * z3fold doesn't export any API and is meant to be used via zpool API.
22 */ 21 */
23 22
24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25 24
26 #include <linux/atomic.h> 25 #include <linux/atomic.h>
27 #include <linux/sched.h> <<
28 #include <linux/cpumask.h> <<
29 #include <linux/list.h> 26 #include <linux/list.h>
30 #include <linux/mm.h> 27 #include <linux/mm.h>
31 #include <linux/module.h> 28 #include <linux/module.h>
32 #include <linux/page-flags.h> <<
33 #include <linux/migrate.h> <<
34 #include <linux/node.h> <<
35 #include <linux/compaction.h> <<
36 #include <linux/percpu.h> <<
37 #include <linux/preempt.h> 29 #include <linux/preempt.h>
38 #include <linux/workqueue.h> <<
39 #include <linux/slab.h> 30 #include <linux/slab.h>
40 #include <linux/spinlock.h> 31 #include <linux/spinlock.h>
41 #include <linux/zpool.h> 32 #include <linux/zpool.h>
42 #include <linux/kmemleak.h> <<
43 33
>> 34 /*****************
>> 35 * Structures
>> 36 *****************/
44 /* 37 /*
45 * NCHUNKS_ORDER determines the internal alloc 38 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
46 * adjusting internal fragmentation. It also 39 * adjusting internal fragmentation. It also determines the number of
47 * freelists maintained in each pool. NCHUNKS_ 40 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
48 * allocation granularity will be in chunks of !! 41 * allocation granularity will be in chunks of size PAGE_SIZE/64. As one chunk
49 * in the beginning of an allocated page are o !! 42 * in allocated page is occupied by z3fold header, NCHUNKS will be calculated
50 * NCHUNKS will be calculated to 63 (or 62 in !! 43 * to 63 which shows the max number of free chunks in z3fold page, also there
51 * which shows the max number of free chunks i !! 44 * will be 63 freelists per pool.
52 * be 63, or 62, respectively, freelists per p <<
53 */ 45 */
54 #define NCHUNKS_ORDER 6 46 #define NCHUNKS_ORDER 6
55 47
56 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ 48 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER)
57 #define CHUNK_SIZE (1 << CHUNK_SHIFT) 49 #define CHUNK_SIZE (1 << CHUNK_SHIFT)
58 #define ZHDR_SIZE_ALIGNED round_up(sizeof(stru !! 50 #define ZHDR_SIZE_ALIGNED CHUNK_SIZE
59 #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> !! 51 #define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT)
60 #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SH !! 52
61 #define NCHUNKS (TOTAL_CHUNKS - ZHDR_C !! 53 #define BUDDY_MASK ((1 << NCHUNKS_ORDER) - 1)
62 <<
63 #define BUDDY_MASK (0x3) <<
64 #define BUDDY_SHIFT 2 <<
65 #define SLOTS_ALIGN (0x40) <<
66 54
67 /***************** <<
68 * Structures <<
69 *****************/ <<
70 struct z3fold_pool; 55 struct z3fold_pool;
>> 56 struct z3fold_ops {
>> 57 int (*evict)(struct z3fold_pool *pool, unsigned long handle);
>> 58 };
>> 59
>> 60 /**
>> 61 * struct z3fold_pool - stores metadata for each z3fold pool
>> 62 * @lock: protects all pool fields and first|last_chunk fields of any
>> 63 * z3fold page in the pool
>> 64 * @unbuddied: array of lists tracking z3fold pages that contain 2- buddies;
>> 65 * the lists each z3fold page is added to depends on the size of
>> 66 * its free region.
>> 67 * @buddied: list tracking the z3fold pages that contain 3 buddies;
>> 68 * these z3fold pages are full
>> 69 * @lru: list tracking the z3fold pages in LRU order by most recently
>> 70 * added buddy.
>> 71 * @pages_nr: number of z3fold pages in the pool.
>> 72 * @ops: pointer to a structure of user defined operations specified at
>> 73 * pool creation time.
>> 74 *
>> 75 * This structure is allocated at pool creation time and maintains metadata
>> 76 * pertaining to a particular z3fold pool.
>> 77 */
>> 78 struct z3fold_pool {
>> 79 spinlock_t lock;
>> 80 struct list_head unbuddied[NCHUNKS];
>> 81 struct list_head buddied;
>> 82 struct list_head lru;
>> 83 u64 pages_nr;
>> 84 const struct z3fold_ops *ops;
>> 85 struct zpool *zpool;
>> 86 const struct zpool_ops *zpool_ops;
>> 87 };
71 88
72 enum buddy { 89 enum buddy {
73 HEADLESS = 0, 90 HEADLESS = 0,
74 FIRST, 91 FIRST,
75 MIDDLE, 92 MIDDLE,
76 LAST, 93 LAST,
77 BUDDIES_MAX = LAST !! 94 BUDDIES_MAX
78 }; <<
79 <<
80 struct z3fold_buddy_slots { <<
81 /* <<
82 * we are using BUDDY_MASK in handle_t <<
83 * be enough slots to hold all possibl <<
84 */ <<
85 unsigned long slot[BUDDY_MASK + 1]; <<
86 unsigned long pool; /* back link */ <<
87 rwlock_t lock; <<
88 }; 95 };
89 #define HANDLE_FLAG_MASK (0x03) <<
90 96
91 /* 97 /*
92 * struct z3fold_header - z3fold page metadata !! 98 * struct z3fold_header - z3fold page metadata occupying the first chunk of each
93 * z3fold page, except fo 99 * z3fold page, except for HEADLESS pages
94 * @buddy: links the z3fold page !! 100 * @buddy: links the z3fold page into the relevant list in the pool
95 * pool <<
96 * @page_lock: per-page lock <<
97 * @refcount: reference count for th <<
98 * @work: work_struct for page l <<
99 * @slots: pointer to the structu <<
100 * @pool: pointer to the contain <<
101 * @cpu: CPU which this page "b <<
102 * @first_chunks: the size of the first 101 * @first_chunks: the size of the first buddy in chunks, 0 if free
103 * @middle_chunks: the size of the middle 102 * @middle_chunks: the size of the middle buddy in chunks, 0 if free
104 * @last_chunks: the size of the last b 103 * @last_chunks: the size of the last buddy in chunks, 0 if free
105 * @first_num: the starting number (f 104 * @first_num: the starting number (for the first handle)
106 * @mapped_count: the number of objects <<
107 */ 105 */
108 struct z3fold_header { 106 struct z3fold_header {
109 struct list_head buddy; 107 struct list_head buddy;
110 spinlock_t page_lock; <<
111 struct kref refcount; <<
112 struct work_struct work; <<
113 struct z3fold_buddy_slots *slots; <<
114 struct z3fold_pool *pool; <<
115 short cpu; <<
116 unsigned short first_chunks; 108 unsigned short first_chunks;
117 unsigned short middle_chunks; 109 unsigned short middle_chunks;
118 unsigned short last_chunks; 110 unsigned short last_chunks;
119 unsigned short start_middle; 111 unsigned short start_middle;
120 unsigned short first_num:2; !! 112 unsigned short first_num:NCHUNKS_ORDER;
121 unsigned short mapped_count:2; <<
122 unsigned short foreign_handles:2; <<
123 }; <<
124 <<
125 /** <<
126 * struct z3fold_pool - stores metadata for ea <<
127 * @name: pool name <<
128 * @lock: protects pool unbuddied lists <<
129 * @stale_lock: protects pool stale page list <<
130 * @unbuddied: per-cpu array of lists trackin <<
131 * buddies; the list each z3fold <<
132 * the size of its free region. <<
133 * @stale: list of pages marked for freei <<
134 * @pages_nr: number of z3fold pages in the <<
135 * @c_handle: cache for z3fold_buddy_slots a <<
136 * @compact_wq: workqueue for page layout back <<
137 * @release_wq: workqueue for safe page releas <<
138 * @work: work_struct for safe page rele <<
139 * <<
140 * This structure is allocated at pool creatio <<
141 * pertaining to a particular z3fold pool. <<
142 */ <<
143 struct z3fold_pool { <<
144 const char *name; <<
145 spinlock_t lock; <<
146 spinlock_t stale_lock; <<
147 struct list_head __percpu *unbuddied; <<
148 struct list_head stale; <<
149 atomic64_t pages_nr; <<
150 struct kmem_cache *c_handle; <<
151 struct workqueue_struct *compact_wq; <<
152 struct workqueue_struct *release_wq; <<
153 struct work_struct work; <<
154 }; 113 };
155 114
156 /* 115 /*
157 * Internal z3fold page flags 116 * Internal z3fold page flags
158 */ 117 */
159 enum z3fold_page_flags { 118 enum z3fold_page_flags {
160 PAGE_HEADLESS = 0, !! 119 UNDER_RECLAIM = 0,
>> 120 PAGE_HEADLESS,
161 MIDDLE_CHUNK_MAPPED, 121 MIDDLE_CHUNK_MAPPED,
162 NEEDS_COMPACTING, <<
163 PAGE_STALE, <<
164 PAGE_CLAIMED, /* by either reclaim or <<
165 PAGE_MIGRATED, /* page is migrated and <<
166 }; 122 };
167 123
168 /* <<
169 * handle flags, go under HANDLE_FLAG_MASK <<
170 */ <<
171 enum z3fold_handle_flags { <<
172 HANDLES_NOFREE = 0, <<
173 }; <<
174 <<
175 /* <<
176 * Forward declarations <<
177 */ <<
178 static struct z3fold_header *__z3fold_alloc(st <<
179 static void compact_page_work(struct work_stru <<
180 <<
181 /***************** 124 /*****************
182 * Helpers 125 * Helpers
183 *****************/ 126 *****************/
184 127
185 /* Converts an allocation size in bytes to siz 128 /* Converts an allocation size in bytes to size in z3fold chunks */
186 static int size_to_chunks(size_t size) 129 static int size_to_chunks(size_t size)
187 { 130 {
188 return (size + CHUNK_SIZE - 1) >> CHUN 131 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
189 } 132 }
190 133
191 #define for_each_unbuddied_list(_iter, _begin) 134 #define for_each_unbuddied_list(_iter, _begin) \
192 for ((_iter) = (_begin); (_iter) < NCH 135 for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
193 136
194 static inline struct z3fold_buddy_slots *alloc <<
195 <<
196 { <<
197 struct z3fold_buddy_slots *slots = kme <<
198 <<
199 <<
200 if (slots) { <<
201 /* It will be freed separately <<
202 kmemleak_not_leak(slots); <<
203 slots->pool = (unsigned long)p <<
204 rwlock_init(&slots->lock); <<
205 } <<
206 <<
207 return slots; <<
208 } <<
209 <<
210 static inline struct z3fold_pool *slots_to_poo <<
211 { <<
212 return (struct z3fold_pool *)(s->pool <<
213 } <<
214 <<
215 static inline struct z3fold_buddy_slots *handl <<
216 { <<
217 return (struct z3fold_buddy_slots *)(h <<
218 } <<
219 <<
220 /* Lock a z3fold page */ <<
221 static inline void z3fold_page_lock(struct z3f <<
222 { <<
223 spin_lock(&zhdr->page_lock); <<
224 } <<
225 <<
226 /* Try to lock a z3fold page */ <<
227 static inline int z3fold_page_trylock(struct z <<
228 { <<
229 return spin_trylock(&zhdr->page_lock); <<
230 } <<
231 <<
232 /* Unlock a z3fold page */ <<
233 static inline void z3fold_page_unlock(struct z <<
234 { <<
235 spin_unlock(&zhdr->page_lock); <<
236 } <<
237 <<
238 /* return locked z3fold page if it's not headl <<
239 static inline struct z3fold_header *get_z3fold <<
240 { <<
241 struct z3fold_buddy_slots *slots; <<
242 struct z3fold_header *zhdr; <<
243 int locked = 0; <<
244 <<
245 if (!(handle & (1 << PAGE_HEADLESS))) <<
246 slots = handle_to_slots(handle <<
247 do { <<
248 unsigned long addr; <<
249 <<
250 read_lock(&slots->lock <<
251 addr = *(unsigned long <<
252 zhdr = (struct z3fold_ <<
253 locked = z3fold_page_t <<
254 read_unlock(&slots->lo <<
255 if (locked) { <<
256 struct page *p <<
257 <<
258 if (!test_bit( <<
259 break; <<
260 z3fold_page_un <<
261 } <<
262 cpu_relax(); <<
263 } while (true); <<
264 } else { <<
265 zhdr = (struct z3fold_header * <<
266 } <<
267 <<
268 return zhdr; <<
269 } <<
270 <<
271 static inline void put_z3fold_header(struct z3 <<
272 { <<
273 struct page *page = virt_to_page(zhdr) <<
274 <<
275 if (!test_bit(PAGE_HEADLESS, &page->pr <<
276 z3fold_page_unlock(zhdr); <<
277 } <<
278 <<
279 static inline void free_handle(unsigned long h <<
280 { <<
281 struct z3fold_buddy_slots *slots; <<
282 int i; <<
283 bool is_free; <<
284 <<
285 if (WARN_ON(*(unsigned long *)handle = <<
286 return; <<
287 <<
288 slots = handle_to_slots(handle); <<
289 write_lock(&slots->lock); <<
290 *(unsigned long *)handle = 0; <<
291 <<
292 if (test_bit(HANDLES_NOFREE, &slots->p <<
293 write_unlock(&slots->lock); <<
294 return; /* simple case, nothin <<
295 } <<
296 <<
297 if (zhdr->slots != slots) <<
298 zhdr->foreign_handles--; <<
299 <<
300 is_free = true; <<
301 for (i = 0; i <= BUDDY_MASK; i++) { <<
302 if (slots->slot[i]) { <<
303 is_free = false; <<
304 break; <<
305 } <<
306 } <<
307 write_unlock(&slots->lock); <<
308 <<
309 if (is_free) { <<
310 struct z3fold_pool *pool = slo <<
311 <<
312 if (zhdr->slots == slots) <<
313 zhdr->slots = NULL; <<
314 kmem_cache_free(pool->c_handle <<
315 } <<
316 } <<
317 <<
318 /* Initializes the z3fold header of a newly al 137 /* Initializes the z3fold header of a newly allocated z3fold page */
319 static struct z3fold_header *init_z3fold_page( !! 138 static struct z3fold_header *init_z3fold_page(struct page *page)
320 struct <<
321 { 139 {
322 struct z3fold_header *zhdr = page_addr 140 struct z3fold_header *zhdr = page_address(page);
323 struct z3fold_buddy_slots *slots; <<
324 141
>> 142 INIT_LIST_HEAD(&page->lru);
>> 143 clear_bit(UNDER_RECLAIM, &page->private);
325 clear_bit(PAGE_HEADLESS, &page->privat 144 clear_bit(PAGE_HEADLESS, &page->private);
326 clear_bit(MIDDLE_CHUNK_MAPPED, &page-> 145 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
327 clear_bit(NEEDS_COMPACTING, &page->pri <<
328 clear_bit(PAGE_STALE, &page->private); <<
329 clear_bit(PAGE_CLAIMED, &page->private <<
330 clear_bit(PAGE_MIGRATED, &page->privat <<
331 if (headless) <<
332 return zhdr; <<
333 <<
334 slots = alloc_slots(pool, gfp); <<
335 if (!slots) <<
336 return NULL; <<
337 146
338 memset(zhdr, 0, sizeof(*zhdr)); !! 147 zhdr->first_chunks = 0;
339 spin_lock_init(&zhdr->page_lock); !! 148 zhdr->middle_chunks = 0;
340 kref_init(&zhdr->refcount); !! 149 zhdr->last_chunks = 0;
341 zhdr->cpu = -1; !! 150 zhdr->first_num = 0;
342 zhdr->slots = slots; !! 151 zhdr->start_middle = 0;
343 zhdr->pool = pool; <<
344 INIT_LIST_HEAD(&zhdr->buddy); 152 INIT_LIST_HEAD(&zhdr->buddy);
345 INIT_WORK(&zhdr->work, compact_page_wo <<
346 return zhdr; 153 return zhdr;
347 } 154 }
348 155
349 /* Resets the struct page fields and frees the 156 /* Resets the struct page fields and frees the page */
350 static void free_z3fold_page(struct page *page !! 157 static void free_z3fold_page(struct z3fold_header *zhdr)
351 { <<
352 if (!headless) { <<
353 lock_page(page); <<
354 __ClearPageMovable(page); <<
355 unlock_page(page); <<
356 } <<
357 __free_page(page); <<
358 } <<
359 <<
360 /* Helper function to build the index */ <<
361 static inline int __idx(struct z3fold_header * <<
362 { 158 {
363 return (bud + zhdr->first_num) & BUDDY !! 159 __free_page(virt_to_page(zhdr));
364 } 160 }
365 161
366 /* 162 /*
367 * Encodes the handle of a particular buddy wi !! 163 * Encodes the handle of a particular buddy within a z3fold page
368 * Zhdr->page_lock should be held as this func !! 164 * Pool lock should be held as this function accesses first_num
369 * if bud != HEADLESS. <<
370 */ 165 */
371 static unsigned long __encode_handle(struct z3 <<
372 struct z3fold_ <<
373 enum buddy bud <<
374 { <<
375 unsigned long h = (unsigned long)zhdr; <<
376 int idx = 0; <<
377 <<
378 /* <<
379 * For a headless page, its handle is <<
380 * PAGE_HEADLESS bit set <<
381 */ <<
382 if (bud == HEADLESS) <<
383 return h | (1 << PAGE_HEADLESS <<
384 <<
385 /* otherwise, return pointer to encode <<
386 idx = __idx(zhdr, bud); <<
387 h += idx; <<
388 if (bud == LAST) <<
389 h |= (zhdr->last_chunks << BUD <<
390 <<
391 write_lock(&slots->lock); <<
392 slots->slot[idx] = h; <<
393 write_unlock(&slots->lock); <<
394 return (unsigned long)&slots->slot[idx <<
395 } <<
396 <<
397 static unsigned long encode_handle(struct z3fo 166 static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud)
398 { 167 {
399 return __encode_handle(zhdr, zhdr->slo !! 168 unsigned long handle;
400 } <<
401 <<
402 /* only for LAST bud, returns zero otherwise * <<
403 static unsigned short handle_to_chunks(unsigne <<
404 { <<
405 struct z3fold_buddy_slots *slots = han <<
406 unsigned long addr; <<
407 <<
408 read_lock(&slots->lock); <<
409 addr = *(unsigned long *)handle; <<
410 read_unlock(&slots->lock); <<
411 return (addr & ~PAGE_MASK) >> BUDDY_SH <<
412 } <<
413 <<
414 /* <<
415 * (handle & BUDDY_MASK) < zhdr->first_num is <<
416 * but that doesn't matter. because the maski <<
417 * correct buddy number. <<
418 */ <<
419 static enum buddy handle_to_buddy(unsigned lon <<
420 { <<
421 struct z3fold_header *zhdr; <<
422 struct z3fold_buddy_slots *slots = han <<
423 unsigned long addr; <<
424 <<
425 read_lock(&slots->lock); <<
426 WARN_ON(handle & (1 << PAGE_HEADLESS)) <<
427 addr = *(unsigned long *)handle; <<
428 read_unlock(&slots->lock); <<
429 zhdr = (struct z3fold_header *)(addr & <<
430 return (addr - zhdr->first_num) & BUDD <<
431 } <<
432 <<
433 static inline struct z3fold_pool *zhdr_to_pool <<
434 { <<
435 return zhdr->pool; <<
436 } <<
437 <<
438 static void __release_z3fold_page(struct z3fol <<
439 { <<
440 struct page *page = virt_to_page(zhdr) <<
441 struct z3fold_pool *pool = zhdr_to_poo <<
442 <<
443 WARN_ON(!list_empty(&zhdr->buddy)); <<
444 set_bit(PAGE_STALE, &page->private); <<
445 clear_bit(NEEDS_COMPACTING, &page->pri <<
446 spin_lock(&pool->lock); <<
447 spin_unlock(&pool->lock); <<
448 <<
449 if (locked) <<
450 z3fold_page_unlock(zhdr); <<
451 <<
452 spin_lock(&pool->stale_lock); <<
453 list_add(&zhdr->buddy, &pool->stale); <<
454 queue_work(pool->release_wq, &pool->wo <<
455 spin_unlock(&pool->stale_lock); <<
456 <<
457 atomic64_dec(&pool->pages_nr); <<
458 } <<
459 <<
460 static void release_z3fold_page_locked(struct <<
461 { <<
462 struct z3fold_header *zhdr = container <<
463 <<
464 WARN_ON(z3fold_page_trylock(zhdr)); <<
465 __release_z3fold_page(zhdr, true); <<
466 } <<
467 <<
468 static void release_z3fold_page_locked_list(st <<
469 { <<
470 struct z3fold_header *zhdr = container <<
471 <<
472 struct z3fold_pool *pool = zhdr_to_poo <<
473 <<
474 spin_lock(&pool->lock); <<
475 list_del_init(&zhdr->buddy); <<
476 spin_unlock(&pool->lock); <<
477 <<
478 WARN_ON(z3fold_page_trylock(zhdr)); <<
479 __release_z3fold_page(zhdr, true); <<
480 } <<
481 169
482 static inline int put_z3fold_locked(struct z3f !! 170 handle = (unsigned long)zhdr;
483 { !! 171 if (bud != HEADLESS)
484 return kref_put(&zhdr->refcount, relea !! 172 handle += (bud + zhdr->first_num) & BUDDY_MASK;
>> 173 return handle;
485 } 174 }
486 175
487 static inline int put_z3fold_locked_list(struc !! 176 /* Returns the z3fold page where a given handle is stored */
>> 177 static struct z3fold_header *handle_to_z3fold_header(unsigned long handle)
488 { 178 {
489 return kref_put(&zhdr->refcount, relea !! 179 return (struct z3fold_header *)(handle & PAGE_MASK);
490 } 180 }
491 181
492 static void free_pages_work(struct work_struct !! 182 /* Returns buddy number */
>> 183 static enum buddy handle_to_buddy(unsigned long handle)
493 { 184 {
494 struct z3fold_pool *pool = container_o !! 185 struct z3fold_header *zhdr = handle_to_z3fold_header(handle);
495 !! 186 return (handle - zhdr->first_num) & BUDDY_MASK;
496 spin_lock(&pool->stale_lock); <<
497 while (!list_empty(&pool->stale)) { <<
498 struct z3fold_header *zhdr = l <<
499 <<
500 struct page *page = virt_to_pa <<
501 <<
502 list_del(&zhdr->buddy); <<
503 if (WARN_ON(!test_bit(PAGE_STA <<
504 continue; <<
505 spin_unlock(&pool->stale_lock) <<
506 cancel_work_sync(&zhdr->work); <<
507 free_z3fold_page(page, false); <<
508 cond_resched(); <<
509 spin_lock(&pool->stale_lock); <<
510 } <<
511 spin_unlock(&pool->stale_lock); <<
512 } 187 }
513 188
514 /* 189 /*
515 * Returns the number of free chunks in a z3fo 190 * Returns the number of free chunks in a z3fold page.
516 * NB: can't be used with HEADLESS pages. 191 * NB: can't be used with HEADLESS pages.
517 */ 192 */
518 static int num_free_chunks(struct z3fold_heade 193 static int num_free_chunks(struct z3fold_header *zhdr)
519 { 194 {
520 int nfree; 195 int nfree;
521 /* 196 /*
522 * If there is a middle object, pick u 197 * If there is a middle object, pick up the bigger free space
523 * either before or after it. Otherwis 198 * either before or after it. Otherwise just subtract the number
524 * of chunks occupied by the first and 199 * of chunks occupied by the first and the last objects.
525 */ 200 */
526 if (zhdr->middle_chunks != 0) { 201 if (zhdr->middle_chunks != 0) {
527 int nfree_before = zhdr->first 202 int nfree_before = zhdr->first_chunks ?
528 0 : zhdr->start_middle !! 203 0 : zhdr->start_middle - 1;
529 int nfree_after = zhdr->last_c 204 int nfree_after = zhdr->last_chunks ?
530 0 : TOTAL_CHUNKS - !! 205 0 : NCHUNKS - zhdr->start_middle - zhdr->middle_chunks;
531 (zhdr->start_m <<
532 nfree = max(nfree_before, nfre 206 nfree = max(nfree_before, nfree_after);
533 } else 207 } else
534 nfree = NCHUNKS - zhdr->first_ 208 nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
535 return nfree; 209 return nfree;
536 } 210 }
537 211
538 /* Add to the appropriate unbuddied list */ !! 212 /*****************
539 static inline void add_to_unbuddied(struct z3f <<
540 struct z3fold_ <<
541 { <<
542 if (zhdr->first_chunks == 0 || zhdr->l <<
543 zhdr->middle_chunks == <<
544 struct list_head *unbuddied; <<
545 int freechunks = num_free_chun <<
546 <<
547 migrate_disable(); <<
548 unbuddied = this_cpu_ptr(pool- <<
549 spin_lock(&pool->lock); <<
550 list_add(&zhdr->buddy, &unbudd <<
551 spin_unlock(&pool->lock); <<
552 zhdr->cpu = smp_processor_id() <<
553 migrate_enable(); <<
554 } <<
555 } <<
556 <<
557 static inline enum buddy get_free_buddy(struct <<
558 { <<
559 enum buddy bud = HEADLESS; <<
560 <<
561 if (zhdr->middle_chunks) { <<
562 if (!zhdr->first_chunks && <<
563 chunks <= zhdr->start_midd <<
564 bud = FIRST; <<
565 else if (!zhdr->last_chunks) <<
566 bud = LAST; <<
567 } else { <<
568 if (!zhdr->first_chunks) <<
569 bud = FIRST; <<
570 else if (!zhdr->last_chunks) <<
571 bud = LAST; <<
572 else <<
573 bud = MIDDLE; <<
574 } <<
575 <<
576 return bud; <<
577 } <<
578 <<
579 static inline void *mchunk_memmove(struct z3fo <<
580 unsigned short <<
581 { <<
582 void *beg = zhdr; <<
583 return memmove(beg + (dst_chunk << CHU <<
584 beg + (zhdr->start_midd <<
585 zhdr->middle_chunks << <<
586 } <<
587 <<
588 static inline bool buddy_single(struct z3fold_ <<
589 { <<
590 return !((zhdr->first_chunks && zhdr-> <<
591 (zhdr->first_chunks && <<
592 (zhdr->middle_chunks & <<
593 } <<
594 <<
595 static struct z3fold_header *compact_single_bu <<
596 { <<
597 struct z3fold_pool *pool = zhdr_to_poo <<
598 void *p = zhdr; <<
599 unsigned long old_handle = 0; <<
600 size_t sz = 0; <<
601 struct z3fold_header *new_zhdr = NULL; <<
602 int first_idx = __idx(zhdr, FIRST); <<
603 int middle_idx = __idx(zhdr, MIDDLE); <<
604 int last_idx = __idx(zhdr, LAST); <<
605 unsigned short *moved_chunks = NULL; <<
606 <<
607 /* <<
608 * No need to protect slots here -- al <<
609 * the page lock is already taken <<
610 */ <<
611 if (zhdr->first_chunks && zhdr->slots- <<
612 p += ZHDR_SIZE_ALIGNED; <<
613 sz = zhdr->first_chunks << CHU <<
614 old_handle = (unsigned long)&z <<
615 moved_chunks = &zhdr->first_ch <<
616 } else if (zhdr->middle_chunks && zhdr <<
617 p += zhdr->start_middle << CHU <<
618 sz = zhdr->middle_chunks << CH <<
619 old_handle = (unsigned long)&z <<
620 moved_chunks = &zhdr->middle_c <<
621 } else if (zhdr->last_chunks && zhdr-> <<
622 p += PAGE_SIZE - (zhdr->last_c <<
623 sz = zhdr->last_chunks << CHUN <<
624 old_handle = (unsigned long)&z <<
625 moved_chunks = &zhdr->last_chu <<
626 } <<
627 <<
628 if (sz > 0) { <<
629 enum buddy new_bud = HEADLESS; <<
630 short chunks = size_to_chunks( <<
631 void *q; <<
632 <<
633 new_zhdr = __z3fold_alloc(pool <<
634 if (!new_zhdr) <<
635 return NULL; <<
636 <<
637 if (WARN_ON(new_zhdr == zhdr)) <<
638 goto out_fail; <<
639 <<
640 new_bud = get_free_buddy(new_z <<
641 q = new_zhdr; <<
642 switch (new_bud) { <<
643 case FIRST: <<
644 new_zhdr->first_chunks <<
645 q += ZHDR_SIZE_ALIGNED <<
646 break; <<
647 case MIDDLE: <<
648 new_zhdr->middle_chunk <<
649 new_zhdr->start_middle <<
650 new_zhdr->firs <<
651 q += new_zhdr->start_m <<
652 break; <<
653 case LAST: <<
654 new_zhdr->last_chunks <<
655 q += PAGE_SIZE - (new_ <<
656 break; <<
657 default: <<
658 goto out_fail; <<
659 } <<
660 new_zhdr->foreign_handles++; <<
661 memcpy(q, p, sz); <<
662 write_lock(&zhdr->slots->lock) <<
663 *(unsigned long *)old_handle = <<
664 __idx(new_zhdr, new_bu <<
665 if (new_bud == LAST) <<
666 *(unsigned long *)old_ <<
667 (new_z <<
668 write_unlock(&zhdr->slots->loc <<
669 add_to_unbuddied(pool, new_zhd <<
670 z3fold_page_unlock(new_zhdr); <<
671 <<
672 *moved_chunks = 0; <<
673 } <<
674 <<
675 return new_zhdr; <<
676 <<
677 out_fail: <<
678 if (new_zhdr && !put_z3fold_locked(new <<
679 add_to_unbuddied(pool, new_zhd <<
680 z3fold_page_unlock(new_zhdr); <<
681 } <<
682 return NULL; <<
683 <<
684 } <<
685 <<
686 #define BIG_CHUNK_GAP 3 <<
687 /* Has to be called with lock held */ <<
688 static int z3fold_compact_page(struct z3fold_h <<
689 { <<
690 struct page *page = virt_to_page(zhdr) <<
691 <<
692 if (test_bit(MIDDLE_CHUNK_MAPPED, &pag <<
693 return 0; /* can't move middle <<
694 <<
695 if (unlikely(PageIsolated(page))) <<
696 return 0; <<
697 <<
698 if (zhdr->middle_chunks == 0) <<
699 return 0; /* nothing to compac <<
700 <<
701 if (zhdr->first_chunks == 0 && zhdr->l <<
702 /* move to the beginning */ <<
703 mchunk_memmove(zhdr, ZHDR_CHUN <<
704 zhdr->first_chunks = zhdr->mid <<
705 zhdr->middle_chunks = 0; <<
706 zhdr->start_middle = 0; <<
707 zhdr->first_num++; <<
708 return 1; <<
709 } <<
710 <<
711 /* <<
712 * moving data is expensive, so let's <<
713 * there's substantial gain (at least <<
714 */ <<
715 if (zhdr->first_chunks != 0 && zhdr->l <<
716 zhdr->start_middle - (zhdr->first_ <<
717 BIG_CHUNK_GAP) { <<
718 mchunk_memmove(zhdr, zhdr->fir <<
719 zhdr->start_middle = zhdr->fir <<
720 return 1; <<
721 } else if (zhdr->last_chunks != 0 && z <<
722 TOTAL_CHUNKS - (zhdr->last_ <<
723 + zhdr <<
724 BIG_CHUNK_GAP) { <<
725 unsigned short new_start = TOT <<
726 zhdr->middle_chunks; <<
727 mchunk_memmove(zhdr, new_start <<
728 zhdr->start_middle = new_start <<
729 return 1; <<
730 } <<
731 <<
732 return 0; <<
733 } <<
734 <<
735 static void do_compact_page(struct z3fold_head <<
736 { <<
737 struct z3fold_pool *pool = zhdr_to_poo <<
738 struct page *page; <<
739 <<
740 page = virt_to_page(zhdr); <<
741 if (locked) <<
742 WARN_ON(z3fold_page_trylock(zh <<
743 else <<
744 z3fold_page_lock(zhdr); <<
745 if (WARN_ON(!test_and_clear_bit(NEEDS_ <<
746 z3fold_page_unlock(zhdr); <<
747 return; <<
748 } <<
749 spin_lock(&pool->lock); <<
750 list_del_init(&zhdr->buddy); <<
751 spin_unlock(&pool->lock); <<
752 <<
753 if (put_z3fold_locked(zhdr)) <<
754 return; <<
755 <<
756 if (test_bit(PAGE_STALE, &page->privat <<
757 test_and_set_bit(PAGE_CLAIMED, &pa <<
758 z3fold_page_unlock(zhdr); <<
759 return; <<
760 } <<
761 <<
762 if (!zhdr->foreign_handles && buddy_si <<
763 zhdr->mapped_count == 0 && compact <<
764 if (!put_z3fold_locked(zhdr)) <<
765 clear_bit(PAGE_CLAIMED <<
766 z3fold_page_unlock(zhd <<
767 } <<
768 return; <<
769 } <<
770 <<
771 z3fold_compact_page(zhdr); <<
772 add_to_unbuddied(pool, zhdr); <<
773 clear_bit(PAGE_CLAIMED, &page->private <<
774 z3fold_page_unlock(zhdr); <<
775 } <<
776 <<
777 static void compact_page_work(struct work_stru <<
778 { <<
779 struct z3fold_header *zhdr = container <<
780 <<
781 <<
782 do_compact_page(zhdr, false); <<
783 } <<
784 <<
785 /* returns _locked_ z3fold page header or NULL <<
786 static inline struct z3fold_header *__z3fold_a <<
787 <<
788 { <<
789 struct z3fold_header *zhdr = NULL; <<
790 struct page *page; <<
791 struct list_head *unbuddied; <<
792 int chunks = size_to_chunks(size), i; <<
793 <<
794 lookup: <<
795 migrate_disable(); <<
796 /* First, try to find an unbuddied z3f <<
797 unbuddied = this_cpu_ptr(pool->unbuddi <<
798 for_each_unbuddied_list(i, chunks) { <<
799 struct list_head *l = &unbuddi <<
800 <<
801 zhdr = list_first_entry_or_nul <<
802 struct <<
803 <<
804 if (!zhdr) <<
805 continue; <<
806 <<
807 /* Re-check under lock. */ <<
808 spin_lock(&pool->lock); <<
809 if (unlikely(zhdr != list_firs <<
810 <<
811 !z3fold_page_trylock(zhdr) <<
812 spin_unlock(&pool->loc <<
813 zhdr = NULL; <<
814 migrate_enable(); <<
815 if (can_sleep) <<
816 cond_resched() <<
817 goto lookup; <<
818 } <<
819 list_del_init(&zhdr->buddy); <<
820 zhdr->cpu = -1; <<
821 spin_unlock(&pool->lock); <<
822 <<
823 page = virt_to_page(zhdr); <<
824 if (test_bit(NEEDS_COMPACTING, <<
825 test_bit(PAGE_CLAIMED, &pa <<
826 z3fold_page_unlock(zhd <<
827 zhdr = NULL; <<
828 migrate_enable(); <<
829 if (can_sleep) <<
830 cond_resched() <<
831 goto lookup; <<
832 } <<
833 <<
834 /* <<
835 * this page could not be remo <<
836 * list while pool lock was he <<
837 * page lock so kref_put could <<
838 * we got here, so it's safe t <<
839 */ <<
840 kref_get(&zhdr->refcount); <<
841 break; <<
842 } <<
843 migrate_enable(); <<
844 <<
845 if (!zhdr) { <<
846 int cpu; <<
847 <<
848 /* look for _exact_ match on o <<
849 for_each_online_cpu(cpu) { <<
850 struct list_head *l; <<
851 <<
852 unbuddied = per_cpu_pt <<
853 spin_lock(&pool->lock) <<
854 l = &unbuddied[chunks] <<
855 <<
856 zhdr = list_first_entr <<
857 <<
858 <<
859 if (!zhdr || !z3fold_p <<
860 spin_unlock(&p <<
861 zhdr = NULL; <<
862 continue; <<
863 } <<
864 list_del_init(&zhdr->b <<
865 zhdr->cpu = -1; <<
866 spin_unlock(&pool->loc <<
867 <<
868 page = virt_to_page(zh <<
869 if (test_bit(NEEDS_COM <<
870 test_bit(PAGE_CLAI <<
871 z3fold_page_un <<
872 zhdr = NULL; <<
873 if (can_sleep) <<
874 cond_r <<
875 continue; <<
876 } <<
877 kref_get(&zhdr->refcou <<
878 break; <<
879 } <<
880 } <<
881 <<
882 if (zhdr && !zhdr->slots) { <<
883 zhdr->slots = alloc_slots(pool <<
884 if (!zhdr->slots) <<
885 goto out_fail; <<
886 } <<
887 return zhdr; <<
888 <<
889 out_fail: <<
890 if (!put_z3fold_locked(zhdr)) { <<
891 add_to_unbuddied(pool, zhdr); <<
892 z3fold_page_unlock(zhdr); <<
893 } <<
894 return NULL; <<
895 } <<
896 <<
897 /* <<
898 * API Functions 213 * API Functions
899 */ !! 214 *****************/
900 <<
901 /** 215 /**
902 * z3fold_create_pool() - create a new z3fold 216 * z3fold_create_pool() - create a new z3fold pool
903 * @name: pool name <<
904 * @gfp: gfp flags when allocating the 217 * @gfp: gfp flags when allocating the z3fold pool structure
>> 218 * @ops: user-defined operations for the z3fold pool
905 * 219 *
906 * Return: pointer to the new z3fold pool or N 220 * Return: pointer to the new z3fold pool or NULL if the metadata allocation
907 * failed. 221 * failed.
908 */ 222 */
909 static struct z3fold_pool *z3fold_create_pool( !! 223 static struct z3fold_pool *z3fold_create_pool(gfp_t gfp,
>> 224 const struct z3fold_ops *ops)
910 { 225 {
911 struct z3fold_pool *pool = NULL; !! 226 struct z3fold_pool *pool;
912 int i, cpu; !! 227 int i;
913 228
914 pool = kzalloc(sizeof(struct z3fold_po 229 pool = kzalloc(sizeof(struct z3fold_pool), gfp);
915 if (!pool) 230 if (!pool)
916 goto out; !! 231 return NULL;
917 pool->c_handle = kmem_cache_create("z3 <<
918 sizeof(struct <<
919 SLOTS_ALIGN, 0 <<
920 if (!pool->c_handle) <<
921 goto out_c; <<
922 spin_lock_init(&pool->lock); 232 spin_lock_init(&pool->lock);
923 spin_lock_init(&pool->stale_lock); !! 233 for_each_unbuddied_list(i, 0)
924 pool->unbuddied = __alloc_percpu(sizeo !! 234 INIT_LIST_HEAD(&pool->unbuddied[i]);
925 __ali !! 235 INIT_LIST_HEAD(&pool->buddied);
926 if (!pool->unbuddied) !! 236 INIT_LIST_HEAD(&pool->lru);
927 goto out_pool; !! 237 pool->pages_nr = 0;
928 for_each_possible_cpu(cpu) { !! 238 pool->ops = ops;
929 struct list_head *unbuddied = <<
930 per_cpu_ptr(po <<
931 for_each_unbuddied_list(i, 0) <<
932 INIT_LIST_HEAD(&unbudd <<
933 } <<
934 INIT_LIST_HEAD(&pool->stale); <<
935 atomic64_set(&pool->pages_nr, 0); <<
936 pool->name = name; <<
937 pool->compact_wq = create_singlethread <<
938 if (!pool->compact_wq) <<
939 goto out_unbuddied; <<
940 pool->release_wq = create_singlethread <<
941 if (!pool->release_wq) <<
942 goto out_wq; <<
943 INIT_WORK(&pool->work, free_pages_work <<
944 return pool; 239 return pool;
945 <<
946 out_wq: <<
947 destroy_workqueue(pool->compact_wq); <<
948 out_unbuddied: <<
949 free_percpu(pool->unbuddied); <<
950 out_pool: <<
951 kmem_cache_destroy(pool->c_handle); <<
952 out_c: <<
953 kfree(pool); <<
954 out: <<
955 return NULL; <<
956 } 240 }
957 241
958 /** 242 /**
959 * z3fold_destroy_pool() - destroys an existin 243 * z3fold_destroy_pool() - destroys an existing z3fold pool
960 * @pool: the z3fold pool to be destroye 244 * @pool: the z3fold pool to be destroyed
961 * 245 *
962 * The pool should be emptied before this func 246 * The pool should be emptied before this function is called.
963 */ 247 */
964 static void z3fold_destroy_pool(struct z3fold_ 248 static void z3fold_destroy_pool(struct z3fold_pool *pool)
965 { 249 {
966 kmem_cache_destroy(pool->c_handle); <<
967 <<
968 /* <<
969 * We need to destroy pool->compact_wq <<
970 * as any pending work on pool->compac <<
971 * queue_work(pool->release_wq, &pool- <<
972 * <<
973 * There are still outstanding pages u <<
974 * so we cannot unregister migration u <<
975 */ <<
976 <<
977 destroy_workqueue(pool->compact_wq); <<
978 destroy_workqueue(pool->release_wq); <<
979 free_percpu(pool->unbuddied); <<
980 kfree(pool); 250 kfree(pool);
981 } 251 }
982 252
983 static const struct movable_operations z3fold_ !! 253 /* Has to be called with lock held */
>> 254 static int z3fold_compact_page(struct z3fold_header *zhdr)
>> 255 {
>> 256 struct page *page = virt_to_page(zhdr);
>> 257 void *beg = zhdr;
>> 258
>> 259
>> 260 if (!test_bit(MIDDLE_CHUNK_MAPPED, &page->private) &&
>> 261 zhdr->middle_chunks != 0 &&
>> 262 zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
>> 263 memmove(beg + ZHDR_SIZE_ALIGNED,
>> 264 beg + (zhdr->start_middle << CHUNK_SHIFT),
>> 265 zhdr->middle_chunks << CHUNK_SHIFT);
>> 266 zhdr->first_chunks = zhdr->middle_chunks;
>> 267 zhdr->middle_chunks = 0;
>> 268 zhdr->start_middle = 0;
>> 269 zhdr->first_num++;
>> 270 return 1;
>> 271 }
>> 272 return 0;
>> 273 }
984 274
985 /** 275 /**
986 * z3fold_alloc() - allocates a region of a gi 276 * z3fold_alloc() - allocates a region of a given size
987 * @pool: z3fold pool from which to allo 277 * @pool: z3fold pool from which to allocate
988 * @size: size in bytes of the desired a 278 * @size: size in bytes of the desired allocation
989 * @gfp: gfp flags used if the pool nee 279 * @gfp: gfp flags used if the pool needs to grow
990 * @handle: handle of the new allocation 280 * @handle: handle of the new allocation
991 * 281 *
992 * This function will attempt to find a free r 282 * This function will attempt to find a free region in the pool large enough to
993 * satisfy the allocation request. A search o 283 * satisfy the allocation request. A search of the unbuddied lists is
994 * performed first. If no suitable free region 284 * performed first. If no suitable free region is found, then a new page is
995 * allocated and added to the pool to satisfy 285 * allocated and added to the pool to satisfy the request.
996 * 286 *
>> 287 * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used
>> 288 * as z3fold pool pages.
>> 289 *
997 * Return: 0 if success and handle is set, oth 290 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
998 * gfp arguments are invalid or -ENOMEM if the 291 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
999 * a new page. 292 * a new page.
1000 */ 293 */
1001 static int z3fold_alloc(struct z3fold_pool *p 294 static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
1002 unsigned long *handle 295 unsigned long *handle)
1003 { 296 {
1004 int chunks = size_to_chunks(size); !! 297 int chunks = 0, i, freechunks;
1005 struct z3fold_header *zhdr = NULL; 298 struct z3fold_header *zhdr = NULL;
1006 struct page *page = NULL; <<
1007 enum buddy bud; 299 enum buddy bud;
1008 bool can_sleep = gfpflags_allow_block !! 300 struct page *page;
1009 301
1010 if (!size || (gfp & __GFP_HIGHMEM)) 302 if (!size || (gfp & __GFP_HIGHMEM))
1011 return -EINVAL; 303 return -EINVAL;
1012 304
1013 if (size > PAGE_SIZE) 305 if (size > PAGE_SIZE)
1014 return -ENOSPC; 306 return -ENOSPC;
1015 307
1016 if (size > PAGE_SIZE - ZHDR_SIZE_ALIG 308 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
1017 bud = HEADLESS; 309 bud = HEADLESS;
1018 else { 310 else {
1019 retry: !! 311 chunks = size_to_chunks(size);
1020 zhdr = __z3fold_alloc(pool, s !! 312 spin_lock(&pool->lock);
1021 if (zhdr) { !! 313
1022 bud = get_free_buddy( !! 314 /* First, try to find an unbuddied z3fold page. */
1023 if (bud == HEADLESS) !! 315 zhdr = NULL;
1024 if (!put_z3fo !! 316 for_each_unbuddied_list(i, chunks) {
1025 z3fol !! 317 if (!list_empty(&pool->unbuddied[i])) {
1026 pr_err("No fr !! 318 zhdr = list_first_entry(&pool->unbuddied[i],
1027 WARN_ON(1); !! 319 struct z3fold_header, buddy);
1028 goto retry; !! 320 page = virt_to_page(zhdr);
>> 321 if (zhdr->first_chunks == 0) {
>> 322 if (zhdr->middle_chunks != 0 &&
>> 323 chunks >= zhdr->start_middle)
>> 324 bud = LAST;
>> 325 else
>> 326 bud = FIRST;
>> 327 } else if (zhdr->last_chunks == 0)
>> 328 bud = LAST;
>> 329 else if (zhdr->middle_chunks == 0)
>> 330 bud = MIDDLE;
>> 331 else {
>> 332 pr_err("No free chunks in unbuddied\n");
>> 333 WARN_ON(1);
>> 334 continue;
>> 335 }
>> 336 list_del(&zhdr->buddy);
>> 337 goto found;
1029 } 338 }
1030 page = virt_to_page(z <<
1031 goto found; <<
1032 } 339 }
1033 bud = FIRST; 340 bud = FIRST;
>> 341 spin_unlock(&pool->lock);
1034 } 342 }
1035 343
>> 344 /* Couldn't find unbuddied z3fold page, create new one */
1036 page = alloc_page(gfp); 345 page = alloc_page(gfp);
1037 if (!page) 346 if (!page)
1038 return -ENOMEM; 347 return -ENOMEM;
1039 !! 348 spin_lock(&pool->lock);
1040 zhdr = init_z3fold_page(page, bud == !! 349 pool->pages_nr++;
1041 if (!zhdr) { !! 350 zhdr = init_z3fold_page(page);
1042 __free_page(page); <<
1043 return -ENOMEM; <<
1044 } <<
1045 atomic64_inc(&pool->pages_nr); <<
1046 351
1047 if (bud == HEADLESS) { 352 if (bud == HEADLESS) {
1048 set_bit(PAGE_HEADLESS, &page- 353 set_bit(PAGE_HEADLESS, &page->private);
1049 goto headless; 354 goto headless;
1050 } 355 }
1051 if (can_sleep) { <<
1052 lock_page(page); <<
1053 __SetPageMovable(page, &z3fol <<
1054 unlock_page(page); <<
1055 } else { <<
1056 WARN_ON(!trylock_page(page)); <<
1057 __SetPageMovable(page, &z3fol <<
1058 unlock_page(page); <<
1059 } <<
1060 z3fold_page_lock(zhdr); <<
1061 356
1062 found: 357 found:
1063 if (bud == FIRST) 358 if (bud == FIRST)
1064 zhdr->first_chunks = chunks; 359 zhdr->first_chunks = chunks;
1065 else if (bud == LAST) 360 else if (bud == LAST)
1066 zhdr->last_chunks = chunks; 361 zhdr->last_chunks = chunks;
1067 else { 362 else {
1068 zhdr->middle_chunks = chunks; 363 zhdr->middle_chunks = chunks;
1069 zhdr->start_middle = zhdr->fi !! 364 zhdr->start_middle = zhdr->first_chunks + 1;
>> 365 }
>> 366
>> 367 if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 ||
>> 368 zhdr->middle_chunks == 0) {
>> 369 /* Add to unbuddied list */
>> 370 freechunks = num_free_chunks(zhdr);
>> 371 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
>> 372 } else {
>> 373 /* Add to buddied list */
>> 374 list_add(&zhdr->buddy, &pool->buddied);
1070 } 375 }
1071 add_to_unbuddied(pool, zhdr); <<
1072 376
1073 headless: 377 headless:
1074 spin_lock(&pool->lock); !! 378 /* Add/move z3fold page to beginning of LRU */
>> 379 if (!list_empty(&page->lru))
>> 380 list_del(&page->lru);
>> 381
>> 382 list_add(&page->lru, &pool->lru);
>> 383
1075 *handle = encode_handle(zhdr, bud); 384 *handle = encode_handle(zhdr, bud);
1076 spin_unlock(&pool->lock); 385 spin_unlock(&pool->lock);
1077 if (bud != HEADLESS) <<
1078 z3fold_page_unlock(zhdr); <<
1079 386
1080 return 0; 387 return 0;
1081 } 388 }
1082 389
1083 /** 390 /**
1084 * z3fold_free() - frees the allocation assoc 391 * z3fold_free() - frees the allocation associated with the given handle
1085 * @pool: pool in which the allocation 392 * @pool: pool in which the allocation resided
1086 * @handle: handle associated with the al 393 * @handle: handle associated with the allocation returned by z3fold_alloc()
1087 * 394 *
1088 * In the case that the z3fold page in which 395 * In the case that the z3fold page in which the allocation resides is under
1089 * reclaim, as indicated by the PAGE_CLAIMED !! 396 * reclaim, as indicated by the PG_reclaim flag being set, this function
1090 * only sets the first|middle|last_chunks to !! 397 * only sets the first|last_chunks to 0. The page is actually freed
1091 * once all buddies are evicted (see z3fold_r !! 398 * once both buddies are evicted (see z3fold_reclaim_page() below).
1092 */ 399 */
1093 static void z3fold_free(struct z3fold_pool *p 400 static void z3fold_free(struct z3fold_pool *pool, unsigned long handle)
1094 { 401 {
1095 struct z3fold_header *zhdr; 402 struct z3fold_header *zhdr;
>> 403 int freechunks;
1096 struct page *page; 404 struct page *page;
1097 enum buddy bud; 405 enum buddy bud;
1098 bool page_claimed; <<
1099 406
1100 zhdr = get_z3fold_header(handle); !! 407 spin_lock(&pool->lock);
>> 408 zhdr = handle_to_z3fold_header(handle);
1101 page = virt_to_page(zhdr); 409 page = virt_to_page(zhdr);
1102 page_claimed = test_and_set_bit(PAGE_ <<
1103 410
1104 if (test_bit(PAGE_HEADLESS, &page->pr 411 if (test_bit(PAGE_HEADLESS, &page->private)) {
1105 /* if a headless page is unde !! 412 /* HEADLESS page stored */
1106 * NB: we use test_and_set_bi !! 413 bud = HEADLESS;
1107 * has not been set before, w !! 414 } else {
1108 * immediately so we don't ca !! 415 bud = handle_to_buddy(handle);
1109 */ !! 416
1110 if (!page_claimed) { !! 417 switch (bud) {
1111 put_z3fold_header(zhd !! 418 case FIRST:
1112 free_z3fold_page(page !! 419 zhdr->first_chunks = 0;
1113 atomic64_dec(&pool->p !! 420 break;
>> 421 case MIDDLE:
>> 422 zhdr->middle_chunks = 0;
>> 423 zhdr->start_middle = 0;
>> 424 break;
>> 425 case LAST:
>> 426 zhdr->last_chunks = 0;
>> 427 break;
>> 428 default:
>> 429 pr_err("%s: unknown bud %d\n", __func__, bud);
>> 430 WARN_ON(1);
>> 431 spin_unlock(&pool->lock);
>> 432 return;
1114 } 433 }
1115 return; <<
1116 } 434 }
1117 435
1118 /* Non-headless case */ !! 436 if (test_bit(UNDER_RECLAIM, &page->private)) {
1119 bud = handle_to_buddy(handle); !! 437 /* z3fold page is under reclaim, reclaim will free */
1120 !! 438 spin_unlock(&pool->lock);
1121 switch (bud) { <<
1122 case FIRST: <<
1123 zhdr->first_chunks = 0; <<
1124 break; <<
1125 case MIDDLE: <<
1126 zhdr->middle_chunks = 0; <<
1127 break; <<
1128 case LAST: <<
1129 zhdr->last_chunks = 0; <<
1130 break; <<
1131 default: <<
1132 pr_err("%s: unknown bud %d\n" <<
1133 WARN_ON(1); <<
1134 put_z3fold_header(zhdr); <<
1135 return; 439 return;
1136 } 440 }
1137 441
1138 if (!page_claimed) !! 442 if (bud != HEADLESS) {
1139 free_handle(handle, zhdr); !! 443 /* Remove from existing buddy list */
1140 if (put_z3fold_locked_list(zhdr)) !! 444 list_del(&zhdr->buddy);
1141 return; <<
1142 if (page_claimed) { <<
1143 /* the page has not been clai <<
1144 put_z3fold_header(zhdr); <<
1145 return; <<
1146 } 445 }
1147 if (test_and_set_bit(NEEDS_COMPACTING !! 446
1148 clear_bit(PAGE_CLAIMED, &page !! 447 if (bud == HEADLESS ||
1149 put_z3fold_header(zhdr); !! 448 (zhdr->first_chunks == 0 && zhdr->middle_chunks == 0 &&
1150 return; !! 449 zhdr->last_chunks == 0)) {
>> 450 /* z3fold page is empty, free */
>> 451 list_del(&page->lru);
>> 452 clear_bit(PAGE_HEADLESS, &page->private);
>> 453 free_z3fold_page(zhdr);
>> 454 pool->pages_nr--;
>> 455 } else {
>> 456 z3fold_compact_page(zhdr);
>> 457 /* Add to the unbuddied list */
>> 458 freechunks = num_free_chunks(zhdr);
>> 459 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
1151 } 460 }
1152 if (zhdr->cpu < 0 || !cpu_online(zhdr !! 461
1153 zhdr->cpu = -1; !! 462 spin_unlock(&pool->lock);
1154 kref_get(&zhdr->refcount); !! 463 }
1155 clear_bit(PAGE_CLAIMED, &page !! 464
1156 do_compact_page(zhdr, true); !! 465 /**
1157 return; !! 466 * z3fold_reclaim_page() - evicts allocations from a pool page and frees it
>> 467 * @pool: pool from which a page will attempt to be evicted
>> 468 * @retires: number of pages on the LRU list for which eviction will
>> 469 * be attempted before failing
>> 470 *
>> 471 * z3fold reclaim is different from normal system reclaim in that it is done
>> 472 * from the bottom, up. This is because only the bottom layer, z3fold, has
>> 473 * information on how the allocations are organized within each z3fold page.
>> 474 * This has the potential to create interesting locking situations between
>> 475 * z3fold and the user, however.
>> 476 *
>> 477 * To avoid these, this is how z3fold_reclaim_page() should be called:
>> 478
>> 479 * The user detects a page should be reclaimed and calls z3fold_reclaim_page().
>> 480 * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and
>> 481 * call the user-defined eviction handler with the pool and handle as
>> 482 * arguments.
>> 483 *
>> 484 * If the handle can not be evicted, the eviction handler should return
>> 485 * non-zero. z3fold_reclaim_page() will add the z3fold page back to the
>> 486 * appropriate list and try the next z3fold page on the LRU up to
>> 487 * a user defined number of retries.
>> 488 *
>> 489 * If the handle is successfully evicted, the eviction handler should
>> 490 * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free()
>> 491 * contains logic to delay freeing the page if the page is under reclaim,
>> 492 * as indicated by the setting of the PG_reclaim flag on the underlying page.
>> 493 *
>> 494 * If all buddies in the z3fold page are successfully evicted, then the
>> 495 * z3fold page can be freed.
>> 496 *
>> 497 * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are
>> 498 * no pages to evict or an eviction handler is not registered, -EAGAIN if
>> 499 * the retry limit was hit.
>> 500 */
>> 501 static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries)
>> 502 {
>> 503 int i, ret = 0, freechunks;
>> 504 struct z3fold_header *zhdr;
>> 505 struct page *page;
>> 506 unsigned long first_handle = 0, middle_handle = 0, last_handle = 0;
>> 507
>> 508 spin_lock(&pool->lock);
>> 509 if (!pool->ops || !pool->ops->evict || list_empty(&pool->lru) ||
>> 510 retries == 0) {
>> 511 spin_unlock(&pool->lock);
>> 512 return -EINVAL;
>> 513 }
>> 514 for (i = 0; i < retries; i++) {
>> 515 page = list_last_entry(&pool->lru, struct page, lru);
>> 516 list_del(&page->lru);
>> 517
>> 518 /* Protect z3fold page against free */
>> 519 set_bit(UNDER_RECLAIM, &page->private);
>> 520 zhdr = page_address(page);
>> 521 if (!test_bit(PAGE_HEADLESS, &page->private)) {
>> 522 list_del(&zhdr->buddy);
>> 523 /*
>> 524 * We need encode the handles before unlocking, since
>> 525 * we can race with free that will set
>> 526 * (first|last)_chunks to 0
>> 527 */
>> 528 first_handle = 0;
>> 529 last_handle = 0;
>> 530 middle_handle = 0;
>> 531 if (zhdr->first_chunks)
>> 532 first_handle = encode_handle(zhdr, FIRST);
>> 533 if (zhdr->middle_chunks)
>> 534 middle_handle = encode_handle(zhdr, MIDDLE);
>> 535 if (zhdr->last_chunks)
>> 536 last_handle = encode_handle(zhdr, LAST);
>> 537 } else {
>> 538 first_handle = encode_handle(zhdr, HEADLESS);
>> 539 last_handle = middle_handle = 0;
>> 540 }
>> 541
>> 542 spin_unlock(&pool->lock);
>> 543
>> 544 /* Issue the eviction callback(s) */
>> 545 if (middle_handle) {
>> 546 ret = pool->ops->evict(pool, middle_handle);
>> 547 if (ret)
>> 548 goto next;
>> 549 }
>> 550 if (first_handle) {
>> 551 ret = pool->ops->evict(pool, first_handle);
>> 552 if (ret)
>> 553 goto next;
>> 554 }
>> 555 if (last_handle) {
>> 556 ret = pool->ops->evict(pool, last_handle);
>> 557 if (ret)
>> 558 goto next;
>> 559 }
>> 560 next:
>> 561 spin_lock(&pool->lock);
>> 562 clear_bit(UNDER_RECLAIM, &page->private);
>> 563 if ((test_bit(PAGE_HEADLESS, &page->private) && ret == 0) ||
>> 564 (zhdr->first_chunks == 0 && zhdr->last_chunks == 0 &&
>> 565 zhdr->middle_chunks == 0)) {
>> 566 /*
>> 567 * All buddies are now free, free the z3fold page and
>> 568 * return success.
>> 569 */
>> 570 clear_bit(PAGE_HEADLESS, &page->private);
>> 571 free_z3fold_page(zhdr);
>> 572 pool->pages_nr--;
>> 573 spin_unlock(&pool->lock);
>> 574 return 0;
>> 575 } else if (!test_bit(PAGE_HEADLESS, &page->private)) {
>> 576 if (zhdr->first_chunks != 0 &&
>> 577 zhdr->last_chunks != 0 &&
>> 578 zhdr->middle_chunks != 0) {
>> 579 /* Full, add to buddied list */
>> 580 list_add(&zhdr->buddy, &pool->buddied);
>> 581 } else {
>> 582 z3fold_compact_page(zhdr);
>> 583 /* add to unbuddied list */
>> 584 freechunks = num_free_chunks(zhdr);
>> 585 list_add(&zhdr->buddy,
>> 586 &pool->unbuddied[freechunks]);
>> 587 }
>> 588 }
>> 589
>> 590 /* add to beginning of LRU */
>> 591 list_add(&page->lru, &pool->lru);
1158 } 592 }
1159 kref_get(&zhdr->refcount); !! 593 spin_unlock(&pool->lock);
1160 clear_bit(PAGE_CLAIMED, &page->privat !! 594 return -EAGAIN;
1161 queue_work_on(zhdr->cpu, pool->compac <<
1162 put_z3fold_header(zhdr); <<
1163 } 595 }
1164 596
1165 /** 597 /**
1166 * z3fold_map() - maps the allocation associa 598 * z3fold_map() - maps the allocation associated with the given handle
1167 * @pool: pool in which the allocation 599 * @pool: pool in which the allocation resides
1168 * @handle: handle associated with the al 600 * @handle: handle associated with the allocation to be mapped
1169 * 601 *
1170 * Extracts the buddy number from handle and 602 * Extracts the buddy number from handle and constructs the pointer to the
1171 * correct starting chunk within the page. 603 * correct starting chunk within the page.
1172 * 604 *
1173 * Returns: a pointer to the mapped allocatio 605 * Returns: a pointer to the mapped allocation
1174 */ 606 */
1175 static void *z3fold_map(struct z3fold_pool *p 607 static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle)
1176 { 608 {
1177 struct z3fold_header *zhdr; 609 struct z3fold_header *zhdr;
1178 struct page *page; 610 struct page *page;
1179 void *addr; 611 void *addr;
1180 enum buddy buddy; 612 enum buddy buddy;
1181 613
1182 zhdr = get_z3fold_header(handle); !! 614 spin_lock(&pool->lock);
>> 615 zhdr = handle_to_z3fold_header(handle);
1183 addr = zhdr; 616 addr = zhdr;
1184 page = virt_to_page(zhdr); 617 page = virt_to_page(zhdr);
1185 618
1186 if (test_bit(PAGE_HEADLESS, &page->pr 619 if (test_bit(PAGE_HEADLESS, &page->private))
1187 goto out; 620 goto out;
1188 621
1189 buddy = handle_to_buddy(handle); 622 buddy = handle_to_buddy(handle);
1190 switch (buddy) { 623 switch (buddy) {
1191 case FIRST: 624 case FIRST:
1192 addr += ZHDR_SIZE_ALIGNED; 625 addr += ZHDR_SIZE_ALIGNED;
1193 break; 626 break;
1194 case MIDDLE: 627 case MIDDLE:
1195 addr += zhdr->start_middle << 628 addr += zhdr->start_middle << CHUNK_SHIFT;
1196 set_bit(MIDDLE_CHUNK_MAPPED, 629 set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1197 break; 630 break;
1198 case LAST: 631 case LAST:
1199 addr += PAGE_SIZE - (handle_t !! 632 addr += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
1200 break; 633 break;
1201 default: 634 default:
1202 pr_err("unknown buddy id %d\n 635 pr_err("unknown buddy id %d\n", buddy);
1203 WARN_ON(1); 636 WARN_ON(1);
1204 addr = NULL; 637 addr = NULL;
1205 break; 638 break;
1206 } 639 }
1207 <<
1208 if (addr) <<
1209 zhdr->mapped_count++; <<
1210 out: 640 out:
1211 put_z3fold_header(zhdr); !! 641 spin_unlock(&pool->lock);
1212 return addr; 642 return addr;
1213 } 643 }
1214 644
1215 /** 645 /**
1216 * z3fold_unmap() - unmaps the allocation ass 646 * z3fold_unmap() - unmaps the allocation associated with the given handle
1217 * @pool: pool in which the allocation 647 * @pool: pool in which the allocation resides
1218 * @handle: handle associated with the al 648 * @handle: handle associated with the allocation to be unmapped
1219 */ 649 */
1220 static void z3fold_unmap(struct z3fold_pool * 650 static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle)
1221 { 651 {
1222 struct z3fold_header *zhdr; 652 struct z3fold_header *zhdr;
1223 struct page *page; 653 struct page *page;
1224 enum buddy buddy; 654 enum buddy buddy;
1225 655
1226 zhdr = get_z3fold_header(handle); !! 656 spin_lock(&pool->lock);
>> 657 zhdr = handle_to_z3fold_header(handle);
1227 page = virt_to_page(zhdr); 658 page = virt_to_page(zhdr);
1228 659
1229 if (test_bit(PAGE_HEADLESS, &page->pr !! 660 if (test_bit(PAGE_HEADLESS, &page->private)) {
>> 661 spin_unlock(&pool->lock);
1230 return; 662 return;
>> 663 }
1231 664
1232 buddy = handle_to_buddy(handle); 665 buddy = handle_to_buddy(handle);
1233 if (buddy == MIDDLE) 666 if (buddy == MIDDLE)
1234 clear_bit(MIDDLE_CHUNK_MAPPED 667 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1235 zhdr->mapped_count--; !! 668 spin_unlock(&pool->lock);
1236 put_z3fold_header(zhdr); <<
1237 } 669 }
1238 670
1239 /** 671 /**
1240 * z3fold_get_pool_pages() - gets the z3fold !! 672 * z3fold_get_pool_size() - gets the z3fold pool size in pages
1241 * @pool: pool whose size is being quer 673 * @pool: pool whose size is being queried
1242 * 674 *
1243 * Returns: size in pages of the given pool. !! 675 * Returns: size in pages of the given pool. The pool lock need not be
>> 676 * taken to access pages_nr.
1244 */ 677 */
1245 static u64 z3fold_get_pool_pages(struct z3fol !! 678 static u64 z3fold_get_pool_size(struct z3fold_pool *pool)
1246 { 679 {
1247 return atomic64_read(&pool->pages_nr) !! 680 return pool->pages_nr;
1248 } 681 }
1249 682
1250 static bool z3fold_page_isolate(struct page * !! 683 /*****************
1251 { !! 684 * zpool
1252 struct z3fold_header *zhdr; !! 685 ****************/
1253 struct z3fold_pool *pool; <<
1254 <<
1255 VM_BUG_ON_PAGE(PageIsolated(page), pa <<
1256 <<
1257 if (test_bit(PAGE_HEADLESS, &page->pr <<
1258 return false; <<
1259 <<
1260 zhdr = page_address(page); <<
1261 z3fold_page_lock(zhdr); <<
1262 if (test_bit(NEEDS_COMPACTING, &page- <<
1263 test_bit(PAGE_STALE, &page->priva <<
1264 goto out; <<
1265 <<
1266 if (zhdr->mapped_count != 0 || zhdr-> <<
1267 goto out; <<
1268 <<
1269 if (test_and_set_bit(PAGE_CLAIMED, &p <<
1270 goto out; <<
1271 pool = zhdr_to_pool(zhdr); <<
1272 spin_lock(&pool->lock); <<
1273 if (!list_empty(&zhdr->buddy)) <<
1274 list_del_init(&zhdr->buddy); <<
1275 spin_unlock(&pool->lock); <<
1276 <<
1277 kref_get(&zhdr->refcount); <<
1278 z3fold_page_unlock(zhdr); <<
1279 return true; <<
1280 <<
1281 out: <<
1282 z3fold_page_unlock(zhdr); <<
1283 return false; <<
1284 } <<
1285 <<
1286 static int z3fold_page_migrate(struct page *n <<
1287 enum migrate_mode mode) <<
1288 { <<
1289 struct z3fold_header *zhdr, *new_zhdr <<
1290 struct z3fold_pool *pool; <<
1291 <<
1292 VM_BUG_ON_PAGE(!PageIsolated(page), p <<
1293 VM_BUG_ON_PAGE(!test_bit(PAGE_CLAIMED <<
1294 VM_BUG_ON_PAGE(!PageLocked(newpage), <<
1295 <<
1296 zhdr = page_address(page); <<
1297 pool = zhdr_to_pool(zhdr); <<
1298 <<
1299 if (!z3fold_page_trylock(zhdr)) <<
1300 return -EAGAIN; <<
1301 if (zhdr->mapped_count != 0 || zhdr-> <<
1302 clear_bit(PAGE_CLAIMED, &page <<
1303 z3fold_page_unlock(zhdr); <<
1304 return -EBUSY; <<
1305 } <<
1306 if (work_pending(&zhdr->work)) { <<
1307 z3fold_page_unlock(zhdr); <<
1308 return -EAGAIN; <<
1309 } <<
1310 new_zhdr = page_address(newpage); <<
1311 memcpy(new_zhdr, zhdr, PAGE_SIZE); <<
1312 newpage->private = page->private; <<
1313 set_bit(PAGE_MIGRATED, &page->private <<
1314 z3fold_page_unlock(zhdr); <<
1315 spin_lock_init(&new_zhdr->page_lock); <<
1316 INIT_WORK(&new_zhdr->work, compact_pa <<
1317 /* <<
1318 * z3fold_page_isolate() ensures that <<
1319 * so we only have to reinitialize it <<
1320 */ <<
1321 INIT_LIST_HEAD(&new_zhdr->buddy); <<
1322 __ClearPageMovable(page); <<
1323 <<
1324 get_page(newpage); <<
1325 z3fold_page_lock(new_zhdr); <<
1326 if (new_zhdr->first_chunks) <<
1327 encode_handle(new_zhdr, FIRST <<
1328 if (new_zhdr->last_chunks) <<
1329 encode_handle(new_zhdr, LAST) <<
1330 if (new_zhdr->middle_chunks) <<
1331 encode_handle(new_zhdr, MIDDL <<
1332 set_bit(NEEDS_COMPACTING, &newpage->p <<
1333 new_zhdr->cpu = smp_processor_id(); <<
1334 __SetPageMovable(newpage, &z3fold_mop <<
1335 z3fold_page_unlock(new_zhdr); <<
1336 <<
1337 queue_work_on(new_zhdr->cpu, pool->co <<
1338 <<
1339 /* PAGE_CLAIMED and PAGE_MIGRATED are <<
1340 page->private = 0; <<
1341 put_page(page); <<
1342 return 0; <<
1343 } <<
1344 686
1345 static void z3fold_page_putback(struct page * !! 687 static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle)
1346 { 688 {
1347 struct z3fold_header *zhdr; !! 689 if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict)
1348 struct z3fold_pool *pool; !! 690 return pool->zpool_ops->evict(pool->zpool, handle);
1349 !! 691 else
1350 zhdr = page_address(page); !! 692 return -ENOENT;
1351 pool = zhdr_to_pool(zhdr); <<
1352 <<
1353 z3fold_page_lock(zhdr); <<
1354 if (!list_empty(&zhdr->buddy)) <<
1355 list_del_init(&zhdr->buddy); <<
1356 INIT_LIST_HEAD(&page->lru); <<
1357 if (put_z3fold_locked(zhdr)) <<
1358 return; <<
1359 if (list_empty(&zhdr->buddy)) <<
1360 add_to_unbuddied(pool, zhdr); <<
1361 clear_bit(PAGE_CLAIMED, &page->privat <<
1362 z3fold_page_unlock(zhdr); <<
1363 } 693 }
1364 694
1365 static const struct movable_operations z3fold !! 695 static const struct z3fold_ops z3fold_zpool_ops = {
1366 .isolate_page = z3fold_page_isolate, !! 696 .evict = z3fold_zpool_evict
1367 .migrate_page = z3fold_page_migrate, <<
1368 .putback_page = z3fold_page_putback, <<
1369 }; 697 };
1370 698
1371 /***************** !! 699 static void *z3fold_zpool_create(const char *name, gfp_t gfp,
1372 * zpool !! 700 const struct zpool_ops *zpool_ops,
1373 ****************/ !! 701 struct zpool *zpool)
1374 <<
1375 static void *z3fold_zpool_create(const char * <<
1376 { 702 {
1377 return z3fold_create_pool(name, gfp); !! 703 struct z3fold_pool *pool;
>> 704
>> 705 pool = z3fold_create_pool(gfp, zpool_ops ? &z3fold_zpool_ops : NULL);
>> 706 if (pool) {
>> 707 pool->zpool = zpool;
>> 708 pool->zpool_ops = zpool_ops;
>> 709 }
>> 710 return pool;
1378 } 711 }
1379 712
1380 static void z3fold_zpool_destroy(void *pool) 713 static void z3fold_zpool_destroy(void *pool)
1381 { 714 {
1382 z3fold_destroy_pool(pool); 715 z3fold_destroy_pool(pool);
1383 } 716 }
1384 717
1385 static int z3fold_zpool_malloc(void *pool, si 718 static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp,
1386 unsigned long *handle 719 unsigned long *handle)
1387 { 720 {
1388 return z3fold_alloc(pool, size, gfp, 721 return z3fold_alloc(pool, size, gfp, handle);
1389 } 722 }
1390 static void z3fold_zpool_free(void *pool, uns 723 static void z3fold_zpool_free(void *pool, unsigned long handle)
1391 { 724 {
1392 z3fold_free(pool, handle); 725 z3fold_free(pool, handle);
1393 } 726 }
1394 727
>> 728 static int z3fold_zpool_shrink(void *pool, unsigned int pages,
>> 729 unsigned int *reclaimed)
>> 730 {
>> 731 unsigned int total = 0;
>> 732 int ret = -EINVAL;
>> 733
>> 734 while (total < pages) {
>> 735 ret = z3fold_reclaim_page(pool, 8);
>> 736 if (ret < 0)
>> 737 break;
>> 738 total++;
>> 739 }
>> 740
>> 741 if (reclaimed)
>> 742 *reclaimed = total;
>> 743
>> 744 return ret;
>> 745 }
>> 746
1395 static void *z3fold_zpool_map(void *pool, uns 747 static void *z3fold_zpool_map(void *pool, unsigned long handle,
1396 enum zpool_mapmode mm 748 enum zpool_mapmode mm)
1397 { 749 {
1398 return z3fold_map(pool, handle); 750 return z3fold_map(pool, handle);
1399 } 751 }
1400 static void z3fold_zpool_unmap(void *pool, un 752 static void z3fold_zpool_unmap(void *pool, unsigned long handle)
1401 { 753 {
1402 z3fold_unmap(pool, handle); 754 z3fold_unmap(pool, handle);
1403 } 755 }
1404 756
1405 static u64 z3fold_zpool_total_pages(void *poo !! 757 static u64 z3fold_zpool_total_size(void *pool)
1406 { 758 {
1407 return z3fold_get_pool_pages(pool); !! 759 return z3fold_get_pool_size(pool) * PAGE_SIZE;
1408 } 760 }
1409 761
1410 static struct zpool_driver z3fold_zpool_drive 762 static struct zpool_driver z3fold_zpool_driver = {
1411 .type = "z3fold", 763 .type = "z3fold",
1412 .sleep_mapped = true, <<
1413 .owner = THIS_MODULE, 764 .owner = THIS_MODULE,
1414 .create = z3fold_zpool_create, 765 .create = z3fold_zpool_create,
1415 .destroy = z3fold_zpool_destroy, 766 .destroy = z3fold_zpool_destroy,
1416 .malloc = z3fold_zpool_malloc, 767 .malloc = z3fold_zpool_malloc,
1417 .free = z3fold_zpool_free, 768 .free = z3fold_zpool_free,
>> 769 .shrink = z3fold_zpool_shrink,
1418 .map = z3fold_zpool_map, 770 .map = z3fold_zpool_map,
1419 .unmap = z3fold_zpool_unmap, 771 .unmap = z3fold_zpool_unmap,
1420 .total_pages = z3fold_zpool_total_pa !! 772 .total_size = z3fold_zpool_total_size,
1421 }; 773 };
1422 774
1423 MODULE_ALIAS("zpool-z3fold"); 775 MODULE_ALIAS("zpool-z3fold");
1424 776
1425 static int __init init_z3fold(void) 777 static int __init init_z3fold(void)
1426 { 778 {
1427 /* !! 779 /* Make sure the z3fold header will fit in one chunk */
1428 * Make sure the z3fold header is not !! 780 BUILD_BUG_ON(sizeof(struct z3fold_header) > ZHDR_SIZE_ALIGNED);
1429 * there has remaining spaces for its <<
1430 */ <<
1431 BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE <<
1432 zpool_register_driver(&z3fold_zpool_d 781 zpool_register_driver(&z3fold_zpool_driver);
1433 782
1434 return 0; 783 return 0;
1435 } 784 }
1436 785
1437 static void __exit exit_z3fold(void) 786 static void __exit exit_z3fold(void)
1438 { 787 {
1439 zpool_unregister_driver(&z3fold_zpool 788 zpool_unregister_driver(&z3fold_zpool_driver);
1440 } 789 }
1441 790
1442 module_init(init_z3fold); 791 module_init(init_z3fold);
1443 module_exit(exit_z3fold); 792 module_exit(exit_z3fold);
1444 793
1445 MODULE_LICENSE("GPL"); 794 MODULE_LICENSE("GPL");
1446 MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail. 795 MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>");
1447 MODULE_DESCRIPTION("3-Fold Allocator for Comp 796 MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages");
1448 797
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.