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> 26 #include <linux/sched.h>
28 #include <linux/cpumask.h> <<
29 #include <linux/list.h> 27 #include <linux/list.h>
30 #include <linux/mm.h> 28 #include <linux/mm.h>
31 #include <linux/module.h> 29 #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> 30 #include <linux/percpu.h>
37 #include <linux/preempt.h> 31 #include <linux/preempt.h>
38 #include <linux/workqueue.h> 32 #include <linux/workqueue.h>
39 #include <linux/slab.h> 33 #include <linux/slab.h>
40 #include <linux/spinlock.h> 34 #include <linux/spinlock.h>
41 #include <linux/zpool.h> 35 #include <linux/zpool.h>
42 #include <linux/kmemleak.h> <<
43 <<
44 /* <<
45 * NCHUNKS_ORDER determines the internal alloc <<
46 * adjusting internal fragmentation. It also <<
47 * freelists maintained in each pool. NCHUNKS_ <<
48 * allocation granularity will be in chunks of <<
49 * in the beginning of an allocated page are o <<
50 * NCHUNKS will be calculated to 63 (or 62 in <<
51 * which shows the max number of free chunks i <<
52 * be 63, or 62, respectively, freelists per p <<
53 */ <<
54 #define NCHUNKS_ORDER 6 <<
55 <<
56 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ <<
57 #define CHUNK_SIZE (1 << CHUNK_SHIFT) <<
58 #define ZHDR_SIZE_ALIGNED round_up(sizeof(stru <<
59 #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> <<
60 #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SH <<
61 #define NCHUNKS (TOTAL_CHUNKS - ZHDR_C <<
62 <<
63 #define BUDDY_MASK (0x3) <<
64 #define BUDDY_SHIFT 2 <<
65 #define SLOTS_ALIGN (0x40) <<
66 36
67 /***************** 37 /*****************
68 * Structures 38 * Structures
69 *****************/ 39 *****************/
70 struct z3fold_pool; 40 struct z3fold_pool;
>> 41 struct z3fold_ops {
>> 42 int (*evict)(struct z3fold_pool *pool, unsigned long handle);
>> 43 };
71 44
72 enum buddy { 45 enum buddy {
73 HEADLESS = 0, 46 HEADLESS = 0,
74 FIRST, 47 FIRST,
75 MIDDLE, 48 MIDDLE,
76 LAST, 49 LAST,
77 BUDDIES_MAX = LAST !! 50 BUDDIES_MAX
78 }; 51 };
79 52
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 }; <<
89 #define HANDLE_FLAG_MASK (0x03) <<
90 <<
91 /* 53 /*
92 * struct z3fold_header - z3fold page metadata 54 * struct z3fold_header - z3fold page metadata occupying first chunks of each
93 * z3fold page, except fo 55 * z3fold page, except for HEADLESS pages
94 * @buddy: links the z3fold page 56 * @buddy: links the z3fold page into the relevant list in the
95 * pool 57 * pool
96 * @page_lock: per-page lock 58 * @page_lock: per-page lock
97 * @refcount: reference count for th 59 * @refcount: reference count for the z3fold page
98 * @work: work_struct for page l 60 * @work: work_struct for page layout optimization
99 * @slots: pointer to the structu !! 61 * @pool: pointer to the pool which this page belongs to
100 * @pool: pointer to the contain <<
101 * @cpu: CPU which this page "b 62 * @cpu: CPU which this page "belongs" to
102 * @first_chunks: the size of the first 63 * @first_chunks: the size of the first buddy in chunks, 0 if free
103 * @middle_chunks: the size of the middle 64 * @middle_chunks: the size of the middle buddy in chunks, 0 if free
104 * @last_chunks: the size of the last b 65 * @last_chunks: the size of the last buddy in chunks, 0 if free
105 * @first_num: the starting number (f 66 * @first_num: the starting number (for the first handle)
106 * @mapped_count: the number of objects <<
107 */ 67 */
108 struct z3fold_header { 68 struct z3fold_header {
109 struct list_head buddy; 69 struct list_head buddy;
110 spinlock_t page_lock; 70 spinlock_t page_lock;
111 struct kref refcount; 71 struct kref refcount;
112 struct work_struct work; 72 struct work_struct work;
113 struct z3fold_buddy_slots *slots; <<
114 struct z3fold_pool *pool; 73 struct z3fold_pool *pool;
115 short cpu; 74 short cpu;
116 unsigned short first_chunks; 75 unsigned short first_chunks;
117 unsigned short middle_chunks; 76 unsigned short middle_chunks;
118 unsigned short last_chunks; 77 unsigned short last_chunks;
119 unsigned short start_middle; 78 unsigned short start_middle;
120 unsigned short first_num:2; 79 unsigned short first_num:2;
121 unsigned short mapped_count:2; <<
122 unsigned short foreign_handles:2; <<
123 }; 80 };
124 81
>> 82 /*
>> 83 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
>> 84 * adjusting internal fragmentation. It also determines the number of
>> 85 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
>> 86 * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks
>> 87 * in the beginning of an allocated page are occupied by z3fold header, so
>> 88 * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y),
>> 89 * which shows the max number of free chunks in z3fold page, also there will
>> 90 * be 63, or 62, respectively, freelists per pool.
>> 91 */
>> 92 #define NCHUNKS_ORDER 6
>> 93
>> 94 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER)
>> 95 #define CHUNK_SIZE (1 << CHUNK_SHIFT)
>> 96 #define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE)
>> 97 #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT)
>> 98 #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SHIFT)
>> 99 #define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT)
>> 100
>> 101 #define BUDDY_MASK (0x3)
>> 102 #define BUDDY_SHIFT 2
>> 103
125 /** 104 /**
126 * struct z3fold_pool - stores metadata for ea 105 * struct z3fold_pool - stores metadata for each z3fold pool
127 * @name: pool name 106 * @name: pool name
128 * @lock: protects pool unbuddied lists !! 107 * @lock: protects pool unbuddied/lru lists
129 * @stale_lock: protects pool stale page list 108 * @stale_lock: protects pool stale page list
130 * @unbuddied: per-cpu array of lists trackin 109 * @unbuddied: per-cpu array of lists tracking z3fold pages that contain 2-
131 * buddies; the list each z3fold 110 * buddies; the list each z3fold page is added to depends on
132 * the size of its free region. 111 * the size of its free region.
>> 112 * @lru: list tracking the z3fold pages in LRU order by most recently
>> 113 * added buddy.
133 * @stale: list of pages marked for freei 114 * @stale: list of pages marked for freeing
134 * @pages_nr: number of z3fold pages in the 115 * @pages_nr: number of z3fold pages in the pool.
135 * @c_handle: cache for z3fold_buddy_slots a !! 116 * @ops: pointer to a structure of user defined operations specified at
>> 117 * pool creation time.
136 * @compact_wq: workqueue for page layout back 118 * @compact_wq: workqueue for page layout background optimization
137 * @release_wq: workqueue for safe page releas 119 * @release_wq: workqueue for safe page release
138 * @work: work_struct for safe page rele 120 * @work: work_struct for safe page release
139 * 121 *
140 * This structure is allocated at pool creatio 122 * This structure is allocated at pool creation time and maintains metadata
141 * pertaining to a particular z3fold pool. 123 * pertaining to a particular z3fold pool.
142 */ 124 */
143 struct z3fold_pool { 125 struct z3fold_pool {
144 const char *name; 126 const char *name;
145 spinlock_t lock; 127 spinlock_t lock;
146 spinlock_t stale_lock; 128 spinlock_t stale_lock;
147 struct list_head __percpu *unbuddied; !! 129 struct list_head *unbuddied;
>> 130 struct list_head lru;
148 struct list_head stale; 131 struct list_head stale;
149 atomic64_t pages_nr; 132 atomic64_t pages_nr;
150 struct kmem_cache *c_handle; !! 133 const struct z3fold_ops *ops;
>> 134 struct zpool *zpool;
>> 135 const struct zpool_ops *zpool_ops;
151 struct workqueue_struct *compact_wq; 136 struct workqueue_struct *compact_wq;
152 struct workqueue_struct *release_wq; 137 struct workqueue_struct *release_wq;
153 struct work_struct work; 138 struct work_struct work;
154 }; 139 };
155 140
156 /* 141 /*
157 * Internal z3fold page flags 142 * Internal z3fold page flags
158 */ 143 */
159 enum z3fold_page_flags { 144 enum z3fold_page_flags {
160 PAGE_HEADLESS = 0, 145 PAGE_HEADLESS = 0,
161 MIDDLE_CHUNK_MAPPED, 146 MIDDLE_CHUNK_MAPPED,
162 NEEDS_COMPACTING, 147 NEEDS_COMPACTING,
163 PAGE_STALE, 148 PAGE_STALE,
164 PAGE_CLAIMED, /* by either reclaim or 149 PAGE_CLAIMED, /* by either reclaim or free */
165 PAGE_MIGRATED, /* page is migrated and <<
166 }; <<
167 <<
168 /* <<
169 * handle flags, go under HANDLE_FLAG_MASK <<
170 */ <<
171 enum z3fold_handle_flags { <<
172 HANDLES_NOFREE = 0, <<
173 }; 150 };
174 151
175 /* <<
176 * Forward declarations <<
177 */ <<
178 static struct z3fold_header *__z3fold_alloc(st <<
179 static void compact_page_work(struct work_stru <<
180 <<
181 /***************** 152 /*****************
182 * Helpers 153 * Helpers
183 *****************/ 154 *****************/
184 155
185 /* Converts an allocation size in bytes to siz 156 /* Converts an allocation size in bytes to size in z3fold chunks */
186 static int size_to_chunks(size_t size) 157 static int size_to_chunks(size_t size)
187 { 158 {
188 return (size + CHUNK_SIZE - 1) >> CHUN 159 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
189 } 160 }
190 161
191 #define for_each_unbuddied_list(_iter, _begin) 162 #define for_each_unbuddied_list(_iter, _begin) \
192 for ((_iter) = (_begin); (_iter) < NCH 163 for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
193 164
194 static inline struct z3fold_buddy_slots *alloc !! 165 static void compact_page_work(struct work_struct *w);
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 } <<
317 166
318 /* Initializes the z3fold header of a newly al 167 /* Initializes the z3fold header of a newly allocated z3fold page */
319 static struct z3fold_header *init_z3fold_page( !! 168 static struct z3fold_header *init_z3fold_page(struct page *page,
320 struct !! 169 struct z3fold_pool *pool)
321 { 170 {
322 struct z3fold_header *zhdr = page_addr 171 struct z3fold_header *zhdr = page_address(page);
323 struct z3fold_buddy_slots *slots; <<
324 172
>> 173 INIT_LIST_HEAD(&page->lru);
325 clear_bit(PAGE_HEADLESS, &page->privat 174 clear_bit(PAGE_HEADLESS, &page->private);
326 clear_bit(MIDDLE_CHUNK_MAPPED, &page-> 175 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
327 clear_bit(NEEDS_COMPACTING, &page->pri 176 clear_bit(NEEDS_COMPACTING, &page->private);
328 clear_bit(PAGE_STALE, &page->private); 177 clear_bit(PAGE_STALE, &page->private);
329 clear_bit(PAGE_CLAIMED, &page->private 178 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; <<
337 179
338 memset(zhdr, 0, sizeof(*zhdr)); <<
339 spin_lock_init(&zhdr->page_lock); 180 spin_lock_init(&zhdr->page_lock);
340 kref_init(&zhdr->refcount); 181 kref_init(&zhdr->refcount);
>> 182 zhdr->first_chunks = 0;
>> 183 zhdr->middle_chunks = 0;
>> 184 zhdr->last_chunks = 0;
>> 185 zhdr->first_num = 0;
>> 186 zhdr->start_middle = 0;
341 zhdr->cpu = -1; 187 zhdr->cpu = -1;
342 zhdr->slots = slots; <<
343 zhdr->pool = pool; 188 zhdr->pool = pool;
344 INIT_LIST_HEAD(&zhdr->buddy); 189 INIT_LIST_HEAD(&zhdr->buddy);
345 INIT_WORK(&zhdr->work, compact_page_wo 190 INIT_WORK(&zhdr->work, compact_page_work);
346 return zhdr; 191 return zhdr;
347 } 192 }
348 193
349 /* Resets the struct page fields and frees the 194 /* Resets the struct page fields and frees the page */
350 static void free_z3fold_page(struct page *page !! 195 static void free_z3fold_page(struct page *page)
351 { 196 {
352 if (!headless) { <<
353 lock_page(page); <<
354 __ClearPageMovable(page); <<
355 unlock_page(page); <<
356 } <<
357 __free_page(page); 197 __free_page(page);
358 } 198 }
359 199
360 /* Helper function to build the index */ !! 200 /* Lock a z3fold page */
361 static inline int __idx(struct z3fold_header * !! 201 static inline void z3fold_page_lock(struct z3fold_header *zhdr)
>> 202 {
>> 203 spin_lock(&zhdr->page_lock);
>> 204 }
>> 205
>> 206 /* Try to lock a z3fold page */
>> 207 static inline int z3fold_page_trylock(struct z3fold_header *zhdr)
362 { 208 {
363 return (bud + zhdr->first_num) & BUDDY !! 209 return spin_trylock(&zhdr->page_lock);
>> 210 }
>> 211
>> 212 /* Unlock a z3fold page */
>> 213 static inline void z3fold_page_unlock(struct z3fold_header *zhdr)
>> 214 {
>> 215 spin_unlock(&zhdr->page_lock);
364 } 216 }
365 217
366 /* 218 /*
367 * Encodes the handle of a particular buddy wi !! 219 * Encodes the handle of a particular buddy within a z3fold page
368 * Zhdr->page_lock should be held as this func !! 220 * Pool lock should be held as this function accesses first_num
369 * if bud != HEADLESS. <<
370 */ 221 */
371 static unsigned long __encode_handle(struct z3 !! 222 static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud)
372 struct z3fold_ <<
373 enum buddy bud <<
374 { 223 {
375 unsigned long h = (unsigned long)zhdr; !! 224 unsigned long handle;
376 int idx = 0; <<
377 225
378 /* !! 226 handle = (unsigned long)zhdr;
379 * For a headless page, its handle is !! 227 if (bud != HEADLESS) {
380 * PAGE_HEADLESS bit set !! 228 handle |= (bud + zhdr->first_num) & BUDDY_MASK;
381 */ !! 229 if (bud == LAST)
382 if (bud == HEADLESS) !! 230 handle |= (zhdr->last_chunks << BUDDY_SHIFT);
383 return h | (1 << PAGE_HEADLESS !! 231 }
384 !! 232 return handle;
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 } 233 }
396 234
397 static unsigned long encode_handle(struct z3fo !! 235 /* Returns the z3fold page where a given handle is stored */
>> 236 static struct z3fold_header *handle_to_z3fold_header(unsigned long handle)
398 { 237 {
399 return __encode_handle(zhdr, zhdr->slo !! 238 return (struct z3fold_header *)(handle & PAGE_MASK);
400 } 239 }
401 240
402 /* only for LAST bud, returns zero otherwise * 241 /* only for LAST bud, returns zero otherwise */
403 static unsigned short handle_to_chunks(unsigne 242 static unsigned short handle_to_chunks(unsigned long handle)
404 { 243 {
405 struct z3fold_buddy_slots *slots = han !! 244 return (handle & ~PAGE_MASK) >> BUDDY_SHIFT;
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 } 245 }
413 246
414 /* 247 /*
415 * (handle & BUDDY_MASK) < zhdr->first_num is 248 * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle
416 * but that doesn't matter. because the maski 249 * but that doesn't matter. because the masking will result in the
417 * correct buddy number. 250 * correct buddy number.
418 */ 251 */
419 static enum buddy handle_to_buddy(unsigned lon 252 static enum buddy handle_to_buddy(unsigned long handle)
420 { 253 {
421 struct z3fold_header *zhdr; !! 254 struct z3fold_header *zhdr = handle_to_z3fold_header(handle);
422 struct z3fold_buddy_slots *slots = han !! 255 return (handle - zhdr->first_num) & BUDDY_MASK;
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 } 256 }
437 257
438 static void __release_z3fold_page(struct z3fol 258 static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked)
439 { 259 {
440 struct page *page = virt_to_page(zhdr) 260 struct page *page = virt_to_page(zhdr);
441 struct z3fold_pool *pool = zhdr_to_poo !! 261 struct z3fold_pool *pool = zhdr->pool;
442 262
443 WARN_ON(!list_empty(&zhdr->buddy)); 263 WARN_ON(!list_empty(&zhdr->buddy));
444 set_bit(PAGE_STALE, &page->private); 264 set_bit(PAGE_STALE, &page->private);
445 clear_bit(NEEDS_COMPACTING, &page->pri 265 clear_bit(NEEDS_COMPACTING, &page->private);
446 spin_lock(&pool->lock); 266 spin_lock(&pool->lock);
>> 267 if (!list_empty(&page->lru))
>> 268 list_del(&page->lru);
447 spin_unlock(&pool->lock); 269 spin_unlock(&pool->lock);
448 <<
449 if (locked) 270 if (locked)
450 z3fold_page_unlock(zhdr); 271 z3fold_page_unlock(zhdr);
451 <<
452 spin_lock(&pool->stale_lock); 272 spin_lock(&pool->stale_lock);
453 list_add(&zhdr->buddy, &pool->stale); 273 list_add(&zhdr->buddy, &pool->stale);
454 queue_work(pool->release_wq, &pool->wo 274 queue_work(pool->release_wq, &pool->work);
455 spin_unlock(&pool->stale_lock); 275 spin_unlock(&pool->stale_lock);
>> 276 }
456 277
457 atomic64_dec(&pool->pages_nr); !! 278 static void __attribute__((__unused__))
>> 279 release_z3fold_page(struct kref *ref)
>> 280 {
>> 281 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
>> 282 refcount);
>> 283 __release_z3fold_page(zhdr, false);
458 } 284 }
459 285
460 static void release_z3fold_page_locked(struct 286 static void release_z3fold_page_locked(struct kref *ref)
461 { 287 {
462 struct z3fold_header *zhdr = container 288 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
463 289 refcount);
464 WARN_ON(z3fold_page_trylock(zhdr)); 290 WARN_ON(z3fold_page_trylock(zhdr));
465 __release_z3fold_page(zhdr, true); 291 __release_z3fold_page(zhdr, true);
466 } 292 }
467 293
468 static void release_z3fold_page_locked_list(st 294 static void release_z3fold_page_locked_list(struct kref *ref)
469 { 295 {
470 struct z3fold_header *zhdr = container 296 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
471 297 refcount);
472 struct z3fold_pool *pool = zhdr_to_poo !! 298 spin_lock(&zhdr->pool->lock);
473 <<
474 spin_lock(&pool->lock); <<
475 list_del_init(&zhdr->buddy); 299 list_del_init(&zhdr->buddy);
476 spin_unlock(&pool->lock); !! 300 spin_unlock(&zhdr->pool->lock);
477 301
478 WARN_ON(z3fold_page_trylock(zhdr)); 302 WARN_ON(z3fold_page_trylock(zhdr));
479 __release_z3fold_page(zhdr, true); 303 __release_z3fold_page(zhdr, true);
480 } 304 }
481 305
482 static inline int put_z3fold_locked(struct z3f <<
483 { <<
484 return kref_put(&zhdr->refcount, relea <<
485 } <<
486 <<
487 static inline int put_z3fold_locked_list(struc <<
488 { <<
489 return kref_put(&zhdr->refcount, relea <<
490 } <<
491 <<
492 static void free_pages_work(struct work_struct 306 static void free_pages_work(struct work_struct *w)
493 { 307 {
494 struct z3fold_pool *pool = container_o 308 struct z3fold_pool *pool = container_of(w, struct z3fold_pool, work);
495 309
496 spin_lock(&pool->stale_lock); 310 spin_lock(&pool->stale_lock);
497 while (!list_empty(&pool->stale)) { 311 while (!list_empty(&pool->stale)) {
498 struct z3fold_header *zhdr = l 312 struct z3fold_header *zhdr = list_first_entry(&pool->stale,
499 313 struct z3fold_header, buddy);
500 struct page *page = virt_to_pa 314 struct page *page = virt_to_page(zhdr);
501 315
502 list_del(&zhdr->buddy); 316 list_del(&zhdr->buddy);
503 if (WARN_ON(!test_bit(PAGE_STA 317 if (WARN_ON(!test_bit(PAGE_STALE, &page->private)))
504 continue; 318 continue;
505 spin_unlock(&pool->stale_lock) 319 spin_unlock(&pool->stale_lock);
506 cancel_work_sync(&zhdr->work); 320 cancel_work_sync(&zhdr->work);
507 free_z3fold_page(page, false); !! 321 free_z3fold_page(page);
508 cond_resched(); 322 cond_resched();
509 spin_lock(&pool->stale_lock); 323 spin_lock(&pool->stale_lock);
510 } 324 }
511 spin_unlock(&pool->stale_lock); 325 spin_unlock(&pool->stale_lock);
512 } 326 }
513 327
514 /* 328 /*
515 * Returns the number of free chunks in a z3fo 329 * Returns the number of free chunks in a z3fold page.
516 * NB: can't be used with HEADLESS pages. 330 * NB: can't be used with HEADLESS pages.
517 */ 331 */
518 static int num_free_chunks(struct z3fold_heade 332 static int num_free_chunks(struct z3fold_header *zhdr)
519 { 333 {
520 int nfree; 334 int nfree;
521 /* 335 /*
522 * If there is a middle object, pick u 336 * If there is a middle object, pick up the bigger free space
523 * either before or after it. Otherwis 337 * either before or after it. Otherwise just subtract the number
524 * of chunks occupied by the first and 338 * of chunks occupied by the first and the last objects.
525 */ 339 */
526 if (zhdr->middle_chunks != 0) { 340 if (zhdr->middle_chunks != 0) {
527 int nfree_before = zhdr->first 341 int nfree_before = zhdr->first_chunks ?
528 0 : zhdr->start_middle 342 0 : zhdr->start_middle - ZHDR_CHUNKS;
529 int nfree_after = zhdr->last_c 343 int nfree_after = zhdr->last_chunks ?
530 0 : TOTAL_CHUNKS - 344 0 : TOTAL_CHUNKS -
531 (zhdr->start_m 345 (zhdr->start_middle + zhdr->middle_chunks);
532 nfree = max(nfree_before, nfre 346 nfree = max(nfree_before, nfree_after);
533 } else 347 } else
534 nfree = NCHUNKS - zhdr->first_ 348 nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
535 return nfree; 349 return nfree;
536 } 350 }
537 351
538 /* Add to the appropriate unbuddied list */ <<
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 352 static inline void *mchunk_memmove(struct z3fold_header *zhdr,
580 unsigned short 353 unsigned short dst_chunk)
581 { 354 {
582 void *beg = zhdr; 355 void *beg = zhdr;
583 return memmove(beg + (dst_chunk << CHU 356 return memmove(beg + (dst_chunk << CHUNK_SHIFT),
584 beg + (zhdr->start_midd 357 beg + (zhdr->start_middle << CHUNK_SHIFT),
585 zhdr->middle_chunks << 358 zhdr->middle_chunks << CHUNK_SHIFT);
586 } 359 }
587 360
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 3 361 #define BIG_CHUNK_GAP 3
687 /* Has to be called with lock held */ 362 /* Has to be called with lock held */
688 static int z3fold_compact_page(struct z3fold_h 363 static int z3fold_compact_page(struct z3fold_header *zhdr)
689 { 364 {
690 struct page *page = virt_to_page(zhdr) 365 struct page *page = virt_to_page(zhdr);
691 366
692 if (test_bit(MIDDLE_CHUNK_MAPPED, &pag 367 if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private))
693 return 0; /* can't move middle 368 return 0; /* can't move middle chunk, it's used */
694 369
695 if (unlikely(PageIsolated(page))) <<
696 return 0; <<
697 <<
698 if (zhdr->middle_chunks == 0) 370 if (zhdr->middle_chunks == 0)
699 return 0; /* nothing to compac 371 return 0; /* nothing to compact */
700 372
701 if (zhdr->first_chunks == 0 && zhdr->l 373 if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
702 /* move to the beginning */ 374 /* move to the beginning */
703 mchunk_memmove(zhdr, ZHDR_CHUN 375 mchunk_memmove(zhdr, ZHDR_CHUNKS);
704 zhdr->first_chunks = zhdr->mid 376 zhdr->first_chunks = zhdr->middle_chunks;
705 zhdr->middle_chunks = 0; 377 zhdr->middle_chunks = 0;
706 zhdr->start_middle = 0; 378 zhdr->start_middle = 0;
707 zhdr->first_num++; 379 zhdr->first_num++;
708 return 1; 380 return 1;
709 } 381 }
710 382
711 /* 383 /*
712 * moving data is expensive, so let's 384 * moving data is expensive, so let's only do that if
713 * there's substantial gain (at least 385 * there's substantial gain (at least BIG_CHUNK_GAP chunks)
714 */ 386 */
715 if (zhdr->first_chunks != 0 && zhdr->l 387 if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 &&
716 zhdr->start_middle - (zhdr->first_ 388 zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >=
717 BIG_CHUNK_GAP) { 389 BIG_CHUNK_GAP) {
718 mchunk_memmove(zhdr, zhdr->fir 390 mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS);
719 zhdr->start_middle = zhdr->fir 391 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
720 return 1; 392 return 1;
721 } else if (zhdr->last_chunks != 0 && z 393 } else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 &&
722 TOTAL_CHUNKS - (zhdr->last_ 394 TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle
723 + zhdr 395 + zhdr->middle_chunks) >=
724 BIG_CHUNK_GAP) { 396 BIG_CHUNK_GAP) {
725 unsigned short new_start = TOT 397 unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks -
726 zhdr->middle_chunks; 398 zhdr->middle_chunks;
727 mchunk_memmove(zhdr, new_start 399 mchunk_memmove(zhdr, new_start);
728 zhdr->start_middle = new_start 400 zhdr->start_middle = new_start;
729 return 1; 401 return 1;
730 } 402 }
731 403
732 return 0; 404 return 0;
733 } 405 }
734 406
735 static void do_compact_page(struct z3fold_head 407 static void do_compact_page(struct z3fold_header *zhdr, bool locked)
736 { 408 {
737 struct z3fold_pool *pool = zhdr_to_poo !! 409 struct z3fold_pool *pool = zhdr->pool;
738 struct page *page; 410 struct page *page;
>> 411 struct list_head *unbuddied;
>> 412 int fchunks;
739 413
740 page = virt_to_page(zhdr); 414 page = virt_to_page(zhdr);
741 if (locked) 415 if (locked)
742 WARN_ON(z3fold_page_trylock(zh 416 WARN_ON(z3fold_page_trylock(zhdr));
743 else 417 else
744 z3fold_page_lock(zhdr); 418 z3fold_page_lock(zhdr);
745 if (WARN_ON(!test_and_clear_bit(NEEDS_ 419 if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) {
746 z3fold_page_unlock(zhdr); 420 z3fold_page_unlock(zhdr);
747 return; 421 return;
748 } 422 }
749 spin_lock(&pool->lock); 423 spin_lock(&pool->lock);
750 list_del_init(&zhdr->buddy); 424 list_del_init(&zhdr->buddy);
751 spin_unlock(&pool->lock); 425 spin_unlock(&pool->lock);
752 426
753 if (put_z3fold_locked(zhdr)) !! 427 if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) {
754 return; !! 428 atomic64_dec(&pool->pages_nr);
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; 429 return;
769 } 430 }
770 431
771 z3fold_compact_page(zhdr); 432 z3fold_compact_page(zhdr);
772 add_to_unbuddied(pool, zhdr); !! 433 unbuddied = get_cpu_ptr(pool->unbuddied);
773 clear_bit(PAGE_CLAIMED, &page->private !! 434 fchunks = num_free_chunks(zhdr);
>> 435 if (fchunks < NCHUNKS &&
>> 436 (!zhdr->first_chunks || !zhdr->middle_chunks ||
>> 437 !zhdr->last_chunks)) {
>> 438 /* the page's not completely free and it's unbuddied */
>> 439 spin_lock(&pool->lock);
>> 440 list_add(&zhdr->buddy, &unbuddied[fchunks]);
>> 441 spin_unlock(&pool->lock);
>> 442 zhdr->cpu = smp_processor_id();
>> 443 }
>> 444 put_cpu_ptr(pool->unbuddied);
774 z3fold_page_unlock(zhdr); 445 z3fold_page_unlock(zhdr);
775 } 446 }
776 447
777 static void compact_page_work(struct work_stru 448 static void compact_page_work(struct work_struct *w)
778 { 449 {
779 struct z3fold_header *zhdr = container 450 struct z3fold_header *zhdr = container_of(w, struct z3fold_header,
780 451 work);
781 452
782 do_compact_page(zhdr, false); 453 do_compact_page(zhdr, false);
783 } 454 }
784 455
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 } <<
896 456
897 /* 457 /*
898 * API Functions 458 * API Functions
899 */ 459 */
900 460
901 /** 461 /**
902 * z3fold_create_pool() - create a new z3fold 462 * z3fold_create_pool() - create a new z3fold pool
903 * @name: pool name 463 * @name: pool name
904 * @gfp: gfp flags when allocating the 464 * @gfp: gfp flags when allocating the z3fold pool structure
>> 465 * @ops: user-defined operations for the z3fold pool
905 * 466 *
906 * Return: pointer to the new z3fold pool or N 467 * Return: pointer to the new z3fold pool or NULL if the metadata allocation
907 * failed. 468 * failed.
908 */ 469 */
909 static struct z3fold_pool *z3fold_create_pool( !! 470 static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp,
>> 471 const struct z3fold_ops *ops)
910 { 472 {
911 struct z3fold_pool *pool = NULL; 473 struct z3fold_pool *pool = NULL;
912 int i, cpu; 474 int i, cpu;
913 475
914 pool = kzalloc(sizeof(struct z3fold_po 476 pool = kzalloc(sizeof(struct z3fold_pool), gfp);
915 if (!pool) 477 if (!pool)
916 goto out; 478 goto out;
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); 479 spin_lock_init(&pool->lock);
923 spin_lock_init(&pool->stale_lock); 480 spin_lock_init(&pool->stale_lock);
924 pool->unbuddied = __alloc_percpu(sizeo !! 481 pool->unbuddied = __alloc_percpu(sizeof(struct list_head)*NCHUNKS, 2);
925 __ali <<
926 if (!pool->unbuddied) 482 if (!pool->unbuddied)
927 goto out_pool; 483 goto out_pool;
928 for_each_possible_cpu(cpu) { 484 for_each_possible_cpu(cpu) {
929 struct list_head *unbuddied = 485 struct list_head *unbuddied =
930 per_cpu_ptr(po 486 per_cpu_ptr(pool->unbuddied, cpu);
931 for_each_unbuddied_list(i, 0) 487 for_each_unbuddied_list(i, 0)
932 INIT_LIST_HEAD(&unbudd 488 INIT_LIST_HEAD(&unbuddied[i]);
933 } 489 }
>> 490 INIT_LIST_HEAD(&pool->lru);
934 INIT_LIST_HEAD(&pool->stale); 491 INIT_LIST_HEAD(&pool->stale);
935 atomic64_set(&pool->pages_nr, 0); 492 atomic64_set(&pool->pages_nr, 0);
936 pool->name = name; 493 pool->name = name;
937 pool->compact_wq = create_singlethread 494 pool->compact_wq = create_singlethread_workqueue(pool->name);
938 if (!pool->compact_wq) 495 if (!pool->compact_wq)
939 goto out_unbuddied; 496 goto out_unbuddied;
940 pool->release_wq = create_singlethread 497 pool->release_wq = create_singlethread_workqueue(pool->name);
941 if (!pool->release_wq) 498 if (!pool->release_wq)
942 goto out_wq; 499 goto out_wq;
943 INIT_WORK(&pool->work, free_pages_work 500 INIT_WORK(&pool->work, free_pages_work);
>> 501 pool->ops = ops;
944 return pool; 502 return pool;
945 503
946 out_wq: 504 out_wq:
947 destroy_workqueue(pool->compact_wq); 505 destroy_workqueue(pool->compact_wq);
948 out_unbuddied: 506 out_unbuddied:
949 free_percpu(pool->unbuddied); 507 free_percpu(pool->unbuddied);
950 out_pool: 508 out_pool:
951 kmem_cache_destroy(pool->c_handle); <<
952 out_c: <<
953 kfree(pool); 509 kfree(pool);
954 out: 510 out:
955 return NULL; 511 return NULL;
956 } 512 }
957 513
958 /** 514 /**
959 * z3fold_destroy_pool() - destroys an existin 515 * z3fold_destroy_pool() - destroys an existing z3fold pool
960 * @pool: the z3fold pool to be destroye 516 * @pool: the z3fold pool to be destroyed
961 * 517 *
962 * The pool should be emptied before this func 518 * The pool should be emptied before this function is called.
963 */ 519 */
964 static void z3fold_destroy_pool(struct z3fold_ 520 static void z3fold_destroy_pool(struct z3fold_pool *pool)
965 { 521 {
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); 522 destroy_workqueue(pool->release_wq);
979 free_percpu(pool->unbuddied); !! 523 destroy_workqueue(pool->compact_wq);
980 kfree(pool); 524 kfree(pool);
981 } 525 }
982 526
983 static const struct movable_operations z3fold_ <<
984 <<
985 /** 527 /**
986 * z3fold_alloc() - allocates a region of a gi 528 * z3fold_alloc() - allocates a region of a given size
987 * @pool: z3fold pool from which to allo 529 * @pool: z3fold pool from which to allocate
988 * @size: size in bytes of the desired a 530 * @size: size in bytes of the desired allocation
989 * @gfp: gfp flags used if the pool nee 531 * @gfp: gfp flags used if the pool needs to grow
990 * @handle: handle of the new allocation 532 * @handle: handle of the new allocation
991 * 533 *
992 * This function will attempt to find a free r 534 * This function will attempt to find a free region in the pool large enough to
993 * satisfy the allocation request. A search o 535 * satisfy the allocation request. A search of the unbuddied lists is
994 * performed first. If no suitable free region 536 * performed first. If no suitable free region is found, then a new page is
995 * allocated and added to the pool to satisfy 537 * allocated and added to the pool to satisfy the request.
996 * 538 *
>> 539 * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used
>> 540 * as z3fold pool pages.
>> 541 *
997 * Return: 0 if success and handle is set, oth 542 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
998 * gfp arguments are invalid or -ENOMEM if the 543 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
999 * a new page. 544 * a new page.
1000 */ 545 */
1001 static int z3fold_alloc(struct z3fold_pool *p 546 static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
1002 unsigned long *handle 547 unsigned long *handle)
1003 { 548 {
1004 int chunks = size_to_chunks(size); !! 549 int chunks = 0, i, freechunks;
1005 struct z3fold_header *zhdr = NULL; 550 struct z3fold_header *zhdr = NULL;
1006 struct page *page = NULL; 551 struct page *page = NULL;
1007 enum buddy bud; 552 enum buddy bud;
1008 bool can_sleep = gfpflags_allow_block !! 553 bool can_sleep = (gfp & __GFP_RECLAIM) == __GFP_RECLAIM;
1009 554
1010 if (!size || (gfp & __GFP_HIGHMEM)) 555 if (!size || (gfp & __GFP_HIGHMEM))
1011 return -EINVAL; 556 return -EINVAL;
1012 557
1013 if (size > PAGE_SIZE) 558 if (size > PAGE_SIZE)
1014 return -ENOSPC; 559 return -ENOSPC;
1015 560
1016 if (size > PAGE_SIZE - ZHDR_SIZE_ALIG 561 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
1017 bud = HEADLESS; 562 bud = HEADLESS;
1018 else { 563 else {
1019 retry: !! 564 struct list_head *unbuddied;
1020 zhdr = __z3fold_alloc(pool, s !! 565 chunks = size_to_chunks(size);
>> 566
>> 567 lookup:
>> 568 /* First, try to find an unbuddied z3fold page. */
>> 569 unbuddied = get_cpu_ptr(pool->unbuddied);
>> 570 for_each_unbuddied_list(i, chunks) {
>> 571 struct list_head *l = &unbuddied[i];
>> 572
>> 573 zhdr = list_first_entry_or_null(READ_ONCE(l),
>> 574 struct z3fold_header, buddy);
>> 575
>> 576 if (!zhdr)
>> 577 continue;
>> 578
>> 579 /* Re-check under lock. */
>> 580 spin_lock(&pool->lock);
>> 581 l = &unbuddied[i];
>> 582 if (unlikely(zhdr != list_first_entry(READ_ONCE(l),
>> 583 struct z3fold_header, buddy)) ||
>> 584 !z3fold_page_trylock(zhdr)) {
>> 585 spin_unlock(&pool->lock);
>> 586 put_cpu_ptr(pool->unbuddied);
>> 587 goto lookup;
>> 588 }
>> 589 list_del_init(&zhdr->buddy);
>> 590 zhdr->cpu = -1;
>> 591 spin_unlock(&pool->lock);
>> 592
>> 593 page = virt_to_page(zhdr);
>> 594 if (test_bit(NEEDS_COMPACTING, &page->private)) {
>> 595 z3fold_page_unlock(zhdr);
>> 596 zhdr = NULL;
>> 597 put_cpu_ptr(pool->unbuddied);
>> 598 if (can_sleep)
>> 599 cond_resched();
>> 600 goto lookup;
>> 601 }
>> 602
>> 603 /*
>> 604 * this page could not be removed from its unbuddied
>> 605 * list while pool lock was held, and then we've taken
>> 606 * page lock so kref_put could not be called before
>> 607 * we got here, so it's safe to just call kref_get()
>> 608 */
>> 609 kref_get(&zhdr->refcount);
>> 610 break;
>> 611 }
>> 612 put_cpu_ptr(pool->unbuddied);
>> 613
1021 if (zhdr) { 614 if (zhdr) {
1022 bud = get_free_buddy( !! 615 if (zhdr->first_chunks == 0) {
1023 if (bud == HEADLESS) !! 616 if (zhdr->middle_chunks != 0 &&
1024 if (!put_z3fo !! 617 chunks >= zhdr->start_middle)
>> 618 bud = LAST;
>> 619 else
>> 620 bud = FIRST;
>> 621 } else if (zhdr->last_chunks == 0)
>> 622 bud = LAST;
>> 623 else if (zhdr->middle_chunks == 0)
>> 624 bud = MIDDLE;
>> 625 else {
>> 626 if (kref_put(&zhdr->refcount,
>> 627 release_z3fold_page_locked))
>> 628 atomic64_dec(&pool->pages_nr);
>> 629 else
1025 z3fol 630 z3fold_page_unlock(zhdr);
1026 pr_err("No fr 631 pr_err("No free chunks in unbuddied\n");
1027 WARN_ON(1); 632 WARN_ON(1);
1028 goto retry; !! 633 goto lookup;
1029 } 634 }
1030 page = virt_to_page(z <<
1031 goto found; 635 goto found;
1032 } 636 }
1033 bud = FIRST; 637 bud = FIRST;
1034 } 638 }
1035 639
1036 page = alloc_page(gfp); !! 640 spin_lock(&pool->stale_lock);
>> 641 zhdr = list_first_entry_or_null(&pool->stale,
>> 642 struct z3fold_header, buddy);
>> 643 /*
>> 644 * Before allocating a page, let's see if we can take one from the
>> 645 * stale pages list. cancel_work_sync() can sleep so we must make
>> 646 * sure it won't be called in case we're in atomic context.
>> 647 */
>> 648 if (zhdr && (can_sleep || !work_pending(&zhdr->work))) {
>> 649 list_del(&zhdr->buddy);
>> 650 spin_unlock(&pool->stale_lock);
>> 651 if (can_sleep)
>> 652 cancel_work_sync(&zhdr->work);
>> 653 page = virt_to_page(zhdr);
>> 654 } else {
>> 655 spin_unlock(&pool->stale_lock);
>> 656 page = alloc_page(gfp);
>> 657 }
>> 658
1037 if (!page) 659 if (!page)
1038 return -ENOMEM; 660 return -ENOMEM;
1039 661
1040 zhdr = init_z3fold_page(page, bud == <<
1041 if (!zhdr) { <<
1042 __free_page(page); <<
1043 return -ENOMEM; <<
1044 } <<
1045 atomic64_inc(&pool->pages_nr); 662 atomic64_inc(&pool->pages_nr);
>> 663 zhdr = init_z3fold_page(page, pool);
1046 664
1047 if (bud == HEADLESS) { 665 if (bud == HEADLESS) {
1048 set_bit(PAGE_HEADLESS, &page- 666 set_bit(PAGE_HEADLESS, &page->private);
1049 goto headless; 667 goto headless;
1050 } 668 }
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); 669 z3fold_page_lock(zhdr);
1061 670
1062 found: 671 found:
1063 if (bud == FIRST) 672 if (bud == FIRST)
1064 zhdr->first_chunks = chunks; 673 zhdr->first_chunks = chunks;
1065 else if (bud == LAST) 674 else if (bud == LAST)
1066 zhdr->last_chunks = chunks; 675 zhdr->last_chunks = chunks;
1067 else { 676 else {
1068 zhdr->middle_chunks = chunks; 677 zhdr->middle_chunks = chunks;
1069 zhdr->start_middle = zhdr->fi 678 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
1070 } 679 }
1071 add_to_unbuddied(pool, zhdr); !! 680
>> 681 if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 ||
>> 682 zhdr->middle_chunks == 0) {
>> 683 struct list_head *unbuddied = get_cpu_ptr(pool->unbuddied);
>> 684
>> 685 /* Add to unbuddied list */
>> 686 freechunks = num_free_chunks(zhdr);
>> 687 spin_lock(&pool->lock);
>> 688 list_add(&zhdr->buddy, &unbuddied[freechunks]);
>> 689 spin_unlock(&pool->lock);
>> 690 zhdr->cpu = smp_processor_id();
>> 691 put_cpu_ptr(pool->unbuddied);
>> 692 }
1072 693
1073 headless: 694 headless:
1074 spin_lock(&pool->lock); 695 spin_lock(&pool->lock);
>> 696 /* Add/move z3fold page to beginning of LRU */
>> 697 if (!list_empty(&page->lru))
>> 698 list_del(&page->lru);
>> 699
>> 700 list_add(&page->lru, &pool->lru);
>> 701
1075 *handle = encode_handle(zhdr, bud); 702 *handle = encode_handle(zhdr, bud);
1076 spin_unlock(&pool->lock); 703 spin_unlock(&pool->lock);
1077 if (bud != HEADLESS) 704 if (bud != HEADLESS)
1078 z3fold_page_unlock(zhdr); 705 z3fold_page_unlock(zhdr);
1079 706
1080 return 0; 707 return 0;
1081 } 708 }
1082 709
1083 /** 710 /**
1084 * z3fold_free() - frees the allocation assoc 711 * z3fold_free() - frees the allocation associated with the given handle
1085 * @pool: pool in which the allocation 712 * @pool: pool in which the allocation resided
1086 * @handle: handle associated with the al 713 * @handle: handle associated with the allocation returned by z3fold_alloc()
1087 * 714 *
1088 * In the case that the z3fold page in which 715 * In the case that the z3fold page in which the allocation resides is under
1089 * reclaim, as indicated by the PAGE_CLAIMED !! 716 * reclaim, as indicated by the PG_reclaim flag being set, this function
1090 * only sets the first|middle|last_chunks to !! 717 * only sets the first|last_chunks to 0. The page is actually freed
1091 * once all buddies are evicted (see z3fold_r !! 718 * once both buddies are evicted (see z3fold_reclaim_page() below).
1092 */ 719 */
1093 static void z3fold_free(struct z3fold_pool *p 720 static void z3fold_free(struct z3fold_pool *pool, unsigned long handle)
1094 { 721 {
1095 struct z3fold_header *zhdr; 722 struct z3fold_header *zhdr;
1096 struct page *page; 723 struct page *page;
1097 enum buddy bud; 724 enum buddy bud;
1098 bool page_claimed; <<
1099 725
1100 zhdr = get_z3fold_header(handle); !! 726 zhdr = handle_to_z3fold_header(handle);
1101 page = virt_to_page(zhdr); 727 page = virt_to_page(zhdr);
1102 page_claimed = test_and_set_bit(PAGE_ <<
1103 728
1104 if (test_bit(PAGE_HEADLESS, &page->pr 729 if (test_bit(PAGE_HEADLESS, &page->private)) {
1105 /* if a headless page is unde 730 /* if a headless page is under reclaim, just leave.
1106 * NB: we use test_and_set_bi 731 * NB: we use test_and_set_bit for a reason: if the bit
1107 * has not been set before, w 732 * has not been set before, we release this page
1108 * immediately so we don't ca 733 * immediately so we don't care about its value any more.
1109 */ 734 */
1110 if (!page_claimed) { !! 735 if (!test_and_set_bit(PAGE_CLAIMED, &page->private)) {
1111 put_z3fold_header(zhd !! 736 spin_lock(&pool->lock);
1112 free_z3fold_page(page !! 737 list_del(&page->lru);
>> 738 spin_unlock(&pool->lock);
>> 739 free_z3fold_page(page);
1113 atomic64_dec(&pool->p 740 atomic64_dec(&pool->pages_nr);
1114 } 741 }
1115 return; 742 return;
1116 } 743 }
1117 744
1118 /* Non-headless case */ 745 /* Non-headless case */
>> 746 z3fold_page_lock(zhdr);
1119 bud = handle_to_buddy(handle); 747 bud = handle_to_buddy(handle);
1120 748
1121 switch (bud) { 749 switch (bud) {
1122 case FIRST: 750 case FIRST:
1123 zhdr->first_chunks = 0; 751 zhdr->first_chunks = 0;
1124 break; 752 break;
1125 case MIDDLE: 753 case MIDDLE:
1126 zhdr->middle_chunks = 0; 754 zhdr->middle_chunks = 0;
1127 break; 755 break;
1128 case LAST: 756 case LAST:
1129 zhdr->last_chunks = 0; 757 zhdr->last_chunks = 0;
1130 break; 758 break;
1131 default: 759 default:
1132 pr_err("%s: unknown bud %d\n" 760 pr_err("%s: unknown bud %d\n", __func__, bud);
1133 WARN_ON(1); 761 WARN_ON(1);
1134 put_z3fold_header(zhdr); !! 762 z3fold_page_unlock(zhdr);
1135 return; 763 return;
1136 } 764 }
1137 765
1138 if (!page_claimed) !! 766 if (kref_put(&zhdr->refcount, release_z3fold_page_locked_list)) {
1139 free_handle(handle, zhdr); !! 767 atomic64_dec(&pool->pages_nr);
1140 if (put_z3fold_locked_list(zhdr)) <<
1141 return; 768 return;
1142 if (page_claimed) { !! 769 }
1143 /* the page has not been clai !! 770 if (test_bit(PAGE_CLAIMED, &page->private)) {
1144 put_z3fold_header(zhdr); !! 771 z3fold_page_unlock(zhdr);
1145 return; 772 return;
1146 } 773 }
1147 if (test_and_set_bit(NEEDS_COMPACTING 774 if (test_and_set_bit(NEEDS_COMPACTING, &page->private)) {
1148 clear_bit(PAGE_CLAIMED, &page !! 775 z3fold_page_unlock(zhdr);
1149 put_z3fold_header(zhdr); <<
1150 return; 776 return;
1151 } 777 }
1152 if (zhdr->cpu < 0 || !cpu_online(zhdr 778 if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) {
>> 779 spin_lock(&pool->lock);
>> 780 list_del_init(&zhdr->buddy);
>> 781 spin_unlock(&pool->lock);
1153 zhdr->cpu = -1; 782 zhdr->cpu = -1;
1154 kref_get(&zhdr->refcount); 783 kref_get(&zhdr->refcount);
1155 clear_bit(PAGE_CLAIMED, &page <<
1156 do_compact_page(zhdr, true); 784 do_compact_page(zhdr, true);
1157 return; 785 return;
1158 } 786 }
1159 kref_get(&zhdr->refcount); 787 kref_get(&zhdr->refcount);
1160 clear_bit(PAGE_CLAIMED, &page->privat <<
1161 queue_work_on(zhdr->cpu, pool->compac 788 queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work);
1162 put_z3fold_header(zhdr); !! 789 z3fold_page_unlock(zhdr);
>> 790 }
>> 791
>> 792 /**
>> 793 * z3fold_reclaim_page() - evicts allocations from a pool page and frees it
>> 794 * @pool: pool from which a page will attempt to be evicted
>> 795 * @retires: number of pages on the LRU list for which eviction will
>> 796 * be attempted before failing
>> 797 *
>> 798 * z3fold reclaim is different from normal system reclaim in that it is done
>> 799 * from the bottom, up. This is because only the bottom layer, z3fold, has
>> 800 * information on how the allocations are organized within each z3fold page.
>> 801 * This has the potential to create interesting locking situations between
>> 802 * z3fold and the user, however.
>> 803 *
>> 804 * To avoid these, this is how z3fold_reclaim_page() should be called:
>> 805
>> 806 * The user detects a page should be reclaimed and calls z3fold_reclaim_page().
>> 807 * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and
>> 808 * call the user-defined eviction handler with the pool and handle as
>> 809 * arguments.
>> 810 *
>> 811 * If the handle can not be evicted, the eviction handler should return
>> 812 * non-zero. z3fold_reclaim_page() will add the z3fold page back to the
>> 813 * appropriate list and try the next z3fold page on the LRU up to
>> 814 * a user defined number of retries.
>> 815 *
>> 816 * If the handle is successfully evicted, the eviction handler should
>> 817 * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free()
>> 818 * contains logic to delay freeing the page if the page is under reclaim,
>> 819 * as indicated by the setting of the PG_reclaim flag on the underlying page.
>> 820 *
>> 821 * If all buddies in the z3fold page are successfully evicted, then the
>> 822 * z3fold page can be freed.
>> 823 *
>> 824 * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are
>> 825 * no pages to evict or an eviction handler is not registered, -EAGAIN if
>> 826 * the retry limit was hit.
>> 827 */
>> 828 static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries)
>> 829 {
>> 830 int i, ret = 0;
>> 831 struct z3fold_header *zhdr = NULL;
>> 832 struct page *page = NULL;
>> 833 struct list_head *pos;
>> 834 unsigned long first_handle = 0, middle_handle = 0, last_handle = 0;
>> 835
>> 836 spin_lock(&pool->lock);
>> 837 if (!pool->ops || !pool->ops->evict || retries == 0) {
>> 838 spin_unlock(&pool->lock);
>> 839 return -EINVAL;
>> 840 }
>> 841 for (i = 0; i < retries; i++) {
>> 842 if (list_empty(&pool->lru)) {
>> 843 spin_unlock(&pool->lock);
>> 844 return -EINVAL;
>> 845 }
>> 846 list_for_each_prev(pos, &pool->lru) {
>> 847 page = list_entry(pos, struct page, lru);
>> 848
>> 849 /* this bit could have been set by free, in which case
>> 850 * we pass over to the next page in the pool.
>> 851 */
>> 852 if (test_and_set_bit(PAGE_CLAIMED, &page->private))
>> 853 continue;
>> 854
>> 855 zhdr = page_address(page);
>> 856 if (test_bit(PAGE_HEADLESS, &page->private))
>> 857 break;
>> 858
>> 859 if (!z3fold_page_trylock(zhdr)) {
>> 860 zhdr = NULL;
>> 861 continue; /* can't evict at this point */
>> 862 }
>> 863 kref_get(&zhdr->refcount);
>> 864 list_del_init(&zhdr->buddy);
>> 865 zhdr->cpu = -1;
>> 866 break;
>> 867 }
>> 868
>> 869 if (!zhdr)
>> 870 break;
>> 871
>> 872 list_del_init(&page->lru);
>> 873 spin_unlock(&pool->lock);
>> 874
>> 875 if (!test_bit(PAGE_HEADLESS, &page->private)) {
>> 876 /*
>> 877 * We need encode the handles before unlocking, since
>> 878 * we can race with free that will set
>> 879 * (first|last)_chunks to 0
>> 880 */
>> 881 first_handle = 0;
>> 882 last_handle = 0;
>> 883 middle_handle = 0;
>> 884 if (zhdr->first_chunks)
>> 885 first_handle = encode_handle(zhdr, FIRST);
>> 886 if (zhdr->middle_chunks)
>> 887 middle_handle = encode_handle(zhdr, MIDDLE);
>> 888 if (zhdr->last_chunks)
>> 889 last_handle = encode_handle(zhdr, LAST);
>> 890 /*
>> 891 * it's safe to unlock here because we hold a
>> 892 * reference to this page
>> 893 */
>> 894 z3fold_page_unlock(zhdr);
>> 895 } else {
>> 896 first_handle = encode_handle(zhdr, HEADLESS);
>> 897 last_handle = middle_handle = 0;
>> 898 }
>> 899
>> 900 /* Issue the eviction callback(s) */
>> 901 if (middle_handle) {
>> 902 ret = pool->ops->evict(pool, middle_handle);
>> 903 if (ret)
>> 904 goto next;
>> 905 }
>> 906 if (first_handle) {
>> 907 ret = pool->ops->evict(pool, first_handle);
>> 908 if (ret)
>> 909 goto next;
>> 910 }
>> 911 if (last_handle) {
>> 912 ret = pool->ops->evict(pool, last_handle);
>> 913 if (ret)
>> 914 goto next;
>> 915 }
>> 916 next:
>> 917 if (test_bit(PAGE_HEADLESS, &page->private)) {
>> 918 if (ret == 0) {
>> 919 free_z3fold_page(page);
>> 920 atomic64_dec(&pool->pages_nr);
>> 921 return 0;
>> 922 }
>> 923 spin_lock(&pool->lock);
>> 924 list_add(&page->lru, &pool->lru);
>> 925 spin_unlock(&pool->lock);
>> 926 } else {
>> 927 z3fold_page_lock(zhdr);
>> 928 clear_bit(PAGE_CLAIMED, &page->private);
>> 929 if (kref_put(&zhdr->refcount,
>> 930 release_z3fold_page_locked)) {
>> 931 atomic64_dec(&pool->pages_nr);
>> 932 return 0;
>> 933 }
>> 934 /*
>> 935 * if we are here, the page is still not completely
>> 936 * free. Take the global pool lock then to be able
>> 937 * to add it back to the lru list
>> 938 */
>> 939 spin_lock(&pool->lock);
>> 940 list_add(&page->lru, &pool->lru);
>> 941 spin_unlock(&pool->lock);
>> 942 z3fold_page_unlock(zhdr);
>> 943 }
>> 944
>> 945 /* We started off locked to we need to lock the pool back */
>> 946 spin_lock(&pool->lock);
>> 947 }
>> 948 spin_unlock(&pool->lock);
>> 949 return -EAGAIN;
1163 } 950 }
1164 951
1165 /** 952 /**
1166 * z3fold_map() - maps the allocation associa 953 * z3fold_map() - maps the allocation associated with the given handle
1167 * @pool: pool in which the allocation 954 * @pool: pool in which the allocation resides
1168 * @handle: handle associated with the al 955 * @handle: handle associated with the allocation to be mapped
1169 * 956 *
1170 * Extracts the buddy number from handle and 957 * Extracts the buddy number from handle and constructs the pointer to the
1171 * correct starting chunk within the page. 958 * correct starting chunk within the page.
1172 * 959 *
1173 * Returns: a pointer to the mapped allocatio 960 * Returns: a pointer to the mapped allocation
1174 */ 961 */
1175 static void *z3fold_map(struct z3fold_pool *p 962 static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle)
1176 { 963 {
1177 struct z3fold_header *zhdr; 964 struct z3fold_header *zhdr;
1178 struct page *page; 965 struct page *page;
1179 void *addr; 966 void *addr;
1180 enum buddy buddy; 967 enum buddy buddy;
1181 968
1182 zhdr = get_z3fold_header(handle); !! 969 zhdr = handle_to_z3fold_header(handle);
1183 addr = zhdr; 970 addr = zhdr;
1184 page = virt_to_page(zhdr); 971 page = virt_to_page(zhdr);
1185 972
1186 if (test_bit(PAGE_HEADLESS, &page->pr 973 if (test_bit(PAGE_HEADLESS, &page->private))
1187 goto out; 974 goto out;
1188 975
>> 976 z3fold_page_lock(zhdr);
1189 buddy = handle_to_buddy(handle); 977 buddy = handle_to_buddy(handle);
1190 switch (buddy) { 978 switch (buddy) {
1191 case FIRST: 979 case FIRST:
1192 addr += ZHDR_SIZE_ALIGNED; 980 addr += ZHDR_SIZE_ALIGNED;
1193 break; 981 break;
1194 case MIDDLE: 982 case MIDDLE:
1195 addr += zhdr->start_middle << 983 addr += zhdr->start_middle << CHUNK_SHIFT;
1196 set_bit(MIDDLE_CHUNK_MAPPED, 984 set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1197 break; 985 break;
1198 case LAST: 986 case LAST:
1199 addr += PAGE_SIZE - (handle_t 987 addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT);
1200 break; 988 break;
1201 default: 989 default:
1202 pr_err("unknown buddy id %d\n 990 pr_err("unknown buddy id %d\n", buddy);
1203 WARN_ON(1); 991 WARN_ON(1);
1204 addr = NULL; 992 addr = NULL;
1205 break; 993 break;
1206 } 994 }
1207 995
1208 if (addr) !! 996 z3fold_page_unlock(zhdr);
1209 zhdr->mapped_count++; <<
1210 out: 997 out:
1211 put_z3fold_header(zhdr); <<
1212 return addr; 998 return addr;
1213 } 999 }
1214 1000
1215 /** 1001 /**
1216 * z3fold_unmap() - unmaps the allocation ass 1002 * z3fold_unmap() - unmaps the allocation associated with the given handle
1217 * @pool: pool in which the allocation 1003 * @pool: pool in which the allocation resides
1218 * @handle: handle associated with the al 1004 * @handle: handle associated with the allocation to be unmapped
1219 */ 1005 */
1220 static void z3fold_unmap(struct z3fold_pool * 1006 static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle)
1221 { 1007 {
1222 struct z3fold_header *zhdr; 1008 struct z3fold_header *zhdr;
1223 struct page *page; 1009 struct page *page;
1224 enum buddy buddy; 1010 enum buddy buddy;
1225 1011
1226 zhdr = get_z3fold_header(handle); !! 1012 zhdr = handle_to_z3fold_header(handle);
1227 page = virt_to_page(zhdr); 1013 page = virt_to_page(zhdr);
1228 1014
1229 if (test_bit(PAGE_HEADLESS, &page->pr 1015 if (test_bit(PAGE_HEADLESS, &page->private))
1230 return; 1016 return;
1231 1017
>> 1018 z3fold_page_lock(zhdr);
1232 buddy = handle_to_buddy(handle); 1019 buddy = handle_to_buddy(handle);
1233 if (buddy == MIDDLE) 1020 if (buddy == MIDDLE)
1234 clear_bit(MIDDLE_CHUNK_MAPPED 1021 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1235 zhdr->mapped_count--; !! 1022 z3fold_page_unlock(zhdr);
1236 put_z3fold_header(zhdr); <<
1237 } 1023 }
1238 1024
1239 /** 1025 /**
1240 * z3fold_get_pool_pages() - gets the z3fold !! 1026 * z3fold_get_pool_size() - gets the z3fold pool size in pages
1241 * @pool: pool whose size is being quer 1027 * @pool: pool whose size is being queried
1242 * 1028 *
1243 * Returns: size in pages of the given pool. 1029 * Returns: size in pages of the given pool.
1244 */ 1030 */
1245 static u64 z3fold_get_pool_pages(struct z3fol !! 1031 static u64 z3fold_get_pool_size(struct z3fold_pool *pool)
1246 { 1032 {
1247 return atomic64_read(&pool->pages_nr) 1033 return atomic64_read(&pool->pages_nr);
1248 } 1034 }
1249 1035
1250 static bool z3fold_page_isolate(struct page * !! 1036 /*****************
1251 { !! 1037 * zpool
1252 struct z3fold_header *zhdr; !! 1038 ****************/
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 1039
1345 static void z3fold_page_putback(struct page * !! 1040 static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle)
1346 { 1041 {
1347 struct z3fold_header *zhdr; !! 1042 if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict)
1348 struct z3fold_pool *pool; !! 1043 return pool->zpool_ops->evict(pool->zpool, handle);
1349 !! 1044 else
1350 zhdr = page_address(page); !! 1045 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 } 1046 }
1364 1047
1365 static const struct movable_operations z3fold !! 1048 static const struct z3fold_ops z3fold_zpool_ops = {
1366 .isolate_page = z3fold_page_isolate, !! 1049 .evict = z3fold_zpool_evict
1367 .migrate_page = z3fold_page_migrate, <<
1368 .putback_page = z3fold_page_putback, <<
1369 }; 1050 };
1370 1051
1371 /***************** !! 1052 static void *z3fold_zpool_create(const char *name, gfp_t gfp,
1372 * zpool !! 1053 const struct zpool_ops *zpool_ops,
1373 ****************/ !! 1054 struct zpool *zpool)
1374 <<
1375 static void *z3fold_zpool_create(const char * <<
1376 { 1055 {
1377 return z3fold_create_pool(name, gfp); !! 1056 struct z3fold_pool *pool;
>> 1057
>> 1058 pool = z3fold_create_pool(name, gfp,
>> 1059 zpool_ops ? &z3fold_zpool_ops : NULL);
>> 1060 if (pool) {
>> 1061 pool->zpool = zpool;
>> 1062 pool->zpool_ops = zpool_ops;
>> 1063 }
>> 1064 return pool;
1378 } 1065 }
1379 1066
1380 static void z3fold_zpool_destroy(void *pool) 1067 static void z3fold_zpool_destroy(void *pool)
1381 { 1068 {
1382 z3fold_destroy_pool(pool); 1069 z3fold_destroy_pool(pool);
1383 } 1070 }
1384 1071
1385 static int z3fold_zpool_malloc(void *pool, si 1072 static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp,
1386 unsigned long *handle 1073 unsigned long *handle)
1387 { 1074 {
1388 return z3fold_alloc(pool, size, gfp, 1075 return z3fold_alloc(pool, size, gfp, handle);
1389 } 1076 }
1390 static void z3fold_zpool_free(void *pool, uns 1077 static void z3fold_zpool_free(void *pool, unsigned long handle)
1391 { 1078 {
1392 z3fold_free(pool, handle); 1079 z3fold_free(pool, handle);
1393 } 1080 }
1394 1081
>> 1082 static int z3fold_zpool_shrink(void *pool, unsigned int pages,
>> 1083 unsigned int *reclaimed)
>> 1084 {
>> 1085 unsigned int total = 0;
>> 1086 int ret = -EINVAL;
>> 1087
>> 1088 while (total < pages) {
>> 1089 ret = z3fold_reclaim_page(pool, 8);
>> 1090 if (ret < 0)
>> 1091 break;
>> 1092 total++;
>> 1093 }
>> 1094
>> 1095 if (reclaimed)
>> 1096 *reclaimed = total;
>> 1097
>> 1098 return ret;
>> 1099 }
>> 1100
1395 static void *z3fold_zpool_map(void *pool, uns 1101 static void *z3fold_zpool_map(void *pool, unsigned long handle,
1396 enum zpool_mapmode mm 1102 enum zpool_mapmode mm)
1397 { 1103 {
1398 return z3fold_map(pool, handle); 1104 return z3fold_map(pool, handle);
1399 } 1105 }
1400 static void z3fold_zpool_unmap(void *pool, un 1106 static void z3fold_zpool_unmap(void *pool, unsigned long handle)
1401 { 1107 {
1402 z3fold_unmap(pool, handle); 1108 z3fold_unmap(pool, handle);
1403 } 1109 }
1404 1110
1405 static u64 z3fold_zpool_total_pages(void *poo !! 1111 static u64 z3fold_zpool_total_size(void *pool)
1406 { 1112 {
1407 return z3fold_get_pool_pages(pool); !! 1113 return z3fold_get_pool_size(pool) * PAGE_SIZE;
1408 } 1114 }
1409 1115
1410 static struct zpool_driver z3fold_zpool_drive 1116 static struct zpool_driver z3fold_zpool_driver = {
1411 .type = "z3fold", 1117 .type = "z3fold",
1412 .sleep_mapped = true, <<
1413 .owner = THIS_MODULE, 1118 .owner = THIS_MODULE,
1414 .create = z3fold_zpool_create, 1119 .create = z3fold_zpool_create,
1415 .destroy = z3fold_zpool_destroy, 1120 .destroy = z3fold_zpool_destroy,
1416 .malloc = z3fold_zpool_malloc, 1121 .malloc = z3fold_zpool_malloc,
1417 .free = z3fold_zpool_free, 1122 .free = z3fold_zpool_free,
>> 1123 .shrink = z3fold_zpool_shrink,
1418 .map = z3fold_zpool_map, 1124 .map = z3fold_zpool_map,
1419 .unmap = z3fold_zpool_unmap, 1125 .unmap = z3fold_zpool_unmap,
1420 .total_pages = z3fold_zpool_total_pa !! 1126 .total_size = z3fold_zpool_total_size,
1421 }; 1127 };
1422 1128
1423 MODULE_ALIAS("zpool-z3fold"); 1129 MODULE_ALIAS("zpool-z3fold");
1424 1130
1425 static int __init init_z3fold(void) 1131 static int __init init_z3fold(void)
1426 { 1132 {
1427 /* !! 1133 /* Make sure the z3fold header is not larger than the page size */
1428 * Make sure the z3fold header is not !! 1134 BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE);
1429 * there has remaining spaces for its <<
1430 */ <<
1431 BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE <<
1432 zpool_register_driver(&z3fold_zpool_d 1135 zpool_register_driver(&z3fold_zpool_driver);
1433 1136
1434 return 0; 1137 return 0;
1435 } 1138 }
1436 1139
1437 static void __exit exit_z3fold(void) 1140 static void __exit exit_z3fold(void)
1438 { 1141 {
1439 zpool_unregister_driver(&z3fold_zpool 1142 zpool_unregister_driver(&z3fold_zpool_driver);
1440 } 1143 }
1441 1144
1442 module_init(init_z3fold); 1145 module_init(init_z3fold);
1443 module_exit(exit_z3fold); 1146 module_exit(exit_z3fold);
1444 1147
1445 MODULE_LICENSE("GPL"); 1148 MODULE_LICENSE("GPL");
1446 MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail. 1149 MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>");
1447 MODULE_DESCRIPTION("3-Fold Allocator for Comp 1150 MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages");
1448 1151
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.