1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /* 2 /*
3 * z3fold.c 3 * z3fold.c
4 * 4 *
5 * Author: Vitaly Wool <vitaly.wool@konsulko.c 5 * Author: Vitaly Wool <vitaly.wool@konsulko.com>
6 * Copyright (C) 2016, Sony Mobile Communicati 6 * Copyright (C) 2016, Sony Mobile Communications Inc.
7 * 7 *
8 * This implementation is based on zbud writte 8 * This implementation is based on zbud written by Seth Jennings.
9 * 9 *
10 * z3fold is an special purpose allocator for 10 * z3fold is an special purpose allocator for storing compressed pages. It
11 * can store up to three compressed pages per 11 * can store up to three compressed pages per page which improves the
12 * compression ratio of zbud while retaining i 12 * compression ratio of zbud while retaining its main concepts (e. g. always
13 * storing an integral number of objects per p 13 * storing an integral number of objects per page) and simplicity.
14 * It still has simple and deterministic recla 14 * It still has simple and deterministic reclaim properties that make it
15 * preferable to a higher density approach (wi 15 * preferable to a higher density approach (with no requirement on integral
16 * number of object per page) when reclaim is 16 * number of object per page) when reclaim is used.
17 * 17 *
18 * As in zbud, pages are divided into "chunks" 18 * As in zbud, pages are divided into "chunks". The size of the chunks is
19 * fixed at compile time and is determined by 19 * fixed at compile time and is determined by NCHUNKS_ORDER below.
20 * 20 *
21 * z3fold doesn't export any API and is meant 21 * z3fold doesn't export any API and is meant to be used via zpool API.
22 */ 22 */
23 23
24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25 25
26 #include <linux/atomic.h> 26 #include <linux/atomic.h>
27 #include <linux/sched.h> 27 #include <linux/sched.h>
28 #include <linux/cpumask.h> 28 #include <linux/cpumask.h>
>> 29 #include <linux/dcache.h>
29 #include <linux/list.h> 30 #include <linux/list.h>
30 #include <linux/mm.h> 31 #include <linux/mm.h>
31 #include <linux/module.h> 32 #include <linux/module.h>
32 #include <linux/page-flags.h> 33 #include <linux/page-flags.h>
33 #include <linux/migrate.h> 34 #include <linux/migrate.h>
34 #include <linux/node.h> 35 #include <linux/node.h>
35 #include <linux/compaction.h> 36 #include <linux/compaction.h>
36 #include <linux/percpu.h> 37 #include <linux/percpu.h>
>> 38 #include <linux/mount.h>
>> 39 #include <linux/fs.h>
37 #include <linux/preempt.h> 40 #include <linux/preempt.h>
38 #include <linux/workqueue.h> 41 #include <linux/workqueue.h>
39 #include <linux/slab.h> 42 #include <linux/slab.h>
40 #include <linux/spinlock.h> 43 #include <linux/spinlock.h>
41 #include <linux/zpool.h> 44 #include <linux/zpool.h>
42 #include <linux/kmemleak.h> <<
43 45
44 /* 46 /*
45 * NCHUNKS_ORDER determines the internal alloc 47 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
46 * adjusting internal fragmentation. It also 48 * adjusting internal fragmentation. It also determines the number of
47 * freelists maintained in each pool. NCHUNKS_ 49 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
48 * allocation granularity will be in chunks of 50 * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks
49 * in the beginning of an allocated page are o 51 * in the beginning of an allocated page are occupied by z3fold header, so
50 * NCHUNKS will be calculated to 63 (or 62 in 52 * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y),
51 * which shows the max number of free chunks i 53 * which shows the max number of free chunks in z3fold page, also there will
52 * be 63, or 62, respectively, freelists per p 54 * be 63, or 62, respectively, freelists per pool.
53 */ 55 */
54 #define NCHUNKS_ORDER 6 56 #define NCHUNKS_ORDER 6
55 57
56 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ 58 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER)
57 #define CHUNK_SIZE (1 << CHUNK_SHIFT) 59 #define CHUNK_SIZE (1 << CHUNK_SHIFT)
58 #define ZHDR_SIZE_ALIGNED round_up(sizeof(stru 60 #define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE)
59 #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> 61 #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT)
60 #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SH 62 #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SHIFT)
61 #define NCHUNKS (TOTAL_CHUNKS - ZHDR_C !! 63 #define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT)
62 64
63 #define BUDDY_MASK (0x3) 65 #define BUDDY_MASK (0x3)
64 #define BUDDY_SHIFT 2 66 #define BUDDY_SHIFT 2
65 #define SLOTS_ALIGN (0x40) 67 #define SLOTS_ALIGN (0x40)
66 68
67 /***************** 69 /*****************
68 * Structures 70 * Structures
69 *****************/ 71 *****************/
70 struct z3fold_pool; 72 struct z3fold_pool;
>> 73 struct z3fold_ops {
>> 74 int (*evict)(struct z3fold_pool *pool, unsigned long handle);
>> 75 };
71 76
72 enum buddy { 77 enum buddy {
73 HEADLESS = 0, 78 HEADLESS = 0,
74 FIRST, 79 FIRST,
75 MIDDLE, 80 MIDDLE,
76 LAST, 81 LAST,
77 BUDDIES_MAX = LAST 82 BUDDIES_MAX = LAST
78 }; 83 };
79 84
80 struct z3fold_buddy_slots { 85 struct z3fold_buddy_slots {
81 /* 86 /*
82 * we are using BUDDY_MASK in handle_t 87 * we are using BUDDY_MASK in handle_to_buddy etc. so there should
83 * be enough slots to hold all possibl 88 * be enough slots to hold all possible variants
84 */ 89 */
85 unsigned long slot[BUDDY_MASK + 1]; 90 unsigned long slot[BUDDY_MASK + 1];
86 unsigned long pool; /* back link */ !! 91 unsigned long pool; /* back link + flags */
87 rwlock_t lock; <<
88 }; 92 };
89 #define HANDLE_FLAG_MASK (0x03) 93 #define HANDLE_FLAG_MASK (0x03)
90 94
91 /* 95 /*
92 * struct z3fold_header - z3fold page metadata 96 * struct z3fold_header - z3fold page metadata occupying first chunks of each
93 * z3fold page, except fo 97 * z3fold page, except for HEADLESS pages
94 * @buddy: links the z3fold page 98 * @buddy: links the z3fold page into the relevant list in the
95 * pool 99 * pool
96 * @page_lock: per-page lock 100 * @page_lock: per-page lock
97 * @refcount: reference count for th 101 * @refcount: reference count for the z3fold page
98 * @work: work_struct for page l 102 * @work: work_struct for page layout optimization
99 * @slots: pointer to the structu 103 * @slots: pointer to the structure holding buddy slots
100 * @pool: pointer to the contain 104 * @pool: pointer to the containing pool
101 * @cpu: CPU which this page "b 105 * @cpu: CPU which this page "belongs" to
102 * @first_chunks: the size of the first 106 * @first_chunks: the size of the first buddy in chunks, 0 if free
103 * @middle_chunks: the size of the middle 107 * @middle_chunks: the size of the middle buddy in chunks, 0 if free
104 * @last_chunks: the size of the last b 108 * @last_chunks: the size of the last buddy in chunks, 0 if free
105 * @first_num: the starting number (f 109 * @first_num: the starting number (for the first handle)
106 * @mapped_count: the number of objects 110 * @mapped_count: the number of objects currently mapped
107 */ 111 */
108 struct z3fold_header { 112 struct z3fold_header {
109 struct list_head buddy; 113 struct list_head buddy;
110 spinlock_t page_lock; 114 spinlock_t page_lock;
111 struct kref refcount; 115 struct kref refcount;
112 struct work_struct work; 116 struct work_struct work;
113 struct z3fold_buddy_slots *slots; 117 struct z3fold_buddy_slots *slots;
114 struct z3fold_pool *pool; 118 struct z3fold_pool *pool;
115 short cpu; 119 short cpu;
116 unsigned short first_chunks; 120 unsigned short first_chunks;
117 unsigned short middle_chunks; 121 unsigned short middle_chunks;
118 unsigned short last_chunks; 122 unsigned short last_chunks;
119 unsigned short start_middle; 123 unsigned short start_middle;
120 unsigned short first_num:2; 124 unsigned short first_num:2;
121 unsigned short mapped_count:2; 125 unsigned short mapped_count:2;
122 unsigned short foreign_handles:2; <<
123 }; 126 };
124 127
125 /** 128 /**
126 * struct z3fold_pool - stores metadata for ea 129 * struct z3fold_pool - stores metadata for each z3fold pool
127 * @name: pool name 130 * @name: pool name
128 * @lock: protects pool unbuddied lists !! 131 * @lock: protects pool unbuddied/lru lists
129 * @stale_lock: protects pool stale page list 132 * @stale_lock: protects pool stale page list
130 * @unbuddied: per-cpu array of lists trackin 133 * @unbuddied: per-cpu array of lists tracking z3fold pages that contain 2-
131 * buddies; the list each z3fold 134 * buddies; the list each z3fold page is added to depends on
132 * the size of its free region. 135 * the size of its free region.
>> 136 * @lru: list tracking the z3fold pages in LRU order by most recently
>> 137 * added buddy.
133 * @stale: list of pages marked for freei 138 * @stale: list of pages marked for freeing
134 * @pages_nr: number of z3fold pages in the 139 * @pages_nr: number of z3fold pages in the pool.
135 * @c_handle: cache for z3fold_buddy_slots a 140 * @c_handle: cache for z3fold_buddy_slots allocation
>> 141 * @ops: pointer to a structure of user defined operations specified at
>> 142 * pool creation time.
136 * @compact_wq: workqueue for page layout back 143 * @compact_wq: workqueue for page layout background optimization
137 * @release_wq: workqueue for safe page releas 144 * @release_wq: workqueue for safe page release
138 * @work: work_struct for safe page rele 145 * @work: work_struct for safe page release
>> 146 * @inode: inode for z3fold pseudo filesystem
139 * 147 *
140 * This structure is allocated at pool creatio 148 * This structure is allocated at pool creation time and maintains metadata
141 * pertaining to a particular z3fold pool. 149 * pertaining to a particular z3fold pool.
142 */ 150 */
143 struct z3fold_pool { 151 struct z3fold_pool {
144 const char *name; 152 const char *name;
145 spinlock_t lock; 153 spinlock_t lock;
146 spinlock_t stale_lock; 154 spinlock_t stale_lock;
147 struct list_head __percpu *unbuddied; !! 155 struct list_head *unbuddied;
>> 156 struct list_head lru;
148 struct list_head stale; 157 struct list_head stale;
149 atomic64_t pages_nr; 158 atomic64_t pages_nr;
150 struct kmem_cache *c_handle; 159 struct kmem_cache *c_handle;
>> 160 const struct z3fold_ops *ops;
>> 161 struct zpool *zpool;
>> 162 const struct zpool_ops *zpool_ops;
151 struct workqueue_struct *compact_wq; 163 struct workqueue_struct *compact_wq;
152 struct workqueue_struct *release_wq; 164 struct workqueue_struct *release_wq;
153 struct work_struct work; 165 struct work_struct work;
>> 166 struct inode *inode;
154 }; 167 };
155 168
156 /* 169 /*
157 * Internal z3fold page flags 170 * Internal z3fold page flags
158 */ 171 */
159 enum z3fold_page_flags { 172 enum z3fold_page_flags {
160 PAGE_HEADLESS = 0, 173 PAGE_HEADLESS = 0,
161 MIDDLE_CHUNK_MAPPED, 174 MIDDLE_CHUNK_MAPPED,
162 NEEDS_COMPACTING, 175 NEEDS_COMPACTING,
163 PAGE_STALE, 176 PAGE_STALE,
164 PAGE_CLAIMED, /* by either reclaim or 177 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 }; 178 };
174 179
175 /* <<
176 * Forward declarations <<
177 */ <<
178 static struct z3fold_header *__z3fold_alloc(st <<
179 static void compact_page_work(struct work_stru <<
180 <<
181 /***************** 180 /*****************
182 * Helpers 181 * Helpers
183 *****************/ 182 *****************/
184 183
185 /* Converts an allocation size in bytes to siz 184 /* Converts an allocation size in bytes to size in z3fold chunks */
186 static int size_to_chunks(size_t size) 185 static int size_to_chunks(size_t size)
187 { 186 {
188 return (size + CHUNK_SIZE - 1) >> CHUN 187 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
189 } 188 }
190 189
191 #define for_each_unbuddied_list(_iter, _begin) 190 #define for_each_unbuddied_list(_iter, _begin) \
192 for ((_iter) = (_begin); (_iter) < NCH 191 for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
193 192
>> 193 static void compact_page_work(struct work_struct *w);
>> 194
194 static inline struct z3fold_buddy_slots *alloc 195 static inline struct z3fold_buddy_slots *alloc_slots(struct z3fold_pool *pool,
195 196 gfp_t gfp)
196 { 197 {
197 struct z3fold_buddy_slots *slots = kme !! 198 struct z3fold_buddy_slots *slots = kmem_cache_alloc(pool->c_handle,
198 !! 199 gfp);
199 200
200 if (slots) { 201 if (slots) {
201 /* It will be freed separately !! 202 memset(slots->slot, 0, sizeof(slots->slot));
202 kmemleak_not_leak(slots); <<
203 slots->pool = (unsigned long)p 203 slots->pool = (unsigned long)pool;
204 rwlock_init(&slots->lock); <<
205 } 204 }
206 205
207 return slots; 206 return slots;
208 } 207 }
209 208
210 static inline struct z3fold_pool *slots_to_poo 209 static inline struct z3fold_pool *slots_to_pool(struct z3fold_buddy_slots *s)
211 { 210 {
212 return (struct z3fold_pool *)(s->pool 211 return (struct z3fold_pool *)(s->pool & ~HANDLE_FLAG_MASK);
213 } 212 }
214 213
215 static inline struct z3fold_buddy_slots *handl 214 static inline struct z3fold_buddy_slots *handle_to_slots(unsigned long handle)
216 { 215 {
217 return (struct z3fold_buddy_slots *)(h 216 return (struct z3fold_buddy_slots *)(handle & ~(SLOTS_ALIGN - 1));
218 } 217 }
219 218
220 /* Lock a z3fold page */ !! 219 static inline void free_handle(unsigned long handle)
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 { 220 {
281 struct z3fold_buddy_slots *slots; 221 struct z3fold_buddy_slots *slots;
282 int i; 222 int i;
283 bool is_free; 223 bool is_free;
284 224
285 if (WARN_ON(*(unsigned long *)handle = !! 225 if (handle & (1 << PAGE_HEADLESS))
286 return; 226 return;
287 227
288 slots = handle_to_slots(handle); !! 228 WARN_ON(*(unsigned long *)handle == 0);
289 write_lock(&slots->lock); <<
290 *(unsigned long *)handle = 0; 229 *(unsigned long *)handle = 0;
291 !! 230 slots = handle_to_slots(handle);
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; 231 is_free = true;
301 for (i = 0; i <= BUDDY_MASK; i++) { 232 for (i = 0; i <= BUDDY_MASK; i++) {
302 if (slots->slot[i]) { 233 if (slots->slot[i]) {
303 is_free = false; 234 is_free = false;
304 break; 235 break;
305 } 236 }
306 } 237 }
307 write_unlock(&slots->lock); <<
308 238
309 if (is_free) { 239 if (is_free) {
310 struct z3fold_pool *pool = slo 240 struct z3fold_pool *pool = slots_to_pool(slots);
311 241
312 if (zhdr->slots == slots) <<
313 zhdr->slots = NULL; <<
314 kmem_cache_free(pool->c_handle 242 kmem_cache_free(pool->c_handle, slots);
315 } 243 }
316 } 244 }
317 245
>> 246 static struct dentry *z3fold_do_mount(struct file_system_type *fs_type,
>> 247 int flags, const char *dev_name, void *data)
>> 248 {
>> 249 static const struct dentry_operations ops = {
>> 250 .d_dname = simple_dname,
>> 251 };
>> 252
>> 253 return mount_pseudo(fs_type, "z3fold:", NULL, &ops, 0x33);
>> 254 }
>> 255
>> 256 static struct file_system_type z3fold_fs = {
>> 257 .name = "z3fold",
>> 258 .mount = z3fold_do_mount,
>> 259 .kill_sb = kill_anon_super,
>> 260 };
>> 261
>> 262 static struct vfsmount *z3fold_mnt;
>> 263 static int z3fold_mount(void)
>> 264 {
>> 265 int ret = 0;
>> 266
>> 267 z3fold_mnt = kern_mount(&z3fold_fs);
>> 268 if (IS_ERR(z3fold_mnt))
>> 269 ret = PTR_ERR(z3fold_mnt);
>> 270
>> 271 return ret;
>> 272 }
>> 273
>> 274 static void z3fold_unmount(void)
>> 275 {
>> 276 kern_unmount(z3fold_mnt);
>> 277 }
>> 278
>> 279 static const struct address_space_operations z3fold_aops;
>> 280 static int z3fold_register_migration(struct z3fold_pool *pool)
>> 281 {
>> 282 pool->inode = alloc_anon_inode(z3fold_mnt->mnt_sb);
>> 283 if (IS_ERR(pool->inode)) {
>> 284 pool->inode = NULL;
>> 285 return 1;
>> 286 }
>> 287
>> 288 pool->inode->i_mapping->private_data = pool;
>> 289 pool->inode->i_mapping->a_ops = &z3fold_aops;
>> 290 return 0;
>> 291 }
>> 292
>> 293 static void z3fold_unregister_migration(struct z3fold_pool *pool)
>> 294 {
>> 295 if (pool->inode)
>> 296 iput(pool->inode);
>> 297 }
>> 298
318 /* Initializes the z3fold header of a newly al 299 /* Initializes the z3fold header of a newly allocated z3fold page */
319 static struct z3fold_header *init_z3fold_page( 300 static struct z3fold_header *init_z3fold_page(struct page *page, bool headless,
320 struct 301 struct z3fold_pool *pool, gfp_t gfp)
321 { 302 {
322 struct z3fold_header *zhdr = page_addr 303 struct z3fold_header *zhdr = page_address(page);
323 struct z3fold_buddy_slots *slots; 304 struct z3fold_buddy_slots *slots;
324 305
>> 306 INIT_LIST_HEAD(&page->lru);
325 clear_bit(PAGE_HEADLESS, &page->privat 307 clear_bit(PAGE_HEADLESS, &page->private);
326 clear_bit(MIDDLE_CHUNK_MAPPED, &page-> 308 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
327 clear_bit(NEEDS_COMPACTING, &page->pri 309 clear_bit(NEEDS_COMPACTING, &page->private);
328 clear_bit(PAGE_STALE, &page->private); 310 clear_bit(PAGE_STALE, &page->private);
329 clear_bit(PAGE_CLAIMED, &page->private 311 clear_bit(PAGE_CLAIMED, &page->private);
330 clear_bit(PAGE_MIGRATED, &page->privat <<
331 if (headless) 312 if (headless)
332 return zhdr; 313 return zhdr;
333 314
334 slots = alloc_slots(pool, gfp); 315 slots = alloc_slots(pool, gfp);
335 if (!slots) 316 if (!slots)
336 return NULL; 317 return NULL;
337 318
338 memset(zhdr, 0, sizeof(*zhdr)); <<
339 spin_lock_init(&zhdr->page_lock); 319 spin_lock_init(&zhdr->page_lock);
340 kref_init(&zhdr->refcount); 320 kref_init(&zhdr->refcount);
>> 321 zhdr->first_chunks = 0;
>> 322 zhdr->middle_chunks = 0;
>> 323 zhdr->last_chunks = 0;
>> 324 zhdr->first_num = 0;
>> 325 zhdr->start_middle = 0;
341 zhdr->cpu = -1; 326 zhdr->cpu = -1;
342 zhdr->slots = slots; 327 zhdr->slots = slots;
343 zhdr->pool = pool; 328 zhdr->pool = pool;
344 INIT_LIST_HEAD(&zhdr->buddy); 329 INIT_LIST_HEAD(&zhdr->buddy);
345 INIT_WORK(&zhdr->work, compact_page_wo 330 INIT_WORK(&zhdr->work, compact_page_work);
346 return zhdr; 331 return zhdr;
347 } 332 }
348 333
349 /* Resets the struct page fields and frees the 334 /* Resets the struct page fields and frees the page */
350 static void free_z3fold_page(struct page *page 335 static void free_z3fold_page(struct page *page, bool headless)
351 { 336 {
352 if (!headless) { 337 if (!headless) {
353 lock_page(page); 338 lock_page(page);
354 __ClearPageMovable(page); 339 __ClearPageMovable(page);
355 unlock_page(page); 340 unlock_page(page);
356 } 341 }
>> 342 ClearPagePrivate(page);
357 __free_page(page); 343 __free_page(page);
358 } 344 }
359 345
>> 346 /* Lock a z3fold page */
>> 347 static inline void z3fold_page_lock(struct z3fold_header *zhdr)
>> 348 {
>> 349 spin_lock(&zhdr->page_lock);
>> 350 }
>> 351
>> 352 /* Try to lock a z3fold page */
>> 353 static inline int z3fold_page_trylock(struct z3fold_header *zhdr)
>> 354 {
>> 355 return spin_trylock(&zhdr->page_lock);
>> 356 }
>> 357
>> 358 /* Unlock a z3fold page */
>> 359 static inline void z3fold_page_unlock(struct z3fold_header *zhdr)
>> 360 {
>> 361 spin_unlock(&zhdr->page_lock);
>> 362 }
>> 363
360 /* Helper function to build the index */ 364 /* Helper function to build the index */
361 static inline int __idx(struct z3fold_header * 365 static inline int __idx(struct z3fold_header *zhdr, enum buddy bud)
362 { 366 {
363 return (bud + zhdr->first_num) & BUDDY 367 return (bud + zhdr->first_num) & BUDDY_MASK;
364 } 368 }
365 369
366 /* 370 /*
367 * Encodes the handle of a particular buddy wi !! 371 * Encodes the handle of a particular buddy within a z3fold page
368 * Zhdr->page_lock should be held as this func !! 372 * Pool lock should be held as this function accesses first_num
369 * if bud != HEADLESS. <<
370 */ 373 */
371 static unsigned long __encode_handle(struct z3 374 static unsigned long __encode_handle(struct z3fold_header *zhdr,
372 struct z3fold_ 375 struct z3fold_buddy_slots *slots,
373 enum buddy bud 376 enum buddy bud)
374 { 377 {
375 unsigned long h = (unsigned long)zhdr; 378 unsigned long h = (unsigned long)zhdr;
376 int idx = 0; 379 int idx = 0;
377 380
378 /* 381 /*
379 * For a headless page, its handle is 382 * For a headless page, its handle is its pointer with the extra
380 * PAGE_HEADLESS bit set 383 * PAGE_HEADLESS bit set
381 */ 384 */
382 if (bud == HEADLESS) 385 if (bud == HEADLESS)
383 return h | (1 << PAGE_HEADLESS 386 return h | (1 << PAGE_HEADLESS);
384 387
385 /* otherwise, return pointer to encode 388 /* otherwise, return pointer to encoded handle */
386 idx = __idx(zhdr, bud); 389 idx = __idx(zhdr, bud);
387 h += idx; 390 h += idx;
388 if (bud == LAST) 391 if (bud == LAST)
389 h |= (zhdr->last_chunks << BUD 392 h |= (zhdr->last_chunks << BUDDY_SHIFT);
390 393
391 write_lock(&slots->lock); <<
392 slots->slot[idx] = h; 394 slots->slot[idx] = h;
393 write_unlock(&slots->lock); <<
394 return (unsigned long)&slots->slot[idx 395 return (unsigned long)&slots->slot[idx];
395 } 396 }
396 397
397 static unsigned long encode_handle(struct z3fo 398 static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud)
398 { 399 {
399 return __encode_handle(zhdr, zhdr->slo 400 return __encode_handle(zhdr, zhdr->slots, bud);
400 } 401 }
401 402
>> 403 /* Returns the z3fold page where a given handle is stored */
>> 404 static inline struct z3fold_header *handle_to_z3fold_header(unsigned long h)
>> 405 {
>> 406 unsigned long addr = h;
>> 407
>> 408 if (!(addr & (1 << PAGE_HEADLESS)))
>> 409 addr = *(unsigned long *)h;
>> 410
>> 411 return (struct z3fold_header *)(addr & PAGE_MASK);
>> 412 }
>> 413
402 /* only for LAST bud, returns zero otherwise * 414 /* only for LAST bud, returns zero otherwise */
403 static unsigned short handle_to_chunks(unsigne 415 static unsigned short handle_to_chunks(unsigned long handle)
404 { 416 {
405 struct z3fold_buddy_slots *slots = han !! 417 unsigned long addr = *(unsigned long *)handle;
406 unsigned long addr; <<
407 418
408 read_lock(&slots->lock); <<
409 addr = *(unsigned long *)handle; <<
410 read_unlock(&slots->lock); <<
411 return (addr & ~PAGE_MASK) >> BUDDY_SH 419 return (addr & ~PAGE_MASK) >> BUDDY_SHIFT;
412 } 420 }
413 421
414 /* 422 /*
415 * (handle & BUDDY_MASK) < zhdr->first_num is 423 * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle
416 * but that doesn't matter. because the maski 424 * but that doesn't matter. because the masking will result in the
417 * correct buddy number. 425 * correct buddy number.
418 */ 426 */
419 static enum buddy handle_to_buddy(unsigned lon 427 static enum buddy handle_to_buddy(unsigned long handle)
420 { 428 {
421 struct z3fold_header *zhdr; 429 struct z3fold_header *zhdr;
422 struct z3fold_buddy_slots *slots = han <<
423 unsigned long addr; 430 unsigned long addr;
424 431
425 read_lock(&slots->lock); <<
426 WARN_ON(handle & (1 << PAGE_HEADLESS)) 432 WARN_ON(handle & (1 << PAGE_HEADLESS));
427 addr = *(unsigned long *)handle; 433 addr = *(unsigned long *)handle;
428 read_unlock(&slots->lock); <<
429 zhdr = (struct z3fold_header *)(addr & 434 zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
430 return (addr - zhdr->first_num) & BUDD 435 return (addr - zhdr->first_num) & BUDDY_MASK;
431 } 436 }
432 437
433 static inline struct z3fold_pool *zhdr_to_pool 438 static inline struct z3fold_pool *zhdr_to_pool(struct z3fold_header *zhdr)
434 { 439 {
435 return zhdr->pool; 440 return zhdr->pool;
436 } 441 }
437 442
438 static void __release_z3fold_page(struct z3fol 443 static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked)
439 { 444 {
440 struct page *page = virt_to_page(zhdr) 445 struct page *page = virt_to_page(zhdr);
441 struct z3fold_pool *pool = zhdr_to_poo 446 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
442 447
443 WARN_ON(!list_empty(&zhdr->buddy)); 448 WARN_ON(!list_empty(&zhdr->buddy));
444 set_bit(PAGE_STALE, &page->private); 449 set_bit(PAGE_STALE, &page->private);
445 clear_bit(NEEDS_COMPACTING, &page->pri 450 clear_bit(NEEDS_COMPACTING, &page->private);
446 spin_lock(&pool->lock); 451 spin_lock(&pool->lock);
>> 452 if (!list_empty(&page->lru))
>> 453 list_del_init(&page->lru);
447 spin_unlock(&pool->lock); 454 spin_unlock(&pool->lock);
448 <<
449 if (locked) 455 if (locked)
450 z3fold_page_unlock(zhdr); 456 z3fold_page_unlock(zhdr);
451 <<
452 spin_lock(&pool->stale_lock); 457 spin_lock(&pool->stale_lock);
453 list_add(&zhdr->buddy, &pool->stale); 458 list_add(&zhdr->buddy, &pool->stale);
454 queue_work(pool->release_wq, &pool->wo 459 queue_work(pool->release_wq, &pool->work);
455 spin_unlock(&pool->stale_lock); 460 spin_unlock(&pool->stale_lock);
>> 461 }
456 462
457 atomic64_dec(&pool->pages_nr); !! 463 static void __attribute__((__unused__))
>> 464 release_z3fold_page(struct kref *ref)
>> 465 {
>> 466 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
>> 467 refcount);
>> 468 __release_z3fold_page(zhdr, false);
458 } 469 }
459 470
460 static void release_z3fold_page_locked(struct 471 static void release_z3fold_page_locked(struct kref *ref)
461 { 472 {
462 struct z3fold_header *zhdr = container 473 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
463 474 refcount);
464 WARN_ON(z3fold_page_trylock(zhdr)); 475 WARN_ON(z3fold_page_trylock(zhdr));
465 __release_z3fold_page(zhdr, true); 476 __release_z3fold_page(zhdr, true);
466 } 477 }
467 478
468 static void release_z3fold_page_locked_list(st 479 static void release_z3fold_page_locked_list(struct kref *ref)
469 { 480 {
470 struct z3fold_header *zhdr = container 481 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
471 482 refcount);
472 struct z3fold_pool *pool = zhdr_to_poo 483 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
473 <<
474 spin_lock(&pool->lock); 484 spin_lock(&pool->lock);
475 list_del_init(&zhdr->buddy); 485 list_del_init(&zhdr->buddy);
476 spin_unlock(&pool->lock); 486 spin_unlock(&pool->lock);
477 487
478 WARN_ON(z3fold_page_trylock(zhdr)); 488 WARN_ON(z3fold_page_trylock(zhdr));
479 __release_z3fold_page(zhdr, true); 489 __release_z3fold_page(zhdr, true);
480 } 490 }
481 491
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 492 static void free_pages_work(struct work_struct *w)
493 { 493 {
494 struct z3fold_pool *pool = container_o 494 struct z3fold_pool *pool = container_of(w, struct z3fold_pool, work);
495 495
496 spin_lock(&pool->stale_lock); 496 spin_lock(&pool->stale_lock);
497 while (!list_empty(&pool->stale)) { 497 while (!list_empty(&pool->stale)) {
498 struct z3fold_header *zhdr = l 498 struct z3fold_header *zhdr = list_first_entry(&pool->stale,
499 499 struct z3fold_header, buddy);
500 struct page *page = virt_to_pa 500 struct page *page = virt_to_page(zhdr);
501 501
502 list_del(&zhdr->buddy); 502 list_del(&zhdr->buddy);
503 if (WARN_ON(!test_bit(PAGE_STA 503 if (WARN_ON(!test_bit(PAGE_STALE, &page->private)))
504 continue; 504 continue;
505 spin_unlock(&pool->stale_lock) 505 spin_unlock(&pool->stale_lock);
506 cancel_work_sync(&zhdr->work); 506 cancel_work_sync(&zhdr->work);
507 free_z3fold_page(page, false); 507 free_z3fold_page(page, false);
508 cond_resched(); 508 cond_resched();
509 spin_lock(&pool->stale_lock); 509 spin_lock(&pool->stale_lock);
510 } 510 }
511 spin_unlock(&pool->stale_lock); 511 spin_unlock(&pool->stale_lock);
512 } 512 }
513 513
514 /* 514 /*
515 * Returns the number of free chunks in a z3fo 515 * Returns the number of free chunks in a z3fold page.
516 * NB: can't be used with HEADLESS pages. 516 * NB: can't be used with HEADLESS pages.
517 */ 517 */
518 static int num_free_chunks(struct z3fold_heade 518 static int num_free_chunks(struct z3fold_header *zhdr)
519 { 519 {
520 int nfree; 520 int nfree;
521 /* 521 /*
522 * If there is a middle object, pick u 522 * If there is a middle object, pick up the bigger free space
523 * either before or after it. Otherwis 523 * either before or after it. Otherwise just subtract the number
524 * of chunks occupied by the first and 524 * of chunks occupied by the first and the last objects.
525 */ 525 */
526 if (zhdr->middle_chunks != 0) { 526 if (zhdr->middle_chunks != 0) {
527 int nfree_before = zhdr->first 527 int nfree_before = zhdr->first_chunks ?
528 0 : zhdr->start_middle 528 0 : zhdr->start_middle - ZHDR_CHUNKS;
529 int nfree_after = zhdr->last_c 529 int nfree_after = zhdr->last_chunks ?
530 0 : TOTAL_CHUNKS - 530 0 : TOTAL_CHUNKS -
531 (zhdr->start_m 531 (zhdr->start_middle + zhdr->middle_chunks);
532 nfree = max(nfree_before, nfre 532 nfree = max(nfree_before, nfree_after);
533 } else 533 } else
534 nfree = NCHUNKS - zhdr->first_ 534 nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
535 return nfree; 535 return nfree;
536 } 536 }
537 537
538 /* Add to the appropriate unbuddied list */ 538 /* Add to the appropriate unbuddied list */
539 static inline void add_to_unbuddied(struct z3f 539 static inline void add_to_unbuddied(struct z3fold_pool *pool,
540 struct z3fold_ 540 struct z3fold_header *zhdr)
541 { 541 {
542 if (zhdr->first_chunks == 0 || zhdr->l 542 if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 ||
543 zhdr->middle_chunks == 543 zhdr->middle_chunks == 0) {
544 struct list_head *unbuddied; !! 544 struct list_head *unbuddied = get_cpu_ptr(pool->unbuddied);
545 int freechunks = num_free_chun <<
546 545
547 migrate_disable(); !! 546 int freechunks = num_free_chunks(zhdr);
548 unbuddied = this_cpu_ptr(pool- <<
549 spin_lock(&pool->lock); 547 spin_lock(&pool->lock);
550 list_add(&zhdr->buddy, &unbudd 548 list_add(&zhdr->buddy, &unbuddied[freechunks]);
551 spin_unlock(&pool->lock); 549 spin_unlock(&pool->lock);
552 zhdr->cpu = smp_processor_id() 550 zhdr->cpu = smp_processor_id();
553 migrate_enable(); !! 551 put_cpu_ptr(pool->unbuddied);
554 } 552 }
555 } 553 }
556 554
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 555 static inline void *mchunk_memmove(struct z3fold_header *zhdr,
580 unsigned short 556 unsigned short dst_chunk)
581 { 557 {
582 void *beg = zhdr; 558 void *beg = zhdr;
583 return memmove(beg + (dst_chunk << CHU 559 return memmove(beg + (dst_chunk << CHUNK_SHIFT),
584 beg + (zhdr->start_midd 560 beg + (zhdr->start_middle << CHUNK_SHIFT),
585 zhdr->middle_chunks << 561 zhdr->middle_chunks << CHUNK_SHIFT);
586 } 562 }
587 563
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 3 564 #define BIG_CHUNK_GAP 3
687 /* Has to be called with lock held */ 565 /* Has to be called with lock held */
688 static int z3fold_compact_page(struct z3fold_h 566 static int z3fold_compact_page(struct z3fold_header *zhdr)
689 { 567 {
690 struct page *page = virt_to_page(zhdr) 568 struct page *page = virt_to_page(zhdr);
691 569
692 if (test_bit(MIDDLE_CHUNK_MAPPED, &pag 570 if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private))
693 return 0; /* can't move middle 571 return 0; /* can't move middle chunk, it's used */
694 572
695 if (unlikely(PageIsolated(page))) 573 if (unlikely(PageIsolated(page)))
696 return 0; 574 return 0;
697 575
698 if (zhdr->middle_chunks == 0) 576 if (zhdr->middle_chunks == 0)
699 return 0; /* nothing to compac 577 return 0; /* nothing to compact */
700 578
701 if (zhdr->first_chunks == 0 && zhdr->l 579 if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
702 /* move to the beginning */ 580 /* move to the beginning */
703 mchunk_memmove(zhdr, ZHDR_CHUN 581 mchunk_memmove(zhdr, ZHDR_CHUNKS);
704 zhdr->first_chunks = zhdr->mid 582 zhdr->first_chunks = zhdr->middle_chunks;
705 zhdr->middle_chunks = 0; 583 zhdr->middle_chunks = 0;
706 zhdr->start_middle = 0; 584 zhdr->start_middle = 0;
707 zhdr->first_num++; 585 zhdr->first_num++;
708 return 1; 586 return 1;
709 } 587 }
710 588
711 /* 589 /*
712 * moving data is expensive, so let's 590 * moving data is expensive, so let's only do that if
713 * there's substantial gain (at least 591 * there's substantial gain (at least BIG_CHUNK_GAP chunks)
714 */ 592 */
715 if (zhdr->first_chunks != 0 && zhdr->l 593 if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 &&
716 zhdr->start_middle - (zhdr->first_ 594 zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >=
717 BIG_CHUNK_GAP) { 595 BIG_CHUNK_GAP) {
718 mchunk_memmove(zhdr, zhdr->fir 596 mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS);
719 zhdr->start_middle = zhdr->fir 597 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
720 return 1; 598 return 1;
721 } else if (zhdr->last_chunks != 0 && z 599 } else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 &&
722 TOTAL_CHUNKS - (zhdr->last_ 600 TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle
723 + zhdr 601 + zhdr->middle_chunks) >=
724 BIG_CHUNK_GAP) { 602 BIG_CHUNK_GAP) {
725 unsigned short new_start = TOT 603 unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks -
726 zhdr->middle_chunks; 604 zhdr->middle_chunks;
727 mchunk_memmove(zhdr, new_start 605 mchunk_memmove(zhdr, new_start);
728 zhdr->start_middle = new_start 606 zhdr->start_middle = new_start;
729 return 1; 607 return 1;
730 } 608 }
731 609
732 return 0; 610 return 0;
733 } 611 }
734 612
735 static void do_compact_page(struct z3fold_head 613 static void do_compact_page(struct z3fold_header *zhdr, bool locked)
736 { 614 {
737 struct z3fold_pool *pool = zhdr_to_poo 615 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
738 struct page *page; 616 struct page *page;
739 617
740 page = virt_to_page(zhdr); 618 page = virt_to_page(zhdr);
741 if (locked) 619 if (locked)
742 WARN_ON(z3fold_page_trylock(zh 620 WARN_ON(z3fold_page_trylock(zhdr));
743 else 621 else
744 z3fold_page_lock(zhdr); 622 z3fold_page_lock(zhdr);
745 if (WARN_ON(!test_and_clear_bit(NEEDS_ 623 if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) {
746 z3fold_page_unlock(zhdr); 624 z3fold_page_unlock(zhdr);
747 return; 625 return;
748 } 626 }
749 spin_lock(&pool->lock); 627 spin_lock(&pool->lock);
750 list_del_init(&zhdr->buddy); 628 list_del_init(&zhdr->buddy);
751 spin_unlock(&pool->lock); 629 spin_unlock(&pool->lock);
752 630
753 if (put_z3fold_locked(zhdr)) !! 631 if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) {
754 return; !! 632 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; 633 return;
760 } 634 }
761 635
762 if (!zhdr->foreign_handles && buddy_si !! 636 if (unlikely(PageIsolated(page) ||
763 zhdr->mapped_count == 0 && compact !! 637 test_bit(PAGE_CLAIMED, &page->private) ||
764 if (!put_z3fold_locked(zhdr)) !! 638 test_bit(PAGE_STALE, &page->private))) {
765 clear_bit(PAGE_CLAIMED !! 639 z3fold_page_unlock(zhdr);
766 z3fold_page_unlock(zhd <<
767 } <<
768 return; 640 return;
769 } 641 }
770 642
771 z3fold_compact_page(zhdr); 643 z3fold_compact_page(zhdr);
772 add_to_unbuddied(pool, zhdr); 644 add_to_unbuddied(pool, zhdr);
773 clear_bit(PAGE_CLAIMED, &page->private <<
774 z3fold_page_unlock(zhdr); 645 z3fold_page_unlock(zhdr);
775 } 646 }
776 647
777 static void compact_page_work(struct work_stru 648 static void compact_page_work(struct work_struct *w)
778 { 649 {
779 struct z3fold_header *zhdr = container 650 struct z3fold_header *zhdr = container_of(w, struct z3fold_header,
780 651 work);
781 652
782 do_compact_page(zhdr, false); 653 do_compact_page(zhdr, false);
783 } 654 }
784 655
785 /* returns _locked_ z3fold page header or NULL 656 /* returns _locked_ z3fold page header or NULL */
786 static inline struct z3fold_header *__z3fold_a 657 static inline struct z3fold_header *__z3fold_alloc(struct z3fold_pool *pool,
787 658 size_t size, bool can_sleep)
788 { 659 {
789 struct z3fold_header *zhdr = NULL; 660 struct z3fold_header *zhdr = NULL;
790 struct page *page; 661 struct page *page;
791 struct list_head *unbuddied; 662 struct list_head *unbuddied;
792 int chunks = size_to_chunks(size), i; 663 int chunks = size_to_chunks(size), i;
793 664
794 lookup: 665 lookup:
795 migrate_disable(); <<
796 /* First, try to find an unbuddied z3f 666 /* First, try to find an unbuddied z3fold page. */
797 unbuddied = this_cpu_ptr(pool->unbuddi !! 667 unbuddied = get_cpu_ptr(pool->unbuddied);
798 for_each_unbuddied_list(i, chunks) { 668 for_each_unbuddied_list(i, chunks) {
799 struct list_head *l = &unbuddi 669 struct list_head *l = &unbuddied[i];
800 670
801 zhdr = list_first_entry_or_nul 671 zhdr = list_first_entry_or_null(READ_ONCE(l),
802 struct 672 struct z3fold_header, buddy);
803 673
804 if (!zhdr) 674 if (!zhdr)
805 continue; 675 continue;
806 676
807 /* Re-check under lock. */ 677 /* Re-check under lock. */
808 spin_lock(&pool->lock); 678 spin_lock(&pool->lock);
>> 679 l = &unbuddied[i];
809 if (unlikely(zhdr != list_firs 680 if (unlikely(zhdr != list_first_entry(READ_ONCE(l),
810 681 struct z3fold_header, buddy)) ||
811 !z3fold_page_trylock(zhdr) 682 !z3fold_page_trylock(zhdr)) {
812 spin_unlock(&pool->loc 683 spin_unlock(&pool->lock);
813 zhdr = NULL; 684 zhdr = NULL;
814 migrate_enable(); !! 685 put_cpu_ptr(pool->unbuddied);
815 if (can_sleep) 686 if (can_sleep)
816 cond_resched() 687 cond_resched();
817 goto lookup; 688 goto lookup;
818 } 689 }
819 list_del_init(&zhdr->buddy); 690 list_del_init(&zhdr->buddy);
820 zhdr->cpu = -1; 691 zhdr->cpu = -1;
821 spin_unlock(&pool->lock); 692 spin_unlock(&pool->lock);
822 693
823 page = virt_to_page(zhdr); 694 page = virt_to_page(zhdr);
824 if (test_bit(NEEDS_COMPACTING, !! 695 if (test_bit(NEEDS_COMPACTING, &page->private)) {
825 test_bit(PAGE_CLAIMED, &pa <<
826 z3fold_page_unlock(zhd 696 z3fold_page_unlock(zhdr);
827 zhdr = NULL; 697 zhdr = NULL;
828 migrate_enable(); !! 698 put_cpu_ptr(pool->unbuddied);
829 if (can_sleep) 699 if (can_sleep)
830 cond_resched() 700 cond_resched();
831 goto lookup; 701 goto lookup;
832 } 702 }
833 703
834 /* 704 /*
835 * this page could not be remo 705 * this page could not be removed from its unbuddied
836 * list while pool lock was he 706 * list while pool lock was held, and then we've taken
837 * page lock so kref_put could 707 * page lock so kref_put could not be called before
838 * we got here, so it's safe t 708 * we got here, so it's safe to just call kref_get()
839 */ 709 */
840 kref_get(&zhdr->refcount); 710 kref_get(&zhdr->refcount);
841 break; 711 break;
842 } 712 }
843 migrate_enable(); !! 713 put_cpu_ptr(pool->unbuddied);
844 714
845 if (!zhdr) { 715 if (!zhdr) {
846 int cpu; 716 int cpu;
847 717
848 /* look for _exact_ match on o 718 /* look for _exact_ match on other cpus' lists */
849 for_each_online_cpu(cpu) { 719 for_each_online_cpu(cpu) {
850 struct list_head *l; 720 struct list_head *l;
851 721
852 unbuddied = per_cpu_pt 722 unbuddied = per_cpu_ptr(pool->unbuddied, cpu);
853 spin_lock(&pool->lock) 723 spin_lock(&pool->lock);
854 l = &unbuddied[chunks] 724 l = &unbuddied[chunks];
855 725
856 zhdr = list_first_entr 726 zhdr = list_first_entry_or_null(READ_ONCE(l),
857 727 struct z3fold_header, buddy);
858 728
859 if (!zhdr || !z3fold_p 729 if (!zhdr || !z3fold_page_trylock(zhdr)) {
860 spin_unlock(&p 730 spin_unlock(&pool->lock);
861 zhdr = NULL; 731 zhdr = NULL;
862 continue; 732 continue;
863 } 733 }
864 list_del_init(&zhdr->b 734 list_del_init(&zhdr->buddy);
865 zhdr->cpu = -1; 735 zhdr->cpu = -1;
866 spin_unlock(&pool->loc 736 spin_unlock(&pool->lock);
867 737
868 page = virt_to_page(zh 738 page = virt_to_page(zhdr);
869 if (test_bit(NEEDS_COM !! 739 if (test_bit(NEEDS_COMPACTING, &page->private)) {
870 test_bit(PAGE_CLAI <<
871 z3fold_page_un 740 z3fold_page_unlock(zhdr);
872 zhdr = NULL; 741 zhdr = NULL;
873 if (can_sleep) 742 if (can_sleep)
874 cond_r 743 cond_resched();
875 continue; 744 continue;
876 } 745 }
877 kref_get(&zhdr->refcou 746 kref_get(&zhdr->refcount);
878 break; 747 break;
879 } 748 }
880 } 749 }
881 750
882 if (zhdr && !zhdr->slots) { <<
883 zhdr->slots = alloc_slots(pool <<
884 if (!zhdr->slots) <<
885 goto out_fail; <<
886 } <<
887 return zhdr; 751 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 } 752 }
896 753
897 /* 754 /*
898 * API Functions 755 * API Functions
899 */ 756 */
900 757
901 /** 758 /**
902 * z3fold_create_pool() - create a new z3fold 759 * z3fold_create_pool() - create a new z3fold pool
903 * @name: pool name 760 * @name: pool name
904 * @gfp: gfp flags when allocating the 761 * @gfp: gfp flags when allocating the z3fold pool structure
>> 762 * @ops: user-defined operations for the z3fold pool
905 * 763 *
906 * Return: pointer to the new z3fold pool or N 764 * Return: pointer to the new z3fold pool or NULL if the metadata allocation
907 * failed. 765 * failed.
908 */ 766 */
909 static struct z3fold_pool *z3fold_create_pool( !! 767 static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp,
>> 768 const struct z3fold_ops *ops)
910 { 769 {
911 struct z3fold_pool *pool = NULL; 770 struct z3fold_pool *pool = NULL;
912 int i, cpu; 771 int i, cpu;
913 772
914 pool = kzalloc(sizeof(struct z3fold_po 773 pool = kzalloc(sizeof(struct z3fold_pool), gfp);
915 if (!pool) 774 if (!pool)
916 goto out; 775 goto out;
917 pool->c_handle = kmem_cache_create("z3 776 pool->c_handle = kmem_cache_create("z3fold_handle",
918 sizeof(struct 777 sizeof(struct z3fold_buddy_slots),
919 SLOTS_ALIGN, 0 778 SLOTS_ALIGN, 0, NULL);
920 if (!pool->c_handle) 779 if (!pool->c_handle)
921 goto out_c; 780 goto out_c;
922 spin_lock_init(&pool->lock); 781 spin_lock_init(&pool->lock);
923 spin_lock_init(&pool->stale_lock); 782 spin_lock_init(&pool->stale_lock);
924 pool->unbuddied = __alloc_percpu(sizeo !! 783 pool->unbuddied = __alloc_percpu(sizeof(struct list_head)*NCHUNKS, 2);
925 __ali <<
926 if (!pool->unbuddied) 784 if (!pool->unbuddied)
927 goto out_pool; 785 goto out_pool;
928 for_each_possible_cpu(cpu) { 786 for_each_possible_cpu(cpu) {
929 struct list_head *unbuddied = 787 struct list_head *unbuddied =
930 per_cpu_ptr(po 788 per_cpu_ptr(pool->unbuddied, cpu);
931 for_each_unbuddied_list(i, 0) 789 for_each_unbuddied_list(i, 0)
932 INIT_LIST_HEAD(&unbudd 790 INIT_LIST_HEAD(&unbuddied[i]);
933 } 791 }
>> 792 INIT_LIST_HEAD(&pool->lru);
934 INIT_LIST_HEAD(&pool->stale); 793 INIT_LIST_HEAD(&pool->stale);
935 atomic64_set(&pool->pages_nr, 0); 794 atomic64_set(&pool->pages_nr, 0);
936 pool->name = name; 795 pool->name = name;
937 pool->compact_wq = create_singlethread 796 pool->compact_wq = create_singlethread_workqueue(pool->name);
938 if (!pool->compact_wq) 797 if (!pool->compact_wq)
939 goto out_unbuddied; 798 goto out_unbuddied;
940 pool->release_wq = create_singlethread 799 pool->release_wq = create_singlethread_workqueue(pool->name);
941 if (!pool->release_wq) 800 if (!pool->release_wq)
942 goto out_wq; 801 goto out_wq;
>> 802 if (z3fold_register_migration(pool))
>> 803 goto out_rwq;
943 INIT_WORK(&pool->work, free_pages_work 804 INIT_WORK(&pool->work, free_pages_work);
>> 805 pool->ops = ops;
944 return pool; 806 return pool;
945 807
>> 808 out_rwq:
>> 809 destroy_workqueue(pool->release_wq);
946 out_wq: 810 out_wq:
947 destroy_workqueue(pool->compact_wq); 811 destroy_workqueue(pool->compact_wq);
948 out_unbuddied: 812 out_unbuddied:
949 free_percpu(pool->unbuddied); 813 free_percpu(pool->unbuddied);
950 out_pool: 814 out_pool:
951 kmem_cache_destroy(pool->c_handle); 815 kmem_cache_destroy(pool->c_handle);
952 out_c: 816 out_c:
953 kfree(pool); 817 kfree(pool);
954 out: 818 out:
955 return NULL; 819 return NULL;
956 } 820 }
957 821
958 /** 822 /**
959 * z3fold_destroy_pool() - destroys an existin 823 * z3fold_destroy_pool() - destroys an existing z3fold pool
960 * @pool: the z3fold pool to be destroye 824 * @pool: the z3fold pool to be destroyed
961 * 825 *
962 * The pool should be emptied before this func 826 * The pool should be emptied before this function is called.
963 */ 827 */
964 static void z3fold_destroy_pool(struct z3fold_ 828 static void z3fold_destroy_pool(struct z3fold_pool *pool)
965 { 829 {
966 kmem_cache_destroy(pool->c_handle); 830 kmem_cache_destroy(pool->c_handle);
967 831
968 /* 832 /*
969 * We need to destroy pool->compact_wq 833 * We need to destroy pool->compact_wq before pool->release_wq,
970 * as any pending work on pool->compac 834 * as any pending work on pool->compact_wq will call
971 * queue_work(pool->release_wq, &pool- 835 * queue_work(pool->release_wq, &pool->work).
972 * 836 *
973 * There are still outstanding pages u 837 * There are still outstanding pages until both workqueues are drained,
974 * so we cannot unregister migration u 838 * so we cannot unregister migration until then.
975 */ 839 */
976 840
977 destroy_workqueue(pool->compact_wq); 841 destroy_workqueue(pool->compact_wq);
978 destroy_workqueue(pool->release_wq); 842 destroy_workqueue(pool->release_wq);
979 free_percpu(pool->unbuddied); !! 843 z3fold_unregister_migration(pool);
980 kfree(pool); 844 kfree(pool);
981 } 845 }
982 846
983 static const struct movable_operations z3fold_ <<
984 <<
985 /** 847 /**
986 * z3fold_alloc() - allocates a region of a gi 848 * z3fold_alloc() - allocates a region of a given size
987 * @pool: z3fold pool from which to allo 849 * @pool: z3fold pool from which to allocate
988 * @size: size in bytes of the desired a 850 * @size: size in bytes of the desired allocation
989 * @gfp: gfp flags used if the pool nee 851 * @gfp: gfp flags used if the pool needs to grow
990 * @handle: handle of the new allocation 852 * @handle: handle of the new allocation
991 * 853 *
992 * This function will attempt to find a free r 854 * This function will attempt to find a free region in the pool large enough to
993 * satisfy the allocation request. A search o 855 * satisfy the allocation request. A search of the unbuddied lists is
994 * performed first. If no suitable free region 856 * performed first. If no suitable free region is found, then a new page is
995 * allocated and added to the pool to satisfy 857 * allocated and added to the pool to satisfy the request.
996 * 858 *
>> 859 * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used
>> 860 * as z3fold pool pages.
>> 861 *
997 * Return: 0 if success and handle is set, oth 862 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
998 * gfp arguments are invalid or -ENOMEM if the 863 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
999 * a new page. 864 * a new page.
1000 */ 865 */
1001 static int z3fold_alloc(struct z3fold_pool *p 866 static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
1002 unsigned long *handle 867 unsigned long *handle)
1003 { 868 {
1004 int chunks = size_to_chunks(size); 869 int chunks = size_to_chunks(size);
1005 struct z3fold_header *zhdr = NULL; 870 struct z3fold_header *zhdr = NULL;
1006 struct page *page = NULL; 871 struct page *page = NULL;
1007 enum buddy bud; 872 enum buddy bud;
1008 bool can_sleep = gfpflags_allow_block 873 bool can_sleep = gfpflags_allow_blocking(gfp);
1009 874
1010 if (!size || (gfp & __GFP_HIGHMEM)) 875 if (!size || (gfp & __GFP_HIGHMEM))
1011 return -EINVAL; 876 return -EINVAL;
1012 877
1013 if (size > PAGE_SIZE) 878 if (size > PAGE_SIZE)
1014 return -ENOSPC; 879 return -ENOSPC;
1015 880
1016 if (size > PAGE_SIZE - ZHDR_SIZE_ALIG 881 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
1017 bud = HEADLESS; 882 bud = HEADLESS;
1018 else { 883 else {
1019 retry: 884 retry:
1020 zhdr = __z3fold_alloc(pool, s 885 zhdr = __z3fold_alloc(pool, size, can_sleep);
1021 if (zhdr) { 886 if (zhdr) {
1022 bud = get_free_buddy( !! 887 if (zhdr->first_chunks == 0) {
1023 if (bud == HEADLESS) !! 888 if (zhdr->middle_chunks != 0 &&
1024 if (!put_z3fo !! 889 chunks >= zhdr->start_middle)
>> 890 bud = LAST;
>> 891 else
>> 892 bud = FIRST;
>> 893 } else if (zhdr->last_chunks == 0)
>> 894 bud = LAST;
>> 895 else if (zhdr->middle_chunks == 0)
>> 896 bud = MIDDLE;
>> 897 else {
>> 898 if (kref_put(&zhdr->refcount,
>> 899 release_z3fold_page_locked))
>> 900 atomic64_dec(&pool->pages_nr);
>> 901 else
1025 z3fol 902 z3fold_page_unlock(zhdr);
1026 pr_err("No fr 903 pr_err("No free chunks in unbuddied\n");
1027 WARN_ON(1); 904 WARN_ON(1);
1028 goto retry; 905 goto retry;
1029 } 906 }
1030 page = virt_to_page(z 907 page = virt_to_page(zhdr);
1031 goto found; 908 goto found;
1032 } 909 }
1033 bud = FIRST; 910 bud = FIRST;
1034 } 911 }
1035 912
1036 page = alloc_page(gfp); !! 913 page = NULL;
>> 914 if (can_sleep) {
>> 915 spin_lock(&pool->stale_lock);
>> 916 zhdr = list_first_entry_or_null(&pool->stale,
>> 917 struct z3fold_header, buddy);
>> 918 /*
>> 919 * Before allocating a page, let's see if we can take one from
>> 920 * the stale pages list. cancel_work_sync() can sleep so we
>> 921 * limit this case to the contexts where we can sleep
>> 922 */
>> 923 if (zhdr) {
>> 924 list_del(&zhdr->buddy);
>> 925 spin_unlock(&pool->stale_lock);
>> 926 cancel_work_sync(&zhdr->work);
>> 927 page = virt_to_page(zhdr);
>> 928 } else {
>> 929 spin_unlock(&pool->stale_lock);
>> 930 }
>> 931 }
>> 932 if (!page)
>> 933 page = alloc_page(gfp);
>> 934
1037 if (!page) 935 if (!page)
1038 return -ENOMEM; 936 return -ENOMEM;
1039 937
1040 zhdr = init_z3fold_page(page, bud == 938 zhdr = init_z3fold_page(page, bud == HEADLESS, pool, gfp);
1041 if (!zhdr) { 939 if (!zhdr) {
1042 __free_page(page); 940 __free_page(page);
1043 return -ENOMEM; 941 return -ENOMEM;
1044 } 942 }
1045 atomic64_inc(&pool->pages_nr); 943 atomic64_inc(&pool->pages_nr);
1046 944
1047 if (bud == HEADLESS) { 945 if (bud == HEADLESS) {
1048 set_bit(PAGE_HEADLESS, &page- 946 set_bit(PAGE_HEADLESS, &page->private);
1049 goto headless; 947 goto headless;
1050 } 948 }
1051 if (can_sleep) { 949 if (can_sleep) {
1052 lock_page(page); 950 lock_page(page);
1053 __SetPageMovable(page, &z3fol !! 951 __SetPageMovable(page, pool->inode->i_mapping);
1054 unlock_page(page); 952 unlock_page(page);
1055 } else { 953 } else {
1056 WARN_ON(!trylock_page(page)); !! 954 if (trylock_page(page)) {
1057 __SetPageMovable(page, &z3fol !! 955 __SetPageMovable(page, pool->inode->i_mapping);
1058 unlock_page(page); !! 956 unlock_page(page);
>> 957 }
1059 } 958 }
1060 z3fold_page_lock(zhdr); 959 z3fold_page_lock(zhdr);
1061 960
1062 found: 961 found:
1063 if (bud == FIRST) 962 if (bud == FIRST)
1064 zhdr->first_chunks = chunks; 963 zhdr->first_chunks = chunks;
1065 else if (bud == LAST) 964 else if (bud == LAST)
1066 zhdr->last_chunks = chunks; 965 zhdr->last_chunks = chunks;
1067 else { 966 else {
1068 zhdr->middle_chunks = chunks; 967 zhdr->middle_chunks = chunks;
1069 zhdr->start_middle = zhdr->fi 968 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
1070 } 969 }
1071 add_to_unbuddied(pool, zhdr); 970 add_to_unbuddied(pool, zhdr);
1072 971
1073 headless: 972 headless:
1074 spin_lock(&pool->lock); 973 spin_lock(&pool->lock);
>> 974 /* Add/move z3fold page to beginning of LRU */
>> 975 if (!list_empty(&page->lru))
>> 976 list_del(&page->lru);
>> 977
>> 978 list_add(&page->lru, &pool->lru);
>> 979
1075 *handle = encode_handle(zhdr, bud); 980 *handle = encode_handle(zhdr, bud);
1076 spin_unlock(&pool->lock); 981 spin_unlock(&pool->lock);
1077 if (bud != HEADLESS) 982 if (bud != HEADLESS)
1078 z3fold_page_unlock(zhdr); 983 z3fold_page_unlock(zhdr);
1079 984
1080 return 0; 985 return 0;
1081 } 986 }
1082 987
1083 /** 988 /**
1084 * z3fold_free() - frees the allocation assoc 989 * z3fold_free() - frees the allocation associated with the given handle
1085 * @pool: pool in which the allocation 990 * @pool: pool in which the allocation resided
1086 * @handle: handle associated with the al 991 * @handle: handle associated with the allocation returned by z3fold_alloc()
1087 * 992 *
1088 * In the case that the z3fold page in which 993 * In the case that the z3fold page in which the allocation resides is under
1089 * reclaim, as indicated by the PAGE_CLAIMED !! 994 * reclaim, as indicated by the PG_reclaim flag being set, this function
1090 * only sets the first|middle|last_chunks to !! 995 * only sets the first|last_chunks to 0. The page is actually freed
1091 * once all buddies are evicted (see z3fold_r !! 996 * once both buddies are evicted (see z3fold_reclaim_page() below).
1092 */ 997 */
1093 static void z3fold_free(struct z3fold_pool *p 998 static void z3fold_free(struct z3fold_pool *pool, unsigned long handle)
1094 { 999 {
1095 struct z3fold_header *zhdr; 1000 struct z3fold_header *zhdr;
1096 struct page *page; 1001 struct page *page;
1097 enum buddy bud; 1002 enum buddy bud;
1098 bool page_claimed; <<
1099 1003
1100 zhdr = get_z3fold_header(handle); !! 1004 zhdr = handle_to_z3fold_header(handle);
1101 page = virt_to_page(zhdr); 1005 page = virt_to_page(zhdr);
1102 page_claimed = test_and_set_bit(PAGE_ <<
1103 1006
1104 if (test_bit(PAGE_HEADLESS, &page->pr 1007 if (test_bit(PAGE_HEADLESS, &page->private)) {
1105 /* if a headless page is unde 1008 /* if a headless page is under reclaim, just leave.
1106 * NB: we use test_and_set_bi 1009 * NB: we use test_and_set_bit for a reason: if the bit
1107 * has not been set before, w 1010 * has not been set before, we release this page
1108 * immediately so we don't ca 1011 * immediately so we don't care about its value any more.
1109 */ 1012 */
1110 if (!page_claimed) { !! 1013 if (!test_and_set_bit(PAGE_CLAIMED, &page->private)) {
1111 put_z3fold_header(zhd !! 1014 spin_lock(&pool->lock);
>> 1015 list_del(&page->lru);
>> 1016 spin_unlock(&pool->lock);
1112 free_z3fold_page(page 1017 free_z3fold_page(page, true);
1113 atomic64_dec(&pool->p 1018 atomic64_dec(&pool->pages_nr);
1114 } 1019 }
1115 return; 1020 return;
1116 } 1021 }
1117 1022
1118 /* Non-headless case */ 1023 /* Non-headless case */
>> 1024 z3fold_page_lock(zhdr);
1119 bud = handle_to_buddy(handle); 1025 bud = handle_to_buddy(handle);
1120 1026
1121 switch (bud) { 1027 switch (bud) {
1122 case FIRST: 1028 case FIRST:
1123 zhdr->first_chunks = 0; 1029 zhdr->first_chunks = 0;
1124 break; 1030 break;
1125 case MIDDLE: 1031 case MIDDLE:
1126 zhdr->middle_chunks = 0; 1032 zhdr->middle_chunks = 0;
1127 break; 1033 break;
1128 case LAST: 1034 case LAST:
1129 zhdr->last_chunks = 0; 1035 zhdr->last_chunks = 0;
1130 break; 1036 break;
1131 default: 1037 default:
1132 pr_err("%s: unknown bud %d\n" 1038 pr_err("%s: unknown bud %d\n", __func__, bud);
1133 WARN_ON(1); 1039 WARN_ON(1);
1134 put_z3fold_header(zhdr); !! 1040 z3fold_page_unlock(zhdr);
1135 return; 1041 return;
1136 } 1042 }
1137 1043
1138 if (!page_claimed) !! 1044 free_handle(handle);
1139 free_handle(handle, zhdr); !! 1045 if (kref_put(&zhdr->refcount, release_z3fold_page_locked_list)) {
1140 if (put_z3fold_locked_list(zhdr)) !! 1046 atomic64_dec(&pool->pages_nr);
1141 return; 1047 return;
1142 if (page_claimed) { !! 1048 }
1143 /* the page has not been clai !! 1049 if (test_bit(PAGE_CLAIMED, &page->private)) {
1144 put_z3fold_header(zhdr); !! 1050 z3fold_page_unlock(zhdr);
1145 return; 1051 return;
1146 } 1052 }
1147 if (test_and_set_bit(NEEDS_COMPACTING !! 1053 if (unlikely(PageIsolated(page)) ||
1148 clear_bit(PAGE_CLAIMED, &page !! 1054 test_and_set_bit(NEEDS_COMPACTING, &page->private)) {
1149 put_z3fold_header(zhdr); !! 1055 z3fold_page_unlock(zhdr);
1150 return; 1056 return;
1151 } 1057 }
1152 if (zhdr->cpu < 0 || !cpu_online(zhdr 1058 if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) {
>> 1059 spin_lock(&pool->lock);
>> 1060 list_del_init(&zhdr->buddy);
>> 1061 spin_unlock(&pool->lock);
1153 zhdr->cpu = -1; 1062 zhdr->cpu = -1;
1154 kref_get(&zhdr->refcount); 1063 kref_get(&zhdr->refcount);
1155 clear_bit(PAGE_CLAIMED, &page <<
1156 do_compact_page(zhdr, true); 1064 do_compact_page(zhdr, true);
1157 return; 1065 return;
1158 } 1066 }
1159 kref_get(&zhdr->refcount); 1067 kref_get(&zhdr->refcount);
1160 clear_bit(PAGE_CLAIMED, &page->privat <<
1161 queue_work_on(zhdr->cpu, pool->compac 1068 queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work);
1162 put_z3fold_header(zhdr); !! 1069 z3fold_page_unlock(zhdr);
>> 1070 }
>> 1071
>> 1072 /**
>> 1073 * z3fold_reclaim_page() - evicts allocations from a pool page and frees it
>> 1074 * @pool: pool from which a page will attempt to be evicted
>> 1075 * @retries: number of pages on the LRU list for which eviction will
>> 1076 * be attempted before failing
>> 1077 *
>> 1078 * z3fold reclaim is different from normal system reclaim in that it is done
>> 1079 * from the bottom, up. This is because only the bottom layer, z3fold, has
>> 1080 * information on how the allocations are organized within each z3fold page.
>> 1081 * This has the potential to create interesting locking situations between
>> 1082 * z3fold and the user, however.
>> 1083 *
>> 1084 * To avoid these, this is how z3fold_reclaim_page() should be called:
>> 1085 *
>> 1086 * The user detects a page should be reclaimed and calls z3fold_reclaim_page().
>> 1087 * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and
>> 1088 * call the user-defined eviction handler with the pool and handle as
>> 1089 * arguments.
>> 1090 *
>> 1091 * If the handle can not be evicted, the eviction handler should return
>> 1092 * non-zero. z3fold_reclaim_page() will add the z3fold page back to the
>> 1093 * appropriate list and try the next z3fold page on the LRU up to
>> 1094 * a user defined number of retries.
>> 1095 *
>> 1096 * If the handle is successfully evicted, the eviction handler should
>> 1097 * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free()
>> 1098 * contains logic to delay freeing the page if the page is under reclaim,
>> 1099 * as indicated by the setting of the PG_reclaim flag on the underlying page.
>> 1100 *
>> 1101 * If all buddies in the z3fold page are successfully evicted, then the
>> 1102 * z3fold page can be freed.
>> 1103 *
>> 1104 * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are
>> 1105 * no pages to evict or an eviction handler is not registered, -EAGAIN if
>> 1106 * the retry limit was hit.
>> 1107 */
>> 1108 static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries)
>> 1109 {
>> 1110 int i, ret = 0;
>> 1111 struct z3fold_header *zhdr = NULL;
>> 1112 struct page *page = NULL;
>> 1113 struct list_head *pos;
>> 1114 struct z3fold_buddy_slots slots;
>> 1115 unsigned long first_handle = 0, middle_handle = 0, last_handle = 0;
>> 1116
>> 1117 spin_lock(&pool->lock);
>> 1118 if (!pool->ops || !pool->ops->evict || retries == 0) {
>> 1119 spin_unlock(&pool->lock);
>> 1120 return -EINVAL;
>> 1121 }
>> 1122 for (i = 0; i < retries; i++) {
>> 1123 if (list_empty(&pool->lru)) {
>> 1124 spin_unlock(&pool->lock);
>> 1125 return -EINVAL;
>> 1126 }
>> 1127 list_for_each_prev(pos, &pool->lru) {
>> 1128 page = list_entry(pos, struct page, lru);
>> 1129
>> 1130 /* this bit could have been set by free, in which case
>> 1131 * we pass over to the next page in the pool.
>> 1132 */
>> 1133 if (test_and_set_bit(PAGE_CLAIMED, &page->private)) {
>> 1134 page = NULL;
>> 1135 continue;
>> 1136 }
>> 1137
>> 1138 if (unlikely(PageIsolated(page))) {
>> 1139 clear_bit(PAGE_CLAIMED, &page->private);
>> 1140 page = NULL;
>> 1141 continue;
>> 1142 }
>> 1143 zhdr = page_address(page);
>> 1144 if (test_bit(PAGE_HEADLESS, &page->private))
>> 1145 break;
>> 1146
>> 1147 if (!z3fold_page_trylock(zhdr)) {
>> 1148 clear_bit(PAGE_CLAIMED, &page->private);
>> 1149 zhdr = NULL;
>> 1150 continue; /* can't evict at this point */
>> 1151 }
>> 1152 kref_get(&zhdr->refcount);
>> 1153 list_del_init(&zhdr->buddy);
>> 1154 zhdr->cpu = -1;
>> 1155 break;
>> 1156 }
>> 1157
>> 1158 if (!zhdr)
>> 1159 break;
>> 1160
>> 1161 list_del_init(&page->lru);
>> 1162 spin_unlock(&pool->lock);
>> 1163
>> 1164 if (!test_bit(PAGE_HEADLESS, &page->private)) {
>> 1165 /*
>> 1166 * We need encode the handles before unlocking, and
>> 1167 * use our local slots structure because z3fold_free
>> 1168 * can zero out zhdr->slots and we can't do much
>> 1169 * about that
>> 1170 */
>> 1171 first_handle = 0;
>> 1172 last_handle = 0;
>> 1173 middle_handle = 0;
>> 1174 if (zhdr->first_chunks)
>> 1175 first_handle = __encode_handle(zhdr, &slots,
>> 1176 FIRST);
>> 1177 if (zhdr->middle_chunks)
>> 1178 middle_handle = __encode_handle(zhdr, &slots,
>> 1179 MIDDLE);
>> 1180 if (zhdr->last_chunks)
>> 1181 last_handle = __encode_handle(zhdr, &slots,
>> 1182 LAST);
>> 1183 /*
>> 1184 * it's safe to unlock here because we hold a
>> 1185 * reference to this page
>> 1186 */
>> 1187 z3fold_page_unlock(zhdr);
>> 1188 } else {
>> 1189 first_handle = __encode_handle(zhdr, &slots, HEADLESS);
>> 1190 last_handle = middle_handle = 0;
>> 1191 }
>> 1192
>> 1193 /* Issue the eviction callback(s) */
>> 1194 if (middle_handle) {
>> 1195 ret = pool->ops->evict(pool, middle_handle);
>> 1196 if (ret)
>> 1197 goto next;
>> 1198 }
>> 1199 if (first_handle) {
>> 1200 ret = pool->ops->evict(pool, first_handle);
>> 1201 if (ret)
>> 1202 goto next;
>> 1203 }
>> 1204 if (last_handle) {
>> 1205 ret = pool->ops->evict(pool, last_handle);
>> 1206 if (ret)
>> 1207 goto next;
>> 1208 }
>> 1209 next:
>> 1210 if (test_bit(PAGE_HEADLESS, &page->private)) {
>> 1211 if (ret == 0) {
>> 1212 free_z3fold_page(page, true);
>> 1213 atomic64_dec(&pool->pages_nr);
>> 1214 return 0;
>> 1215 }
>> 1216 spin_lock(&pool->lock);
>> 1217 list_add(&page->lru, &pool->lru);
>> 1218 spin_unlock(&pool->lock);
>> 1219 clear_bit(PAGE_CLAIMED, &page->private);
>> 1220 } else {
>> 1221 z3fold_page_lock(zhdr);
>> 1222 if (kref_put(&zhdr->refcount,
>> 1223 release_z3fold_page_locked)) {
>> 1224 atomic64_dec(&pool->pages_nr);
>> 1225 return 0;
>> 1226 }
>> 1227 /*
>> 1228 * if we are here, the page is still not completely
>> 1229 * free. Take the global pool lock then to be able
>> 1230 * to add it back to the lru list
>> 1231 */
>> 1232 spin_lock(&pool->lock);
>> 1233 list_add(&page->lru, &pool->lru);
>> 1234 spin_unlock(&pool->lock);
>> 1235 z3fold_page_unlock(zhdr);
>> 1236 clear_bit(PAGE_CLAIMED, &page->private);
>> 1237 }
>> 1238
>> 1239 /* We started off locked to we need to lock the pool back */
>> 1240 spin_lock(&pool->lock);
>> 1241 }
>> 1242 spin_unlock(&pool->lock);
>> 1243 return -EAGAIN;
1163 } 1244 }
1164 1245
1165 /** 1246 /**
1166 * z3fold_map() - maps the allocation associa 1247 * z3fold_map() - maps the allocation associated with the given handle
1167 * @pool: pool in which the allocation 1248 * @pool: pool in which the allocation resides
1168 * @handle: handle associated with the al 1249 * @handle: handle associated with the allocation to be mapped
1169 * 1250 *
1170 * Extracts the buddy number from handle and 1251 * Extracts the buddy number from handle and constructs the pointer to the
1171 * correct starting chunk within the page. 1252 * correct starting chunk within the page.
1172 * 1253 *
1173 * Returns: a pointer to the mapped allocatio 1254 * Returns: a pointer to the mapped allocation
1174 */ 1255 */
1175 static void *z3fold_map(struct z3fold_pool *p 1256 static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle)
1176 { 1257 {
1177 struct z3fold_header *zhdr; 1258 struct z3fold_header *zhdr;
1178 struct page *page; 1259 struct page *page;
1179 void *addr; 1260 void *addr;
1180 enum buddy buddy; 1261 enum buddy buddy;
1181 1262
1182 zhdr = get_z3fold_header(handle); !! 1263 zhdr = handle_to_z3fold_header(handle);
1183 addr = zhdr; 1264 addr = zhdr;
1184 page = virt_to_page(zhdr); 1265 page = virt_to_page(zhdr);
1185 1266
1186 if (test_bit(PAGE_HEADLESS, &page->pr 1267 if (test_bit(PAGE_HEADLESS, &page->private))
1187 goto out; 1268 goto out;
1188 1269
>> 1270 z3fold_page_lock(zhdr);
1189 buddy = handle_to_buddy(handle); 1271 buddy = handle_to_buddy(handle);
1190 switch (buddy) { 1272 switch (buddy) {
1191 case FIRST: 1273 case FIRST:
1192 addr += ZHDR_SIZE_ALIGNED; 1274 addr += ZHDR_SIZE_ALIGNED;
1193 break; 1275 break;
1194 case MIDDLE: 1276 case MIDDLE:
1195 addr += zhdr->start_middle << 1277 addr += zhdr->start_middle << CHUNK_SHIFT;
1196 set_bit(MIDDLE_CHUNK_MAPPED, 1278 set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1197 break; 1279 break;
1198 case LAST: 1280 case LAST:
1199 addr += PAGE_SIZE - (handle_t 1281 addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT);
1200 break; 1282 break;
1201 default: 1283 default:
1202 pr_err("unknown buddy id %d\n 1284 pr_err("unknown buddy id %d\n", buddy);
1203 WARN_ON(1); 1285 WARN_ON(1);
1204 addr = NULL; 1286 addr = NULL;
1205 break; 1287 break;
1206 } 1288 }
1207 1289
1208 if (addr) 1290 if (addr)
1209 zhdr->mapped_count++; 1291 zhdr->mapped_count++;
>> 1292 z3fold_page_unlock(zhdr);
1210 out: 1293 out:
1211 put_z3fold_header(zhdr); <<
1212 return addr; 1294 return addr;
1213 } 1295 }
1214 1296
1215 /** 1297 /**
1216 * z3fold_unmap() - unmaps the allocation ass 1298 * z3fold_unmap() - unmaps the allocation associated with the given handle
1217 * @pool: pool in which the allocation 1299 * @pool: pool in which the allocation resides
1218 * @handle: handle associated with the al 1300 * @handle: handle associated with the allocation to be unmapped
1219 */ 1301 */
1220 static void z3fold_unmap(struct z3fold_pool * 1302 static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle)
1221 { 1303 {
1222 struct z3fold_header *zhdr; 1304 struct z3fold_header *zhdr;
1223 struct page *page; 1305 struct page *page;
1224 enum buddy buddy; 1306 enum buddy buddy;
1225 1307
1226 zhdr = get_z3fold_header(handle); !! 1308 zhdr = handle_to_z3fold_header(handle);
1227 page = virt_to_page(zhdr); 1309 page = virt_to_page(zhdr);
1228 1310
1229 if (test_bit(PAGE_HEADLESS, &page->pr 1311 if (test_bit(PAGE_HEADLESS, &page->private))
1230 return; 1312 return;
1231 1313
>> 1314 z3fold_page_lock(zhdr);
1232 buddy = handle_to_buddy(handle); 1315 buddy = handle_to_buddy(handle);
1233 if (buddy == MIDDLE) 1316 if (buddy == MIDDLE)
1234 clear_bit(MIDDLE_CHUNK_MAPPED 1317 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1235 zhdr->mapped_count--; 1318 zhdr->mapped_count--;
1236 put_z3fold_header(zhdr); !! 1319 z3fold_page_unlock(zhdr);
1237 } 1320 }
1238 1321
1239 /** 1322 /**
1240 * z3fold_get_pool_pages() - gets the z3fold !! 1323 * z3fold_get_pool_size() - gets the z3fold pool size in pages
1241 * @pool: pool whose size is being quer 1324 * @pool: pool whose size is being queried
1242 * 1325 *
1243 * Returns: size in pages of the given pool. 1326 * Returns: size in pages of the given pool.
1244 */ 1327 */
1245 static u64 z3fold_get_pool_pages(struct z3fol !! 1328 static u64 z3fold_get_pool_size(struct z3fold_pool *pool)
1246 { 1329 {
1247 return atomic64_read(&pool->pages_nr) 1330 return atomic64_read(&pool->pages_nr);
1248 } 1331 }
1249 1332
1250 static bool z3fold_page_isolate(struct page * 1333 static bool z3fold_page_isolate(struct page *page, isolate_mode_t mode)
1251 { 1334 {
1252 struct z3fold_header *zhdr; 1335 struct z3fold_header *zhdr;
1253 struct z3fold_pool *pool; 1336 struct z3fold_pool *pool;
1254 1337
>> 1338 VM_BUG_ON_PAGE(!PageMovable(page), page);
1255 VM_BUG_ON_PAGE(PageIsolated(page), pa 1339 VM_BUG_ON_PAGE(PageIsolated(page), page);
1256 1340
1257 if (test_bit(PAGE_HEADLESS, &page->pr !! 1341 if (test_bit(PAGE_HEADLESS, &page->private) ||
>> 1342 test_bit(PAGE_CLAIMED, &page->private))
1258 return false; 1343 return false;
1259 1344
1260 zhdr = page_address(page); 1345 zhdr = page_address(page);
1261 z3fold_page_lock(zhdr); 1346 z3fold_page_lock(zhdr);
1262 if (test_bit(NEEDS_COMPACTING, &page- 1347 if (test_bit(NEEDS_COMPACTING, &page->private) ||
1263 test_bit(PAGE_STALE, &page->priva 1348 test_bit(PAGE_STALE, &page->private))
1264 goto out; 1349 goto out;
1265 1350
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); 1351 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 1352
>> 1353 if (zhdr->mapped_count == 0) {
>> 1354 kref_get(&zhdr->refcount);
>> 1355 if (!list_empty(&zhdr->buddy))
>> 1356 list_del_init(&zhdr->buddy);
>> 1357 spin_lock(&pool->lock);
>> 1358 if (!list_empty(&page->lru))
>> 1359 list_del(&page->lru);
>> 1360 spin_unlock(&pool->lock);
>> 1361 z3fold_page_unlock(zhdr);
>> 1362 return true;
>> 1363 }
1281 out: 1364 out:
1282 z3fold_page_unlock(zhdr); 1365 z3fold_page_unlock(zhdr);
1283 return false; 1366 return false;
1284 } 1367 }
1285 1368
1286 static int z3fold_page_migrate(struct page *n !! 1369 static int z3fold_page_migrate(struct address_space *mapping, struct page *newpage,
1287 enum migrate_mode mode) !! 1370 struct page *page, enum migrate_mode mode)
1288 { 1371 {
1289 struct z3fold_header *zhdr, *new_zhdr 1372 struct z3fold_header *zhdr, *new_zhdr;
1290 struct z3fold_pool *pool; 1373 struct z3fold_pool *pool;
>> 1374 struct address_space *new_mapping;
1291 1375
>> 1376 VM_BUG_ON_PAGE(!PageMovable(page), page);
1292 VM_BUG_ON_PAGE(!PageIsolated(page), p 1377 VM_BUG_ON_PAGE(!PageIsolated(page), page);
1293 VM_BUG_ON_PAGE(!test_bit(PAGE_CLAIMED <<
1294 VM_BUG_ON_PAGE(!PageLocked(newpage), 1378 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
1295 1379
1296 zhdr = page_address(page); 1380 zhdr = page_address(page);
1297 pool = zhdr_to_pool(zhdr); 1381 pool = zhdr_to_pool(zhdr);
1298 1382
1299 if (!z3fold_page_trylock(zhdr)) !! 1383 if (!z3fold_page_trylock(zhdr)) {
1300 return -EAGAIN; 1384 return -EAGAIN;
1301 if (zhdr->mapped_count != 0 || zhdr-> !! 1385 }
1302 clear_bit(PAGE_CLAIMED, &page !! 1386 if (zhdr->mapped_count != 0) {
1303 z3fold_page_unlock(zhdr); 1387 z3fold_page_unlock(zhdr);
1304 return -EBUSY; 1388 return -EBUSY;
1305 } 1389 }
1306 if (work_pending(&zhdr->work)) { 1390 if (work_pending(&zhdr->work)) {
1307 z3fold_page_unlock(zhdr); 1391 z3fold_page_unlock(zhdr);
1308 return -EAGAIN; 1392 return -EAGAIN;
1309 } 1393 }
1310 new_zhdr = page_address(newpage); 1394 new_zhdr = page_address(newpage);
1311 memcpy(new_zhdr, zhdr, PAGE_SIZE); 1395 memcpy(new_zhdr, zhdr, PAGE_SIZE);
1312 newpage->private = page->private; 1396 newpage->private = page->private;
1313 set_bit(PAGE_MIGRATED, &page->private !! 1397 page->private = 0;
1314 z3fold_page_unlock(zhdr); 1398 z3fold_page_unlock(zhdr);
1315 spin_lock_init(&new_zhdr->page_lock); 1399 spin_lock_init(&new_zhdr->page_lock);
1316 INIT_WORK(&new_zhdr->work, compact_pa 1400 INIT_WORK(&new_zhdr->work, compact_page_work);
1317 /* 1401 /*
1318 * z3fold_page_isolate() ensures that 1402 * z3fold_page_isolate() ensures that new_zhdr->buddy is empty,
1319 * so we only have to reinitialize it 1403 * so we only have to reinitialize it.
1320 */ 1404 */
1321 INIT_LIST_HEAD(&new_zhdr->buddy); 1405 INIT_LIST_HEAD(&new_zhdr->buddy);
>> 1406 new_mapping = page_mapping(page);
1322 __ClearPageMovable(page); 1407 __ClearPageMovable(page);
>> 1408 ClearPagePrivate(page);
1323 1409
1324 get_page(newpage); 1410 get_page(newpage);
1325 z3fold_page_lock(new_zhdr); 1411 z3fold_page_lock(new_zhdr);
1326 if (new_zhdr->first_chunks) 1412 if (new_zhdr->first_chunks)
1327 encode_handle(new_zhdr, FIRST 1413 encode_handle(new_zhdr, FIRST);
1328 if (new_zhdr->last_chunks) 1414 if (new_zhdr->last_chunks)
1329 encode_handle(new_zhdr, LAST) 1415 encode_handle(new_zhdr, LAST);
1330 if (new_zhdr->middle_chunks) 1416 if (new_zhdr->middle_chunks)
1331 encode_handle(new_zhdr, MIDDL 1417 encode_handle(new_zhdr, MIDDLE);
1332 set_bit(NEEDS_COMPACTING, &newpage->p 1418 set_bit(NEEDS_COMPACTING, &newpage->private);
1333 new_zhdr->cpu = smp_processor_id(); 1419 new_zhdr->cpu = smp_processor_id();
1334 __SetPageMovable(newpage, &z3fold_mop !! 1420 spin_lock(&pool->lock);
>> 1421 list_add(&newpage->lru, &pool->lru);
>> 1422 spin_unlock(&pool->lock);
>> 1423 __SetPageMovable(newpage, new_mapping);
1335 z3fold_page_unlock(new_zhdr); 1424 z3fold_page_unlock(new_zhdr);
1336 1425
1337 queue_work_on(new_zhdr->cpu, pool->co 1426 queue_work_on(new_zhdr->cpu, pool->compact_wq, &new_zhdr->work);
1338 1427
1339 /* PAGE_CLAIMED and PAGE_MIGRATED are !! 1428 page_mapcount_reset(page);
1340 page->private = 0; <<
1341 put_page(page); 1429 put_page(page);
1342 return 0; 1430 return 0;
1343 } 1431 }
1344 1432
1345 static void z3fold_page_putback(struct page * 1433 static void z3fold_page_putback(struct page *page)
1346 { 1434 {
1347 struct z3fold_header *zhdr; 1435 struct z3fold_header *zhdr;
1348 struct z3fold_pool *pool; 1436 struct z3fold_pool *pool;
1349 1437
1350 zhdr = page_address(page); 1438 zhdr = page_address(page);
1351 pool = zhdr_to_pool(zhdr); 1439 pool = zhdr_to_pool(zhdr);
1352 1440
1353 z3fold_page_lock(zhdr); 1441 z3fold_page_lock(zhdr);
1354 if (!list_empty(&zhdr->buddy)) 1442 if (!list_empty(&zhdr->buddy))
1355 list_del_init(&zhdr->buddy); 1443 list_del_init(&zhdr->buddy);
1356 INIT_LIST_HEAD(&page->lru); 1444 INIT_LIST_HEAD(&page->lru);
1357 if (put_z3fold_locked(zhdr)) !! 1445 if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) {
>> 1446 atomic64_dec(&pool->pages_nr);
1358 return; 1447 return;
1359 if (list_empty(&zhdr->buddy)) !! 1448 }
1360 add_to_unbuddied(pool, zhdr); !! 1449 spin_lock(&pool->lock);
1361 clear_bit(PAGE_CLAIMED, &page->privat !! 1450 list_add(&page->lru, &pool->lru);
>> 1451 spin_unlock(&pool->lock);
1362 z3fold_page_unlock(zhdr); 1452 z3fold_page_unlock(zhdr);
1363 } 1453 }
1364 1454
1365 static const struct movable_operations z3fold !! 1455 static const struct address_space_operations z3fold_aops = {
1366 .isolate_page = z3fold_page_isolate, 1456 .isolate_page = z3fold_page_isolate,
1367 .migrate_page = z3fold_page_migrate, !! 1457 .migratepage = z3fold_page_migrate,
1368 .putback_page = z3fold_page_putback, 1458 .putback_page = z3fold_page_putback,
1369 }; 1459 };
1370 1460
1371 /***************** 1461 /*****************
1372 * zpool 1462 * zpool
1373 ****************/ 1463 ****************/
1374 1464
1375 static void *z3fold_zpool_create(const char * !! 1465 static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle)
1376 { 1466 {
1377 return z3fold_create_pool(name, gfp); !! 1467 if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict)
>> 1468 return pool->zpool_ops->evict(pool->zpool, handle);
>> 1469 else
>> 1470 return -ENOENT;
>> 1471 }
>> 1472
>> 1473 static const struct z3fold_ops z3fold_zpool_ops = {
>> 1474 .evict = z3fold_zpool_evict
>> 1475 };
>> 1476
>> 1477 static void *z3fold_zpool_create(const char *name, gfp_t gfp,
>> 1478 const struct zpool_ops *zpool_ops,
>> 1479 struct zpool *zpool)
>> 1480 {
>> 1481 struct z3fold_pool *pool;
>> 1482
>> 1483 pool = z3fold_create_pool(name, gfp,
>> 1484 zpool_ops ? &z3fold_zpool_ops : NULL);
>> 1485 if (pool) {
>> 1486 pool->zpool = zpool;
>> 1487 pool->zpool_ops = zpool_ops;
>> 1488 }
>> 1489 return pool;
1378 } 1490 }
1379 1491
1380 static void z3fold_zpool_destroy(void *pool) 1492 static void z3fold_zpool_destroy(void *pool)
1381 { 1493 {
1382 z3fold_destroy_pool(pool); 1494 z3fold_destroy_pool(pool);
1383 } 1495 }
1384 1496
1385 static int z3fold_zpool_malloc(void *pool, si 1497 static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp,
1386 unsigned long *handle 1498 unsigned long *handle)
1387 { 1499 {
1388 return z3fold_alloc(pool, size, gfp, 1500 return z3fold_alloc(pool, size, gfp, handle);
1389 } 1501 }
1390 static void z3fold_zpool_free(void *pool, uns 1502 static void z3fold_zpool_free(void *pool, unsigned long handle)
1391 { 1503 {
1392 z3fold_free(pool, handle); 1504 z3fold_free(pool, handle);
1393 } 1505 }
1394 1506
>> 1507 static int z3fold_zpool_shrink(void *pool, unsigned int pages,
>> 1508 unsigned int *reclaimed)
>> 1509 {
>> 1510 unsigned int total = 0;
>> 1511 int ret = -EINVAL;
>> 1512
>> 1513 while (total < pages) {
>> 1514 ret = z3fold_reclaim_page(pool, 8);
>> 1515 if (ret < 0)
>> 1516 break;
>> 1517 total++;
>> 1518 }
>> 1519
>> 1520 if (reclaimed)
>> 1521 *reclaimed = total;
>> 1522
>> 1523 return ret;
>> 1524 }
>> 1525
1395 static void *z3fold_zpool_map(void *pool, uns 1526 static void *z3fold_zpool_map(void *pool, unsigned long handle,
1396 enum zpool_mapmode mm 1527 enum zpool_mapmode mm)
1397 { 1528 {
1398 return z3fold_map(pool, handle); 1529 return z3fold_map(pool, handle);
1399 } 1530 }
1400 static void z3fold_zpool_unmap(void *pool, un 1531 static void z3fold_zpool_unmap(void *pool, unsigned long handle)
1401 { 1532 {
1402 z3fold_unmap(pool, handle); 1533 z3fold_unmap(pool, handle);
1403 } 1534 }
1404 1535
1405 static u64 z3fold_zpool_total_pages(void *poo !! 1536 static u64 z3fold_zpool_total_size(void *pool)
1406 { 1537 {
1407 return z3fold_get_pool_pages(pool); !! 1538 return z3fold_get_pool_size(pool) * PAGE_SIZE;
1408 } 1539 }
1409 1540
1410 static struct zpool_driver z3fold_zpool_drive 1541 static struct zpool_driver z3fold_zpool_driver = {
1411 .type = "z3fold", 1542 .type = "z3fold",
1412 .sleep_mapped = true, <<
1413 .owner = THIS_MODULE, 1543 .owner = THIS_MODULE,
1414 .create = z3fold_zpool_create, 1544 .create = z3fold_zpool_create,
1415 .destroy = z3fold_zpool_destroy, 1545 .destroy = z3fold_zpool_destroy,
1416 .malloc = z3fold_zpool_malloc, 1546 .malloc = z3fold_zpool_malloc,
1417 .free = z3fold_zpool_free, 1547 .free = z3fold_zpool_free,
>> 1548 .shrink = z3fold_zpool_shrink,
1418 .map = z3fold_zpool_map, 1549 .map = z3fold_zpool_map,
1419 .unmap = z3fold_zpool_unmap, 1550 .unmap = z3fold_zpool_unmap,
1420 .total_pages = z3fold_zpool_total_pa !! 1551 .total_size = z3fold_zpool_total_size,
1421 }; 1552 };
1422 1553
1423 MODULE_ALIAS("zpool-z3fold"); 1554 MODULE_ALIAS("zpool-z3fold");
1424 1555
1425 static int __init init_z3fold(void) 1556 static int __init init_z3fold(void)
1426 { 1557 {
1427 /* !! 1558 int ret;
1428 * Make sure the z3fold header is not !! 1559
1429 * there has remaining spaces for its !! 1560 /* Make sure the z3fold header is not larger than the page size */
1430 */ !! 1561 BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE);
1431 BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE !! 1562 ret = z3fold_mount();
>> 1563 if (ret)
>> 1564 return ret;
>> 1565
1432 zpool_register_driver(&z3fold_zpool_d 1566 zpool_register_driver(&z3fold_zpool_driver);
1433 1567
1434 return 0; 1568 return 0;
1435 } 1569 }
1436 1570
1437 static void __exit exit_z3fold(void) 1571 static void __exit exit_z3fold(void)
1438 { 1572 {
>> 1573 z3fold_unmount();
1439 zpool_unregister_driver(&z3fold_zpool 1574 zpool_unregister_driver(&z3fold_zpool_driver);
1440 } 1575 }
1441 1576
1442 module_init(init_z3fold); 1577 module_init(init_z3fold);
1443 module_exit(exit_z3fold); 1578 module_exit(exit_z3fold);
1444 1579
1445 MODULE_LICENSE("GPL"); 1580 MODULE_LICENSE("GPL");
1446 MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail. 1581 MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>");
1447 MODULE_DESCRIPTION("3-Fold Allocator for Comp 1582 MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages");
1448 1583
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.