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/list.h> 29 #include <linux/list.h>
30 #include <linux/mm.h> 30 #include <linux/mm.h>
31 #include <linux/module.h> 31 #include <linux/module.h>
32 #include <linux/page-flags.h> 32 #include <linux/page-flags.h>
33 #include <linux/migrate.h> 33 #include <linux/migrate.h>
34 #include <linux/node.h> 34 #include <linux/node.h>
35 #include <linux/compaction.h> 35 #include <linux/compaction.h>
36 #include <linux/percpu.h> 36 #include <linux/percpu.h>
37 #include <linux/preempt.h> 37 #include <linux/preempt.h>
38 #include <linux/workqueue.h> 38 #include <linux/workqueue.h>
39 #include <linux/slab.h> 39 #include <linux/slab.h>
40 #include <linux/spinlock.h> 40 #include <linux/spinlock.h>
41 #include <linux/zpool.h> 41 #include <linux/zpool.h>
42 #include <linux/kmemleak.h> 42 #include <linux/kmemleak.h>
43 43
44 /* 44 /*
45 * NCHUNKS_ORDER determines the internal alloc 45 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
46 * adjusting internal fragmentation. It also 46 * adjusting internal fragmentation. It also determines the number of
47 * freelists maintained in each pool. NCHUNKS_ 47 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
48 * allocation granularity will be in chunks of 48 * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks
49 * in the beginning of an allocated page are o 49 * in the beginning of an allocated page are occupied by z3fold header, so
50 * NCHUNKS will be calculated to 63 (or 62 in 50 * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y),
51 * which shows the max number of free chunks i 51 * which shows the max number of free chunks in z3fold page, also there will
52 * be 63, or 62, respectively, freelists per p 52 * be 63, or 62, respectively, freelists per pool.
53 */ 53 */
54 #define NCHUNKS_ORDER 6 54 #define NCHUNKS_ORDER 6
55 55
56 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ 56 #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER)
57 #define CHUNK_SIZE (1 << CHUNK_SHIFT) 57 #define CHUNK_SIZE (1 << CHUNK_SHIFT)
58 #define ZHDR_SIZE_ALIGNED round_up(sizeof(stru 58 #define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE)
59 #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> 59 #define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT)
60 #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SH 60 #define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SHIFT)
61 #define NCHUNKS (TOTAL_CHUNKS - ZHDR_C 61 #define NCHUNKS (TOTAL_CHUNKS - ZHDR_CHUNKS)
62 62
63 #define BUDDY_MASK (0x3) 63 #define BUDDY_MASK (0x3)
64 #define BUDDY_SHIFT 2 64 #define BUDDY_SHIFT 2
65 #define SLOTS_ALIGN (0x40) 65 #define SLOTS_ALIGN (0x40)
66 66
67 /***************** 67 /*****************
68 * Structures 68 * Structures
69 *****************/ 69 *****************/
70 struct z3fold_pool; 70 struct z3fold_pool;
71 71
72 enum buddy { 72 enum buddy {
73 HEADLESS = 0, 73 HEADLESS = 0,
74 FIRST, 74 FIRST,
75 MIDDLE, 75 MIDDLE,
76 LAST, 76 LAST,
77 BUDDIES_MAX = LAST 77 BUDDIES_MAX = LAST
78 }; 78 };
79 79
80 struct z3fold_buddy_slots { 80 struct z3fold_buddy_slots {
81 /* 81 /*
82 * we are using BUDDY_MASK in handle_t 82 * we are using BUDDY_MASK in handle_to_buddy etc. so there should
83 * be enough slots to hold all possibl 83 * be enough slots to hold all possible variants
84 */ 84 */
85 unsigned long slot[BUDDY_MASK + 1]; 85 unsigned long slot[BUDDY_MASK + 1];
86 unsigned long pool; /* back link */ 86 unsigned long pool; /* back link */
87 rwlock_t lock; 87 rwlock_t lock;
88 }; 88 };
89 #define HANDLE_FLAG_MASK (0x03) 89 #define HANDLE_FLAG_MASK (0x03)
90 90
91 /* 91 /*
92 * struct z3fold_header - z3fold page metadata 92 * struct z3fold_header - z3fold page metadata occupying first chunks of each
93 * z3fold page, except fo 93 * z3fold page, except for HEADLESS pages
94 * @buddy: links the z3fold page 94 * @buddy: links the z3fold page into the relevant list in the
95 * pool 95 * pool
96 * @page_lock: per-page lock 96 * @page_lock: per-page lock
97 * @refcount: reference count for th 97 * @refcount: reference count for the z3fold page
98 * @work: work_struct for page l 98 * @work: work_struct for page layout optimization
99 * @slots: pointer to the structu 99 * @slots: pointer to the structure holding buddy slots
100 * @pool: pointer to the contain 100 * @pool: pointer to the containing pool
101 * @cpu: CPU which this page "b 101 * @cpu: CPU which this page "belongs" to
102 * @first_chunks: the size of the first 102 * @first_chunks: the size of the first buddy in chunks, 0 if free
103 * @middle_chunks: the size of the middle 103 * @middle_chunks: the size of the middle buddy in chunks, 0 if free
104 * @last_chunks: the size of the last b 104 * @last_chunks: the size of the last buddy in chunks, 0 if free
105 * @first_num: the starting number (f 105 * @first_num: the starting number (for the first handle)
106 * @mapped_count: the number of objects 106 * @mapped_count: the number of objects currently mapped
107 */ 107 */
108 struct z3fold_header { 108 struct z3fold_header {
109 struct list_head buddy; 109 struct list_head buddy;
110 spinlock_t page_lock; 110 spinlock_t page_lock;
111 struct kref refcount; 111 struct kref refcount;
112 struct work_struct work; 112 struct work_struct work;
113 struct z3fold_buddy_slots *slots; 113 struct z3fold_buddy_slots *slots;
114 struct z3fold_pool *pool; 114 struct z3fold_pool *pool;
115 short cpu; 115 short cpu;
116 unsigned short first_chunks; 116 unsigned short first_chunks;
117 unsigned short middle_chunks; 117 unsigned short middle_chunks;
118 unsigned short last_chunks; 118 unsigned short last_chunks;
119 unsigned short start_middle; 119 unsigned short start_middle;
120 unsigned short first_num:2; 120 unsigned short first_num:2;
121 unsigned short mapped_count:2; 121 unsigned short mapped_count:2;
122 unsigned short foreign_handles:2; 122 unsigned short foreign_handles:2;
123 }; 123 };
124 124
125 /** 125 /**
126 * struct z3fold_pool - stores metadata for ea 126 * struct z3fold_pool - stores metadata for each z3fold pool
127 * @name: pool name 127 * @name: pool name
128 * @lock: protects pool unbuddied lists 128 * @lock: protects pool unbuddied lists
129 * @stale_lock: protects pool stale page list 129 * @stale_lock: protects pool stale page list
130 * @unbuddied: per-cpu array of lists trackin 130 * @unbuddied: per-cpu array of lists tracking z3fold pages that contain 2-
131 * buddies; the list each z3fold 131 * buddies; the list each z3fold page is added to depends on
132 * the size of its free region. 132 * the size of its free region.
133 * @stale: list of pages marked for freei 133 * @stale: list of pages marked for freeing
134 * @pages_nr: number of z3fold pages in the 134 * @pages_nr: number of z3fold pages in the pool.
135 * @c_handle: cache for z3fold_buddy_slots a 135 * @c_handle: cache for z3fold_buddy_slots allocation
136 * @compact_wq: workqueue for page layout back 136 * @compact_wq: workqueue for page layout background optimization
137 * @release_wq: workqueue for safe page releas 137 * @release_wq: workqueue for safe page release
138 * @work: work_struct for safe page rele 138 * @work: work_struct for safe page release
139 * 139 *
140 * This structure is allocated at pool creatio 140 * This structure is allocated at pool creation time and maintains metadata
141 * pertaining to a particular z3fold pool. 141 * pertaining to a particular z3fold pool.
142 */ 142 */
143 struct z3fold_pool { 143 struct z3fold_pool {
144 const char *name; 144 const char *name;
145 spinlock_t lock; 145 spinlock_t lock;
146 spinlock_t stale_lock; 146 spinlock_t stale_lock;
147 struct list_head __percpu *unbuddied; 147 struct list_head __percpu *unbuddied;
148 struct list_head stale; 148 struct list_head stale;
149 atomic64_t pages_nr; 149 atomic64_t pages_nr;
150 struct kmem_cache *c_handle; 150 struct kmem_cache *c_handle;
151 struct workqueue_struct *compact_wq; 151 struct workqueue_struct *compact_wq;
152 struct workqueue_struct *release_wq; 152 struct workqueue_struct *release_wq;
153 struct work_struct work; 153 struct work_struct work;
154 }; 154 };
155 155
156 /* 156 /*
157 * Internal z3fold page flags 157 * Internal z3fold page flags
158 */ 158 */
159 enum z3fold_page_flags { 159 enum z3fold_page_flags {
160 PAGE_HEADLESS = 0, 160 PAGE_HEADLESS = 0,
161 MIDDLE_CHUNK_MAPPED, 161 MIDDLE_CHUNK_MAPPED,
162 NEEDS_COMPACTING, 162 NEEDS_COMPACTING,
163 PAGE_STALE, 163 PAGE_STALE,
164 PAGE_CLAIMED, /* by either reclaim or 164 PAGE_CLAIMED, /* by either reclaim or free */
165 PAGE_MIGRATED, /* page is migrated and 165 PAGE_MIGRATED, /* page is migrated and soon to be released */
166 }; 166 };
167 167
168 /* 168 /*
169 * handle flags, go under HANDLE_FLAG_MASK 169 * handle flags, go under HANDLE_FLAG_MASK
170 */ 170 */
171 enum z3fold_handle_flags { 171 enum z3fold_handle_flags {
172 HANDLES_NOFREE = 0, 172 HANDLES_NOFREE = 0,
173 }; 173 };
174 174
175 /* 175 /*
176 * Forward declarations 176 * Forward declarations
177 */ 177 */
178 static struct z3fold_header *__z3fold_alloc(st 178 static struct z3fold_header *__z3fold_alloc(struct z3fold_pool *, size_t, bool);
179 static void compact_page_work(struct work_stru 179 static void compact_page_work(struct work_struct *w);
180 180
181 /***************** 181 /*****************
182 * Helpers 182 * Helpers
183 *****************/ 183 *****************/
184 184
185 /* Converts an allocation size in bytes to siz 185 /* Converts an allocation size in bytes to size in z3fold chunks */
186 static int size_to_chunks(size_t size) 186 static int size_to_chunks(size_t size)
187 { 187 {
188 return (size + CHUNK_SIZE - 1) >> CHUN 188 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
189 } 189 }
190 190
191 #define for_each_unbuddied_list(_iter, _begin) 191 #define for_each_unbuddied_list(_iter, _begin) \
192 for ((_iter) = (_begin); (_iter) < NCH 192 for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
193 193
194 static inline struct z3fold_buddy_slots *alloc 194 static inline struct z3fold_buddy_slots *alloc_slots(struct z3fold_pool *pool,
195 195 gfp_t gfp)
196 { 196 {
197 struct z3fold_buddy_slots *slots = kme 197 struct z3fold_buddy_slots *slots = kmem_cache_zalloc(pool->c_handle,
198 198 gfp);
199 199
200 if (slots) { 200 if (slots) {
201 /* It will be freed separately 201 /* It will be freed separately in free_handle(). */
202 kmemleak_not_leak(slots); 202 kmemleak_not_leak(slots);
203 slots->pool = (unsigned long)p 203 slots->pool = (unsigned long)pool;
204 rwlock_init(&slots->lock); 204 rwlock_init(&slots->lock);
205 } 205 }
206 206
207 return slots; 207 return slots;
208 } 208 }
209 209
210 static inline struct z3fold_pool *slots_to_poo 210 static inline struct z3fold_pool *slots_to_pool(struct z3fold_buddy_slots *s)
211 { 211 {
212 return (struct z3fold_pool *)(s->pool 212 return (struct z3fold_pool *)(s->pool & ~HANDLE_FLAG_MASK);
213 } 213 }
214 214
215 static inline struct z3fold_buddy_slots *handl 215 static inline struct z3fold_buddy_slots *handle_to_slots(unsigned long handle)
216 { 216 {
217 return (struct z3fold_buddy_slots *)(h 217 return (struct z3fold_buddy_slots *)(handle & ~(SLOTS_ALIGN - 1));
218 } 218 }
219 219
220 /* Lock a z3fold page */ 220 /* Lock a z3fold page */
221 static inline void z3fold_page_lock(struct z3f 221 static inline void z3fold_page_lock(struct z3fold_header *zhdr)
222 { 222 {
223 spin_lock(&zhdr->page_lock); 223 spin_lock(&zhdr->page_lock);
224 } 224 }
225 225
226 /* Try to lock a z3fold page */ 226 /* Try to lock a z3fold page */
227 static inline int z3fold_page_trylock(struct z 227 static inline int z3fold_page_trylock(struct z3fold_header *zhdr)
228 { 228 {
229 return spin_trylock(&zhdr->page_lock); 229 return spin_trylock(&zhdr->page_lock);
230 } 230 }
231 231
232 /* Unlock a z3fold page */ 232 /* Unlock a z3fold page */
233 static inline void z3fold_page_unlock(struct z 233 static inline void z3fold_page_unlock(struct z3fold_header *zhdr)
234 { 234 {
235 spin_unlock(&zhdr->page_lock); 235 spin_unlock(&zhdr->page_lock);
236 } 236 }
237 237
238 /* return locked z3fold page if it's not headl 238 /* return locked z3fold page if it's not headless */
239 static inline struct z3fold_header *get_z3fold 239 static inline struct z3fold_header *get_z3fold_header(unsigned long handle)
240 { 240 {
241 struct z3fold_buddy_slots *slots; 241 struct z3fold_buddy_slots *slots;
242 struct z3fold_header *zhdr; 242 struct z3fold_header *zhdr;
243 int locked = 0; 243 int locked = 0;
244 244
245 if (!(handle & (1 << PAGE_HEADLESS))) 245 if (!(handle & (1 << PAGE_HEADLESS))) {
246 slots = handle_to_slots(handle 246 slots = handle_to_slots(handle);
247 do { 247 do {
248 unsigned long addr; 248 unsigned long addr;
249 249
250 read_lock(&slots->lock 250 read_lock(&slots->lock);
251 addr = *(unsigned long 251 addr = *(unsigned long *)handle;
252 zhdr = (struct z3fold_ 252 zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
253 locked = z3fold_page_t 253 locked = z3fold_page_trylock(zhdr);
254 read_unlock(&slots->lo 254 read_unlock(&slots->lock);
255 if (locked) { 255 if (locked) {
256 struct page *p 256 struct page *page = virt_to_page(zhdr);
257 257
258 if (!test_bit( 258 if (!test_bit(PAGE_MIGRATED, &page->private))
259 break; 259 break;
260 z3fold_page_un 260 z3fold_page_unlock(zhdr);
261 } 261 }
262 cpu_relax(); 262 cpu_relax();
263 } while (true); 263 } while (true);
264 } else { 264 } else {
265 zhdr = (struct z3fold_header * 265 zhdr = (struct z3fold_header *)(handle & PAGE_MASK);
266 } 266 }
267 267
268 return zhdr; 268 return zhdr;
269 } 269 }
270 270
271 static inline void put_z3fold_header(struct z3 271 static inline void put_z3fold_header(struct z3fold_header *zhdr)
272 { 272 {
273 struct page *page = virt_to_page(zhdr) 273 struct page *page = virt_to_page(zhdr);
274 274
275 if (!test_bit(PAGE_HEADLESS, &page->pr 275 if (!test_bit(PAGE_HEADLESS, &page->private))
276 z3fold_page_unlock(zhdr); 276 z3fold_page_unlock(zhdr);
277 } 277 }
278 278
279 static inline void free_handle(unsigned long h 279 static inline void free_handle(unsigned long handle, struct z3fold_header *zhdr)
280 { 280 {
281 struct z3fold_buddy_slots *slots; 281 struct z3fold_buddy_slots *slots;
282 int i; 282 int i;
283 bool is_free; 283 bool is_free;
284 284
285 if (WARN_ON(*(unsigned long *)handle = 285 if (WARN_ON(*(unsigned long *)handle == 0))
286 return; 286 return;
287 287
288 slots = handle_to_slots(handle); 288 slots = handle_to_slots(handle);
289 write_lock(&slots->lock); 289 write_lock(&slots->lock);
290 *(unsigned long *)handle = 0; 290 *(unsigned long *)handle = 0;
291 291
292 if (test_bit(HANDLES_NOFREE, &slots->p 292 if (test_bit(HANDLES_NOFREE, &slots->pool)) {
293 write_unlock(&slots->lock); 293 write_unlock(&slots->lock);
294 return; /* simple case, nothin 294 return; /* simple case, nothing else to do */
295 } 295 }
296 296
297 if (zhdr->slots != slots) 297 if (zhdr->slots != slots)
298 zhdr->foreign_handles--; 298 zhdr->foreign_handles--;
299 299
300 is_free = true; 300 is_free = true;
301 for (i = 0; i <= BUDDY_MASK; i++) { 301 for (i = 0; i <= BUDDY_MASK; i++) {
302 if (slots->slot[i]) { 302 if (slots->slot[i]) {
303 is_free = false; 303 is_free = false;
304 break; 304 break;
305 } 305 }
306 } 306 }
307 write_unlock(&slots->lock); 307 write_unlock(&slots->lock);
308 308
309 if (is_free) { 309 if (is_free) {
310 struct z3fold_pool *pool = slo 310 struct z3fold_pool *pool = slots_to_pool(slots);
311 311
312 if (zhdr->slots == slots) 312 if (zhdr->slots == slots)
313 zhdr->slots = NULL; 313 zhdr->slots = NULL;
314 kmem_cache_free(pool->c_handle 314 kmem_cache_free(pool->c_handle, slots);
315 } 315 }
316 } 316 }
317 317
318 /* Initializes the z3fold header of a newly al 318 /* Initializes the z3fold header of a newly allocated z3fold page */
319 static struct z3fold_header *init_z3fold_page( 319 static struct z3fold_header *init_z3fold_page(struct page *page, bool headless,
320 struct 320 struct z3fold_pool *pool, gfp_t gfp)
321 { 321 {
322 struct z3fold_header *zhdr = page_addr 322 struct z3fold_header *zhdr = page_address(page);
323 struct z3fold_buddy_slots *slots; 323 struct z3fold_buddy_slots *slots;
324 324
325 clear_bit(PAGE_HEADLESS, &page->privat 325 clear_bit(PAGE_HEADLESS, &page->private);
326 clear_bit(MIDDLE_CHUNK_MAPPED, &page-> 326 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
327 clear_bit(NEEDS_COMPACTING, &page->pri 327 clear_bit(NEEDS_COMPACTING, &page->private);
328 clear_bit(PAGE_STALE, &page->private); 328 clear_bit(PAGE_STALE, &page->private);
329 clear_bit(PAGE_CLAIMED, &page->private 329 clear_bit(PAGE_CLAIMED, &page->private);
330 clear_bit(PAGE_MIGRATED, &page->privat 330 clear_bit(PAGE_MIGRATED, &page->private);
331 if (headless) 331 if (headless)
332 return zhdr; 332 return zhdr;
333 333
334 slots = alloc_slots(pool, gfp); 334 slots = alloc_slots(pool, gfp);
335 if (!slots) 335 if (!slots)
336 return NULL; 336 return NULL;
337 337
338 memset(zhdr, 0, sizeof(*zhdr)); 338 memset(zhdr, 0, sizeof(*zhdr));
339 spin_lock_init(&zhdr->page_lock); 339 spin_lock_init(&zhdr->page_lock);
340 kref_init(&zhdr->refcount); 340 kref_init(&zhdr->refcount);
341 zhdr->cpu = -1; 341 zhdr->cpu = -1;
342 zhdr->slots = slots; 342 zhdr->slots = slots;
343 zhdr->pool = pool; 343 zhdr->pool = pool;
344 INIT_LIST_HEAD(&zhdr->buddy); 344 INIT_LIST_HEAD(&zhdr->buddy);
345 INIT_WORK(&zhdr->work, compact_page_wo 345 INIT_WORK(&zhdr->work, compact_page_work);
346 return zhdr; 346 return zhdr;
347 } 347 }
348 348
349 /* Resets the struct page fields and frees the 349 /* Resets the struct page fields and frees the page */
350 static void free_z3fold_page(struct page *page 350 static void free_z3fold_page(struct page *page, bool headless)
351 { 351 {
352 if (!headless) { 352 if (!headless) {
353 lock_page(page); 353 lock_page(page);
354 __ClearPageMovable(page); 354 __ClearPageMovable(page);
355 unlock_page(page); 355 unlock_page(page);
356 } 356 }
357 __free_page(page); 357 __free_page(page);
358 } 358 }
359 359
360 /* Helper function to build the index */ 360 /* Helper function to build the index */
361 static inline int __idx(struct z3fold_header * 361 static inline int __idx(struct z3fold_header *zhdr, enum buddy bud)
362 { 362 {
363 return (bud + zhdr->first_num) & BUDDY 363 return (bud + zhdr->first_num) & BUDDY_MASK;
364 } 364 }
365 365
366 /* 366 /*
367 * Encodes the handle of a particular buddy wi 367 * Encodes the handle of a particular buddy within a z3fold page.
368 * Zhdr->page_lock should be held as this func 368 * Zhdr->page_lock should be held as this function accesses first_num
369 * if bud != HEADLESS. 369 * if bud != HEADLESS.
370 */ 370 */
371 static unsigned long __encode_handle(struct z3 371 static unsigned long __encode_handle(struct z3fold_header *zhdr,
372 struct z3fold_ 372 struct z3fold_buddy_slots *slots,
373 enum buddy bud 373 enum buddy bud)
374 { 374 {
375 unsigned long h = (unsigned long)zhdr; 375 unsigned long h = (unsigned long)zhdr;
376 int idx = 0; 376 int idx = 0;
377 377
378 /* 378 /*
379 * For a headless page, its handle is 379 * For a headless page, its handle is its pointer with the extra
380 * PAGE_HEADLESS bit set 380 * PAGE_HEADLESS bit set
381 */ 381 */
382 if (bud == HEADLESS) 382 if (bud == HEADLESS)
383 return h | (1 << PAGE_HEADLESS 383 return h | (1 << PAGE_HEADLESS);
384 384
385 /* otherwise, return pointer to encode 385 /* otherwise, return pointer to encoded handle */
386 idx = __idx(zhdr, bud); 386 idx = __idx(zhdr, bud);
387 h += idx; 387 h += idx;
388 if (bud == LAST) 388 if (bud == LAST)
389 h |= (zhdr->last_chunks << BUD 389 h |= (zhdr->last_chunks << BUDDY_SHIFT);
390 390
391 write_lock(&slots->lock); 391 write_lock(&slots->lock);
392 slots->slot[idx] = h; 392 slots->slot[idx] = h;
393 write_unlock(&slots->lock); 393 write_unlock(&slots->lock);
394 return (unsigned long)&slots->slot[idx 394 return (unsigned long)&slots->slot[idx];
395 } 395 }
396 396
397 static unsigned long encode_handle(struct z3fo 397 static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud)
398 { 398 {
399 return __encode_handle(zhdr, zhdr->slo 399 return __encode_handle(zhdr, zhdr->slots, bud);
400 } 400 }
401 401
402 /* only for LAST bud, returns zero otherwise * 402 /* only for LAST bud, returns zero otherwise */
403 static unsigned short handle_to_chunks(unsigne 403 static unsigned short handle_to_chunks(unsigned long handle)
404 { 404 {
405 struct z3fold_buddy_slots *slots = han 405 struct z3fold_buddy_slots *slots = handle_to_slots(handle);
406 unsigned long addr; 406 unsigned long addr;
407 407
408 read_lock(&slots->lock); 408 read_lock(&slots->lock);
409 addr = *(unsigned long *)handle; 409 addr = *(unsigned long *)handle;
410 read_unlock(&slots->lock); 410 read_unlock(&slots->lock);
411 return (addr & ~PAGE_MASK) >> BUDDY_SH 411 return (addr & ~PAGE_MASK) >> BUDDY_SHIFT;
412 } 412 }
413 413
414 /* 414 /*
415 * (handle & BUDDY_MASK) < zhdr->first_num is 415 * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle
416 * but that doesn't matter. because the maski 416 * but that doesn't matter. because the masking will result in the
417 * correct buddy number. 417 * correct buddy number.
418 */ 418 */
419 static enum buddy handle_to_buddy(unsigned lon 419 static enum buddy handle_to_buddy(unsigned long handle)
420 { 420 {
421 struct z3fold_header *zhdr; 421 struct z3fold_header *zhdr;
422 struct z3fold_buddy_slots *slots = han 422 struct z3fold_buddy_slots *slots = handle_to_slots(handle);
423 unsigned long addr; 423 unsigned long addr;
424 424
425 read_lock(&slots->lock); 425 read_lock(&slots->lock);
426 WARN_ON(handle & (1 << PAGE_HEADLESS)) 426 WARN_ON(handle & (1 << PAGE_HEADLESS));
427 addr = *(unsigned long *)handle; 427 addr = *(unsigned long *)handle;
428 read_unlock(&slots->lock); 428 read_unlock(&slots->lock);
429 zhdr = (struct z3fold_header *)(addr & 429 zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
430 return (addr - zhdr->first_num) & BUDD 430 return (addr - zhdr->first_num) & BUDDY_MASK;
431 } 431 }
432 432
433 static inline struct z3fold_pool *zhdr_to_pool 433 static inline struct z3fold_pool *zhdr_to_pool(struct z3fold_header *zhdr)
434 { 434 {
435 return zhdr->pool; 435 return zhdr->pool;
436 } 436 }
437 437
438 static void __release_z3fold_page(struct z3fol 438 static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked)
439 { 439 {
440 struct page *page = virt_to_page(zhdr) 440 struct page *page = virt_to_page(zhdr);
441 struct z3fold_pool *pool = zhdr_to_poo 441 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
442 442
443 WARN_ON(!list_empty(&zhdr->buddy)); 443 WARN_ON(!list_empty(&zhdr->buddy));
444 set_bit(PAGE_STALE, &page->private); 444 set_bit(PAGE_STALE, &page->private);
445 clear_bit(NEEDS_COMPACTING, &page->pri 445 clear_bit(NEEDS_COMPACTING, &page->private);
446 spin_lock(&pool->lock); 446 spin_lock(&pool->lock);
447 spin_unlock(&pool->lock); 447 spin_unlock(&pool->lock);
448 448
449 if (locked) 449 if (locked)
450 z3fold_page_unlock(zhdr); 450 z3fold_page_unlock(zhdr);
451 451
452 spin_lock(&pool->stale_lock); 452 spin_lock(&pool->stale_lock);
453 list_add(&zhdr->buddy, &pool->stale); 453 list_add(&zhdr->buddy, &pool->stale);
454 queue_work(pool->release_wq, &pool->wo 454 queue_work(pool->release_wq, &pool->work);
455 spin_unlock(&pool->stale_lock); 455 spin_unlock(&pool->stale_lock);
456 456
457 atomic64_dec(&pool->pages_nr); 457 atomic64_dec(&pool->pages_nr);
458 } 458 }
459 459
460 static void release_z3fold_page_locked(struct 460 static void release_z3fold_page_locked(struct kref *ref)
461 { 461 {
462 struct z3fold_header *zhdr = container 462 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
463 463 refcount);
464 WARN_ON(z3fold_page_trylock(zhdr)); 464 WARN_ON(z3fold_page_trylock(zhdr));
465 __release_z3fold_page(zhdr, true); 465 __release_z3fold_page(zhdr, true);
466 } 466 }
467 467
468 static void release_z3fold_page_locked_list(st 468 static void release_z3fold_page_locked_list(struct kref *ref)
469 { 469 {
470 struct z3fold_header *zhdr = container 470 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
471 471 refcount);
472 struct z3fold_pool *pool = zhdr_to_poo 472 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
473 473
474 spin_lock(&pool->lock); 474 spin_lock(&pool->lock);
475 list_del_init(&zhdr->buddy); 475 list_del_init(&zhdr->buddy);
476 spin_unlock(&pool->lock); 476 spin_unlock(&pool->lock);
477 477
478 WARN_ON(z3fold_page_trylock(zhdr)); 478 WARN_ON(z3fold_page_trylock(zhdr));
479 __release_z3fold_page(zhdr, true); 479 __release_z3fold_page(zhdr, true);
480 } 480 }
481 481
482 static inline int put_z3fold_locked(struct z3f 482 static inline int put_z3fold_locked(struct z3fold_header *zhdr)
483 { 483 {
484 return kref_put(&zhdr->refcount, relea 484 return kref_put(&zhdr->refcount, release_z3fold_page_locked);
485 } 485 }
486 486
487 static inline int put_z3fold_locked_list(struc 487 static inline int put_z3fold_locked_list(struct z3fold_header *zhdr)
488 { 488 {
489 return kref_put(&zhdr->refcount, relea 489 return kref_put(&zhdr->refcount, release_z3fold_page_locked_list);
490 } 490 }
491 491
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;
545 int freechunks = num_free_chun 545 int freechunks = num_free_chunks(zhdr);
546 546
547 migrate_disable(); 547 migrate_disable();
548 unbuddied = this_cpu_ptr(pool- 548 unbuddied = this_cpu_ptr(pool->unbuddied);
549 spin_lock(&pool->lock); 549 spin_lock(&pool->lock);
550 list_add(&zhdr->buddy, &unbudd 550 list_add(&zhdr->buddy, &unbuddied[freechunks]);
551 spin_unlock(&pool->lock); 551 spin_unlock(&pool->lock);
552 zhdr->cpu = smp_processor_id() 552 zhdr->cpu = smp_processor_id();
553 migrate_enable(); 553 migrate_enable();
554 } 554 }
555 } 555 }
556 556
557 static inline enum buddy get_free_buddy(struct 557 static inline enum buddy get_free_buddy(struct z3fold_header *zhdr, int chunks)
558 { 558 {
559 enum buddy bud = HEADLESS; 559 enum buddy bud = HEADLESS;
560 560
561 if (zhdr->middle_chunks) { 561 if (zhdr->middle_chunks) {
562 if (!zhdr->first_chunks && 562 if (!zhdr->first_chunks &&
563 chunks <= zhdr->start_midd 563 chunks <= zhdr->start_middle - ZHDR_CHUNKS)
564 bud = FIRST; 564 bud = FIRST;
565 else if (!zhdr->last_chunks) 565 else if (!zhdr->last_chunks)
566 bud = LAST; 566 bud = LAST;
567 } else { 567 } else {
568 if (!zhdr->first_chunks) 568 if (!zhdr->first_chunks)
569 bud = FIRST; 569 bud = FIRST;
570 else if (!zhdr->last_chunks) 570 else if (!zhdr->last_chunks)
571 bud = LAST; 571 bud = LAST;
572 else 572 else
573 bud = MIDDLE; 573 bud = MIDDLE;
574 } 574 }
575 575
576 return bud; 576 return bud;
577 } 577 }
578 578
579 static inline void *mchunk_memmove(struct z3fo 579 static inline void *mchunk_memmove(struct z3fold_header *zhdr,
580 unsigned short 580 unsigned short dst_chunk)
581 { 581 {
582 void *beg = zhdr; 582 void *beg = zhdr;
583 return memmove(beg + (dst_chunk << CHU 583 return memmove(beg + (dst_chunk << CHUNK_SHIFT),
584 beg + (zhdr->start_midd 584 beg + (zhdr->start_middle << CHUNK_SHIFT),
585 zhdr->middle_chunks << 585 zhdr->middle_chunks << CHUNK_SHIFT);
586 } 586 }
587 587
588 static inline bool buddy_single(struct z3fold_ 588 static inline bool buddy_single(struct z3fold_header *zhdr)
589 { 589 {
590 return !((zhdr->first_chunks && zhdr-> 590 return !((zhdr->first_chunks && zhdr->middle_chunks) ||
591 (zhdr->first_chunks && 591 (zhdr->first_chunks && zhdr->last_chunks) ||
592 (zhdr->middle_chunks & 592 (zhdr->middle_chunks && zhdr->last_chunks));
593 } 593 }
594 594
595 static struct z3fold_header *compact_single_bu 595 static struct z3fold_header *compact_single_buddy(struct z3fold_header *zhdr)
596 { 596 {
597 struct z3fold_pool *pool = zhdr_to_poo 597 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
598 void *p = zhdr; 598 void *p = zhdr;
599 unsigned long old_handle = 0; 599 unsigned long old_handle = 0;
600 size_t sz = 0; 600 size_t sz = 0;
601 struct z3fold_header *new_zhdr = NULL; 601 struct z3fold_header *new_zhdr = NULL;
602 int first_idx = __idx(zhdr, FIRST); 602 int first_idx = __idx(zhdr, FIRST);
603 int middle_idx = __idx(zhdr, MIDDLE); 603 int middle_idx = __idx(zhdr, MIDDLE);
604 int last_idx = __idx(zhdr, LAST); 604 int last_idx = __idx(zhdr, LAST);
605 unsigned short *moved_chunks = NULL; 605 unsigned short *moved_chunks = NULL;
606 606
607 /* 607 /*
608 * No need to protect slots here -- al 608 * No need to protect slots here -- all the slots are "local" and
609 * the page lock is already taken 609 * the page lock is already taken
610 */ 610 */
611 if (zhdr->first_chunks && zhdr->slots- 611 if (zhdr->first_chunks && zhdr->slots->slot[first_idx]) {
612 p += ZHDR_SIZE_ALIGNED; 612 p += ZHDR_SIZE_ALIGNED;
613 sz = zhdr->first_chunks << CHU 613 sz = zhdr->first_chunks << CHUNK_SHIFT;
614 old_handle = (unsigned long)&z 614 old_handle = (unsigned long)&zhdr->slots->slot[first_idx];
615 moved_chunks = &zhdr->first_ch 615 moved_chunks = &zhdr->first_chunks;
616 } else if (zhdr->middle_chunks && zhdr 616 } else if (zhdr->middle_chunks && zhdr->slots->slot[middle_idx]) {
617 p += zhdr->start_middle << CHU 617 p += zhdr->start_middle << CHUNK_SHIFT;
618 sz = zhdr->middle_chunks << CH 618 sz = zhdr->middle_chunks << CHUNK_SHIFT;
619 old_handle = (unsigned long)&z 619 old_handle = (unsigned long)&zhdr->slots->slot[middle_idx];
620 moved_chunks = &zhdr->middle_c 620 moved_chunks = &zhdr->middle_chunks;
621 } else if (zhdr->last_chunks && zhdr-> 621 } else if (zhdr->last_chunks && zhdr->slots->slot[last_idx]) {
622 p += PAGE_SIZE - (zhdr->last_c 622 p += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
623 sz = zhdr->last_chunks << CHUN 623 sz = zhdr->last_chunks << CHUNK_SHIFT;
624 old_handle = (unsigned long)&z 624 old_handle = (unsigned long)&zhdr->slots->slot[last_idx];
625 moved_chunks = &zhdr->last_chu 625 moved_chunks = &zhdr->last_chunks;
626 } 626 }
627 627
628 if (sz > 0) { 628 if (sz > 0) {
629 enum buddy new_bud = HEADLESS; 629 enum buddy new_bud = HEADLESS;
630 short chunks = size_to_chunks( 630 short chunks = size_to_chunks(sz);
631 void *q; 631 void *q;
632 632
633 new_zhdr = __z3fold_alloc(pool 633 new_zhdr = __z3fold_alloc(pool, sz, false);
634 if (!new_zhdr) 634 if (!new_zhdr)
635 return NULL; 635 return NULL;
636 636
637 if (WARN_ON(new_zhdr == zhdr)) 637 if (WARN_ON(new_zhdr == zhdr))
638 goto out_fail; 638 goto out_fail;
639 639
640 new_bud = get_free_buddy(new_z 640 new_bud = get_free_buddy(new_zhdr, chunks);
641 q = new_zhdr; 641 q = new_zhdr;
642 switch (new_bud) { 642 switch (new_bud) {
643 case FIRST: 643 case FIRST:
644 new_zhdr->first_chunks 644 new_zhdr->first_chunks = chunks;
645 q += ZHDR_SIZE_ALIGNED 645 q += ZHDR_SIZE_ALIGNED;
646 break; 646 break;
647 case MIDDLE: 647 case MIDDLE:
648 new_zhdr->middle_chunk 648 new_zhdr->middle_chunks = chunks;
649 new_zhdr->start_middle 649 new_zhdr->start_middle =
650 new_zhdr->firs 650 new_zhdr->first_chunks + ZHDR_CHUNKS;
651 q += new_zhdr->start_m 651 q += new_zhdr->start_middle << CHUNK_SHIFT;
652 break; 652 break;
653 case LAST: 653 case LAST:
654 new_zhdr->last_chunks 654 new_zhdr->last_chunks = chunks;
655 q += PAGE_SIZE - (new_ 655 q += PAGE_SIZE - (new_zhdr->last_chunks << CHUNK_SHIFT);
656 break; 656 break;
657 default: 657 default:
658 goto out_fail; 658 goto out_fail;
659 } 659 }
660 new_zhdr->foreign_handles++; 660 new_zhdr->foreign_handles++;
661 memcpy(q, p, sz); 661 memcpy(q, p, sz);
662 write_lock(&zhdr->slots->lock) 662 write_lock(&zhdr->slots->lock);
663 *(unsigned long *)old_handle = 663 *(unsigned long *)old_handle = (unsigned long)new_zhdr +
664 __idx(new_zhdr, new_bu 664 __idx(new_zhdr, new_bud);
665 if (new_bud == LAST) 665 if (new_bud == LAST)
666 *(unsigned long *)old_ 666 *(unsigned long *)old_handle |=
667 (new_z 667 (new_zhdr->last_chunks << BUDDY_SHIFT);
668 write_unlock(&zhdr->slots->loc 668 write_unlock(&zhdr->slots->lock);
669 add_to_unbuddied(pool, new_zhd 669 add_to_unbuddied(pool, new_zhdr);
670 z3fold_page_unlock(new_zhdr); 670 z3fold_page_unlock(new_zhdr);
671 671
672 *moved_chunks = 0; 672 *moved_chunks = 0;
673 } 673 }
674 674
675 return new_zhdr; 675 return new_zhdr;
676 676
677 out_fail: 677 out_fail:
678 if (new_zhdr && !put_z3fold_locked(new 678 if (new_zhdr && !put_z3fold_locked(new_zhdr)) {
679 add_to_unbuddied(pool, new_zhd 679 add_to_unbuddied(pool, new_zhdr);
680 z3fold_page_unlock(new_zhdr); 680 z3fold_page_unlock(new_zhdr);
681 } 681 }
682 return NULL; 682 return NULL;
683 683
684 } 684 }
685 685
686 #define BIG_CHUNK_GAP 3 686 #define BIG_CHUNK_GAP 3
687 /* Has to be called with lock held */ 687 /* Has to be called with lock held */
688 static int z3fold_compact_page(struct z3fold_h 688 static int z3fold_compact_page(struct z3fold_header *zhdr)
689 { 689 {
690 struct page *page = virt_to_page(zhdr) 690 struct page *page = virt_to_page(zhdr);
691 691
692 if (test_bit(MIDDLE_CHUNK_MAPPED, &pag 692 if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private))
693 return 0; /* can't move middle 693 return 0; /* can't move middle chunk, it's used */
694 694
695 if (unlikely(PageIsolated(page))) 695 if (unlikely(PageIsolated(page)))
696 return 0; 696 return 0;
697 697
698 if (zhdr->middle_chunks == 0) 698 if (zhdr->middle_chunks == 0)
699 return 0; /* nothing to compac 699 return 0; /* nothing to compact */
700 700
701 if (zhdr->first_chunks == 0 && zhdr->l 701 if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
702 /* move to the beginning */ 702 /* move to the beginning */
703 mchunk_memmove(zhdr, ZHDR_CHUN 703 mchunk_memmove(zhdr, ZHDR_CHUNKS);
704 zhdr->first_chunks = zhdr->mid 704 zhdr->first_chunks = zhdr->middle_chunks;
705 zhdr->middle_chunks = 0; 705 zhdr->middle_chunks = 0;
706 zhdr->start_middle = 0; 706 zhdr->start_middle = 0;
707 zhdr->first_num++; 707 zhdr->first_num++;
708 return 1; 708 return 1;
709 } 709 }
710 710
711 /* 711 /*
712 * moving data is expensive, so let's 712 * moving data is expensive, so let's only do that if
713 * there's substantial gain (at least 713 * there's substantial gain (at least BIG_CHUNK_GAP chunks)
714 */ 714 */
715 if (zhdr->first_chunks != 0 && zhdr->l 715 if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 &&
716 zhdr->start_middle - (zhdr->first_ 716 zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >=
717 BIG_CHUNK_GAP) { 717 BIG_CHUNK_GAP) {
718 mchunk_memmove(zhdr, zhdr->fir 718 mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS);
719 zhdr->start_middle = zhdr->fir 719 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
720 return 1; 720 return 1;
721 } else if (zhdr->last_chunks != 0 && z 721 } else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 &&
722 TOTAL_CHUNKS - (zhdr->last_ 722 TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle
723 + zhdr 723 + zhdr->middle_chunks) >=
724 BIG_CHUNK_GAP) { 724 BIG_CHUNK_GAP) {
725 unsigned short new_start = TOT 725 unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks -
726 zhdr->middle_chunks; 726 zhdr->middle_chunks;
727 mchunk_memmove(zhdr, new_start 727 mchunk_memmove(zhdr, new_start);
728 zhdr->start_middle = new_start 728 zhdr->start_middle = new_start;
729 return 1; 729 return 1;
730 } 730 }
731 731
732 return 0; 732 return 0;
733 } 733 }
734 734
735 static void do_compact_page(struct z3fold_head 735 static void do_compact_page(struct z3fold_header *zhdr, bool locked)
736 { 736 {
737 struct z3fold_pool *pool = zhdr_to_poo 737 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
738 struct page *page; 738 struct page *page;
739 739
740 page = virt_to_page(zhdr); 740 page = virt_to_page(zhdr);
741 if (locked) 741 if (locked)
742 WARN_ON(z3fold_page_trylock(zh 742 WARN_ON(z3fold_page_trylock(zhdr));
743 else 743 else
744 z3fold_page_lock(zhdr); 744 z3fold_page_lock(zhdr);
745 if (WARN_ON(!test_and_clear_bit(NEEDS_ 745 if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) {
746 z3fold_page_unlock(zhdr); 746 z3fold_page_unlock(zhdr);
747 return; 747 return;
748 } 748 }
749 spin_lock(&pool->lock); 749 spin_lock(&pool->lock);
750 list_del_init(&zhdr->buddy); 750 list_del_init(&zhdr->buddy);
751 spin_unlock(&pool->lock); 751 spin_unlock(&pool->lock);
752 752
753 if (put_z3fold_locked(zhdr)) 753 if (put_z3fold_locked(zhdr))
754 return; 754 return;
755 755
756 if (test_bit(PAGE_STALE, &page->privat 756 if (test_bit(PAGE_STALE, &page->private) ||
757 test_and_set_bit(PAGE_CLAIMED, &pa 757 test_and_set_bit(PAGE_CLAIMED, &page->private)) {
758 z3fold_page_unlock(zhdr); 758 z3fold_page_unlock(zhdr);
759 return; 759 return;
760 } 760 }
761 761
762 if (!zhdr->foreign_handles && buddy_si 762 if (!zhdr->foreign_handles && buddy_single(zhdr) &&
763 zhdr->mapped_count == 0 && compact 763 zhdr->mapped_count == 0 && compact_single_buddy(zhdr)) {
764 if (!put_z3fold_locked(zhdr)) 764 if (!put_z3fold_locked(zhdr)) {
765 clear_bit(PAGE_CLAIMED 765 clear_bit(PAGE_CLAIMED, &page->private);
766 z3fold_page_unlock(zhd 766 z3fold_page_unlock(zhdr);
767 } 767 }
768 return; 768 return;
769 } 769 }
770 770
771 z3fold_compact_page(zhdr); 771 z3fold_compact_page(zhdr);
772 add_to_unbuddied(pool, zhdr); 772 add_to_unbuddied(pool, zhdr);
773 clear_bit(PAGE_CLAIMED, &page->private 773 clear_bit(PAGE_CLAIMED, &page->private);
774 z3fold_page_unlock(zhdr); 774 z3fold_page_unlock(zhdr);
775 } 775 }
776 776
777 static void compact_page_work(struct work_stru 777 static void compact_page_work(struct work_struct *w)
778 { 778 {
779 struct z3fold_header *zhdr = container 779 struct z3fold_header *zhdr = container_of(w, struct z3fold_header,
780 780 work);
781 781
782 do_compact_page(zhdr, false); 782 do_compact_page(zhdr, false);
783 } 783 }
784 784
785 /* returns _locked_ z3fold page header or NULL 785 /* returns _locked_ z3fold page header or NULL */
786 static inline struct z3fold_header *__z3fold_a 786 static inline struct z3fold_header *__z3fold_alloc(struct z3fold_pool *pool,
787 787 size_t size, bool can_sleep)
788 { 788 {
789 struct z3fold_header *zhdr = NULL; 789 struct z3fold_header *zhdr = NULL;
790 struct page *page; 790 struct page *page;
791 struct list_head *unbuddied; 791 struct list_head *unbuddied;
792 int chunks = size_to_chunks(size), i; 792 int chunks = size_to_chunks(size), i;
793 793
794 lookup: 794 lookup:
795 migrate_disable(); 795 migrate_disable();
796 /* First, try to find an unbuddied z3f 796 /* First, try to find an unbuddied z3fold page. */
797 unbuddied = this_cpu_ptr(pool->unbuddi 797 unbuddied = this_cpu_ptr(pool->unbuddied);
798 for_each_unbuddied_list(i, chunks) { 798 for_each_unbuddied_list(i, chunks) {
799 struct list_head *l = &unbuddi 799 struct list_head *l = &unbuddied[i];
800 800
801 zhdr = list_first_entry_or_nul 801 zhdr = list_first_entry_or_null(READ_ONCE(l),
802 struct 802 struct z3fold_header, buddy);
803 803
804 if (!zhdr) 804 if (!zhdr)
805 continue; 805 continue;
806 806
807 /* Re-check under lock. */ 807 /* Re-check under lock. */
808 spin_lock(&pool->lock); 808 spin_lock(&pool->lock);
809 if (unlikely(zhdr != list_firs 809 if (unlikely(zhdr != list_first_entry(READ_ONCE(l),
810 810 struct z3fold_header, buddy)) ||
811 !z3fold_page_trylock(zhdr) 811 !z3fold_page_trylock(zhdr)) {
812 spin_unlock(&pool->loc 812 spin_unlock(&pool->lock);
813 zhdr = NULL; 813 zhdr = NULL;
814 migrate_enable(); 814 migrate_enable();
815 if (can_sleep) 815 if (can_sleep)
816 cond_resched() 816 cond_resched();
817 goto lookup; 817 goto lookup;
818 } 818 }
819 list_del_init(&zhdr->buddy); 819 list_del_init(&zhdr->buddy);
820 zhdr->cpu = -1; 820 zhdr->cpu = -1;
821 spin_unlock(&pool->lock); 821 spin_unlock(&pool->lock);
822 822
823 page = virt_to_page(zhdr); 823 page = virt_to_page(zhdr);
824 if (test_bit(NEEDS_COMPACTING, 824 if (test_bit(NEEDS_COMPACTING, &page->private) ||
825 test_bit(PAGE_CLAIMED, &pa 825 test_bit(PAGE_CLAIMED, &page->private)) {
826 z3fold_page_unlock(zhd 826 z3fold_page_unlock(zhdr);
827 zhdr = NULL; 827 zhdr = NULL;
828 migrate_enable(); 828 migrate_enable();
829 if (can_sleep) 829 if (can_sleep)
830 cond_resched() 830 cond_resched();
831 goto lookup; 831 goto lookup;
832 } 832 }
833 833
834 /* 834 /*
835 * this page could not be remo 835 * this page could not be removed from its unbuddied
836 * list while pool lock was he 836 * list while pool lock was held, and then we've taken
837 * page lock so kref_put could 837 * page lock so kref_put could not be called before
838 * we got here, so it's safe t 838 * we got here, so it's safe to just call kref_get()
839 */ 839 */
840 kref_get(&zhdr->refcount); 840 kref_get(&zhdr->refcount);
841 break; 841 break;
842 } 842 }
843 migrate_enable(); 843 migrate_enable();
844 844
845 if (!zhdr) { 845 if (!zhdr) {
846 int cpu; 846 int cpu;
847 847
848 /* look for _exact_ match on o 848 /* look for _exact_ match on other cpus' lists */
849 for_each_online_cpu(cpu) { 849 for_each_online_cpu(cpu) {
850 struct list_head *l; 850 struct list_head *l;
851 851
852 unbuddied = per_cpu_pt 852 unbuddied = per_cpu_ptr(pool->unbuddied, cpu);
853 spin_lock(&pool->lock) 853 spin_lock(&pool->lock);
854 l = &unbuddied[chunks] 854 l = &unbuddied[chunks];
855 855
856 zhdr = list_first_entr 856 zhdr = list_first_entry_or_null(READ_ONCE(l),
857 857 struct z3fold_header, buddy);
858 858
859 if (!zhdr || !z3fold_p 859 if (!zhdr || !z3fold_page_trylock(zhdr)) {
860 spin_unlock(&p 860 spin_unlock(&pool->lock);
861 zhdr = NULL; 861 zhdr = NULL;
862 continue; 862 continue;
863 } 863 }
864 list_del_init(&zhdr->b 864 list_del_init(&zhdr->buddy);
865 zhdr->cpu = -1; 865 zhdr->cpu = -1;
866 spin_unlock(&pool->loc 866 spin_unlock(&pool->lock);
867 867
868 page = virt_to_page(zh 868 page = virt_to_page(zhdr);
869 if (test_bit(NEEDS_COM 869 if (test_bit(NEEDS_COMPACTING, &page->private) ||
870 test_bit(PAGE_CLAI 870 test_bit(PAGE_CLAIMED, &page->private)) {
871 z3fold_page_un 871 z3fold_page_unlock(zhdr);
872 zhdr = NULL; 872 zhdr = NULL;
873 if (can_sleep) 873 if (can_sleep)
874 cond_r 874 cond_resched();
875 continue; 875 continue;
876 } 876 }
877 kref_get(&zhdr->refcou 877 kref_get(&zhdr->refcount);
878 break; 878 break;
879 } 879 }
880 } 880 }
881 881
882 if (zhdr && !zhdr->slots) { 882 if (zhdr && !zhdr->slots) {
883 zhdr->slots = alloc_slots(pool 883 zhdr->slots = alloc_slots(pool, GFP_ATOMIC);
884 if (!zhdr->slots) 884 if (!zhdr->slots)
885 goto out_fail; 885 goto out_fail;
886 } 886 }
887 return zhdr; 887 return zhdr;
888 888
889 out_fail: 889 out_fail:
890 if (!put_z3fold_locked(zhdr)) { 890 if (!put_z3fold_locked(zhdr)) {
891 add_to_unbuddied(pool, zhdr); 891 add_to_unbuddied(pool, zhdr);
892 z3fold_page_unlock(zhdr); 892 z3fold_page_unlock(zhdr);
893 } 893 }
894 return NULL; 894 return NULL;
895 } 895 }
896 896
897 /* 897 /*
898 * API Functions 898 * API Functions
899 */ 899 */
900 900
901 /** 901 /**
902 * z3fold_create_pool() - create a new z3fold 902 * z3fold_create_pool() - create a new z3fold pool
903 * @name: pool name 903 * @name: pool name
904 * @gfp: gfp flags when allocating the 904 * @gfp: gfp flags when allocating the z3fold pool structure
905 * 905 *
906 * Return: pointer to the new z3fold pool or N 906 * Return: pointer to the new z3fold pool or NULL if the metadata allocation
907 * failed. 907 * failed.
908 */ 908 */
909 static struct z3fold_pool *z3fold_create_pool( 909 static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp)
910 { 910 {
911 struct z3fold_pool *pool = NULL; 911 struct z3fold_pool *pool = NULL;
912 int i, cpu; 912 int i, cpu;
913 913
914 pool = kzalloc(sizeof(struct z3fold_po 914 pool = kzalloc(sizeof(struct z3fold_pool), gfp);
915 if (!pool) 915 if (!pool)
916 goto out; 916 goto out;
917 pool->c_handle = kmem_cache_create("z3 917 pool->c_handle = kmem_cache_create("z3fold_handle",
918 sizeof(struct 918 sizeof(struct z3fold_buddy_slots),
919 SLOTS_ALIGN, 0 919 SLOTS_ALIGN, 0, NULL);
920 if (!pool->c_handle) 920 if (!pool->c_handle)
921 goto out_c; 921 goto out_c;
922 spin_lock_init(&pool->lock); 922 spin_lock_init(&pool->lock);
923 spin_lock_init(&pool->stale_lock); 923 spin_lock_init(&pool->stale_lock);
924 pool->unbuddied = __alloc_percpu(sizeo 924 pool->unbuddied = __alloc_percpu(sizeof(struct list_head) * NCHUNKS,
925 __ali 925 __alignof__(struct list_head));
926 if (!pool->unbuddied) 926 if (!pool->unbuddied)
927 goto out_pool; 927 goto out_pool;
928 for_each_possible_cpu(cpu) { 928 for_each_possible_cpu(cpu) {
929 struct list_head *unbuddied = 929 struct list_head *unbuddied =
930 per_cpu_ptr(po 930 per_cpu_ptr(pool->unbuddied, cpu);
931 for_each_unbuddied_list(i, 0) 931 for_each_unbuddied_list(i, 0)
932 INIT_LIST_HEAD(&unbudd 932 INIT_LIST_HEAD(&unbuddied[i]);
933 } 933 }
934 INIT_LIST_HEAD(&pool->stale); 934 INIT_LIST_HEAD(&pool->stale);
935 atomic64_set(&pool->pages_nr, 0); 935 atomic64_set(&pool->pages_nr, 0);
936 pool->name = name; 936 pool->name = name;
937 pool->compact_wq = create_singlethread 937 pool->compact_wq = create_singlethread_workqueue(pool->name);
938 if (!pool->compact_wq) 938 if (!pool->compact_wq)
939 goto out_unbuddied; 939 goto out_unbuddied;
940 pool->release_wq = create_singlethread 940 pool->release_wq = create_singlethread_workqueue(pool->name);
941 if (!pool->release_wq) 941 if (!pool->release_wq)
942 goto out_wq; 942 goto out_wq;
943 INIT_WORK(&pool->work, free_pages_work 943 INIT_WORK(&pool->work, free_pages_work);
944 return pool; 944 return pool;
945 945
946 out_wq: 946 out_wq:
947 destroy_workqueue(pool->compact_wq); 947 destroy_workqueue(pool->compact_wq);
948 out_unbuddied: 948 out_unbuddied:
949 free_percpu(pool->unbuddied); 949 free_percpu(pool->unbuddied);
950 out_pool: 950 out_pool:
951 kmem_cache_destroy(pool->c_handle); 951 kmem_cache_destroy(pool->c_handle);
952 out_c: 952 out_c:
953 kfree(pool); 953 kfree(pool);
954 out: 954 out:
955 return NULL; 955 return NULL;
956 } 956 }
957 957
958 /** 958 /**
959 * z3fold_destroy_pool() - destroys an existin 959 * z3fold_destroy_pool() - destroys an existing z3fold pool
960 * @pool: the z3fold pool to be destroye 960 * @pool: the z3fold pool to be destroyed
961 * 961 *
962 * The pool should be emptied before this func 962 * The pool should be emptied before this function is called.
963 */ 963 */
964 static void z3fold_destroy_pool(struct z3fold_ 964 static void z3fold_destroy_pool(struct z3fold_pool *pool)
965 { 965 {
966 kmem_cache_destroy(pool->c_handle); 966 kmem_cache_destroy(pool->c_handle);
967 967
968 /* 968 /*
969 * We need to destroy pool->compact_wq 969 * We need to destroy pool->compact_wq before pool->release_wq,
970 * as any pending work on pool->compac 970 * as any pending work on pool->compact_wq will call
971 * queue_work(pool->release_wq, &pool- 971 * queue_work(pool->release_wq, &pool->work).
972 * 972 *
973 * There are still outstanding pages u 973 * There are still outstanding pages until both workqueues are drained,
974 * so we cannot unregister migration u 974 * so we cannot unregister migration until then.
975 */ 975 */
976 976
977 destroy_workqueue(pool->compact_wq); 977 destroy_workqueue(pool->compact_wq);
978 destroy_workqueue(pool->release_wq); 978 destroy_workqueue(pool->release_wq);
979 free_percpu(pool->unbuddied); 979 free_percpu(pool->unbuddied);
980 kfree(pool); 980 kfree(pool);
981 } 981 }
982 982
983 static const struct movable_operations z3fold_ 983 static const struct movable_operations z3fold_mops;
984 984
985 /** 985 /**
986 * z3fold_alloc() - allocates a region of a gi 986 * z3fold_alloc() - allocates a region of a given size
987 * @pool: z3fold pool from which to allo 987 * @pool: z3fold pool from which to allocate
988 * @size: size in bytes of the desired a 988 * @size: size in bytes of the desired allocation
989 * @gfp: gfp flags used if the pool nee 989 * @gfp: gfp flags used if the pool needs to grow
990 * @handle: handle of the new allocation 990 * @handle: handle of the new allocation
991 * 991 *
992 * This function will attempt to find a free r 992 * This function will attempt to find a free region in the pool large enough to
993 * satisfy the allocation request. A search o 993 * satisfy the allocation request. A search of the unbuddied lists is
994 * performed first. If no suitable free region 994 * performed first. If no suitable free region is found, then a new page is
995 * allocated and added to the pool to satisfy 995 * allocated and added to the pool to satisfy the request.
996 * 996 *
997 * Return: 0 if success and handle is set, oth 997 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
998 * gfp arguments are invalid or -ENOMEM if the 998 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
999 * a new page. 999 * a new page.
1000 */ 1000 */
1001 static int z3fold_alloc(struct z3fold_pool *p 1001 static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
1002 unsigned long *handle 1002 unsigned long *handle)
1003 { 1003 {
1004 int chunks = size_to_chunks(size); 1004 int chunks = size_to_chunks(size);
1005 struct z3fold_header *zhdr = NULL; 1005 struct z3fold_header *zhdr = NULL;
1006 struct page *page = NULL; 1006 struct page *page = NULL;
1007 enum buddy bud; 1007 enum buddy bud;
1008 bool can_sleep = gfpflags_allow_block 1008 bool can_sleep = gfpflags_allow_blocking(gfp);
1009 1009
1010 if (!size || (gfp & __GFP_HIGHMEM)) 1010 if (!size || (gfp & __GFP_HIGHMEM))
1011 return -EINVAL; 1011 return -EINVAL;
1012 1012
1013 if (size > PAGE_SIZE) 1013 if (size > PAGE_SIZE)
1014 return -ENOSPC; 1014 return -ENOSPC;
1015 1015
1016 if (size > PAGE_SIZE - ZHDR_SIZE_ALIG 1016 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
1017 bud = HEADLESS; 1017 bud = HEADLESS;
1018 else { 1018 else {
1019 retry: 1019 retry:
1020 zhdr = __z3fold_alloc(pool, s 1020 zhdr = __z3fold_alloc(pool, size, can_sleep);
1021 if (zhdr) { 1021 if (zhdr) {
1022 bud = get_free_buddy( 1022 bud = get_free_buddy(zhdr, chunks);
1023 if (bud == HEADLESS) 1023 if (bud == HEADLESS) {
1024 if (!put_z3fo 1024 if (!put_z3fold_locked(zhdr))
1025 z3fol 1025 z3fold_page_unlock(zhdr);
1026 pr_err("No fr 1026 pr_err("No free chunks in unbuddied\n");
1027 WARN_ON(1); 1027 WARN_ON(1);
1028 goto retry; 1028 goto retry;
1029 } 1029 }
1030 page = virt_to_page(z 1030 page = virt_to_page(zhdr);
1031 goto found; 1031 goto found;
1032 } 1032 }
1033 bud = FIRST; 1033 bud = FIRST;
1034 } 1034 }
1035 1035
1036 page = alloc_page(gfp); 1036 page = alloc_page(gfp);
1037 if (!page) 1037 if (!page)
1038 return -ENOMEM; 1038 return -ENOMEM;
1039 1039
1040 zhdr = init_z3fold_page(page, bud == 1040 zhdr = init_z3fold_page(page, bud == HEADLESS, pool, gfp);
1041 if (!zhdr) { 1041 if (!zhdr) {
1042 __free_page(page); 1042 __free_page(page);
1043 return -ENOMEM; 1043 return -ENOMEM;
1044 } 1044 }
1045 atomic64_inc(&pool->pages_nr); 1045 atomic64_inc(&pool->pages_nr);
1046 1046
1047 if (bud == HEADLESS) { 1047 if (bud == HEADLESS) {
1048 set_bit(PAGE_HEADLESS, &page- 1048 set_bit(PAGE_HEADLESS, &page->private);
1049 goto headless; 1049 goto headless;
1050 } 1050 }
1051 if (can_sleep) { 1051 if (can_sleep) {
1052 lock_page(page); 1052 lock_page(page);
1053 __SetPageMovable(page, &z3fol 1053 __SetPageMovable(page, &z3fold_mops);
1054 unlock_page(page); 1054 unlock_page(page);
1055 } else { 1055 } else {
1056 WARN_ON(!trylock_page(page)); 1056 WARN_ON(!trylock_page(page));
1057 __SetPageMovable(page, &z3fol 1057 __SetPageMovable(page, &z3fold_mops);
1058 unlock_page(page); 1058 unlock_page(page);
1059 } 1059 }
1060 z3fold_page_lock(zhdr); 1060 z3fold_page_lock(zhdr);
1061 1061
1062 found: 1062 found:
1063 if (bud == FIRST) 1063 if (bud == FIRST)
1064 zhdr->first_chunks = chunks; 1064 zhdr->first_chunks = chunks;
1065 else if (bud == LAST) 1065 else if (bud == LAST)
1066 zhdr->last_chunks = chunks; 1066 zhdr->last_chunks = chunks;
1067 else { 1067 else {
1068 zhdr->middle_chunks = chunks; 1068 zhdr->middle_chunks = chunks;
1069 zhdr->start_middle = zhdr->fi 1069 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
1070 } 1070 }
1071 add_to_unbuddied(pool, zhdr); 1071 add_to_unbuddied(pool, zhdr);
1072 1072
1073 headless: 1073 headless:
1074 spin_lock(&pool->lock); 1074 spin_lock(&pool->lock);
1075 *handle = encode_handle(zhdr, bud); 1075 *handle = encode_handle(zhdr, bud);
1076 spin_unlock(&pool->lock); 1076 spin_unlock(&pool->lock);
1077 if (bud != HEADLESS) 1077 if (bud != HEADLESS)
1078 z3fold_page_unlock(zhdr); 1078 z3fold_page_unlock(zhdr);
1079 1079
1080 return 0; 1080 return 0;
1081 } 1081 }
1082 1082
1083 /** 1083 /**
1084 * z3fold_free() - frees the allocation assoc 1084 * z3fold_free() - frees the allocation associated with the given handle
1085 * @pool: pool in which the allocation 1085 * @pool: pool in which the allocation resided
1086 * @handle: handle associated with the al 1086 * @handle: handle associated with the allocation returned by z3fold_alloc()
1087 * 1087 *
1088 * In the case that the z3fold page in which 1088 * In the case that the z3fold page in which the allocation resides is under
1089 * reclaim, as indicated by the PAGE_CLAIMED 1089 * reclaim, as indicated by the PAGE_CLAIMED flag being set, this function
1090 * only sets the first|middle|last_chunks to 1090 * only sets the first|middle|last_chunks to 0. The page is actually freed
1091 * once all buddies are evicted (see z3fold_r 1091 * once all buddies are evicted (see z3fold_reclaim_page() below).
1092 */ 1092 */
1093 static void z3fold_free(struct z3fold_pool *p 1093 static void z3fold_free(struct z3fold_pool *pool, unsigned long handle)
1094 { 1094 {
1095 struct z3fold_header *zhdr; 1095 struct z3fold_header *zhdr;
1096 struct page *page; 1096 struct page *page;
1097 enum buddy bud; 1097 enum buddy bud;
1098 bool page_claimed; 1098 bool page_claimed;
1099 1099
1100 zhdr = get_z3fold_header(handle); 1100 zhdr = get_z3fold_header(handle);
1101 page = virt_to_page(zhdr); 1101 page = virt_to_page(zhdr);
1102 page_claimed = test_and_set_bit(PAGE_ 1102 page_claimed = test_and_set_bit(PAGE_CLAIMED, &page->private);
1103 1103
1104 if (test_bit(PAGE_HEADLESS, &page->pr 1104 if (test_bit(PAGE_HEADLESS, &page->private)) {
1105 /* if a headless page is unde 1105 /* if a headless page is under reclaim, just leave.
1106 * NB: we use test_and_set_bi 1106 * NB: we use test_and_set_bit for a reason: if the bit
1107 * has not been set before, w 1107 * has not been set before, we release this page
1108 * immediately so we don't ca 1108 * immediately so we don't care about its value any more.
1109 */ 1109 */
1110 if (!page_claimed) { 1110 if (!page_claimed) {
1111 put_z3fold_header(zhd 1111 put_z3fold_header(zhdr);
1112 free_z3fold_page(page 1112 free_z3fold_page(page, true);
1113 atomic64_dec(&pool->p 1113 atomic64_dec(&pool->pages_nr);
1114 } 1114 }
1115 return; 1115 return;
1116 } 1116 }
1117 1117
1118 /* Non-headless case */ 1118 /* Non-headless case */
1119 bud = handle_to_buddy(handle); 1119 bud = handle_to_buddy(handle);
1120 1120
1121 switch (bud) { 1121 switch (bud) {
1122 case FIRST: 1122 case FIRST:
1123 zhdr->first_chunks = 0; 1123 zhdr->first_chunks = 0;
1124 break; 1124 break;
1125 case MIDDLE: 1125 case MIDDLE:
1126 zhdr->middle_chunks = 0; 1126 zhdr->middle_chunks = 0;
1127 break; 1127 break;
1128 case LAST: 1128 case LAST:
1129 zhdr->last_chunks = 0; 1129 zhdr->last_chunks = 0;
1130 break; 1130 break;
1131 default: 1131 default:
1132 pr_err("%s: unknown bud %d\n" 1132 pr_err("%s: unknown bud %d\n", __func__, bud);
1133 WARN_ON(1); 1133 WARN_ON(1);
1134 put_z3fold_header(zhdr); 1134 put_z3fold_header(zhdr);
1135 return; 1135 return;
1136 } 1136 }
1137 1137
1138 if (!page_claimed) 1138 if (!page_claimed)
1139 free_handle(handle, zhdr); 1139 free_handle(handle, zhdr);
1140 if (put_z3fold_locked_list(zhdr)) 1140 if (put_z3fold_locked_list(zhdr))
1141 return; 1141 return;
1142 if (page_claimed) { 1142 if (page_claimed) {
1143 /* the page has not been clai 1143 /* the page has not been claimed by us */
1144 put_z3fold_header(zhdr); 1144 put_z3fold_header(zhdr);
1145 return; 1145 return;
1146 } 1146 }
1147 if (test_and_set_bit(NEEDS_COMPACTING 1147 if (test_and_set_bit(NEEDS_COMPACTING, &page->private)) {
1148 clear_bit(PAGE_CLAIMED, &page 1148 clear_bit(PAGE_CLAIMED, &page->private);
1149 put_z3fold_header(zhdr); 1149 put_z3fold_header(zhdr);
1150 return; 1150 return;
1151 } 1151 }
1152 if (zhdr->cpu < 0 || !cpu_online(zhdr 1152 if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) {
1153 zhdr->cpu = -1; 1153 zhdr->cpu = -1;
1154 kref_get(&zhdr->refcount); 1154 kref_get(&zhdr->refcount);
1155 clear_bit(PAGE_CLAIMED, &page 1155 clear_bit(PAGE_CLAIMED, &page->private);
1156 do_compact_page(zhdr, true); 1156 do_compact_page(zhdr, true);
1157 return; 1157 return;
1158 } 1158 }
1159 kref_get(&zhdr->refcount); 1159 kref_get(&zhdr->refcount);
1160 clear_bit(PAGE_CLAIMED, &page->privat 1160 clear_bit(PAGE_CLAIMED, &page->private);
1161 queue_work_on(zhdr->cpu, pool->compac 1161 queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work);
1162 put_z3fold_header(zhdr); 1162 put_z3fold_header(zhdr);
1163 } 1163 }
1164 1164
1165 /** 1165 /**
1166 * z3fold_map() - maps the allocation associa 1166 * z3fold_map() - maps the allocation associated with the given handle
1167 * @pool: pool in which the allocation 1167 * @pool: pool in which the allocation resides
1168 * @handle: handle associated with the al 1168 * @handle: handle associated with the allocation to be mapped
1169 * 1169 *
1170 * Extracts the buddy number from handle and 1170 * Extracts the buddy number from handle and constructs the pointer to the
1171 * correct starting chunk within the page. 1171 * correct starting chunk within the page.
1172 * 1172 *
1173 * Returns: a pointer to the mapped allocatio 1173 * Returns: a pointer to the mapped allocation
1174 */ 1174 */
1175 static void *z3fold_map(struct z3fold_pool *p 1175 static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle)
1176 { 1176 {
1177 struct z3fold_header *zhdr; 1177 struct z3fold_header *zhdr;
1178 struct page *page; 1178 struct page *page;
1179 void *addr; 1179 void *addr;
1180 enum buddy buddy; 1180 enum buddy buddy;
1181 1181
1182 zhdr = get_z3fold_header(handle); 1182 zhdr = get_z3fold_header(handle);
1183 addr = zhdr; 1183 addr = zhdr;
1184 page = virt_to_page(zhdr); 1184 page = virt_to_page(zhdr);
1185 1185
1186 if (test_bit(PAGE_HEADLESS, &page->pr 1186 if (test_bit(PAGE_HEADLESS, &page->private))
1187 goto out; 1187 goto out;
1188 1188
1189 buddy = handle_to_buddy(handle); 1189 buddy = handle_to_buddy(handle);
1190 switch (buddy) { 1190 switch (buddy) {
1191 case FIRST: 1191 case FIRST:
1192 addr += ZHDR_SIZE_ALIGNED; 1192 addr += ZHDR_SIZE_ALIGNED;
1193 break; 1193 break;
1194 case MIDDLE: 1194 case MIDDLE:
1195 addr += zhdr->start_middle << 1195 addr += zhdr->start_middle << CHUNK_SHIFT;
1196 set_bit(MIDDLE_CHUNK_MAPPED, 1196 set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1197 break; 1197 break;
1198 case LAST: 1198 case LAST:
1199 addr += PAGE_SIZE - (handle_t 1199 addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT);
1200 break; 1200 break;
1201 default: 1201 default:
1202 pr_err("unknown buddy id %d\n 1202 pr_err("unknown buddy id %d\n", buddy);
1203 WARN_ON(1); 1203 WARN_ON(1);
1204 addr = NULL; 1204 addr = NULL;
1205 break; 1205 break;
1206 } 1206 }
1207 1207
1208 if (addr) 1208 if (addr)
1209 zhdr->mapped_count++; 1209 zhdr->mapped_count++;
1210 out: 1210 out:
1211 put_z3fold_header(zhdr); 1211 put_z3fold_header(zhdr);
1212 return addr; 1212 return addr;
1213 } 1213 }
1214 1214
1215 /** 1215 /**
1216 * z3fold_unmap() - unmaps the allocation ass 1216 * z3fold_unmap() - unmaps the allocation associated with the given handle
1217 * @pool: pool in which the allocation 1217 * @pool: pool in which the allocation resides
1218 * @handle: handle associated with the al 1218 * @handle: handle associated with the allocation to be unmapped
1219 */ 1219 */
1220 static void z3fold_unmap(struct z3fold_pool * 1220 static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle)
1221 { 1221 {
1222 struct z3fold_header *zhdr; 1222 struct z3fold_header *zhdr;
1223 struct page *page; 1223 struct page *page;
1224 enum buddy buddy; 1224 enum buddy buddy;
1225 1225
1226 zhdr = get_z3fold_header(handle); 1226 zhdr = get_z3fold_header(handle);
1227 page = virt_to_page(zhdr); 1227 page = virt_to_page(zhdr);
1228 1228
1229 if (test_bit(PAGE_HEADLESS, &page->pr 1229 if (test_bit(PAGE_HEADLESS, &page->private))
1230 return; 1230 return;
1231 1231
1232 buddy = handle_to_buddy(handle); 1232 buddy = handle_to_buddy(handle);
1233 if (buddy == MIDDLE) 1233 if (buddy == MIDDLE)
1234 clear_bit(MIDDLE_CHUNK_MAPPED 1234 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1235 zhdr->mapped_count--; 1235 zhdr->mapped_count--;
1236 put_z3fold_header(zhdr); 1236 put_z3fold_header(zhdr);
1237 } 1237 }
1238 1238
1239 /** 1239 /**
1240 * z3fold_get_pool_pages() - gets the z3fold 1240 * z3fold_get_pool_pages() - gets the z3fold pool size in pages
1241 * @pool: pool whose size is being quer 1241 * @pool: pool whose size is being queried
1242 * 1242 *
1243 * Returns: size in pages of the given pool. 1243 * Returns: size in pages of the given pool.
1244 */ 1244 */
1245 static u64 z3fold_get_pool_pages(struct z3fol 1245 static u64 z3fold_get_pool_pages(struct z3fold_pool *pool)
1246 { 1246 {
1247 return atomic64_read(&pool->pages_nr) 1247 return atomic64_read(&pool->pages_nr);
1248 } 1248 }
1249 1249
1250 static bool z3fold_page_isolate(struct page * 1250 static bool z3fold_page_isolate(struct page *page, isolate_mode_t mode)
1251 { 1251 {
1252 struct z3fold_header *zhdr; 1252 struct z3fold_header *zhdr;
1253 struct z3fold_pool *pool; 1253 struct z3fold_pool *pool;
1254 1254
1255 VM_BUG_ON_PAGE(PageIsolated(page), pa 1255 VM_BUG_ON_PAGE(PageIsolated(page), page);
1256 1256
1257 if (test_bit(PAGE_HEADLESS, &page->pr 1257 if (test_bit(PAGE_HEADLESS, &page->private))
1258 return false; 1258 return false;
1259 1259
1260 zhdr = page_address(page); 1260 zhdr = page_address(page);
1261 z3fold_page_lock(zhdr); 1261 z3fold_page_lock(zhdr);
1262 if (test_bit(NEEDS_COMPACTING, &page- 1262 if (test_bit(NEEDS_COMPACTING, &page->private) ||
1263 test_bit(PAGE_STALE, &page->priva 1263 test_bit(PAGE_STALE, &page->private))
1264 goto out; 1264 goto out;
1265 1265
1266 if (zhdr->mapped_count != 0 || zhdr-> 1266 if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0)
1267 goto out; 1267 goto out;
1268 1268
1269 if (test_and_set_bit(PAGE_CLAIMED, &p 1269 if (test_and_set_bit(PAGE_CLAIMED, &page->private))
1270 goto out; 1270 goto out;
1271 pool = zhdr_to_pool(zhdr); 1271 pool = zhdr_to_pool(zhdr);
1272 spin_lock(&pool->lock); 1272 spin_lock(&pool->lock);
1273 if (!list_empty(&zhdr->buddy)) 1273 if (!list_empty(&zhdr->buddy))
1274 list_del_init(&zhdr->buddy); 1274 list_del_init(&zhdr->buddy);
1275 spin_unlock(&pool->lock); 1275 spin_unlock(&pool->lock);
1276 1276
1277 kref_get(&zhdr->refcount); 1277 kref_get(&zhdr->refcount);
1278 z3fold_page_unlock(zhdr); 1278 z3fold_page_unlock(zhdr);
1279 return true; 1279 return true;
1280 1280
1281 out: 1281 out:
1282 z3fold_page_unlock(zhdr); 1282 z3fold_page_unlock(zhdr);
1283 return false; 1283 return false;
1284 } 1284 }
1285 1285
1286 static int z3fold_page_migrate(struct page *n 1286 static int z3fold_page_migrate(struct page *newpage, struct page *page,
1287 enum migrate_mode mode) 1287 enum migrate_mode mode)
1288 { 1288 {
1289 struct z3fold_header *zhdr, *new_zhdr 1289 struct z3fold_header *zhdr, *new_zhdr;
1290 struct z3fold_pool *pool; 1290 struct z3fold_pool *pool;
1291 1291
1292 VM_BUG_ON_PAGE(!PageIsolated(page), p 1292 VM_BUG_ON_PAGE(!PageIsolated(page), page);
1293 VM_BUG_ON_PAGE(!test_bit(PAGE_CLAIMED 1293 VM_BUG_ON_PAGE(!test_bit(PAGE_CLAIMED, &page->private), page);
1294 VM_BUG_ON_PAGE(!PageLocked(newpage), 1294 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
1295 1295
1296 zhdr = page_address(page); 1296 zhdr = page_address(page);
1297 pool = zhdr_to_pool(zhdr); 1297 pool = zhdr_to_pool(zhdr);
1298 1298
1299 if (!z3fold_page_trylock(zhdr)) 1299 if (!z3fold_page_trylock(zhdr))
1300 return -EAGAIN; 1300 return -EAGAIN;
1301 if (zhdr->mapped_count != 0 || zhdr-> 1301 if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0) {
1302 clear_bit(PAGE_CLAIMED, &page 1302 clear_bit(PAGE_CLAIMED, &page->private);
1303 z3fold_page_unlock(zhdr); 1303 z3fold_page_unlock(zhdr);
1304 return -EBUSY; 1304 return -EBUSY;
1305 } 1305 }
1306 if (work_pending(&zhdr->work)) { 1306 if (work_pending(&zhdr->work)) {
1307 z3fold_page_unlock(zhdr); 1307 z3fold_page_unlock(zhdr);
1308 return -EAGAIN; 1308 return -EAGAIN;
1309 } 1309 }
1310 new_zhdr = page_address(newpage); 1310 new_zhdr = page_address(newpage);
1311 memcpy(new_zhdr, zhdr, PAGE_SIZE); 1311 memcpy(new_zhdr, zhdr, PAGE_SIZE);
1312 newpage->private = page->private; 1312 newpage->private = page->private;
1313 set_bit(PAGE_MIGRATED, &page->private 1313 set_bit(PAGE_MIGRATED, &page->private);
1314 z3fold_page_unlock(zhdr); 1314 z3fold_page_unlock(zhdr);
1315 spin_lock_init(&new_zhdr->page_lock); 1315 spin_lock_init(&new_zhdr->page_lock);
1316 INIT_WORK(&new_zhdr->work, compact_pa 1316 INIT_WORK(&new_zhdr->work, compact_page_work);
1317 /* 1317 /*
1318 * z3fold_page_isolate() ensures that 1318 * z3fold_page_isolate() ensures that new_zhdr->buddy is empty,
1319 * so we only have to reinitialize it 1319 * so we only have to reinitialize it.
1320 */ 1320 */
1321 INIT_LIST_HEAD(&new_zhdr->buddy); 1321 INIT_LIST_HEAD(&new_zhdr->buddy);
1322 __ClearPageMovable(page); 1322 __ClearPageMovable(page);
1323 1323
1324 get_page(newpage); 1324 get_page(newpage);
1325 z3fold_page_lock(new_zhdr); 1325 z3fold_page_lock(new_zhdr);
1326 if (new_zhdr->first_chunks) 1326 if (new_zhdr->first_chunks)
1327 encode_handle(new_zhdr, FIRST 1327 encode_handle(new_zhdr, FIRST);
1328 if (new_zhdr->last_chunks) 1328 if (new_zhdr->last_chunks)
1329 encode_handle(new_zhdr, LAST) 1329 encode_handle(new_zhdr, LAST);
1330 if (new_zhdr->middle_chunks) 1330 if (new_zhdr->middle_chunks)
1331 encode_handle(new_zhdr, MIDDL 1331 encode_handle(new_zhdr, MIDDLE);
1332 set_bit(NEEDS_COMPACTING, &newpage->p 1332 set_bit(NEEDS_COMPACTING, &newpage->private);
1333 new_zhdr->cpu = smp_processor_id(); 1333 new_zhdr->cpu = smp_processor_id();
1334 __SetPageMovable(newpage, &z3fold_mop 1334 __SetPageMovable(newpage, &z3fold_mops);
1335 z3fold_page_unlock(new_zhdr); 1335 z3fold_page_unlock(new_zhdr);
1336 1336
1337 queue_work_on(new_zhdr->cpu, pool->co 1337 queue_work_on(new_zhdr->cpu, pool->compact_wq, &new_zhdr->work);
1338 1338
1339 /* PAGE_CLAIMED and PAGE_MIGRATED are 1339 /* PAGE_CLAIMED and PAGE_MIGRATED are cleared now. */
1340 page->private = 0; 1340 page->private = 0;
1341 put_page(page); 1341 put_page(page);
1342 return 0; 1342 return 0;
1343 } 1343 }
1344 1344
1345 static void z3fold_page_putback(struct page * 1345 static void z3fold_page_putback(struct page *page)
1346 { 1346 {
1347 struct z3fold_header *zhdr; 1347 struct z3fold_header *zhdr;
1348 struct z3fold_pool *pool; 1348 struct z3fold_pool *pool;
1349 1349
1350 zhdr = page_address(page); 1350 zhdr = page_address(page);
1351 pool = zhdr_to_pool(zhdr); 1351 pool = zhdr_to_pool(zhdr);
1352 1352
1353 z3fold_page_lock(zhdr); 1353 z3fold_page_lock(zhdr);
1354 if (!list_empty(&zhdr->buddy)) 1354 if (!list_empty(&zhdr->buddy))
1355 list_del_init(&zhdr->buddy); 1355 list_del_init(&zhdr->buddy);
1356 INIT_LIST_HEAD(&page->lru); 1356 INIT_LIST_HEAD(&page->lru);
1357 if (put_z3fold_locked(zhdr)) 1357 if (put_z3fold_locked(zhdr))
1358 return; 1358 return;
1359 if (list_empty(&zhdr->buddy)) 1359 if (list_empty(&zhdr->buddy))
1360 add_to_unbuddied(pool, zhdr); 1360 add_to_unbuddied(pool, zhdr);
1361 clear_bit(PAGE_CLAIMED, &page->privat 1361 clear_bit(PAGE_CLAIMED, &page->private);
1362 z3fold_page_unlock(zhdr); 1362 z3fold_page_unlock(zhdr);
1363 } 1363 }
1364 1364
1365 static const struct movable_operations z3fold 1365 static const struct movable_operations z3fold_mops = {
1366 .isolate_page = z3fold_page_isolate, 1366 .isolate_page = z3fold_page_isolate,
1367 .migrate_page = z3fold_page_migrate, 1367 .migrate_page = z3fold_page_migrate,
1368 .putback_page = z3fold_page_putback, 1368 .putback_page = z3fold_page_putback,
1369 }; 1369 };
1370 1370
1371 /***************** 1371 /*****************
1372 * zpool 1372 * zpool
1373 ****************/ 1373 ****************/
1374 1374
1375 static void *z3fold_zpool_create(const char * 1375 static void *z3fold_zpool_create(const char *name, gfp_t gfp)
1376 { 1376 {
1377 return z3fold_create_pool(name, gfp); 1377 return z3fold_create_pool(name, gfp);
1378 } 1378 }
1379 1379
1380 static void z3fold_zpool_destroy(void *pool) 1380 static void z3fold_zpool_destroy(void *pool)
1381 { 1381 {
1382 z3fold_destroy_pool(pool); 1382 z3fold_destroy_pool(pool);
1383 } 1383 }
1384 1384
1385 static int z3fold_zpool_malloc(void *pool, si 1385 static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp,
1386 unsigned long *handle 1386 unsigned long *handle)
1387 { 1387 {
1388 return z3fold_alloc(pool, size, gfp, 1388 return z3fold_alloc(pool, size, gfp, handle);
1389 } 1389 }
1390 static void z3fold_zpool_free(void *pool, uns 1390 static void z3fold_zpool_free(void *pool, unsigned long handle)
1391 { 1391 {
1392 z3fold_free(pool, handle); 1392 z3fold_free(pool, handle);
1393 } 1393 }
1394 1394
1395 static void *z3fold_zpool_map(void *pool, uns 1395 static void *z3fold_zpool_map(void *pool, unsigned long handle,
1396 enum zpool_mapmode mm 1396 enum zpool_mapmode mm)
1397 { 1397 {
1398 return z3fold_map(pool, handle); 1398 return z3fold_map(pool, handle);
1399 } 1399 }
1400 static void z3fold_zpool_unmap(void *pool, un 1400 static void z3fold_zpool_unmap(void *pool, unsigned long handle)
1401 { 1401 {
1402 z3fold_unmap(pool, handle); 1402 z3fold_unmap(pool, handle);
1403 } 1403 }
1404 1404
1405 static u64 z3fold_zpool_total_pages(void *poo 1405 static u64 z3fold_zpool_total_pages(void *pool)
1406 { 1406 {
1407 return z3fold_get_pool_pages(pool); 1407 return z3fold_get_pool_pages(pool);
1408 } 1408 }
1409 1409
1410 static struct zpool_driver z3fold_zpool_drive 1410 static struct zpool_driver z3fold_zpool_driver = {
1411 .type = "z3fold", 1411 .type = "z3fold",
1412 .sleep_mapped = true, 1412 .sleep_mapped = true,
1413 .owner = THIS_MODULE, 1413 .owner = THIS_MODULE,
1414 .create = z3fold_zpool_create, 1414 .create = z3fold_zpool_create,
1415 .destroy = z3fold_zpool_destroy, 1415 .destroy = z3fold_zpool_destroy,
1416 .malloc = z3fold_zpool_malloc, 1416 .malloc = z3fold_zpool_malloc,
1417 .free = z3fold_zpool_free, 1417 .free = z3fold_zpool_free,
1418 .map = z3fold_zpool_map, 1418 .map = z3fold_zpool_map,
1419 .unmap = z3fold_zpool_unmap, 1419 .unmap = z3fold_zpool_unmap,
1420 .total_pages = z3fold_zpool_total_pa 1420 .total_pages = z3fold_zpool_total_pages,
1421 }; 1421 };
1422 1422
1423 MODULE_ALIAS("zpool-z3fold"); 1423 MODULE_ALIAS("zpool-z3fold");
1424 1424
1425 static int __init init_z3fold(void) 1425 static int __init init_z3fold(void)
1426 { 1426 {
1427 /* 1427 /*
1428 * Make sure the z3fold header is not 1428 * Make sure the z3fold header is not larger than the page size and
1429 * there has remaining spaces for its 1429 * there has remaining spaces for its buddy.
1430 */ 1430 */
1431 BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE 1431 BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE - CHUNK_SIZE);
1432 zpool_register_driver(&z3fold_zpool_d 1432 zpool_register_driver(&z3fold_zpool_driver);
1433 1433
1434 return 0; 1434 return 0;
1435 } 1435 }
1436 1436
1437 static void __exit exit_z3fold(void) 1437 static void __exit exit_z3fold(void)
1438 { 1438 {
1439 zpool_unregister_driver(&z3fold_zpool 1439 zpool_unregister_driver(&z3fold_zpool_driver);
1440 } 1440 }
1441 1441
1442 module_init(init_z3fold); 1442 module_init(init_z3fold);
1443 module_exit(exit_z3fold); 1443 module_exit(exit_z3fold);
1444 1444
1445 MODULE_LICENSE("GPL"); 1445 MODULE_LICENSE("GPL");
1446 MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail. 1446 MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>");
1447 MODULE_DESCRIPTION("3-Fold Allocator for Comp 1447 MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages");
1448 1448
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.