Commit | Line | Data |
---|---|---|
61989a80 NG |
1 | /* |
2 | * zsmalloc memory allocator | |
3 | * | |
4 | * Copyright (C) 2011 Nitin Gupta | |
31fc00bb | 5 | * Copyright (C) 2012, 2013 Minchan Kim |
61989a80 NG |
6 | * |
7 | * This code is released using a dual license strategy: BSD/GPL | |
8 | * You can choose the license that better fits your requirements. | |
9 | * | |
10 | * Released under the terms of 3-clause BSD License | |
11 | * Released under the terms of GNU General Public License Version 2.0 | |
12 | */ | |
13 | ||
2db51dae | 14 | /* |
2db51dae NG |
15 | * Following is how we use various fields and flags of underlying |
16 | * struct page(s) to form a zspage. | |
17 | * | |
18 | * Usage of struct page fields: | |
1cac41cb MB |
19 | * page->private: points to zspage |
20 | * page->freelist(index): links together all component pages of a zspage | |
21 | * For the huge page, this is always 0, so we use this field | |
22 | * to store handle. | |
2db51dae NG |
23 | * |
24 | * Usage of struct page flags: | |
25 | * PG_private: identifies the first component page | |
26 | * PG_private2: identifies the last component page | |
1cac41cb | 27 | * PG_owner_priv_1: indentifies the huge component page |
2db51dae NG |
28 | * |
29 | */ | |
30 | ||
61989a80 NG |
31 | #include <linux/module.h> |
32 | #include <linux/kernel.h> | |
312fcae2 | 33 | #include <linux/sched.h> |
61989a80 NG |
34 | #include <linux/bitops.h> |
35 | #include <linux/errno.h> | |
36 | #include <linux/highmem.h> | |
61989a80 NG |
37 | #include <linux/string.h> |
38 | #include <linux/slab.h> | |
39 | #include <asm/tlbflush.h> | |
40 | #include <asm/pgtable.h> | |
41 | #include <linux/cpumask.h> | |
42 | #include <linux/cpu.h> | |
0cbb613f | 43 | #include <linux/vmalloc.h> |
759b26b2 | 44 | #include <linux/preempt.h> |
0959c63f SJ |
45 | #include <linux/spinlock.h> |
46 | #include <linux/types.h> | |
0f050d99 | 47 | #include <linux/debugfs.h> |
bcf1647d | 48 | #include <linux/zsmalloc.h> |
c795779d | 49 | #include <linux/zpool.h> |
1cac41cb MB |
50 | #include <linux/mount.h> |
51 | #include <linux/compaction.h> | |
52 | #include <linux/pagemap.h> | |
53 | #include <linux/swap.h> | |
54 | #include <linux/jiffies.h> | |
55 | ||
56 | #define ZSPAGE_MAGIC 0x58 | |
0959c63f SJ |
57 | |
58 | /* | |
59 | * This must be power of 2 and greater than of equal to sizeof(link_free). | |
60 | * These two conditions ensure that any 'struct link_free' itself doesn't | |
61 | * span more than 1 page which avoids complex case of mapping 2 pages simply | |
62 | * to restore link_free pointer values. | |
63 | */ | |
64 | #define ZS_ALIGN 8 | |
65 | ||
66 | /* | |
67 | * A single 'zspage' is composed of up to 2^N discontiguous 0-order (single) | |
68 | * pages. ZS_MAX_ZSPAGE_ORDER defines upper limit on N. | |
69 | */ | |
70 | #define ZS_MAX_ZSPAGE_ORDER 2 | |
71 | #define ZS_MAX_PAGES_PER_ZSPAGE (_AC(1, UL) << ZS_MAX_ZSPAGE_ORDER) | |
72 | ||
2e40e163 MK |
73 | #define ZS_HANDLE_SIZE (sizeof(unsigned long)) |
74 | ||
0959c63f SJ |
75 | /* |
76 | * Object location (<PFN>, <obj_idx>) is encoded as | |
c3e3e88a | 77 | * as single (unsigned long) handle value. |
0959c63f | 78 | * |
1cac41cb | 79 | * Note that object index <obj_idx> starts from 0. |
0959c63f SJ |
80 | * |
81 | * This is made more complicated by various memory models and PAE. | |
82 | */ | |
83 | ||
84 | #ifndef MAX_PHYSMEM_BITS | |
85 | #ifdef CONFIG_HIGHMEM64G | |
86 | #define MAX_PHYSMEM_BITS 36 | |
87 | #else /* !CONFIG_HIGHMEM64G */ | |
88 | /* | |
89 | * If this definition of MAX_PHYSMEM_BITS is used, OBJ_INDEX_BITS will just | |
90 | * be PAGE_SHIFT | |
91 | */ | |
92 | #define MAX_PHYSMEM_BITS BITS_PER_LONG | |
93 | #endif | |
94 | #endif | |
95 | #define _PFN_BITS (MAX_PHYSMEM_BITS - PAGE_SHIFT) | |
312fcae2 MK |
96 | |
97 | /* | |
98 | * Memory for allocating for handle keeps object position by | |
99 | * encoding <page, obj_idx> and the encoded value has a room | |
100 | * in least bit(ie, look at obj_to_location). | |
101 | * We use the bit to synchronize between object access by | |
102 | * user and migration. | |
103 | */ | |
104 | #define HANDLE_PIN_BIT 0 | |
105 | ||
106 | /* | |
107 | * Head in allocated object should have OBJ_ALLOCATED_TAG | |
108 | * to identify the object was allocated or not. | |
109 | * It's okay to add the status bit in the least bit because | |
110 | * header keeps handle which is 4byte-aligned address so we | |
111 | * have room for two bit at least. | |
112 | */ | |
113 | #define OBJ_ALLOCATED_TAG 1 | |
114 | #define OBJ_TAG_BITS 1 | |
115 | #define OBJ_INDEX_BITS (BITS_PER_LONG - _PFN_BITS - OBJ_TAG_BITS) | |
0959c63f SJ |
116 | #define OBJ_INDEX_MASK ((_AC(1, UL) << OBJ_INDEX_BITS) - 1) |
117 | ||
118 | #define MAX(a, b) ((a) >= (b) ? (a) : (b)) | |
119 | /* ZS_MIN_ALLOC_SIZE must be multiple of ZS_ALIGN */ | |
120 | #define ZS_MIN_ALLOC_SIZE \ | |
121 | MAX(32, (ZS_MAX_PAGES_PER_ZSPAGE << PAGE_SHIFT >> OBJ_INDEX_BITS)) | |
2e40e163 | 122 | /* each chunk includes extra space to keep handle */ |
7b60a685 | 123 | #define ZS_MAX_ALLOC_SIZE PAGE_SIZE |
0959c63f SJ |
124 | |
125 | /* | |
7eb52512 | 126 | * On systems with 4K page size, this gives 255 size classes! There is a |
0959c63f SJ |
127 | * trader-off here: |
128 | * - Large number of size classes is potentially wasteful as free page are | |
129 | * spread across these classes | |
130 | * - Small number of size classes causes large internal fragmentation | |
131 | * - Probably its better to use specific size classes (empirically | |
132 | * determined). NOTE: all those class sizes must be set as multiple of | |
133 | * ZS_ALIGN to make sure link_free itself never has to span 2 pages. | |
134 | * | |
135 | * ZS_MIN_ALLOC_SIZE and ZS_SIZE_CLASS_DELTA must be multiple of ZS_ALIGN | |
136 | * (reason above) | |
137 | */ | |
1cac41cb | 138 | #define ZS_SIZE_CLASS_DELTA (PAGE_SIZE >> CLASS_BITS) |
0959c63f SJ |
139 | |
140 | /* | |
141 | * We do not maintain any list for completely empty or full pages | |
142 | */ | |
143 | enum fullness_group { | |
0959c63f | 144 | ZS_EMPTY, |
1cac41cb MB |
145 | ZS_ALMOST_EMPTY, |
146 | ZS_ALMOST_FULL, | |
147 | ZS_FULL, | |
148 | NR_ZS_FULLNESS, | |
0959c63f SJ |
149 | }; |
150 | ||
0f050d99 | 151 | enum zs_stat_type { |
1cac41cb MB |
152 | CLASS_EMPTY, |
153 | CLASS_ALMOST_EMPTY, | |
154 | CLASS_ALMOST_FULL, | |
155 | CLASS_FULL, | |
0f050d99 GM |
156 | OBJ_ALLOCATED, |
157 | OBJ_USED, | |
1cac41cb | 158 | NR_ZS_STAT_TYPE, |
0f050d99 GM |
159 | }; |
160 | ||
0f050d99 GM |
161 | struct zs_size_stat { |
162 | unsigned long objs[NR_ZS_STAT_TYPE]; | |
163 | }; | |
164 | ||
57244594 SS |
165 | #ifdef CONFIG_ZSMALLOC_STAT |
166 | static struct dentry *zs_stat_root; | |
0f050d99 GM |
167 | #endif |
168 | ||
1cac41cb MB |
169 | #ifdef CONFIG_COMPACTION |
170 | static struct vfsmount *zsmalloc_mnt; | |
171 | #endif | |
172 | ||
173 | #ifdef CONFIG_ZSWAP_MIGRATION_SUPPORT | |
174 | static int zs_page_migration_enabled = 1; | |
175 | #else | |
176 | static int zs_page_migration_enabled; | |
177 | #endif | |
178 | ||
40f9fb8c MG |
179 | /* |
180 | * number of size_classes | |
181 | */ | |
182 | static int zs_size_classes; | |
183 | ||
0959c63f SJ |
184 | /* |
185 | * We assign a page to ZS_ALMOST_EMPTY fullness group when: | |
186 | * n <= N / f, where | |
187 | * n = number of allocated objects | |
188 | * N = total number of objects zspage can store | |
6dd9737e | 189 | * f = fullness_threshold_frac |
0959c63f SJ |
190 | * |
191 | * Similarly, we assign zspage to: | |
192 | * ZS_ALMOST_FULL when n > N / f | |
193 | * ZS_EMPTY when n == 0 | |
194 | * ZS_FULL when n == N | |
195 | * | |
196 | * (see: fix_fullness_group()) | |
197 | */ | |
198 | static const int fullness_threshold_frac = 4; | |
2a89387e | 199 | static size_t huge_class_size; |
0959c63f SJ |
200 | |
201 | struct size_class { | |
57244594 | 202 | spinlock_t lock; |
1cac41cb | 203 | struct list_head fullness_list[NR_ZS_FULLNESS]; |
0959c63f SJ |
204 | /* |
205 | * Size of objects stored in this class. Must be multiple | |
206 | * of ZS_ALIGN. | |
207 | */ | |
208 | int size; | |
1cac41cb | 209 | int objs_per_zspage; |
0959c63f SJ |
210 | /* Number of PAGE_SIZE sized pages to combine to form a 'zspage' */ |
211 | int pages_per_zspage; | |
0959c63f | 212 | |
1cac41cb MB |
213 | unsigned int index; |
214 | struct zs_size_stat stats; | |
0959c63f SJ |
215 | }; |
216 | ||
1cac41cb MB |
217 | /* huge object: pages_per_zspage == 1 && maxobj_per_zspage == 1 */ |
218 | static void SetPageHugeObject(struct page *page) | |
219 | { | |
220 | SetPageOwnerPriv1(page); | |
221 | } | |
222 | ||
223 | static void ClearPageHugeObject(struct page *page) | |
224 | { | |
225 | ClearPageOwnerPriv1(page); | |
226 | } | |
227 | ||
228 | static int PageHugeObject(struct page *page) | |
229 | { | |
230 | return PageOwnerPriv1(page); | |
231 | } | |
232 | ||
0959c63f SJ |
233 | /* |
234 | * Placed within free objects to form a singly linked list. | |
1cac41cb | 235 | * For every zspage, zspage->freeobj gives head of this list. |
0959c63f SJ |
236 | * |
237 | * This must be power of 2 and less than or equal to ZS_ALIGN | |
238 | */ | |
239 | struct link_free { | |
2e40e163 MK |
240 | union { |
241 | /* | |
1cac41cb | 242 | * Free object index; |
2e40e163 MK |
243 | * It's valid for non-allocated object |
244 | */ | |
1cac41cb | 245 | unsigned long next; |
2e40e163 MK |
246 | /* |
247 | * Handle of allocated object. | |
248 | */ | |
249 | unsigned long handle; | |
250 | }; | |
0959c63f SJ |
251 | }; |
252 | ||
253 | struct zs_pool { | |
6f3526d6 | 254 | const char *name; |
0f050d99 | 255 | |
40f9fb8c | 256 | struct size_class **size_class; |
2e40e163 | 257 | struct kmem_cache *handle_cachep; |
1cac41cb | 258 | struct kmem_cache *zspage_cachep; |
0959c63f | 259 | |
13de8933 | 260 | atomic_long_t pages_allocated; |
0f050d99 | 261 | |
7d3f3938 | 262 | struct zs_pool_stats stats; |
ab9d306d SS |
263 | |
264 | /* Compact classes */ | |
265 | struct shrinker shrinker; | |
266 | /* | |
267 | * To signify that register_shrinker() was successful | |
268 | * and unregister_shrinker() will not Oops. | |
269 | */ | |
270 | bool shrinker_enabled; | |
0f050d99 GM |
271 | #ifdef CONFIG_ZSMALLOC_STAT |
272 | struct dentry *stat_dentry; | |
273 | #endif | |
1cac41cb MB |
274 | #ifdef CONFIG_COMPACTION |
275 | struct inode *inode; | |
276 | struct work_struct free_work; | |
277 | #endif | |
0959c63f | 278 | }; |
61989a80 NG |
279 | |
280 | /* | |
281 | * A zspage's class index and fullness group | |
282 | * are encoded in its (first)page->mapping | |
283 | */ | |
1cac41cb MB |
284 | #define FULLNESS_BITS 2 |
285 | #define CLASS_BITS 8 | |
286 | #define ISOLATED_BITS 3 | |
287 | #define MAGIC_VAL_BITS 8 | |
288 | ||
289 | struct zspage { | |
290 | struct { | |
291 | unsigned int fullness:FULLNESS_BITS; | |
292 | unsigned int class:CLASS_BITS; | |
293 | unsigned int isolated:ISOLATED_BITS; | |
294 | unsigned int magic:MAGIC_VAL_BITS; | |
295 | }; | |
296 | unsigned int inuse; | |
297 | unsigned int freeobj; | |
298 | struct page *first_page; | |
299 | struct list_head list; /* fullness list */ | |
300 | #ifdef CONFIG_COMPACTION | |
301 | rwlock_t lock; | |
302 | #endif | |
303 | }; | |
61989a80 | 304 | |
f553646a | 305 | struct mapping_area { |
1b945aee | 306 | #ifdef CONFIG_PGTABLE_MAPPING |
f553646a SJ |
307 | struct vm_struct *vm; /* vm area for mapping object that span pages */ |
308 | #else | |
309 | char *vm_buf; /* copy buffer for objects that span pages */ | |
310 | #endif | |
311 | char *vm_addr; /* address of kmap_atomic()'ed pages */ | |
312 | enum zs_mapmode vm_mm; /* mapping mode */ | |
313 | }; | |
314 | ||
1cac41cb MB |
315 | #ifdef CONFIG_COMPACTION |
316 | static int zs_register_migration(struct zs_pool *pool); | |
317 | static void zs_unregister_migration(struct zs_pool *pool); | |
318 | static void migrate_lock_init(struct zspage *zspage); | |
319 | static void migrate_read_lock(struct zspage *zspage); | |
320 | static void migrate_read_unlock(struct zspage *zspage); | |
321 | static void kick_deferred_free(struct zs_pool *pool); | |
322 | static void init_deferred_free(struct zs_pool *pool); | |
323 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage); | |
324 | #else | |
325 | static int zsmalloc_mount(void) { return 0; } | |
326 | static void zsmalloc_unmount(void) {} | |
327 | static int zs_register_migration(struct zs_pool *pool) { return 0; } | |
328 | static void zs_unregister_migration(struct zs_pool *pool) {} | |
329 | static void migrate_lock_init(struct zspage *zspage) {} | |
330 | static void migrate_read_lock(struct zspage *zspage) {} | |
331 | static void migrate_read_unlock(struct zspage *zspage) {} | |
332 | static void kick_deferred_free(struct zs_pool *pool) {} | |
333 | static void init_deferred_free(struct zs_pool *pool) {} | |
334 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage) {} | |
335 | #endif | |
336 | ||
337 | static int create_cache(struct zs_pool *pool) | |
2e40e163 MK |
338 | { |
339 | pool->handle_cachep = kmem_cache_create("zs_handle", ZS_HANDLE_SIZE, | |
340 | 0, 0, NULL); | |
1cac41cb MB |
341 | if (!pool->handle_cachep) |
342 | return 1; | |
343 | ||
344 | pool->zspage_cachep = kmem_cache_create("zspage", sizeof(struct zspage), | |
345 | 0, 0, NULL); | |
346 | if (!pool->zspage_cachep) { | |
347 | kmem_cache_destroy(pool->handle_cachep); | |
348 | pool->handle_cachep = NULL; | |
349 | return 1; | |
350 | } | |
351 | ||
352 | return 0; | |
2e40e163 MK |
353 | } |
354 | ||
1cac41cb | 355 | static void destroy_cache(struct zs_pool *pool) |
2e40e163 | 356 | { |
1cac41cb | 357 | kmem_cache_destroy(pool->zspage_cachep); |
2e40e163 MK |
358 | } |
359 | ||
1e92b056 | 360 | static unsigned long cache_alloc_handle(struct zs_pool *pool, gfp_t gfp) |
2e40e163 MK |
361 | { |
362 | return (unsigned long)kmem_cache_alloc(pool->handle_cachep, | |
1e92b056 | 363 | gfp & ~(__GFP_HIGHMEM|__GFP_MOVABLE)); |
2e40e163 MK |
364 | } |
365 | ||
1cac41cb | 366 | static void cache_free_handle(struct zs_pool *pool, unsigned long handle) |
2e40e163 MK |
367 | { |
368 | kmem_cache_free(pool->handle_cachep, (void *)handle); | |
369 | } | |
370 | ||
1cac41cb MB |
371 | static struct zspage *cache_alloc_zspage(struct zs_pool *pool, gfp_t flags) |
372 | { | |
373 | return kmem_cache_alloc(pool->zspage_cachep, | |
374 | flags & ~(__GFP_HIGHMEM|__GFP_MOVABLE)); | |
375 | }; | |
376 | ||
377 | static void cache_free_zspage(struct zs_pool *pool, struct zspage *zspage) | |
378 | { | |
379 | kmem_cache_free(pool->zspage_cachep, zspage); | |
380 | } | |
381 | ||
2e40e163 MK |
382 | static void record_obj(unsigned long handle, unsigned long obj) |
383 | { | |
bddaf791 JL |
384 | /* |
385 | * lsb of @obj represents handle lock while other bits | |
386 | * represent object value the handle is pointing so | |
387 | * updating shouldn't do store tearing. | |
388 | */ | |
389 | WRITE_ONCE(*(unsigned long *)handle, obj); | |
2e40e163 MK |
390 | } |
391 | ||
c795779d DS |
392 | /* zpool driver */ |
393 | ||
394 | #ifdef CONFIG_ZPOOL | |
395 | ||
6f3526d6 | 396 | static void *zs_zpool_create(const char *name, gfp_t gfp, |
78672779 | 397 | const struct zpool_ops *zpool_ops, |
479305fd | 398 | struct zpool *zpool) |
c795779d | 399 | { |
1e92b056 SS |
400 | /* |
401 | * Ignore global gfp flags: zs_malloc() may be invoked from | |
402 | * different contexts and its caller must provide a valid | |
403 | * gfp mask. | |
404 | */ | |
405 | return zs_create_pool(name); | |
c795779d DS |
406 | } |
407 | ||
408 | static void zs_zpool_destroy(void *pool) | |
409 | { | |
410 | zs_destroy_pool(pool); | |
411 | } | |
412 | ||
413 | static int zs_zpool_malloc(void *pool, size_t size, gfp_t gfp, | |
414 | unsigned long *handle) | |
415 | { | |
1e92b056 | 416 | *handle = zs_malloc(pool, size, gfp); |
c795779d DS |
417 | return *handle ? 0 : -1; |
418 | } | |
419 | static void zs_zpool_free(void *pool, unsigned long handle) | |
420 | { | |
421 | zs_free(pool, handle); | |
422 | } | |
423 | ||
424 | static int zs_zpool_shrink(void *pool, unsigned int pages, | |
425 | unsigned int *reclaimed) | |
426 | { | |
427 | return -EINVAL; | |
428 | } | |
429 | ||
430 | static void *zs_zpool_map(void *pool, unsigned long handle, | |
431 | enum zpool_mapmode mm) | |
432 | { | |
433 | enum zs_mapmode zs_mm; | |
434 | ||
435 | switch (mm) { | |
436 | case ZPOOL_MM_RO: | |
437 | zs_mm = ZS_MM_RO; | |
438 | break; | |
1cac41cb MB |
439 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
440 | case ZPOOL_MM_RO_NOWAIT: | |
441 | zs_mm = ZS_MM_RO_NOWAIT; | |
442 | break; | |
443 | #endif | |
c795779d DS |
444 | case ZPOOL_MM_WO: |
445 | zs_mm = ZS_MM_WO; | |
446 | break; | |
447 | case ZPOOL_MM_RW: /* fallthru */ | |
448 | default: | |
449 | zs_mm = ZS_MM_RW; | |
450 | break; | |
451 | } | |
452 | ||
453 | return zs_map_object(pool, handle, zs_mm); | |
454 | } | |
455 | static void zs_zpool_unmap(void *pool, unsigned long handle) | |
456 | { | |
457 | zs_unmap_object(pool, handle); | |
458 | } | |
459 | ||
460 | static u64 zs_zpool_total_size(void *pool) | |
461 | { | |
722cdc17 | 462 | return zs_get_total_pages(pool) << PAGE_SHIFT; |
c795779d DS |
463 | } |
464 | ||
465 | static struct zpool_driver zs_zpool_driver = { | |
466 | .type = "zsmalloc", | |
467 | .owner = THIS_MODULE, | |
468 | .create = zs_zpool_create, | |
469 | .destroy = zs_zpool_destroy, | |
470 | .malloc = zs_zpool_malloc, | |
471 | .free = zs_zpool_free, | |
472 | .shrink = zs_zpool_shrink, | |
473 | .map = zs_zpool_map, | |
474 | .unmap = zs_zpool_unmap, | |
475 | .total_size = zs_zpool_total_size, | |
476 | }; | |
477 | ||
137f8cff | 478 | MODULE_ALIAS("zpool-zsmalloc"); |
c795779d DS |
479 | #endif /* CONFIG_ZPOOL */ |
480 | ||
248ca1b0 MK |
481 | static unsigned int get_maxobj_per_zspage(int size, int pages_per_zspage) |
482 | { | |
483 | return pages_per_zspage * PAGE_SIZE / size; | |
484 | } | |
485 | ||
61989a80 NG |
486 | /* per-cpu VM mapping areas for zspage accesses that cross page boundaries */ |
487 | static DEFINE_PER_CPU(struct mapping_area, zs_map_area); | |
488 | ||
1cac41cb MB |
489 | static bool is_zspage_isolated(struct zspage *zspage) |
490 | { | |
491 | return zspage->isolated; | |
492 | } | |
493 | ||
7b516409 | 494 | static __maybe_unused int is_first_page(struct page *page) |
61989a80 | 495 | { |
a27545bf | 496 | return PagePrivate(page); |
61989a80 NG |
497 | } |
498 | ||
1cac41cb MB |
499 | /* Protected by class->lock */ |
500 | static inline int get_zspage_inuse(struct zspage *zspage) | |
501 | { | |
502 | return zspage->inuse; | |
503 | } | |
504 | ||
505 | static inline void set_zspage_inuse(struct zspage *zspage, int val) | |
506 | { | |
507 | zspage->inuse = val; | |
508 | } | |
509 | ||
510 | static inline void mod_zspage_inuse(struct zspage *zspage, int val) | |
511 | { | |
512 | zspage->inuse += val; | |
513 | } | |
514 | ||
515 | static inline struct page *get_first_page(struct zspage *zspage) | |
516 | { | |
517 | struct page *first_page = zspage->first_page; | |
518 | ||
519 | VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); | |
520 | return first_page; | |
521 | } | |
522 | ||
523 | static inline int get_first_obj_offset(struct page *page) | |
524 | { | |
525 | return page->units; | |
526 | } | |
527 | ||
528 | static inline void set_first_obj_offset(struct page *page, int offset) | |
529 | { | |
530 | page->units = offset; | |
531 | } | |
532 | ||
533 | static inline unsigned int get_freeobj(struct zspage *zspage) | |
61989a80 | 534 | { |
1cac41cb | 535 | return zspage->freeobj; |
61989a80 NG |
536 | } |
537 | ||
1cac41cb MB |
538 | static inline void set_freeobj(struct zspage *zspage, unsigned int obj) |
539 | { | |
540 | zspage->freeobj = obj; | |
541 | } | |
542 | ||
543 | static void get_zspage_mapping(struct zspage *zspage, | |
544 | unsigned int *class_idx, | |
61989a80 NG |
545 | enum fullness_group *fullness) |
546 | { | |
1cac41cb | 547 | VM_BUG_ON(zspage->magic != ZSPAGE_MAGIC); |
61989a80 | 548 | |
1cac41cb MB |
549 | *fullness = zspage->fullness; |
550 | *class_idx = zspage->class; | |
61989a80 NG |
551 | } |
552 | ||
1cac41cb MB |
553 | static void set_zspage_mapping(struct zspage *zspage, |
554 | unsigned int class_idx, | |
61989a80 NG |
555 | enum fullness_group fullness) |
556 | { | |
1cac41cb MB |
557 | zspage->class = class_idx; |
558 | zspage->fullness = fullness; | |
61989a80 NG |
559 | } |
560 | ||
c3e3e88a NC |
561 | /* |
562 | * zsmalloc divides the pool into various size classes where each | |
563 | * class maintains a list of zspages where each zspage is divided | |
564 | * into equal sized chunks. Each allocation falls into one of these | |
565 | * classes depending on its size. This function returns index of the | |
566 | * size class which has chunk size big enough to hold the give size. | |
567 | */ | |
61989a80 NG |
568 | static int get_size_class_index(int size) |
569 | { | |
570 | int idx = 0; | |
571 | ||
572 | if (likely(size > ZS_MIN_ALLOC_SIZE)) | |
573 | idx = DIV_ROUND_UP(size - ZS_MIN_ALLOC_SIZE, | |
574 | ZS_SIZE_CLASS_DELTA); | |
575 | ||
7b60a685 | 576 | return min(zs_size_classes - 1, idx); |
61989a80 NG |
577 | } |
578 | ||
dd4bae3c | 579 | /* type can be of enum type zs_stat_type or fullness_group */ |
248ca1b0 | 580 | static inline void zs_stat_inc(struct size_class *class, |
dd4bae3c | 581 | int type, unsigned long cnt) |
248ca1b0 | 582 | { |
1cac41cb | 583 | class->stats.objs[type] += cnt; |
248ca1b0 MK |
584 | } |
585 | ||
dd4bae3c | 586 | /* type can be of enum type zs_stat_type or fullness_group */ |
248ca1b0 | 587 | static inline void zs_stat_dec(struct size_class *class, |
dd4bae3c | 588 | int type, unsigned long cnt) |
248ca1b0 | 589 | { |
1cac41cb | 590 | class->stats.objs[type] -= cnt; |
248ca1b0 MK |
591 | } |
592 | ||
dd4bae3c | 593 | /* type can be of enum type zs_stat_type or fullness_group */ |
248ca1b0 | 594 | static inline unsigned long zs_stat_get(struct size_class *class, |
dd4bae3c | 595 | int type) |
248ca1b0 | 596 | { |
1cac41cb | 597 | return class->stats.objs[type]; |
248ca1b0 MK |
598 | } |
599 | ||
57244594 SS |
600 | #ifdef CONFIG_ZSMALLOC_STAT |
601 | ||
248ca1b0 MK |
602 | static int __init zs_stat_init(void) |
603 | { | |
604 | if (!debugfs_initialized()) | |
605 | return -ENODEV; | |
606 | ||
607 | zs_stat_root = debugfs_create_dir("zsmalloc", NULL); | |
608 | if (!zs_stat_root) | |
609 | return -ENOMEM; | |
610 | ||
611 | return 0; | |
612 | } | |
613 | ||
614 | static void __exit zs_stat_exit(void) | |
615 | { | |
616 | debugfs_remove_recursive(zs_stat_root); | |
617 | } | |
618 | ||
1cac41cb MB |
619 | static unsigned long zs_can_compact(struct size_class *class); |
620 | ||
248ca1b0 MK |
621 | static int zs_stats_size_show(struct seq_file *s, void *v) |
622 | { | |
623 | int i; | |
624 | struct zs_pool *pool = s->private; | |
625 | struct size_class *class; | |
626 | int objs_per_zspage; | |
627 | unsigned long class_almost_full, class_almost_empty; | |
1cac41cb | 628 | unsigned long obj_allocated, obj_used, pages_used, freeable; |
248ca1b0 MK |
629 | unsigned long total_class_almost_full = 0, total_class_almost_empty = 0; |
630 | unsigned long total_objs = 0, total_used_objs = 0, total_pages = 0; | |
1cac41cb | 631 | unsigned long total_freeable = 0; |
248ca1b0 | 632 | |
1cac41cb | 633 | seq_printf(s, " %5s %5s %11s %12s %13s %10s %10s %16s %8s\n", |
248ca1b0 MK |
634 | "class", "size", "almost_full", "almost_empty", |
635 | "obj_allocated", "obj_used", "pages_used", | |
1cac41cb | 636 | "pages_per_zspage", "freeable"); |
248ca1b0 MK |
637 | |
638 | for (i = 0; i < zs_size_classes; i++) { | |
639 | class = pool->size_class[i]; | |
640 | ||
641 | if (class->index != i) | |
642 | continue; | |
643 | ||
644 | spin_lock(&class->lock); | |
645 | class_almost_full = zs_stat_get(class, CLASS_ALMOST_FULL); | |
646 | class_almost_empty = zs_stat_get(class, CLASS_ALMOST_EMPTY); | |
647 | obj_allocated = zs_stat_get(class, OBJ_ALLOCATED); | |
648 | obj_used = zs_stat_get(class, OBJ_USED); | |
1cac41cb | 649 | freeable = zs_can_compact(class); |
248ca1b0 MK |
650 | spin_unlock(&class->lock); |
651 | ||
652 | objs_per_zspage = get_maxobj_per_zspage(class->size, | |
653 | class->pages_per_zspage); | |
654 | pages_used = obj_allocated / objs_per_zspage * | |
655 | class->pages_per_zspage; | |
656 | ||
1cac41cb MB |
657 | seq_printf(s, " %5u %5u %11lu %12lu %13lu" |
658 | " %10lu %10lu %16d %8lu\n", | |
248ca1b0 MK |
659 | i, class->size, class_almost_full, class_almost_empty, |
660 | obj_allocated, obj_used, pages_used, | |
1cac41cb | 661 | class->pages_per_zspage, freeable); |
248ca1b0 MK |
662 | |
663 | total_class_almost_full += class_almost_full; | |
664 | total_class_almost_empty += class_almost_empty; | |
665 | total_objs += obj_allocated; | |
666 | total_used_objs += obj_used; | |
667 | total_pages += pages_used; | |
1cac41cb | 668 | total_freeable += freeable; |
248ca1b0 MK |
669 | } |
670 | ||
671 | seq_puts(s, "\n"); | |
1cac41cb | 672 | seq_printf(s, " %5s %5s %11lu %12lu %13lu %10lu %10lu %16s %8lu\n", |
248ca1b0 MK |
673 | "Total", "", total_class_almost_full, |
674 | total_class_almost_empty, total_objs, | |
1cac41cb | 675 | total_used_objs, total_pages, "", total_freeable); |
248ca1b0 MK |
676 | |
677 | return 0; | |
678 | } | |
679 | ||
680 | static int zs_stats_size_open(struct inode *inode, struct file *file) | |
681 | { | |
682 | return single_open(file, zs_stats_size_show, inode->i_private); | |
683 | } | |
684 | ||
685 | static const struct file_operations zs_stat_size_ops = { | |
686 | .open = zs_stats_size_open, | |
687 | .read = seq_read, | |
688 | .llseek = seq_lseek, | |
689 | .release = single_release, | |
690 | }; | |
691 | ||
1cac41cb | 692 | static int zs_pool_stat_create(struct zs_pool *pool, const char *name) |
248ca1b0 MK |
693 | { |
694 | struct dentry *entry; | |
695 | ||
696 | if (!zs_stat_root) | |
697 | return -ENODEV; | |
698 | ||
699 | entry = debugfs_create_dir(name, zs_stat_root); | |
700 | if (!entry) { | |
701 | pr_warn("debugfs dir <%s> creation failed\n", name); | |
702 | return -ENOMEM; | |
703 | } | |
704 | pool->stat_dentry = entry; | |
705 | ||
706 | entry = debugfs_create_file("classes", S_IFREG | S_IRUGO, | |
707 | pool->stat_dentry, pool, &zs_stat_size_ops); | |
708 | if (!entry) { | |
709 | pr_warn("%s: debugfs file entry <%s> creation failed\n", | |
710 | name, "classes"); | |
711 | return -ENOMEM; | |
712 | } | |
713 | ||
714 | return 0; | |
715 | } | |
716 | ||
717 | static void zs_pool_stat_destroy(struct zs_pool *pool) | |
718 | { | |
719 | debugfs_remove_recursive(pool->stat_dentry); | |
720 | } | |
721 | ||
722 | #else /* CONFIG_ZSMALLOC_STAT */ | |
248ca1b0 MK |
723 | static int __init zs_stat_init(void) |
724 | { | |
725 | return 0; | |
726 | } | |
727 | ||
728 | static void __exit zs_stat_exit(void) | |
729 | { | |
730 | } | |
731 | ||
1cac41cb | 732 | static inline int zs_pool_stat_create(struct zs_pool *pool, const char *name) |
248ca1b0 MK |
733 | { |
734 | return 0; | |
735 | } | |
736 | ||
737 | static inline void zs_pool_stat_destroy(struct zs_pool *pool) | |
738 | { | |
739 | } | |
248ca1b0 MK |
740 | #endif |
741 | ||
742 | ||
1cac41cb | 743 | |
c3e3e88a NC |
744 | /* |
745 | * For each size class, zspages are divided into different groups | |
746 | * depending on how "full" they are. This was done so that we could | |
747 | * easily find empty or nearly empty zspages when we try to shrink | |
748 | * the pool (not yet implemented). This function returns fullness | |
749 | * status of the given page. | |
750 | */ | |
1cac41cb MB |
751 | static enum fullness_group get_fullness_group(struct size_class *class, |
752 | struct zspage *zspage) | |
61989a80 | 753 | { |
1cac41cb | 754 | int inuse, objs_per_zspage; |
61989a80 | 755 | enum fullness_group fg; |
61989a80 | 756 | |
1cac41cb MB |
757 | inuse = get_zspage_inuse(zspage); |
758 | objs_per_zspage = class->objs_per_zspage; | |
61989a80 NG |
759 | |
760 | if (inuse == 0) | |
761 | fg = ZS_EMPTY; | |
1cac41cb | 762 | else if (inuse == objs_per_zspage) |
61989a80 | 763 | fg = ZS_FULL; |
1cac41cb | 764 | else if (inuse <= 3 * objs_per_zspage / fullness_threshold_frac) |
61989a80 NG |
765 | fg = ZS_ALMOST_EMPTY; |
766 | else | |
767 | fg = ZS_ALMOST_FULL; | |
768 | ||
769 | return fg; | |
770 | } | |
771 | ||
c3e3e88a NC |
772 | /* |
773 | * Each size class maintains various freelists and zspages are assigned | |
774 | * to one of these freelists based on the number of live objects they | |
775 | * have. This functions inserts the given zspage into the freelist | |
776 | * identified by <class, fullness_group>. | |
777 | */ | |
1cac41cb MB |
778 | static void insert_zspage(struct size_class *class, |
779 | struct zspage *zspage, | |
61989a80 NG |
780 | enum fullness_group fullness) |
781 | { | |
1cac41cb | 782 | struct zspage *head; |
58f17117 | 783 | |
1cac41cb MB |
784 | zs_stat_inc(class, fullness, 1); |
785 | head = list_first_entry_or_null(&class->fullness_list[fullness], | |
786 | struct zspage, list); | |
58f17117 | 787 | /* |
1cac41cb MB |
788 | * We want to see more ZS_FULL pages and less almost empty/full. |
789 | * Put pages with higher ->inuse first. | |
58f17117 | 790 | */ |
1cac41cb MB |
791 | if (head) { |
792 | if (get_zspage_inuse(zspage) < get_zspage_inuse(head)) { | |
793 | list_add(&zspage->list, &head->list); | |
794 | return; | |
795 | } | |
796 | } | |
797 | list_add(&zspage->list, &class->fullness_list[fullness]); | |
61989a80 NG |
798 | } |
799 | ||
c3e3e88a NC |
800 | /* |
801 | * This function removes the given zspage from the freelist identified | |
802 | * by <class, fullness_group>. | |
803 | */ | |
1cac41cb MB |
804 | static void remove_zspage(struct size_class *class, |
805 | struct zspage *zspage, | |
61989a80 NG |
806 | enum fullness_group fullness) |
807 | { | |
1cac41cb MB |
808 | VM_BUG_ON(list_empty(&class->fullness_list[fullness])); |
809 | VM_BUG_ON(is_zspage_isolated(zspage)); | |
61989a80 | 810 | |
1cac41cb MB |
811 | list_del_init(&zspage->list); |
812 | zs_stat_dec(class, fullness, 1); | |
61989a80 NG |
813 | } |
814 | ||
c3e3e88a NC |
815 | /* |
816 | * Each size class maintains zspages in different fullness groups depending | |
817 | * on the number of live objects they contain. When allocating or freeing | |
818 | * objects, the fullness status of the page can change, say, from ALMOST_FULL | |
819 | * to ALMOST_EMPTY when freeing an object. This function checks if such | |
820 | * a status change has occurred for the given page and accordingly moves the | |
821 | * page from the freelist of the old fullness group to that of the new | |
822 | * fullness group. | |
823 | */ | |
c7806261 | 824 | static enum fullness_group fix_fullness_group(struct size_class *class, |
1cac41cb | 825 | struct zspage *zspage) |
61989a80 NG |
826 | { |
827 | int class_idx; | |
61989a80 NG |
828 | enum fullness_group currfg, newfg; |
829 | ||
1cac41cb MB |
830 | get_zspage_mapping(zspage, &class_idx, &currfg); |
831 | newfg = get_fullness_group(class, zspage); | |
61989a80 NG |
832 | if (newfg == currfg) |
833 | goto out; | |
834 | ||
1cac41cb MB |
835 | if (!is_zspage_isolated(zspage)) { |
836 | remove_zspage(class, zspage, currfg); | |
837 | insert_zspage(class, zspage, newfg); | |
838 | } | |
839 | ||
840 | set_zspage_mapping(zspage, class_idx, newfg); | |
61989a80 NG |
841 | |
842 | out: | |
843 | return newfg; | |
844 | } | |
845 | ||
846 | /* | |
847 | * We have to decide on how many pages to link together | |
848 | * to form a zspage for each size class. This is important | |
849 | * to reduce wastage due to unusable space left at end of | |
850 | * each zspage which is given as: | |
888fa374 YX |
851 | * wastage = Zp % class_size |
852 | * usage = Zp - wastage | |
61989a80 NG |
853 | * where Zp = zspage size = k * PAGE_SIZE where k = 1, 2, ... |
854 | * | |
855 | * For example, for size class of 3/8 * PAGE_SIZE, we should | |
856 | * link together 3 PAGE_SIZE sized pages to form a zspage | |
857 | * since then we can perfectly fit in 8 such objects. | |
858 | */ | |
2e3b6154 | 859 | static int get_pages_per_zspage(int class_size) |
61989a80 NG |
860 | { |
861 | int i, max_usedpc = 0; | |
862 | /* zspage order which gives maximum used size per KB */ | |
863 | int max_usedpc_order = 1; | |
864 | ||
84d4faab | 865 | for (i = 1; i <= ZS_MAX_PAGES_PER_ZSPAGE; i++) { |
61989a80 NG |
866 | int zspage_size; |
867 | int waste, usedpc; | |
868 | ||
869 | zspage_size = i * PAGE_SIZE; | |
870 | waste = zspage_size % class_size; | |
871 | usedpc = (zspage_size - waste) * 100 / zspage_size; | |
872 | ||
873 | if (usedpc > max_usedpc) { | |
874 | max_usedpc = usedpc; | |
875 | max_usedpc_order = i; | |
876 | } | |
877 | } | |
878 | ||
879 | return max_usedpc_order; | |
880 | } | |
881 | ||
1cac41cb | 882 | static struct zspage *get_zspage(struct page *page) |
61989a80 | 883 | { |
1cac41cb MB |
884 | struct zspage *zspage = (struct zspage *)page->private; |
885 | ||
886 | VM_BUG_ON(zspage->magic != ZSPAGE_MAGIC); | |
887 | return zspage; | |
61989a80 NG |
888 | } |
889 | ||
890 | static struct page *get_next_page(struct page *page) | |
891 | { | |
1cac41cb MB |
892 | if (unlikely(PageHugeObject(page))) |
893 | return NULL; | |
61989a80 | 894 | |
1cac41cb MB |
895 | return page->freelist; |
896 | } | |
61989a80 | 897 | |
1cac41cb MB |
898 | /** |
899 | * obj_to_location - get (<page>, <obj_idx>) from encoded object value | |
900 | * @page: page object resides in zspage | |
901 | * @obj_idx: object index | |
902 | */ | |
903 | static void obj_to_location(unsigned long obj, struct page **page, | |
904 | unsigned int *obj_idx) | |
905 | { | |
906 | obj >>= OBJ_TAG_BITS; | |
907 | *page = pfn_to_page(obj >> OBJ_INDEX_BITS); | |
908 | *obj_idx = (obj & OBJ_INDEX_MASK); | |
61989a80 NG |
909 | } |
910 | ||
1cac41cb MB |
911 | /** |
912 | * location_to_obj - get obj value encoded from (<page>, <obj_idx>) | |
913 | * @page: page object resides in zspage | |
914 | * @obj_idx: object index | |
67296874 | 915 | */ |
1cac41cb | 916 | static unsigned long location_to_obj(struct page *page, unsigned int obj_idx) |
61989a80 | 917 | { |
312fcae2 | 918 | unsigned long obj; |
61989a80 | 919 | |
312fcae2 | 920 | obj = page_to_pfn(page) << OBJ_INDEX_BITS; |
1cac41cb | 921 | obj |= obj_idx & OBJ_INDEX_MASK; |
312fcae2 | 922 | obj <<= OBJ_TAG_BITS; |
61989a80 | 923 | |
1cac41cb | 924 | return obj; |
61989a80 NG |
925 | } |
926 | ||
2e40e163 MK |
927 | static unsigned long handle_to_obj(unsigned long handle) |
928 | { | |
929 | return *(unsigned long *)handle; | |
930 | } | |
931 | ||
1cac41cb | 932 | static unsigned long obj_to_head(struct page *page, void *obj) |
312fcae2 | 933 | { |
1cac41cb MB |
934 | if (unlikely(PageHugeObject(page))) { |
935 | VM_BUG_ON_PAGE(!is_first_page(page), page); | |
936 | return page->index; | |
7b60a685 MK |
937 | } else |
938 | return *(unsigned long *)obj; | |
312fcae2 MK |
939 | } |
940 | ||
1cac41cb | 941 | static inline int testpin_tag(unsigned long handle) |
61989a80 | 942 | { |
1cac41cb | 943 | return bit_spin_is_locked(HANDLE_PIN_BIT, (unsigned long *)handle); |
61989a80 NG |
944 | } |
945 | ||
312fcae2 MK |
946 | static inline int trypin_tag(unsigned long handle) |
947 | { | |
1cac41cb | 948 | return bit_spin_trylock(HANDLE_PIN_BIT, (unsigned long *)handle); |
312fcae2 MK |
949 | } |
950 | ||
951 | static void pin_tag(unsigned long handle) | |
952 | { | |
1cac41cb | 953 | bit_spin_lock(HANDLE_PIN_BIT, (unsigned long *)handle); |
312fcae2 MK |
954 | } |
955 | ||
956 | static void unpin_tag(unsigned long handle) | |
957 | { | |
1cac41cb | 958 | bit_spin_unlock(HANDLE_PIN_BIT, (unsigned long *)handle); |
312fcae2 MK |
959 | } |
960 | ||
f4477e90 NG |
961 | static void reset_page(struct page *page) |
962 | { | |
1cac41cb | 963 | __ClearPageMovable(page); |
f4477e90 NG |
964 | clear_bit(PG_private, &page->flags); |
965 | clear_bit(PG_private_2, &page->flags); | |
966 | set_page_private(page, 0); | |
22b751c3 | 967 | page_mapcount_reset(page); |
1cac41cb MB |
968 | ClearPageHugeObject(page); |
969 | page->freelist = NULL; | |
970 | } | |
971 | ||
972 | /* | |
973 | * To prevent zspage destroy during migration, zspage freeing should | |
974 | * hold locks of all pages in the zspage. | |
975 | */ | |
976 | void lock_zspage(struct zspage *zspage) | |
977 | { | |
978 | struct page *page = get_first_page(zspage); | |
979 | ||
980 | do { | |
981 | lock_page(page); | |
982 | } while ((page = get_next_page(page)) != NULL); | |
f4477e90 NG |
983 | } |
984 | ||
1cac41cb | 985 | int trylock_zspage(struct zspage *zspage) |
61989a80 | 986 | { |
1cac41cb | 987 | struct page *cursor, *fail; |
61989a80 | 988 | |
1cac41cb MB |
989 | for (cursor = get_first_page(zspage); cursor != NULL; cursor = |
990 | get_next_page(cursor)) { | |
991 | if (!trylock_page(cursor)) { | |
992 | fail = cursor; | |
993 | goto unlock; | |
994 | } | |
995 | } | |
61989a80 | 996 | |
1cac41cb MB |
997 | return 1; |
998 | unlock: | |
999 | for (cursor = get_first_page(zspage); cursor != fail; cursor = | |
1000 | get_next_page(cursor)) | |
1001 | unlock_page(cursor); | |
61989a80 | 1002 | |
1cac41cb MB |
1003 | return 0; |
1004 | } | |
61989a80 | 1005 | |
1cac41cb MB |
1006 | static void __free_zspage(struct zs_pool *pool, struct size_class *class, |
1007 | struct zspage *zspage) | |
1008 | { | |
1009 | struct page *page, *next; | |
1010 | enum fullness_group fg; | |
1011 | unsigned int class_idx; | |
1012 | ||
1013 | get_zspage_mapping(zspage, &class_idx, &fg); | |
1014 | ||
1015 | assert_spin_locked(&class->lock); | |
1016 | ||
1017 | VM_BUG_ON(get_zspage_inuse(zspage)); | |
1018 | VM_BUG_ON(fg != ZS_EMPTY); | |
1019 | ||
1020 | next = page = get_first_page(zspage); | |
1021 | do { | |
1022 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
1023 | next = get_next_page(page); | |
1024 | reset_page(page); | |
1025 | unlock_page(page); | |
1026 | put_page(page); | |
1027 | page = next; | |
1028 | } while (page != NULL); | |
1029 | ||
1030 | cache_free_zspage(pool, zspage); | |
61989a80 | 1031 | |
1cac41cb MB |
1032 | zs_stat_dec(class, OBJ_ALLOCATED, get_maxobj_per_zspage( |
1033 | class->size, class->pages_per_zspage)); | |
1034 | atomic_long_sub(class->pages_per_zspage, | |
1035 | &pool->pages_allocated); | |
1036 | } | |
1037 | ||
1038 | static void free_zspage(struct zs_pool *pool, struct size_class *class, | |
1039 | struct zspage *zspage) | |
1040 | { | |
1041 | VM_BUG_ON(get_zspage_inuse(zspage)); | |
1042 | VM_BUG_ON(list_empty(&zspage->list)); | |
1043 | ||
1044 | if (!trylock_zspage(zspage)) { | |
1045 | kick_deferred_free(pool); | |
1046 | return; | |
61989a80 | 1047 | } |
1cac41cb MB |
1048 | |
1049 | remove_zspage(class, zspage, ZS_EMPTY); | |
1050 | __free_zspage(pool, class, zspage); | |
61989a80 NG |
1051 | } |
1052 | ||
1053 | /* Initialize a newly allocated zspage */ | |
1cac41cb | 1054 | static void init_zspage(struct size_class *class, struct zspage *zspage) |
61989a80 | 1055 | { |
1cac41cb | 1056 | unsigned int freeobj = 1; |
61989a80 | 1057 | unsigned long off = 0; |
1cac41cb | 1058 | struct page *page = get_first_page(zspage); |
61989a80 | 1059 | |
61989a80 NG |
1060 | while (page) { |
1061 | struct page *next_page; | |
1062 | struct link_free *link; | |
af4ee5e9 | 1063 | void *vaddr; |
61989a80 | 1064 | |
1cac41cb | 1065 | set_first_obj_offset(page, off); |
61989a80 | 1066 | |
af4ee5e9 MK |
1067 | vaddr = kmap_atomic(page); |
1068 | link = (struct link_free *)vaddr + off / sizeof(*link); | |
5538c562 DS |
1069 | |
1070 | while ((off += class->size) < PAGE_SIZE) { | |
1cac41cb | 1071 | link->next = freeobj++ << OBJ_ALLOCATED_TAG; |
5538c562 | 1072 | link += class->size / sizeof(*link); |
61989a80 NG |
1073 | } |
1074 | ||
1075 | /* | |
1076 | * We now come to the last (full or partial) object on this | |
1077 | * page, which must point to the first object on the next | |
1078 | * page (if present) | |
1079 | */ | |
1080 | next_page = get_next_page(page); | |
1cac41cb MB |
1081 | if (next_page) { |
1082 | link->next = freeobj++ << OBJ_ALLOCATED_TAG; | |
1083 | } else { | |
1084 | /* | |
1085 | * Reset OBJ_ALLOCATED_TAG bit to last link to tell | |
1086 | * whether it's allocated object or not. | |
1087 | */ | |
1088 | link->next = -1 << OBJ_ALLOCATED_TAG; | |
1089 | } | |
af4ee5e9 | 1090 | kunmap_atomic(vaddr); |
61989a80 | 1091 | page = next_page; |
5538c562 | 1092 | off %= PAGE_SIZE; |
61989a80 | 1093 | } |
1cac41cb MB |
1094 | |
1095 | set_freeobj(zspage, 0); | |
61989a80 NG |
1096 | } |
1097 | ||
1cac41cb MB |
1098 | static void create_page_chain(struct size_class *class, struct zspage *zspage, |
1099 | struct page *pages[]) | |
61989a80 | 1100 | { |
1cac41cb MB |
1101 | int i; |
1102 | struct page *page; | |
1103 | struct page *prev_page = NULL; | |
1104 | int nr_pages = class->pages_per_zspage; | |
61989a80 NG |
1105 | |
1106 | /* | |
1107 | * Allocate individual pages and link them together as: | |
1cac41cb MB |
1108 | * 1. all pages are linked together using page->freelist |
1109 | * 2. each sub-page point to zspage using page->private | |
61989a80 | 1110 | * |
1cac41cb MB |
1111 | * we set PG_private to identify the first page (i.e. no other sub-page |
1112 | * has this flag set) and PG_private_2 to identify the last page. | |
61989a80 | 1113 | */ |
1cac41cb MB |
1114 | for (i = 0; i < nr_pages; i++) { |
1115 | page = pages[i]; | |
1116 | set_page_private(page, (unsigned long)zspage); | |
1117 | page->freelist = NULL; | |
1118 | if (i == 0) { | |
1119 | zspage->first_page = page; | |
a27545bf | 1120 | SetPagePrivate(page); |
1cac41cb MB |
1121 | if (unlikely(class->objs_per_zspage == 1 && |
1122 | class->pages_per_zspage == 1)) | |
1123 | SetPageHugeObject(page); | |
1124 | } else { | |
1125 | prev_page->freelist = page; | |
61989a80 | 1126 | } |
1cac41cb | 1127 | if (i == nr_pages - 1) |
a27545bf | 1128 | SetPagePrivate2(page); |
61989a80 NG |
1129 | prev_page = page; |
1130 | } | |
1cac41cb | 1131 | } |
61989a80 | 1132 | |
1cac41cb MB |
1133 | /* |
1134 | * Allocate a zspage for the given size class | |
1135 | */ | |
1136 | static struct zspage *alloc_zspage(struct zs_pool *pool, | |
1137 | struct size_class *class, | |
1138 | gfp_t gfp) | |
1139 | { | |
1140 | int i; | |
1141 | struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE]; | |
1142 | struct zspage *zspage = cache_alloc_zspage(pool, gfp); | |
61989a80 | 1143 | |
1cac41cb MB |
1144 | if (!zspage) |
1145 | return NULL; | |
61989a80 | 1146 | |
1cac41cb MB |
1147 | memset(zspage, 0, sizeof(struct zspage)); |
1148 | zspage->magic = ZSPAGE_MAGIC; | |
1149 | migrate_lock_init(zspage); | |
61989a80 | 1150 | |
1cac41cb MB |
1151 | for (i = 0; i < class->pages_per_zspage; i++) { |
1152 | struct page *page; | |
1153 | ||
1154 | page = alloc_page(gfp); | |
1155 | if (!page) { | |
1156 | while (--i >= 0) | |
1157 | __free_page(pages[i]); | |
1158 | cache_free_zspage(pool, zspage); | |
1159 | return NULL; | |
1160 | } | |
1161 | pages[i] = page; | |
61989a80 NG |
1162 | } |
1163 | ||
1cac41cb MB |
1164 | create_page_chain(class, zspage, pages); |
1165 | init_zspage(class, zspage); | |
1166 | ||
1167 | return zspage; | |
61989a80 NG |
1168 | } |
1169 | ||
1cac41cb | 1170 | static struct zspage *find_get_zspage(struct size_class *class) |
61989a80 NG |
1171 | { |
1172 | int i; | |
1cac41cb | 1173 | struct zspage *zspage; |
61989a80 | 1174 | |
1cac41cb MB |
1175 | for (i = ZS_ALMOST_FULL; i >= ZS_EMPTY; i--) { |
1176 | zspage = list_first_entry_or_null(&class->fullness_list[i], | |
1177 | struct zspage, list); | |
1178 | if (zspage) | |
61989a80 NG |
1179 | break; |
1180 | } | |
1181 | ||
1cac41cb | 1182 | return zspage; |
61989a80 NG |
1183 | } |
1184 | ||
1b945aee | 1185 | #ifdef CONFIG_PGTABLE_MAPPING |
f553646a SJ |
1186 | static inline int __zs_cpu_up(struct mapping_area *area) |
1187 | { | |
1188 | /* | |
1189 | * Make sure we don't leak memory if a cpu UP notification | |
1190 | * and zs_init() race and both call zs_cpu_up() on the same cpu | |
1191 | */ | |
1192 | if (area->vm) | |
1193 | return 0; | |
1194 | area->vm = alloc_vm_area(PAGE_SIZE * 2, NULL); | |
1195 | if (!area->vm) | |
1196 | return -ENOMEM; | |
1197 | return 0; | |
1198 | } | |
1199 | ||
1200 | static inline void __zs_cpu_down(struct mapping_area *area) | |
1201 | { | |
1202 | if (area->vm) | |
1203 | free_vm_area(area->vm); | |
1204 | area->vm = NULL; | |
1205 | } | |
1206 | ||
1207 | static inline void *__zs_map_object(struct mapping_area *area, | |
1208 | struct page *pages[2], int off, int size) | |
1209 | { | |
f6f8ed47 | 1210 | BUG_ON(map_vm_area(area->vm, PAGE_KERNEL, pages)); |
f553646a SJ |
1211 | area->vm_addr = area->vm->addr; |
1212 | return area->vm_addr + off; | |
1213 | } | |
1214 | ||
1215 | static inline void __zs_unmap_object(struct mapping_area *area, | |
1216 | struct page *pages[2], int off, int size) | |
1217 | { | |
1218 | unsigned long addr = (unsigned long)area->vm_addr; | |
f553646a | 1219 | |
d95abbbb | 1220 | unmap_kernel_range(addr, PAGE_SIZE * 2); |
f553646a SJ |
1221 | } |
1222 | ||
1b945aee | 1223 | #else /* CONFIG_PGTABLE_MAPPING */ |
f553646a SJ |
1224 | |
1225 | static inline int __zs_cpu_up(struct mapping_area *area) | |
1226 | { | |
1227 | /* | |
1228 | * Make sure we don't leak memory if a cpu UP notification | |
1229 | * and zs_init() race and both call zs_cpu_up() on the same cpu | |
1230 | */ | |
1231 | if (area->vm_buf) | |
1232 | return 0; | |
40f9fb8c | 1233 | area->vm_buf = kmalloc(ZS_MAX_ALLOC_SIZE, GFP_KERNEL); |
f553646a SJ |
1234 | if (!area->vm_buf) |
1235 | return -ENOMEM; | |
1236 | return 0; | |
1237 | } | |
1238 | ||
1239 | static inline void __zs_cpu_down(struct mapping_area *area) | |
1240 | { | |
40f9fb8c | 1241 | kfree(area->vm_buf); |
f553646a SJ |
1242 | area->vm_buf = NULL; |
1243 | } | |
1244 | ||
1245 | static void *__zs_map_object(struct mapping_area *area, | |
1246 | struct page *pages[2], int off, int size) | |
5f601902 | 1247 | { |
5f601902 SJ |
1248 | int sizes[2]; |
1249 | void *addr; | |
f553646a | 1250 | char *buf = area->vm_buf; |
5f601902 | 1251 | |
f553646a SJ |
1252 | /* disable page faults to match kmap_atomic() return conditions */ |
1253 | pagefault_disable(); | |
1254 | ||
1255 | /* no read fastpath */ | |
1256 | if (area->vm_mm == ZS_MM_WO) | |
1257 | goto out; | |
5f601902 SJ |
1258 | |
1259 | sizes[0] = PAGE_SIZE - off; | |
1260 | sizes[1] = size - sizes[0]; | |
1261 | ||
5f601902 SJ |
1262 | /* copy object to per-cpu buffer */ |
1263 | addr = kmap_atomic(pages[0]); | |
1264 | memcpy(buf, addr + off, sizes[0]); | |
1265 | kunmap_atomic(addr); | |
1266 | addr = kmap_atomic(pages[1]); | |
1267 | memcpy(buf + sizes[0], addr, sizes[1]); | |
1268 | kunmap_atomic(addr); | |
f553646a SJ |
1269 | out: |
1270 | return area->vm_buf; | |
5f601902 SJ |
1271 | } |
1272 | ||
f553646a SJ |
1273 | static void __zs_unmap_object(struct mapping_area *area, |
1274 | struct page *pages[2], int off, int size) | |
5f601902 | 1275 | { |
5f601902 SJ |
1276 | int sizes[2]; |
1277 | void *addr; | |
2e40e163 | 1278 | char *buf; |
5f601902 | 1279 | |
f553646a SJ |
1280 | /* no write fastpath */ |
1281 | if (area->vm_mm == ZS_MM_RO) | |
1282 | goto out; | |
5f601902 | 1283 | |
7b60a685 | 1284 | buf = area->vm_buf; |
1cac41cb MB |
1285 | buf = buf + ZS_HANDLE_SIZE; |
1286 | size -= ZS_HANDLE_SIZE; | |
1287 | off += ZS_HANDLE_SIZE; | |
2e40e163 | 1288 | |
5f601902 SJ |
1289 | sizes[0] = PAGE_SIZE - off; |
1290 | sizes[1] = size - sizes[0]; | |
1291 | ||
1292 | /* copy per-cpu buffer to object */ | |
1293 | addr = kmap_atomic(pages[0]); | |
1294 | memcpy(addr + off, buf, sizes[0]); | |
1295 | kunmap_atomic(addr); | |
1296 | addr = kmap_atomic(pages[1]); | |
1297 | memcpy(addr, buf + sizes[0], sizes[1]); | |
1298 | kunmap_atomic(addr); | |
f553646a SJ |
1299 | |
1300 | out: | |
1301 | /* enable page faults to match kunmap_atomic() return conditions */ | |
1302 | pagefault_enable(); | |
5f601902 | 1303 | } |
61989a80 | 1304 | |
1b945aee | 1305 | #endif /* CONFIG_PGTABLE_MAPPING */ |
f553646a | 1306 | |
61989a80 NG |
1307 | static int zs_cpu_notifier(struct notifier_block *nb, unsigned long action, |
1308 | void *pcpu) | |
1309 | { | |
f553646a | 1310 | int ret, cpu = (long)pcpu; |
61989a80 NG |
1311 | struct mapping_area *area; |
1312 | ||
1313 | switch (action) { | |
1314 | case CPU_UP_PREPARE: | |
1315 | area = &per_cpu(zs_map_area, cpu); | |
f553646a SJ |
1316 | ret = __zs_cpu_up(area); |
1317 | if (ret) | |
1318 | return notifier_from_errno(ret); | |
61989a80 NG |
1319 | break; |
1320 | case CPU_DEAD: | |
1321 | case CPU_UP_CANCELED: | |
1322 | area = &per_cpu(zs_map_area, cpu); | |
f553646a | 1323 | __zs_cpu_down(area); |
61989a80 NG |
1324 | break; |
1325 | } | |
1326 | ||
1327 | return NOTIFY_OK; | |
1328 | } | |
1329 | ||
1330 | static struct notifier_block zs_cpu_nb = { | |
1331 | .notifier_call = zs_cpu_notifier | |
1332 | }; | |
1333 | ||
b1b00a5b | 1334 | static int zs_register_cpu_notifier(void) |
61989a80 | 1335 | { |
b1b00a5b | 1336 | int cpu, uninitialized_var(ret); |
61989a80 | 1337 | |
f0e71fcd SB |
1338 | cpu_notifier_register_begin(); |
1339 | ||
1340 | __register_cpu_notifier(&zs_cpu_nb); | |
61989a80 NG |
1341 | for_each_online_cpu(cpu) { |
1342 | ret = zs_cpu_notifier(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
b1b00a5b SS |
1343 | if (notifier_to_errno(ret)) |
1344 | break; | |
61989a80 | 1345 | } |
f0e71fcd SB |
1346 | |
1347 | cpu_notifier_register_done(); | |
b1b00a5b SS |
1348 | return notifier_to_errno(ret); |
1349 | } | |
f0e71fcd | 1350 | |
66cdef66 | 1351 | static void zs_unregister_cpu_notifier(void) |
40f9fb8c | 1352 | { |
66cdef66 | 1353 | int cpu; |
40f9fb8c | 1354 | |
66cdef66 | 1355 | cpu_notifier_register_begin(); |
40f9fb8c | 1356 | |
66cdef66 GM |
1357 | for_each_online_cpu(cpu) |
1358 | zs_cpu_notifier(NULL, CPU_DEAD, (void *)(long)cpu); | |
1359 | __unregister_cpu_notifier(&zs_cpu_nb); | |
40f9fb8c | 1360 | |
66cdef66 | 1361 | cpu_notifier_register_done(); |
b1b00a5b SS |
1362 | } |
1363 | ||
66cdef66 | 1364 | static void init_zs_size_classes(void) |
b1b00a5b | 1365 | { |
66cdef66 | 1366 | int nr; |
c795779d | 1367 | |
66cdef66 GM |
1368 | nr = (ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) / ZS_SIZE_CLASS_DELTA + 1; |
1369 | if ((ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) % ZS_SIZE_CLASS_DELTA) | |
1370 | nr += 1; | |
40f9fb8c | 1371 | |
66cdef66 | 1372 | zs_size_classes = nr; |
61989a80 NG |
1373 | } |
1374 | ||
9eec4cd5 JK |
1375 | static bool can_merge(struct size_class *prev, int size, int pages_per_zspage) |
1376 | { | |
1377 | if (prev->pages_per_zspage != pages_per_zspage) | |
1378 | return false; | |
1379 | ||
1380 | if (get_maxobj_per_zspage(prev->size, prev->pages_per_zspage) | |
1381 | != get_maxobj_per_zspage(size, pages_per_zspage)) | |
1382 | return false; | |
1383 | ||
1384 | return true; | |
1385 | } | |
1386 | ||
1cac41cb | 1387 | static bool zspage_full(struct size_class *class, struct zspage *zspage) |
312fcae2 | 1388 | { |
1cac41cb | 1389 | return get_zspage_inuse(zspage) == class->objs_per_zspage; |
312fcae2 MK |
1390 | } |
1391 | ||
66cdef66 GM |
1392 | unsigned long zs_get_total_pages(struct zs_pool *pool) |
1393 | { | |
1394 | return atomic_long_read(&pool->pages_allocated); | |
1395 | } | |
1396 | EXPORT_SYMBOL_GPL(zs_get_total_pages); | |
1397 | ||
4bbc0bc0 | 1398 | /** |
66cdef66 GM |
1399 | * zs_map_object - get address of allocated object from handle. |
1400 | * @pool: pool from which the object was allocated | |
1401 | * @handle: handle returned from zs_malloc | |
4bbc0bc0 | 1402 | * |
66cdef66 GM |
1403 | * Before using an object allocated from zs_malloc, it must be mapped using |
1404 | * this function. When done with the object, it must be unmapped using | |
1405 | * zs_unmap_object. | |
4bbc0bc0 | 1406 | * |
66cdef66 GM |
1407 | * Only one object can be mapped per cpu at a time. There is no protection |
1408 | * against nested mappings. | |
1409 | * | |
1410 | * This function returns with preemption and page faults disabled. | |
4bbc0bc0 | 1411 | */ |
66cdef66 GM |
1412 | void *zs_map_object(struct zs_pool *pool, unsigned long handle, |
1413 | enum zs_mapmode mm) | |
61989a80 | 1414 | { |
1cac41cb | 1415 | struct zspage *zspage; |
66cdef66 | 1416 | struct page *page; |
1cac41cb MB |
1417 | unsigned long obj, off; |
1418 | unsigned int obj_idx; | |
61989a80 | 1419 | |
66cdef66 GM |
1420 | unsigned int class_idx; |
1421 | enum fullness_group fg; | |
1422 | struct size_class *class; | |
1423 | struct mapping_area *area; | |
1424 | struct page *pages[2]; | |
2e40e163 | 1425 | void *ret; |
61989a80 | 1426 | |
9eec4cd5 | 1427 | /* |
66cdef66 GM |
1428 | * Because we use per-cpu mapping areas shared among the |
1429 | * pools/users, we can't allow mapping in interrupt context | |
1430 | * because it can corrupt another users mappings. | |
9eec4cd5 | 1431 | */ |
1cac41cb | 1432 | WARN_ON_ONCE(in_interrupt()); |
61989a80 | 1433 | |
312fcae2 | 1434 | /* From now on, migration cannot move the object */ |
1cac41cb MB |
1435 | #ifdef CONFIG_ZSWAP_SAME_PAGE_SHARING |
1436 | if (mm == ZS_MM_RO_NOWAIT) { | |
1437 | if (!trypin_tag(handle)) | |
1438 | return NULL; | |
1439 | } else | |
1440 | pin_tag(handle); | |
1441 | #else | |
312fcae2 | 1442 | pin_tag(handle); |
1cac41cb | 1443 | #endif |
312fcae2 | 1444 | |
2e40e163 MK |
1445 | obj = handle_to_obj(handle); |
1446 | obj_to_location(obj, &page, &obj_idx); | |
1cac41cb MB |
1447 | zspage = get_zspage(page); |
1448 | ||
1449 | /* migration cannot move any subpage in this zspage */ | |
1450 | migrate_read_lock(zspage); | |
1451 | ||
1452 | get_zspage_mapping(zspage, &class_idx, &fg); | |
66cdef66 | 1453 | class = pool->size_class[class_idx]; |
1cac41cb | 1454 | off = (class->size * obj_idx) & ~PAGE_MASK; |
df8b5bb9 | 1455 | |
66cdef66 GM |
1456 | area = &get_cpu_var(zs_map_area); |
1457 | area->vm_mm = mm; | |
1458 | if (off + class->size <= PAGE_SIZE) { | |
1459 | /* this object is contained entirely within a page */ | |
1460 | area->vm_addr = kmap_atomic(page); | |
2e40e163 MK |
1461 | ret = area->vm_addr + off; |
1462 | goto out; | |
61989a80 NG |
1463 | } |
1464 | ||
66cdef66 GM |
1465 | /* this object spans two pages */ |
1466 | pages[0] = page; | |
1467 | pages[1] = get_next_page(page); | |
1468 | BUG_ON(!pages[1]); | |
9eec4cd5 | 1469 | |
2e40e163 MK |
1470 | ret = __zs_map_object(area, pages, off, class->size); |
1471 | out: | |
1cac41cb | 1472 | if (likely(!PageHugeObject(page))) |
7b60a685 MK |
1473 | ret += ZS_HANDLE_SIZE; |
1474 | ||
1475 | return ret; | |
61989a80 | 1476 | } |
66cdef66 | 1477 | EXPORT_SYMBOL_GPL(zs_map_object); |
61989a80 | 1478 | |
66cdef66 | 1479 | void zs_unmap_object(struct zs_pool *pool, unsigned long handle) |
61989a80 | 1480 | { |
1cac41cb | 1481 | struct zspage *zspage; |
66cdef66 | 1482 | struct page *page; |
1cac41cb MB |
1483 | unsigned long obj, off; |
1484 | unsigned int obj_idx; | |
61989a80 | 1485 | |
66cdef66 GM |
1486 | unsigned int class_idx; |
1487 | enum fullness_group fg; | |
1488 | struct size_class *class; | |
1489 | struct mapping_area *area; | |
9eec4cd5 | 1490 | |
2e40e163 MK |
1491 | obj = handle_to_obj(handle); |
1492 | obj_to_location(obj, &page, &obj_idx); | |
1cac41cb MB |
1493 | zspage = get_zspage(page); |
1494 | get_zspage_mapping(zspage, &class_idx, &fg); | |
66cdef66 | 1495 | class = pool->size_class[class_idx]; |
1cac41cb | 1496 | off = (class->size * obj_idx) & ~PAGE_MASK; |
61989a80 | 1497 | |
66cdef66 GM |
1498 | area = this_cpu_ptr(&zs_map_area); |
1499 | if (off + class->size <= PAGE_SIZE) | |
1500 | kunmap_atomic(area->vm_addr); | |
1501 | else { | |
1502 | struct page *pages[2]; | |
40f9fb8c | 1503 | |
66cdef66 GM |
1504 | pages[0] = page; |
1505 | pages[1] = get_next_page(page); | |
1506 | BUG_ON(!pages[1]); | |
1507 | ||
1508 | __zs_unmap_object(area, pages, off, class->size); | |
1509 | } | |
1510 | put_cpu_var(zs_map_area); | |
1cac41cb MB |
1511 | |
1512 | migrate_read_unlock(zspage); | |
312fcae2 | 1513 | unpin_tag(handle); |
61989a80 | 1514 | } |
66cdef66 | 1515 | EXPORT_SYMBOL_GPL(zs_unmap_object); |
61989a80 | 1516 | |
2a89387e SS |
1517 | /** |
1518 | * zs_huge_class_size() - Returns the size (in bytes) of the first huge | |
1519 | * zsmalloc &size_class. | |
1520 | * @pool: zsmalloc pool to use | |
1521 | * | |
1522 | * The function returns the size of the first huge class - any object of equal | |
1523 | * or bigger size will be stored in zspage consisting of a single physical | |
1524 | * page. | |
1525 | * | |
1526 | * Context: Any context. | |
1527 | * | |
1528 | * Return: the size (in bytes) of the first huge zsmalloc &size_class. | |
1529 | */ | |
1530 | size_t zs_huge_class_size(struct zs_pool *pool) | |
1531 | { | |
1532 | return huge_class_size; | |
1533 | } | |
1534 | EXPORT_SYMBOL_GPL(zs_huge_class_size); | |
1535 | ||
1cac41cb MB |
1536 | static unsigned long obj_malloc(struct size_class *class, |
1537 | struct zspage *zspage, unsigned long handle) | |
c7806261 | 1538 | { |
1cac41cb | 1539 | int i, nr_page, offset; |
c7806261 MK |
1540 | unsigned long obj; |
1541 | struct link_free *link; | |
1542 | ||
1543 | struct page *m_page; | |
1cac41cb | 1544 | unsigned long m_offset; |
c7806261 MK |
1545 | void *vaddr; |
1546 | ||
312fcae2 | 1547 | handle |= OBJ_ALLOCATED_TAG; |
1cac41cb MB |
1548 | obj = get_freeobj(zspage); |
1549 | ||
1550 | offset = obj * class->size; | |
1551 | nr_page = offset >> PAGE_SHIFT; | |
1552 | m_offset = offset & ~PAGE_MASK; | |
1553 | m_page = get_first_page(zspage); | |
1554 | ||
1555 | for (i = 0; i < nr_page; i++) | |
1556 | m_page = get_next_page(m_page); | |
c7806261 MK |
1557 | |
1558 | vaddr = kmap_atomic(m_page); | |
1559 | link = (struct link_free *)vaddr + m_offset / sizeof(*link); | |
1cac41cb MB |
1560 | set_freeobj(zspage, link->next >> OBJ_ALLOCATED_TAG); |
1561 | if (likely(!PageHugeObject(m_page))) | |
7b60a685 MK |
1562 | /* record handle in the header of allocated chunk */ |
1563 | link->handle = handle; | |
1564 | else | |
1cac41cb MB |
1565 | /* record handle to page->index */ |
1566 | zspage->first_page->index = handle; | |
1567 | ||
c7806261 | 1568 | kunmap_atomic(vaddr); |
1cac41cb | 1569 | mod_zspage_inuse(zspage, 1); |
c7806261 MK |
1570 | zs_stat_inc(class, OBJ_USED, 1); |
1571 | ||
1cac41cb MB |
1572 | obj = location_to_obj(m_page, obj); |
1573 | ||
c7806261 MK |
1574 | return obj; |
1575 | } | |
1576 | ||
1577 | ||
61989a80 NG |
1578 | /** |
1579 | * zs_malloc - Allocate block of given size from pool. | |
1580 | * @pool: pool to allocate from | |
1581 | * @size: size of block to allocate | |
61989a80 | 1582 | * |
00a61d86 | 1583 | * On success, handle to the allocated object is returned, |
c2344348 | 1584 | * otherwise 0. |
61989a80 NG |
1585 | * Allocation requests with size > ZS_MAX_ALLOC_SIZE will fail. |
1586 | */ | |
1e92b056 | 1587 | unsigned long zs_malloc(struct zs_pool *pool, size_t size, gfp_t gfp) |
61989a80 | 1588 | { |
2e40e163 | 1589 | unsigned long handle, obj; |
61989a80 | 1590 | struct size_class *class; |
1cac41cb MB |
1591 | enum fullness_group newfg; |
1592 | struct zspage *zspage; | |
61989a80 | 1593 | |
7b60a685 | 1594 | if (unlikely(!size || size > ZS_MAX_ALLOC_SIZE)) |
2e40e163 MK |
1595 | return 0; |
1596 | ||
1e92b056 | 1597 | handle = cache_alloc_handle(pool, gfp); |
2e40e163 | 1598 | if (!handle) |
c2344348 | 1599 | return 0; |
61989a80 | 1600 | |
2e40e163 MK |
1601 | /* extra space in chunk to keep the handle */ |
1602 | size += ZS_HANDLE_SIZE; | |
9eec4cd5 | 1603 | class = pool->size_class[get_size_class_index(size)]; |
61989a80 NG |
1604 | |
1605 | spin_lock(&class->lock); | |
1cac41cb | 1606 | zspage = find_get_zspage(class); |
61989a80 | 1607 | |
1cac41cb MB |
1608 | if (likely(zspage)) { |
1609 | obj = obj_malloc(class, zspage, handle); | |
1610 | /* Now move the zspage to another fullness group, if required */ | |
1611 | fix_fullness_group(class, zspage); | |
1612 | record_obj(handle, obj); | |
61989a80 | 1613 | spin_unlock(&class->lock); |
61989a80 | 1614 | |
1cac41cb MB |
1615 | return handle; |
1616 | } | |
0f050d99 | 1617 | |
1cac41cb MB |
1618 | spin_unlock(&class->lock); |
1619 | ||
1e92b056 | 1620 | zspage = alloc_zspage(pool, class, gfp); |
1cac41cb MB |
1621 | if (!zspage) { |
1622 | cache_free_handle(pool, handle); | |
1623 | return 0; | |
61989a80 NG |
1624 | } |
1625 | ||
1cac41cb MB |
1626 | spin_lock(&class->lock); |
1627 | obj = obj_malloc(class, zspage, handle); | |
1628 | newfg = get_fullness_group(class, zspage); | |
1629 | insert_zspage(class, zspage, newfg); | |
1630 | set_zspage_mapping(zspage, class->index, newfg); | |
2e40e163 | 1631 | record_obj(handle, obj); |
1cac41cb MB |
1632 | atomic_long_add(class->pages_per_zspage, |
1633 | &pool->pages_allocated); | |
1634 | zs_stat_inc(class, OBJ_ALLOCATED, get_maxobj_per_zspage( | |
1635 | class->size, class->pages_per_zspage)); | |
1636 | ||
1637 | /* We completely set up zspage so mark them as movable */ | |
1638 | SetZsPageMovable(pool, zspage); | |
61989a80 NG |
1639 | spin_unlock(&class->lock); |
1640 | ||
2e40e163 | 1641 | return handle; |
61989a80 NG |
1642 | } |
1643 | EXPORT_SYMBOL_GPL(zs_malloc); | |
1644 | ||
1cac41cb | 1645 | static void obj_free(struct size_class *class, unsigned long obj) |
61989a80 NG |
1646 | { |
1647 | struct link_free *link; | |
1cac41cb MB |
1648 | struct zspage *zspage; |
1649 | struct page *f_page; | |
1650 | unsigned long f_offset; | |
1651 | unsigned int f_objidx; | |
af4ee5e9 | 1652 | void *vaddr; |
61989a80 | 1653 | |
312fcae2 | 1654 | obj &= ~OBJ_ALLOCATED_TAG; |
2e40e163 | 1655 | obj_to_location(obj, &f_page, &f_objidx); |
1cac41cb MB |
1656 | f_offset = (class->size * f_objidx) & ~PAGE_MASK; |
1657 | zspage = get_zspage(f_page); | |
61989a80 | 1658 | |
c7806261 | 1659 | vaddr = kmap_atomic(f_page); |
61989a80 NG |
1660 | |
1661 | /* Insert this object in containing zspage's freelist */ | |
af4ee5e9 | 1662 | link = (struct link_free *)(vaddr + f_offset); |
1cac41cb | 1663 | link->next = get_freeobj(zspage) << OBJ_ALLOCATED_TAG; |
af4ee5e9 | 1664 | kunmap_atomic(vaddr); |
1cac41cb MB |
1665 | set_freeobj(zspage, f_objidx); |
1666 | mod_zspage_inuse(zspage, -1); | |
0f050d99 | 1667 | zs_stat_dec(class, OBJ_USED, 1); |
c7806261 MK |
1668 | } |
1669 | ||
1670 | void zs_free(struct zs_pool *pool, unsigned long handle) | |
1671 | { | |
1cac41cb MB |
1672 | struct zspage *zspage; |
1673 | struct page *f_page; | |
1674 | unsigned long obj; | |
1675 | unsigned int f_objidx; | |
c7806261 MK |
1676 | int class_idx; |
1677 | struct size_class *class; | |
1678 | enum fullness_group fullness; | |
1cac41cb | 1679 | bool isolated; |
c7806261 MK |
1680 | |
1681 | if (unlikely(!handle)) | |
1682 | return; | |
1683 | ||
312fcae2 | 1684 | pin_tag(handle); |
c7806261 | 1685 | obj = handle_to_obj(handle); |
c7806261 | 1686 | obj_to_location(obj, &f_page, &f_objidx); |
1cac41cb MB |
1687 | zspage = get_zspage(f_page); |
1688 | ||
1689 | migrate_read_lock(zspage); | |
c7806261 | 1690 | |
1cac41cb | 1691 | get_zspage_mapping(zspage, &class_idx, &fullness); |
c7806261 MK |
1692 | class = pool->size_class[class_idx]; |
1693 | ||
1694 | spin_lock(&class->lock); | |
1cac41cb MB |
1695 | obj_free(class, obj); |
1696 | fullness = fix_fullness_group(class, zspage); | |
1697 | if (fullness != ZS_EMPTY) { | |
1698 | migrate_read_unlock(zspage); | |
1699 | goto out; | |
312fcae2 | 1700 | } |
1cac41cb MB |
1701 | |
1702 | isolated = is_zspage_isolated(zspage); | |
1703 | migrate_read_unlock(zspage); | |
1704 | /* If zspage is isolated, zs_page_putback will free the zspage */ | |
1705 | if (likely(!isolated)) | |
1706 | free_zspage(pool, class, zspage); | |
1707 | out: | |
1708 | ||
61989a80 | 1709 | spin_unlock(&class->lock); |
312fcae2 | 1710 | unpin_tag(handle); |
1cac41cb | 1711 | cache_free_handle(pool, handle); |
312fcae2 MK |
1712 | } |
1713 | EXPORT_SYMBOL_GPL(zs_free); | |
1714 | ||
1cac41cb MB |
1715 | static void zs_object_copy(struct size_class *class, unsigned long dst, |
1716 | unsigned long src) | |
312fcae2 MK |
1717 | { |
1718 | struct page *s_page, *d_page; | |
1cac41cb | 1719 | unsigned int s_objidx, d_objidx; |
312fcae2 MK |
1720 | unsigned long s_off, d_off; |
1721 | void *s_addr, *d_addr; | |
1722 | int s_size, d_size, size; | |
1723 | int written = 0; | |
1724 | ||
1725 | s_size = d_size = class->size; | |
1726 | ||
1727 | obj_to_location(src, &s_page, &s_objidx); | |
1728 | obj_to_location(dst, &d_page, &d_objidx); | |
1729 | ||
1cac41cb MB |
1730 | s_off = (class->size * s_objidx) & ~PAGE_MASK; |
1731 | d_off = (class->size * d_objidx) & ~PAGE_MASK; | |
312fcae2 MK |
1732 | |
1733 | if (s_off + class->size > PAGE_SIZE) | |
1734 | s_size = PAGE_SIZE - s_off; | |
1735 | ||
1736 | if (d_off + class->size > PAGE_SIZE) | |
1737 | d_size = PAGE_SIZE - d_off; | |
1738 | ||
1739 | s_addr = kmap_atomic(s_page); | |
1740 | d_addr = kmap_atomic(d_page); | |
1741 | ||
1742 | while (1) { | |
1743 | size = min(s_size, d_size); | |
1744 | memcpy(d_addr + d_off, s_addr + s_off, size); | |
1745 | written += size; | |
1746 | ||
1747 | if (written == class->size) | |
1748 | break; | |
1749 | ||
495819ea SS |
1750 | s_off += size; |
1751 | s_size -= size; | |
1752 | d_off += size; | |
1753 | d_size -= size; | |
1754 | ||
1755 | if (s_off >= PAGE_SIZE) { | |
312fcae2 MK |
1756 | kunmap_atomic(d_addr); |
1757 | kunmap_atomic(s_addr); | |
1758 | s_page = get_next_page(s_page); | |
312fcae2 MK |
1759 | s_addr = kmap_atomic(s_page); |
1760 | d_addr = kmap_atomic(d_page); | |
1761 | s_size = class->size - written; | |
1762 | s_off = 0; | |
312fcae2 MK |
1763 | } |
1764 | ||
495819ea | 1765 | if (d_off >= PAGE_SIZE) { |
312fcae2 MK |
1766 | kunmap_atomic(d_addr); |
1767 | d_page = get_next_page(d_page); | |
312fcae2 MK |
1768 | d_addr = kmap_atomic(d_page); |
1769 | d_size = class->size - written; | |
1770 | d_off = 0; | |
312fcae2 MK |
1771 | } |
1772 | } | |
1773 | ||
1774 | kunmap_atomic(d_addr); | |
1775 | kunmap_atomic(s_addr); | |
1776 | } | |
1777 | ||
1778 | /* | |
1779 | * Find alloced object in zspage from index object and | |
1780 | * return handle. | |
1781 | */ | |
1cac41cb MB |
1782 | static unsigned long find_alloced_obj(struct size_class *class, |
1783 | struct page *page, int index) | |
312fcae2 MK |
1784 | { |
1785 | unsigned long head; | |
1786 | int offset = 0; | |
1787 | unsigned long handle = 0; | |
1788 | void *addr = kmap_atomic(page); | |
1789 | ||
1cac41cb | 1790 | offset = get_first_obj_offset(page); |
312fcae2 MK |
1791 | offset += class->size * index; |
1792 | ||
1793 | while (offset < PAGE_SIZE) { | |
1cac41cb | 1794 | head = obj_to_head(page, addr + offset); |
312fcae2 MK |
1795 | if (head & OBJ_ALLOCATED_TAG) { |
1796 | handle = head & ~OBJ_ALLOCATED_TAG; | |
1797 | if (trypin_tag(handle)) | |
1798 | break; | |
1799 | handle = 0; | |
1800 | } | |
1801 | ||
1802 | offset += class->size; | |
1803 | index++; | |
1804 | } | |
1805 | ||
1806 | kunmap_atomic(addr); | |
1807 | return handle; | |
1808 | } | |
1809 | ||
1810 | struct zs_compact_control { | |
1cac41cb | 1811 | /* Source spage for migration which could be a subpage of zspage */ |
312fcae2 MK |
1812 | struct page *s_page; |
1813 | /* Destination page for migration which should be a first page | |
1814 | * of zspage. */ | |
1815 | struct page *d_page; | |
1816 | /* Starting object index within @s_page which used for live object | |
1817 | * in the subpage. */ | |
1818 | int index; | |
312fcae2 MK |
1819 | }; |
1820 | ||
1821 | static int migrate_zspage(struct zs_pool *pool, struct size_class *class, | |
1822 | struct zs_compact_control *cc) | |
1823 | { | |
1824 | unsigned long used_obj, free_obj; | |
1825 | unsigned long handle; | |
1826 | struct page *s_page = cc->s_page; | |
1827 | struct page *d_page = cc->d_page; | |
1828 | unsigned long index = cc->index; | |
312fcae2 MK |
1829 | int ret = 0; |
1830 | ||
1831 | while (1) { | |
1cac41cb | 1832 | handle = find_alloced_obj(class, s_page, index); |
312fcae2 MK |
1833 | if (!handle) { |
1834 | s_page = get_next_page(s_page); | |
1835 | if (!s_page) | |
1836 | break; | |
1837 | index = 0; | |
1838 | continue; | |
1839 | } | |
1840 | ||
1841 | /* Stop if there is no more space */ | |
1cac41cb | 1842 | if (zspage_full(class, get_zspage(d_page))) { |
312fcae2 MK |
1843 | unpin_tag(handle); |
1844 | ret = -ENOMEM; | |
1845 | break; | |
1846 | } | |
1847 | ||
1848 | used_obj = handle_to_obj(handle); | |
1cac41cb MB |
1849 | free_obj = obj_malloc(class, get_zspage(d_page), handle); |
1850 | zs_object_copy(class, free_obj, used_obj); | |
312fcae2 | 1851 | index++; |
bddaf791 JL |
1852 | /* |
1853 | * record_obj updates handle's value to free_obj and it will | |
1854 | * invalidate lock bit(ie, HANDLE_PIN_BIT) of handle, which | |
1855 | * breaks synchronization using pin_tag(e,g, zs_free) so | |
1856 | * let's keep the lock bit. | |
1857 | */ | |
1858 | free_obj |= BIT(HANDLE_PIN_BIT); | |
312fcae2 MK |
1859 | record_obj(handle, free_obj); |
1860 | unpin_tag(handle); | |
1cac41cb | 1861 | obj_free(class, used_obj); |
312fcae2 MK |
1862 | } |
1863 | ||
1864 | /* Remember last position in this iteration */ | |
1865 | cc->s_page = s_page; | |
1866 | cc->index = index; | |
312fcae2 MK |
1867 | |
1868 | return ret; | |
1869 | } | |
1870 | ||
1cac41cb | 1871 | static struct zspage *isolate_zspage(struct size_class *class, bool source) |
312fcae2 MK |
1872 | { |
1873 | int i; | |
1cac41cb MB |
1874 | struct zspage *zspage; |
1875 | enum fullness_group fg[2] = {ZS_ALMOST_EMPTY, ZS_ALMOST_FULL}; | |
312fcae2 | 1876 | |
1cac41cb MB |
1877 | if (!source) { |
1878 | fg[0] = ZS_ALMOST_FULL; | |
1879 | fg[1] = ZS_ALMOST_EMPTY; | |
1880 | } | |
1881 | ||
1882 | for (i = 0; i < 2; i++) { | |
1883 | zspage = list_first_entry_or_null(&class->fullness_list[fg[i]], | |
1884 | struct zspage, list); | |
1885 | if (zspage) { | |
1886 | VM_BUG_ON(is_zspage_isolated(zspage)); | |
1887 | remove_zspage(class, zspage, fg[i]); | |
1888 | return zspage; | |
312fcae2 MK |
1889 | } |
1890 | } | |
1891 | ||
1cac41cb | 1892 | return zspage; |
312fcae2 MK |
1893 | } |
1894 | ||
860c707d | 1895 | /* |
1cac41cb | 1896 | * putback_zspage - add @zspage into right class's fullness list |
860c707d | 1897 | * @class: destination class |
1cac41cb | 1898 | * @zspage: target page |
860c707d | 1899 | * |
1cac41cb | 1900 | * Return @zspage's fullness_group |
860c707d | 1901 | */ |
1cac41cb MB |
1902 | static enum fullness_group putback_zspage(struct size_class *class, |
1903 | struct zspage *zspage) | |
312fcae2 | 1904 | { |
312fcae2 MK |
1905 | enum fullness_group fullness; |
1906 | ||
1cac41cb | 1907 | VM_BUG_ON(is_zspage_isolated(zspage)); |
312fcae2 | 1908 | |
1cac41cb MB |
1909 | fullness = get_fullness_group(class, zspage); |
1910 | insert_zspage(class, zspage, fullness); | |
1911 | set_zspage_mapping(zspage, class->index, fullness); | |
839373e6 | 1912 | |
1cac41cb MB |
1913 | return fullness; |
1914 | } | |
1915 | ||
1916 | #ifdef CONFIG_COMPACTION | |
1917 | static struct dentry *zs_mount(struct file_system_type *fs_type, | |
1918 | int flags, const char *dev_name, void *data) | |
1919 | { | |
1920 | static const struct dentry_operations ops = { | |
1921 | .d_dname = simple_dname, | |
1922 | }; | |
1923 | ||
1924 | return mount_pseudo(fs_type, "zsmalloc:", NULL, &ops, ZSMALLOC_MAGIC); | |
1925 | } | |
1926 | ||
1927 | static struct file_system_type zsmalloc_fs = { | |
1928 | .name = "zsmalloc", | |
1929 | .mount = zs_mount, | |
1930 | .kill_sb = kill_anon_super, | |
1931 | }; | |
1932 | ||
1933 | static int zsmalloc_mount(void) | |
1934 | { | |
1935 | int ret = 0; | |
1936 | ||
1937 | zsmalloc_mnt = kern_mount(&zsmalloc_fs); | |
1938 | if (IS_ERR(zsmalloc_mnt)) | |
1939 | ret = PTR_ERR(zsmalloc_mnt); | |
1940 | ||
1941 | return ret; | |
1942 | } | |
1943 | ||
1944 | static void zsmalloc_unmount(void) | |
1945 | { | |
1946 | kern_unmount(zsmalloc_mnt); | |
1947 | } | |
1948 | ||
1949 | static void migrate_lock_init(struct zspage *zspage) | |
1950 | { | |
1951 | rwlock_init(&zspage->lock); | |
1952 | } | |
1953 | ||
1954 | static void migrate_read_lock(struct zspage *zspage) | |
1955 | { | |
1956 | read_lock(&zspage->lock); | |
1957 | } | |
1958 | ||
1959 | static void migrate_read_unlock(struct zspage *zspage) | |
1960 | { | |
1961 | read_unlock(&zspage->lock); | |
1962 | } | |
1963 | ||
1964 | static void migrate_write_lock(struct zspage *zspage) | |
1965 | { | |
1966 | write_lock(&zspage->lock); | |
1967 | } | |
1968 | ||
1969 | static void migrate_write_unlock(struct zspage *zspage) | |
1970 | { | |
1971 | write_unlock(&zspage->lock); | |
1972 | } | |
312fcae2 | 1973 | |
1cac41cb MB |
1974 | /* Number of isolated subpage for *page migration* in this zspage */ |
1975 | static void inc_zspage_isolation(struct zspage *zspage) | |
1976 | { | |
1977 | zspage->isolated++; | |
1978 | } | |
1979 | ||
1980 | static void dec_zspage_isolation(struct zspage *zspage) | |
1981 | { | |
1982 | zspage->isolated--; | |
1983 | } | |
1984 | ||
1985 | static void replace_sub_page(struct size_class *class, struct zspage *zspage, | |
1986 | struct page *newpage, struct page *oldpage) | |
1987 | { | |
1988 | struct page *page; | |
1989 | struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE] = {NULL, }; | |
1990 | int idx = 0; | |
1991 | ||
1992 | page = get_first_page(zspage); | |
1993 | do { | |
1994 | if (page == oldpage) | |
1995 | pages[idx] = newpage; | |
1996 | else | |
1997 | pages[idx] = page; | |
1998 | idx++; | |
1999 | } while ((page = get_next_page(page)) != NULL); | |
2000 | ||
2001 | create_page_chain(class, zspage, pages); | |
2002 | set_first_obj_offset(newpage, get_first_obj_offset(oldpage)); | |
2003 | if (unlikely(PageHugeObject(oldpage))) | |
2004 | newpage->index = oldpage->index; | |
2005 | __SetPageMovable(newpage, page_mapping(oldpage)); | |
2006 | } | |
2007 | ||
2008 | bool zs_page_isolate(struct page *page, isolate_mode_t mode) | |
2009 | { | |
2010 | struct zs_pool *pool; | |
2011 | struct size_class *class; | |
2012 | int class_idx; | |
2013 | enum fullness_group fullness; | |
2014 | struct zspage *zspage; | |
2015 | struct address_space *mapping; | |
2016 | ||
2017 | /* | |
2018 | * Page is locked so zspage couldn't be destroyed. For detail, look at | |
2019 | * lock_zspage in free_zspage. | |
2020 | */ | |
2021 | if (!zs_page_migration_enabled) | |
2022 | return false; | |
2023 | ||
2024 | VM_BUG_ON_PAGE(!PageMovable(page), page); | |
2025 | VM_BUG_ON_PAGE(PageIsolated(page), page); | |
2026 | ||
2027 | zspage = get_zspage(page); | |
2028 | ||
2029 | /* | |
2030 | * Without class lock, fullness could be stale while class_idx is okay | |
2031 | * because class_idx is constant unless page is freed so we should get | |
2032 | * fullness again under class lock. | |
2033 | */ | |
2034 | get_zspage_mapping(zspage, &class_idx, &fullness); | |
2035 | mapping = page_mapping(page); | |
2036 | pool = mapping->private_data; | |
2037 | class = pool->size_class[class_idx]; | |
2038 | ||
2039 | spin_lock(&class->lock); | |
2040 | if (get_zspage_inuse(zspage) == 0) { | |
2041 | spin_unlock(&class->lock); | |
2042 | return false; | |
13de8933 | 2043 | } |
860c707d | 2044 | |
1cac41cb MB |
2045 | /* zspage is isolated for object migration */ |
2046 | if (list_empty(&zspage->list) && !is_zspage_isolated(zspage)) { | |
2047 | spin_unlock(&class->lock); | |
2048 | return false; | |
2049 | } | |
2050 | ||
2051 | /* | |
2052 | * If this is first time isolation for the zspage, isolate zspage from | |
2053 | * size_class to prevent further object allocation from the zspage. | |
2054 | */ | |
2055 | if (!list_empty(&zspage->list) && !is_zspage_isolated(zspage)) { | |
2056 | get_zspage_mapping(zspage, &class_idx, &fullness); | |
2057 | remove_zspage(class, zspage, fullness); | |
2058 | } | |
2059 | ||
2060 | inc_zspage_isolation(zspage); | |
2061 | spin_unlock(&class->lock); | |
2062 | ||
2063 | return true; | |
2064 | } | |
2065 | ||
2066 | int zs_page_migrate(struct address_space *mapping, struct page *newpage, | |
2067 | struct page *page, enum migrate_mode mode) | |
2068 | { | |
2069 | struct zs_pool *pool; | |
2070 | struct size_class *class; | |
2071 | int class_idx; | |
2072 | enum fullness_group fullness; | |
2073 | struct zspage *zspage; | |
2074 | struct page *dummy; | |
2075 | void *s_addr, *d_addr, *addr; | |
2076 | int offset, pos; | |
2077 | unsigned long handle, head; | |
2078 | unsigned long old_obj, new_obj; | |
2079 | unsigned int obj_idx; | |
2080 | int ret = -EAGAIN; | |
2081 | ||
2082 | VM_BUG_ON_PAGE(!PageMovable(page), page); | |
2083 | VM_BUG_ON_PAGE(!PageIsolated(page), page); | |
2084 | ||
2085 | zspage = get_zspage(page); | |
2086 | ||
2087 | /* Concurrent compactor cannot migrate any subpage in zspage */ | |
2088 | migrate_write_lock(zspage); | |
2089 | get_zspage_mapping(zspage, &class_idx, &fullness); | |
2090 | pool = mapping->private_data; | |
2091 | class = pool->size_class[class_idx]; | |
2092 | offset = get_first_obj_offset(page); | |
2093 | ||
2094 | spin_lock(&class->lock); | |
2095 | if (!get_zspage_inuse(zspage)) { | |
2096 | ret = -EBUSY; | |
2097 | goto unlock_class; | |
2098 | } | |
2099 | ||
2100 | pos = offset; | |
2101 | s_addr = kmap_atomic(page); | |
2102 | while (pos < PAGE_SIZE) { | |
2103 | head = obj_to_head(page, s_addr + pos); | |
2104 | if (head & OBJ_ALLOCATED_TAG) { | |
2105 | handle = head & ~OBJ_ALLOCATED_TAG; | |
2106 | if (!trypin_tag(handle)) | |
2107 | goto unpin_objects; | |
2108 | } | |
2109 | pos += class->size; | |
2110 | } | |
2111 | ||
2112 | /* | |
2113 | * Here, any user cannot access all objects in the zspage so let's move. | |
2114 | */ | |
2115 | d_addr = kmap_atomic(newpage); | |
2116 | memcpy(d_addr, s_addr, PAGE_SIZE); | |
2117 | kunmap_atomic(d_addr); | |
2118 | ||
2119 | for (addr = s_addr + offset; addr < s_addr + pos; | |
2120 | addr += class->size) { | |
2121 | head = obj_to_head(page, addr); | |
2122 | if (head & OBJ_ALLOCATED_TAG) { | |
2123 | handle = head & ~OBJ_ALLOCATED_TAG; | |
2124 | if (!testpin_tag(handle)) | |
2125 | BUG(); | |
2126 | ||
2127 | old_obj = handle_to_obj(handle); | |
2128 | obj_to_location(old_obj, &dummy, &obj_idx); | |
2129 | new_obj = (unsigned long)location_to_obj(newpage, | |
2130 | obj_idx); | |
2131 | new_obj |= BIT(HANDLE_PIN_BIT); | |
2132 | record_obj(handle, new_obj); | |
2133 | } | |
2134 | } | |
2135 | ||
2136 | replace_sub_page(class, zspage, newpage, page); | |
2137 | get_page(newpage); | |
2138 | ||
2139 | dec_zspage_isolation(zspage); | |
2140 | ||
2141 | /* | |
2142 | * Page migration is done so let's putback isolated zspage to | |
2143 | * the list if @page is final isolated subpage in the zspage. | |
2144 | */ | |
2145 | if (!is_zspage_isolated(zspage)) | |
2146 | putback_zspage(class, zspage); | |
2147 | ||
2148 | reset_page(page); | |
2149 | put_page(page); | |
2150 | page = newpage; | |
2151 | ||
2152 | ret = 0; | |
2153 | unpin_objects: | |
2154 | for (addr = s_addr + offset; addr < s_addr + pos; | |
2155 | addr += class->size) { | |
2156 | head = obj_to_head(page, addr); | |
2157 | if (head & OBJ_ALLOCATED_TAG) { | |
2158 | handle = head & ~OBJ_ALLOCATED_TAG; | |
2159 | if (!testpin_tag(handle)) | |
2160 | BUG(); | |
2161 | unpin_tag(handle); | |
2162 | } | |
2163 | } | |
2164 | kunmap_atomic(s_addr); | |
2165 | unlock_class: | |
2166 | spin_unlock(&class->lock); | |
2167 | migrate_write_unlock(zspage); | |
2168 | ||
2169 | return ret; | |
2170 | } | |
2171 | ||
2172 | void zs_page_putback(struct page *page) | |
2173 | { | |
2174 | struct zs_pool *pool; | |
2175 | struct size_class *class; | |
2176 | int class_idx; | |
2177 | enum fullness_group fg; | |
2178 | struct address_space *mapping; | |
2179 | struct zspage *zspage; | |
2180 | ||
2181 | VM_BUG_ON_PAGE(!PageMovable(page), page); | |
2182 | VM_BUG_ON_PAGE(!PageIsolated(page), page); | |
2183 | ||
2184 | zspage = get_zspage(page); | |
2185 | get_zspage_mapping(zspage, &class_idx, &fg); | |
2186 | mapping = page_mapping(page); | |
2187 | pool = mapping->private_data; | |
2188 | class = pool->size_class[class_idx]; | |
2189 | ||
2190 | spin_lock(&class->lock); | |
2191 | dec_zspage_isolation(zspage); | |
2192 | if (!is_zspage_isolated(zspage)) { | |
2193 | fg = putback_zspage(class, zspage); | |
2194 | /* | |
2195 | * Due to page_lock, we cannot free zspage immediately | |
2196 | * so let's defer. | |
2197 | */ | |
2198 | if (fg == ZS_EMPTY) | |
2199 | schedule_work(&pool->free_work); | |
2200 | } | |
2201 | spin_unlock(&class->lock); | |
61989a80 | 2202 | } |
312fcae2 | 2203 | |
1cac41cb MB |
2204 | const struct address_space_operations zsmalloc_aops = { |
2205 | .isolate_page = zs_page_isolate, | |
2206 | .migratepage = zs_page_migrate, | |
2207 | .putback_page = zs_page_putback, | |
2208 | }; | |
2209 | ||
2210 | static int zs_register_migration(struct zs_pool *pool) | |
2211 | { | |
2212 | pool->inode = alloc_anon_inode(zsmalloc_mnt->mnt_sb); | |
2213 | if (IS_ERR(pool->inode)) { | |
2214 | pool->inode = NULL; | |
2215 | return 1; | |
2216 | } | |
2217 | ||
2218 | pool->inode->i_mapping->private_data = pool; | |
2219 | pool->inode->i_mapping->a_ops = &zsmalloc_aops; | |
2220 | return 0; | |
2221 | } | |
2222 | ||
2223 | static void zs_unregister_migration(struct zs_pool *pool) | |
2224 | { | |
2225 | flush_work(&pool->free_work); | |
2226 | if (pool->inode) | |
2227 | iput(pool->inode); | |
2228 | } | |
2229 | ||
2230 | /* | |
2231 | * Caller should hold page_lock of all pages in the zspage | |
2232 | * In here, we cannot use zspage meta data. | |
2233 | */ | |
2234 | static void async_free_zspage(struct work_struct *work) | |
312fcae2 | 2235 | { |
ad9d5e17 | 2236 | int i; |
1cac41cb MB |
2237 | struct size_class *class; |
2238 | unsigned int class_idx; | |
2239 | enum fullness_group fullness; | |
2240 | struct zspage *zspage, *tmp; | |
2241 | LIST_HEAD(free_pages); | |
2242 | struct zs_pool *pool = container_of(work, struct zs_pool, | |
2243 | free_work); | |
ad9d5e17 | 2244 | |
1cac41cb MB |
2245 | for (i = 0; i < zs_size_classes; i++) { |
2246 | class = pool->size_class[i]; | |
2247 | if (class->index != i) | |
ad9d5e17 | 2248 | continue; |
312fcae2 | 2249 | |
1cac41cb MB |
2250 | spin_lock(&class->lock); |
2251 | list_splice_init(&class->fullness_list[ZS_EMPTY], &free_pages); | |
2252 | spin_unlock(&class->lock); | |
ad9d5e17 | 2253 | } |
312fcae2 | 2254 | |
1cac41cb MB |
2255 | |
2256 | list_for_each_entry_safe(zspage, tmp, &free_pages, list) { | |
2257 | list_del(&zspage->list); | |
2258 | lock_zspage(zspage); | |
2259 | ||
2260 | get_zspage_mapping(zspage, &class_idx, &fullness); | |
2261 | VM_BUG_ON(fullness != ZS_EMPTY); | |
2262 | class = pool->size_class[class_idx]; | |
2263 | spin_lock(&class->lock); | |
2264 | __free_zspage(pool, pool->size_class[class_idx], zspage); | |
2265 | spin_unlock(&class->lock); | |
2266 | } | |
2267 | }; | |
2268 | ||
2269 | static void kick_deferred_free(struct zs_pool *pool) | |
2270 | { | |
2271 | schedule_work(&pool->free_work); | |
312fcae2 MK |
2272 | } |
2273 | ||
1cac41cb MB |
2274 | static void init_deferred_free(struct zs_pool *pool) |
2275 | { | |
2276 | INIT_WORK(&pool->free_work, async_free_zspage); | |
2277 | } | |
2278 | ||
2279 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage) | |
2280 | { | |
2281 | struct page *page = get_first_page(zspage); | |
2282 | ||
2283 | do { | |
2284 | WARN_ON(!trylock_page(page)); | |
2285 | __SetPageMovable(page, pool->inode->i_mapping); | |
2286 | unlock_page(page); | |
2287 | } while ((page = get_next_page(page)) != NULL); | |
2288 | } | |
2289 | #endif | |
2290 | ||
04f05909 SS |
2291 | /* |
2292 | * | |
2293 | * Based on the number of unused allocated objects calculate | |
2294 | * and return the number of pages that we can free. | |
04f05909 SS |
2295 | */ |
2296 | static unsigned long zs_can_compact(struct size_class *class) | |
2297 | { | |
2298 | unsigned long obj_wasted; | |
1d77f0a5 SS |
2299 | unsigned long obj_allocated = zs_stat_get(class, OBJ_ALLOCATED); |
2300 | unsigned long obj_used = zs_stat_get(class, OBJ_USED); | |
04f05909 | 2301 | |
1d77f0a5 SS |
2302 | if (obj_allocated <= obj_used) |
2303 | return 0; | |
04f05909 | 2304 | |
1d77f0a5 | 2305 | obj_wasted = obj_allocated - obj_used; |
04f05909 SS |
2306 | obj_wasted /= get_maxobj_per_zspage(class->size, |
2307 | class->pages_per_zspage); | |
2308 | ||
6cbf16b3 | 2309 | return obj_wasted * class->pages_per_zspage; |
04f05909 SS |
2310 | } |
2311 | ||
7d3f3938 | 2312 | static void __zs_compact(struct zs_pool *pool, struct size_class *class) |
312fcae2 | 2313 | { |
312fcae2 | 2314 | struct zs_compact_control cc; |
1cac41cb MB |
2315 | struct zspage *src_zspage; |
2316 | struct zspage *dst_zspage = NULL; | |
312fcae2 | 2317 | |
312fcae2 | 2318 | spin_lock(&class->lock); |
1cac41cb | 2319 | while ((src_zspage = isolate_zspage(class, true))) { |
312fcae2 | 2320 | |
04f05909 SS |
2321 | if (!zs_can_compact(class)) |
2322 | break; | |
2323 | ||
312fcae2 | 2324 | cc.index = 0; |
1cac41cb | 2325 | cc.s_page = get_first_page(src_zspage); |
312fcae2 | 2326 | |
1cac41cb MB |
2327 | while ((dst_zspage = isolate_zspage(class, false))) { |
2328 | cc.d_page = get_first_page(dst_zspage); | |
312fcae2 | 2329 | /* |
0dc63d48 SS |
2330 | * If there is no more space in dst_page, resched |
2331 | * and see if anyone had allocated another zspage. | |
312fcae2 MK |
2332 | */ |
2333 | if (!migrate_zspage(pool, class, &cc)) | |
2334 | break; | |
2335 | ||
1cac41cb | 2336 | putback_zspage(class, dst_zspage); |
312fcae2 MK |
2337 | } |
2338 | ||
2339 | /* Stop if we couldn't find slot */ | |
1cac41cb | 2340 | if (dst_zspage == NULL) |
312fcae2 MK |
2341 | break; |
2342 | ||
1cac41cb MB |
2343 | putback_zspage(class, dst_zspage); |
2344 | if (putback_zspage(class, src_zspage) == ZS_EMPTY) { | |
2345 | free_zspage(pool, class, src_zspage); | |
6cbf16b3 | 2346 | pool->stats.pages_compacted += class->pages_per_zspage; |
1cac41cb | 2347 | } |
312fcae2 | 2348 | spin_unlock(&class->lock); |
312fcae2 MK |
2349 | cond_resched(); |
2350 | spin_lock(&class->lock); | |
2351 | } | |
2352 | ||
1cac41cb MB |
2353 | if (src_zspage) |
2354 | putback_zspage(class, src_zspage); | |
312fcae2 | 2355 | |
7d3f3938 | 2356 | spin_unlock(&class->lock); |
312fcae2 MK |
2357 | } |
2358 | ||
2359 | unsigned long zs_compact(struct zs_pool *pool) | |
2360 | { | |
2361 | int i; | |
312fcae2 MK |
2362 | struct size_class *class; |
2363 | ||
2364 | for (i = zs_size_classes - 1; i >= 0; i--) { | |
2365 | class = pool->size_class[i]; | |
2366 | if (!class) | |
2367 | continue; | |
2368 | if (class->index != i) | |
2369 | continue; | |
7d3f3938 | 2370 | __zs_compact(pool, class); |
312fcae2 MK |
2371 | } |
2372 | ||
860c707d | 2373 | return pool->stats.pages_compacted; |
312fcae2 MK |
2374 | } |
2375 | EXPORT_SYMBOL_GPL(zs_compact); | |
61989a80 | 2376 | |
7d3f3938 SS |
2377 | void zs_pool_stats(struct zs_pool *pool, struct zs_pool_stats *stats) |
2378 | { | |
2379 | memcpy(stats, &pool->stats, sizeof(struct zs_pool_stats)); | |
2380 | } | |
2381 | EXPORT_SYMBOL_GPL(zs_pool_stats); | |
2382 | ||
ab9d306d SS |
2383 | static unsigned long zs_shrinker_scan(struct shrinker *shrinker, |
2384 | struct shrink_control *sc) | |
2385 | { | |
2386 | unsigned long pages_freed; | |
2387 | struct zs_pool *pool = container_of(shrinker, struct zs_pool, | |
2388 | shrinker); | |
2389 | ||
2390 | pages_freed = pool->stats.pages_compacted; | |
2391 | /* | |
2392 | * Compact classes and calculate compaction delta. | |
2393 | * Can run concurrently with a manually triggered | |
2394 | * (by user) compaction. | |
2395 | */ | |
2396 | pages_freed = zs_compact(pool) - pages_freed; | |
2397 | ||
2398 | return pages_freed ? pages_freed : SHRINK_STOP; | |
2399 | } | |
2400 | ||
1cac41cb MB |
2401 | #define ZS_SHRINKER_THRESHOLD 1024 |
2402 | #define ZS_SHRINKER_INTERVAL 10 | |
2403 | ||
ab9d306d SS |
2404 | static unsigned long zs_shrinker_count(struct shrinker *shrinker, |
2405 | struct shrink_control *sc) | |
2406 | { | |
2407 | int i; | |
2408 | struct size_class *class; | |
2409 | unsigned long pages_to_free = 0; | |
2410 | struct zs_pool *pool = container_of(shrinker, struct zs_pool, | |
2411 | shrinker); | |
1cac41cb MB |
2412 | static unsigned long time_stamp; |
2413 | ||
2414 | if (!current_is_kswapd() || time_is_after_jiffies(time_stamp)) | |
2415 | return 0; | |
ab9d306d | 2416 | |
ab9d306d SS |
2417 | for (i = zs_size_classes - 1; i >= 0; i--) { |
2418 | class = pool->size_class[i]; | |
2419 | if (!class) | |
2420 | continue; | |
2421 | if (class->index != i) | |
2422 | continue; | |
2423 | ||
ab9d306d | 2424 | pages_to_free += zs_can_compact(class); |
ab9d306d SS |
2425 | } |
2426 | ||
1cac41cb MB |
2427 | if (pages_to_free > ZS_SHRINKER_THRESHOLD) |
2428 | time_stamp = jiffies + (ZS_SHRINKER_INTERVAL * HZ); | |
2429 | else | |
2430 | pages_to_free = 0; | |
2431 | ||
ab9d306d SS |
2432 | return pages_to_free; |
2433 | } | |
2434 | ||
2435 | static void zs_unregister_shrinker(struct zs_pool *pool) | |
2436 | { | |
2437 | if (pool->shrinker_enabled) { | |
2438 | unregister_shrinker(&pool->shrinker); | |
2439 | pool->shrinker_enabled = false; | |
2440 | } | |
2441 | } | |
2442 | ||
2443 | static int zs_register_shrinker(struct zs_pool *pool) | |
2444 | { | |
2445 | pool->shrinker.scan_objects = zs_shrinker_scan; | |
2446 | pool->shrinker.count_objects = zs_shrinker_count; | |
2447 | pool->shrinker.batch = 0; | |
2448 | pool->shrinker.seeks = DEFAULT_SEEKS; | |
2449 | ||
2450 | return register_shrinker(&pool->shrinker); | |
2451 | } | |
2452 | ||
00a61d86 | 2453 | /** |
66cdef66 GM |
2454 | * zs_create_pool - Creates an allocation pool to work from. |
2455 | * @flags: allocation flags used to allocate pool metadata | |
166cfda7 | 2456 | * |
66cdef66 GM |
2457 | * This function must be called before anything when using |
2458 | * the zsmalloc allocator. | |
166cfda7 | 2459 | * |
66cdef66 GM |
2460 | * On success, a pointer to the newly created pool is returned, |
2461 | * otherwise NULL. | |
396b7fd6 | 2462 | */ |
1e92b056 | 2463 | struct zs_pool *zs_create_pool(const char *name) |
61989a80 | 2464 | { |
66cdef66 GM |
2465 | int i; |
2466 | struct zs_pool *pool; | |
2467 | struct size_class *prev_class = NULL; | |
61989a80 | 2468 | |
66cdef66 GM |
2469 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
2470 | if (!pool) | |
2471 | return NULL; | |
61989a80 | 2472 | |
1cac41cb | 2473 | init_deferred_free(pool); |
66cdef66 GM |
2474 | pool->size_class = kcalloc(zs_size_classes, sizeof(struct size_class *), |
2475 | GFP_KERNEL); | |
2476 | if (!pool->size_class) { | |
2477 | kfree(pool); | |
2478 | return NULL; | |
2479 | } | |
61989a80 | 2480 | |
2e40e163 MK |
2481 | pool->name = kstrdup(name, GFP_KERNEL); |
2482 | if (!pool->name) | |
2483 | goto err; | |
2484 | ||
1cac41cb | 2485 | if (create_cache(pool)) |
2e40e163 MK |
2486 | goto err; |
2487 | ||
c60369f0 | 2488 | /* |
66cdef66 GM |
2489 | * Iterate reversly, because, size of size_class that we want to use |
2490 | * for merging should be larger or equal to current size. | |
c60369f0 | 2491 | */ |
66cdef66 GM |
2492 | for (i = zs_size_classes - 1; i >= 0; i--) { |
2493 | int size; | |
2494 | int pages_per_zspage; | |
2a89387e | 2495 | int objs_per_zspage; |
66cdef66 | 2496 | struct size_class *class; |
1cac41cb | 2497 | int fullness = 0; |
c60369f0 | 2498 | |
66cdef66 GM |
2499 | size = ZS_MIN_ALLOC_SIZE + i * ZS_SIZE_CLASS_DELTA; |
2500 | if (size > ZS_MAX_ALLOC_SIZE) | |
2501 | size = ZS_MAX_ALLOC_SIZE; | |
2502 | pages_per_zspage = get_pages_per_zspage(size); | |
2a89387e SS |
2503 | objs_per_zspage = pages_per_zspage * PAGE_SIZE / size; |
2504 | ||
2505 | /* | |
2506 | * We iterate from biggest down to smallest classes, | |
2507 | * so huge_class_size holds the size of the first huge | |
2508 | * class. Any object bigger than or equal to that will | |
2509 | * endup in the huge class. | |
2510 | */ | |
2511 | if (pages_per_zspage != 1 && objs_per_zspage != 1 && | |
2512 | !huge_class_size) { | |
2513 | huge_class_size = size; | |
2514 | /* | |
2515 | * The object uses ZS_HANDLE_SIZE bytes to store the | |
2516 | * handle. We need to subtract it, because zs_malloc() | |
2517 | * unconditionally adds handle size before it performs | |
2518 | * size class search - so object may be smaller than | |
2519 | * huge class size, yet it still can end up in the huge | |
2520 | * class because it grows by ZS_HANDLE_SIZE extra bytes | |
2521 | * right before class lookup. | |
2522 | */ | |
2523 | huge_class_size -= (ZS_HANDLE_SIZE - 1); | |
2524 | } | |
61989a80 | 2525 | |
66cdef66 GM |
2526 | /* |
2527 | * size_class is used for normal zsmalloc operation such | |
2528 | * as alloc/free for that size. Although it is natural that we | |
2529 | * have one size_class for each size, there is a chance that we | |
2530 | * can get more memory utilization if we use one size_class for | |
2531 | * many different sizes whose size_class have same | |
2532 | * characteristics. So, we makes size_class point to | |
2533 | * previous size_class if possible. | |
2534 | */ | |
2535 | if (prev_class) { | |
2536 | if (can_merge(prev_class, size, pages_per_zspage)) { | |
2537 | pool->size_class[i] = prev_class; | |
2538 | continue; | |
2539 | } | |
2540 | } | |
2541 | ||
2542 | class = kzalloc(sizeof(struct size_class), GFP_KERNEL); | |
2543 | if (!class) | |
2544 | goto err; | |
2545 | ||
2546 | class->size = size; | |
2547 | class->index = i; | |
2548 | class->pages_per_zspage = pages_per_zspage; | |
1cac41cb MB |
2549 | class->objs_per_zspage = class->pages_per_zspage * |
2550 | PAGE_SIZE / class->size; | |
66cdef66 GM |
2551 | spin_lock_init(&class->lock); |
2552 | pool->size_class[i] = class; | |
1cac41cb MB |
2553 | for (fullness = ZS_EMPTY; fullness < NR_ZS_FULLNESS; |
2554 | fullness++) | |
2555 | INIT_LIST_HEAD(&class->fullness_list[fullness]); | |
66cdef66 GM |
2556 | |
2557 | prev_class = class; | |
61989a80 NG |
2558 | } |
2559 | ||
1cac41cb MB |
2560 | if (zs_pool_stat_create(pool, name)) |
2561 | goto err; | |
2562 | ||
2563 | if (zs_register_migration(pool)) | |
0f050d99 GM |
2564 | goto err; |
2565 | ||
ab9d306d SS |
2566 | /* |
2567 | * Not critical, we still can use the pool | |
2568 | * and user can trigger compaction manually. | |
2569 | */ | |
2570 | if (zs_register_shrinker(pool) == 0) | |
2571 | pool->shrinker_enabled = true; | |
66cdef66 GM |
2572 | return pool; |
2573 | ||
2574 | err: | |
2575 | zs_destroy_pool(pool); | |
2576 | return NULL; | |
61989a80 | 2577 | } |
66cdef66 | 2578 | EXPORT_SYMBOL_GPL(zs_create_pool); |
61989a80 | 2579 | |
66cdef66 | 2580 | void zs_destroy_pool(struct zs_pool *pool) |
61989a80 | 2581 | { |
66cdef66 | 2582 | int i; |
61989a80 | 2583 | |
ab9d306d | 2584 | zs_unregister_shrinker(pool); |
1cac41cb | 2585 | zs_unregister_migration(pool); |
0f050d99 GM |
2586 | zs_pool_stat_destroy(pool); |
2587 | ||
66cdef66 GM |
2588 | for (i = 0; i < zs_size_classes; i++) { |
2589 | int fg; | |
2590 | struct size_class *class = pool->size_class[i]; | |
61989a80 | 2591 | |
66cdef66 GM |
2592 | if (!class) |
2593 | continue; | |
61989a80 | 2594 | |
66cdef66 GM |
2595 | if (class->index != i) |
2596 | continue; | |
61989a80 | 2597 | |
1cac41cb MB |
2598 | for (fg = ZS_EMPTY; fg < NR_ZS_FULLNESS; fg++) { |
2599 | if (!list_empty(&class->fullness_list[fg])) { | |
66cdef66 GM |
2600 | pr_info("Freeing non-empty class with size %db, fullness group %d\n", |
2601 | class->size, fg); | |
2602 | } | |
2603 | } | |
2604 | kfree(class); | |
2605 | } | |
f553646a | 2606 | |
1cac41cb | 2607 | destroy_cache(pool); |
66cdef66 | 2608 | kfree(pool->size_class); |
0f050d99 | 2609 | kfree(pool->name); |
66cdef66 GM |
2610 | kfree(pool); |
2611 | } | |
2612 | EXPORT_SYMBOL_GPL(zs_destroy_pool); | |
b7418510 | 2613 | |
66cdef66 GM |
2614 | static int __init zs_init(void) |
2615 | { | |
1cac41cb MB |
2616 | int ret; |
2617 | ||
2618 | ret = zsmalloc_mount(); | |
2619 | if (ret) | |
2620 | goto out; | |
2621 | ||
2622 | ret = zs_register_cpu_notifier(); | |
66cdef66 | 2623 | |
0f050d99 GM |
2624 | if (ret) |
2625 | goto notifier_fail; | |
66cdef66 GM |
2626 | |
2627 | init_zs_size_classes(); | |
2628 | ||
2629 | #ifdef CONFIG_ZPOOL | |
2630 | zpool_register_driver(&zs_zpool_driver); | |
2631 | #endif | |
0f050d99 GM |
2632 | |
2633 | ret = zs_stat_init(); | |
2634 | if (ret) { | |
2635 | pr_err("zs stat initialization failed\n"); | |
2636 | goto stat_fail; | |
2637 | } | |
66cdef66 | 2638 | return 0; |
0f050d99 GM |
2639 | |
2640 | stat_fail: | |
2641 | #ifdef CONFIG_ZPOOL | |
2642 | zpool_unregister_driver(&zs_zpool_driver); | |
2643 | #endif | |
2644 | notifier_fail: | |
2645 | zs_unregister_cpu_notifier(); | |
1cac41cb MB |
2646 | zsmalloc_unmount(); |
2647 | out: | |
0f050d99 | 2648 | return ret; |
61989a80 | 2649 | } |
61989a80 | 2650 | |
66cdef66 | 2651 | static void __exit zs_exit(void) |
61989a80 | 2652 | { |
66cdef66 GM |
2653 | #ifdef CONFIG_ZPOOL |
2654 | zpool_unregister_driver(&zs_zpool_driver); | |
2655 | #endif | |
1cac41cb | 2656 | zsmalloc_unmount(); |
66cdef66 | 2657 | zs_unregister_cpu_notifier(); |
0f050d99 GM |
2658 | |
2659 | zs_stat_exit(); | |
61989a80 | 2660 | } |
069f101f BH |
2661 | |
2662 | module_init(zs_init); | |
2663 | module_exit(zs_exit); | |
2664 | ||
2665 | MODULE_LICENSE("Dual BSD/GPL"); | |
2666 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); |