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: | |
3783689a | 19 | * page->private: points to zspage |
48b4800a MK |
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. | |
fd854463 | 23 | * page->units: first object offset in a subpage of zspage |
2db51dae NG |
24 | * |
25 | * Usage of struct page flags: | |
26 | * PG_private: identifies the first component page | |
27 | * PG_private2: identifies the last component page | |
48b4800a | 28 | * PG_owner_priv_1: indentifies the huge component page |
2db51dae NG |
29 | * |
30 | */ | |
31 | ||
4abaac9b DS |
32 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
33 | ||
40bf5a9d GM |
34 | #define CREATE_TRACE_POINTS |
35 | ||
61989a80 NG |
36 | #include <linux/module.h> |
37 | #include <linux/kernel.h> | |
312fcae2 | 38 | #include <linux/sched.h> |
61989a80 NG |
39 | #include <linux/bitops.h> |
40 | #include <linux/errno.h> | |
41 | #include <linux/highmem.h> | |
61989a80 NG |
42 | #include <linux/string.h> |
43 | #include <linux/slab.h> | |
44 | #include <asm/tlbflush.h> | |
45 | #include <asm/pgtable.h> | |
46 | #include <linux/cpumask.h> | |
47 | #include <linux/cpu.h> | |
0cbb613f | 48 | #include <linux/vmalloc.h> |
759b26b2 | 49 | #include <linux/preempt.h> |
0959c63f SJ |
50 | #include <linux/spinlock.h> |
51 | #include <linux/types.h> | |
0f050d99 | 52 | #include <linux/debugfs.h> |
bcf1647d | 53 | #include <linux/zsmalloc.h> |
c795779d | 54 | #include <linux/zpool.h> |
48b4800a | 55 | #include <linux/mount.h> |
dd4123f3 | 56 | #include <linux/migrate.h> |
48b4800a | 57 | #include <linux/pagemap.h> |
40bf5a9d | 58 | #include <trace/events/zsmalloc.h> |
48b4800a MK |
59 | |
60 | #define ZSPAGE_MAGIC 0x58 | |
0959c63f SJ |
61 | |
62 | /* | |
63 | * This must be power of 2 and greater than of equal to sizeof(link_free). | |
64 | * These two conditions ensure that any 'struct link_free' itself doesn't | |
65 | * span more than 1 page which avoids complex case of mapping 2 pages simply | |
66 | * to restore link_free pointer values. | |
67 | */ | |
68 | #define ZS_ALIGN 8 | |
69 | ||
70 | /* | |
71 | * A single 'zspage' is composed of up to 2^N discontiguous 0-order (single) | |
72 | * pages. ZS_MAX_ZSPAGE_ORDER defines upper limit on N. | |
73 | */ | |
74 | #define ZS_MAX_ZSPAGE_ORDER 2 | |
75 | #define ZS_MAX_PAGES_PER_ZSPAGE (_AC(1, UL) << ZS_MAX_ZSPAGE_ORDER) | |
76 | ||
2e40e163 MK |
77 | #define ZS_HANDLE_SIZE (sizeof(unsigned long)) |
78 | ||
0959c63f SJ |
79 | /* |
80 | * Object location (<PFN>, <obj_idx>) is encoded as | |
c3e3e88a | 81 | * as single (unsigned long) handle value. |
0959c63f | 82 | * |
bfd093f5 | 83 | * Note that object index <obj_idx> starts from 0. |
0959c63f SJ |
84 | * |
85 | * This is made more complicated by various memory models and PAE. | |
86 | */ | |
87 | ||
88 | #ifndef MAX_PHYSMEM_BITS | |
89 | #ifdef CONFIG_HIGHMEM64G | |
90 | #define MAX_PHYSMEM_BITS 36 | |
91 | #else /* !CONFIG_HIGHMEM64G */ | |
92 | /* | |
93 | * If this definition of MAX_PHYSMEM_BITS is used, OBJ_INDEX_BITS will just | |
94 | * be PAGE_SHIFT | |
95 | */ | |
96 | #define MAX_PHYSMEM_BITS BITS_PER_LONG | |
97 | #endif | |
98 | #endif | |
99 | #define _PFN_BITS (MAX_PHYSMEM_BITS - PAGE_SHIFT) | |
312fcae2 MK |
100 | |
101 | /* | |
102 | * Memory for allocating for handle keeps object position by | |
103 | * encoding <page, obj_idx> and the encoded value has a room | |
104 | * in least bit(ie, look at obj_to_location). | |
105 | * We use the bit to synchronize between object access by | |
106 | * user and migration. | |
107 | */ | |
108 | #define HANDLE_PIN_BIT 0 | |
109 | ||
110 | /* | |
111 | * Head in allocated object should have OBJ_ALLOCATED_TAG | |
112 | * to identify the object was allocated or not. | |
113 | * It's okay to add the status bit in the least bit because | |
114 | * header keeps handle which is 4byte-aligned address so we | |
115 | * have room for two bit at least. | |
116 | */ | |
117 | #define OBJ_ALLOCATED_TAG 1 | |
118 | #define OBJ_TAG_BITS 1 | |
119 | #define OBJ_INDEX_BITS (BITS_PER_LONG - _PFN_BITS - OBJ_TAG_BITS) | |
0959c63f SJ |
120 | #define OBJ_INDEX_MASK ((_AC(1, UL) << OBJ_INDEX_BITS) - 1) |
121 | ||
122 | #define MAX(a, b) ((a) >= (b) ? (a) : (b)) | |
123 | /* ZS_MIN_ALLOC_SIZE must be multiple of ZS_ALIGN */ | |
124 | #define ZS_MIN_ALLOC_SIZE \ | |
125 | MAX(32, (ZS_MAX_PAGES_PER_ZSPAGE << PAGE_SHIFT >> OBJ_INDEX_BITS)) | |
2e40e163 | 126 | /* each chunk includes extra space to keep handle */ |
7b60a685 | 127 | #define ZS_MAX_ALLOC_SIZE PAGE_SIZE |
0959c63f SJ |
128 | |
129 | /* | |
7eb52512 | 130 | * On systems with 4K page size, this gives 255 size classes! There is a |
0959c63f SJ |
131 | * trader-off here: |
132 | * - Large number of size classes is potentially wasteful as free page are | |
133 | * spread across these classes | |
134 | * - Small number of size classes causes large internal fragmentation | |
135 | * - Probably its better to use specific size classes (empirically | |
136 | * determined). NOTE: all those class sizes must be set as multiple of | |
137 | * ZS_ALIGN to make sure link_free itself never has to span 2 pages. | |
138 | * | |
139 | * ZS_MIN_ALLOC_SIZE and ZS_SIZE_CLASS_DELTA must be multiple of ZS_ALIGN | |
140 | * (reason above) | |
141 | */ | |
3783689a | 142 | #define ZS_SIZE_CLASS_DELTA (PAGE_SIZE >> CLASS_BITS) |
0959c63f | 143 | |
0959c63f | 144 | enum fullness_group { |
0959c63f | 145 | ZS_EMPTY, |
48b4800a MK |
146 | ZS_ALMOST_EMPTY, |
147 | ZS_ALMOST_FULL, | |
148 | ZS_FULL, | |
149 | NR_ZS_FULLNESS, | |
0959c63f SJ |
150 | }; |
151 | ||
0f050d99 | 152 | enum zs_stat_type { |
48b4800a MK |
153 | CLASS_EMPTY, |
154 | CLASS_ALMOST_EMPTY, | |
155 | CLASS_ALMOST_FULL, | |
156 | CLASS_FULL, | |
0f050d99 GM |
157 | OBJ_ALLOCATED, |
158 | OBJ_USED, | |
48b4800a | 159 | NR_ZS_STAT_TYPE, |
0f050d99 GM |
160 | }; |
161 | ||
0f050d99 GM |
162 | struct zs_size_stat { |
163 | unsigned long objs[NR_ZS_STAT_TYPE]; | |
164 | }; | |
165 | ||
57244594 SS |
166 | #ifdef CONFIG_ZSMALLOC_STAT |
167 | static struct dentry *zs_stat_root; | |
0f050d99 GM |
168 | #endif |
169 | ||
48b4800a MK |
170 | #ifdef CONFIG_COMPACTION |
171 | static struct vfsmount *zsmalloc_mnt; | |
172 | #endif | |
173 | ||
40f9fb8c MG |
174 | /* |
175 | * number of size_classes | |
176 | */ | |
177 | static int zs_size_classes; | |
178 | ||
0959c63f SJ |
179 | /* |
180 | * We assign a page to ZS_ALMOST_EMPTY fullness group when: | |
181 | * n <= N / f, where | |
182 | * n = number of allocated objects | |
183 | * N = total number of objects zspage can store | |
6dd9737e | 184 | * f = fullness_threshold_frac |
0959c63f SJ |
185 | * |
186 | * Similarly, we assign zspage to: | |
187 | * ZS_ALMOST_FULL when n > N / f | |
188 | * ZS_EMPTY when n == 0 | |
189 | * ZS_FULL when n == N | |
190 | * | |
191 | * (see: fix_fullness_group()) | |
192 | */ | |
193 | static const int fullness_threshold_frac = 4; | |
194 | ||
195 | struct size_class { | |
57244594 | 196 | spinlock_t lock; |
48b4800a | 197 | struct list_head fullness_list[NR_ZS_FULLNESS]; |
0959c63f SJ |
198 | /* |
199 | * Size of objects stored in this class. Must be multiple | |
200 | * of ZS_ALIGN. | |
201 | */ | |
202 | int size; | |
1fc6e27d | 203 | int objs_per_zspage; |
7dfa4612 WY |
204 | /* Number of PAGE_SIZE sized pages to combine to form a 'zspage' */ |
205 | int pages_per_zspage; | |
48b4800a MK |
206 | |
207 | unsigned int index; | |
208 | struct zs_size_stat stats; | |
0959c63f SJ |
209 | }; |
210 | ||
48b4800a MK |
211 | /* huge object: pages_per_zspage == 1 && maxobj_per_zspage == 1 */ |
212 | static void SetPageHugeObject(struct page *page) | |
213 | { | |
214 | SetPageOwnerPriv1(page); | |
215 | } | |
216 | ||
217 | static void ClearPageHugeObject(struct page *page) | |
218 | { | |
219 | ClearPageOwnerPriv1(page); | |
220 | } | |
221 | ||
222 | static int PageHugeObject(struct page *page) | |
223 | { | |
224 | return PageOwnerPriv1(page); | |
225 | } | |
226 | ||
0959c63f SJ |
227 | /* |
228 | * Placed within free objects to form a singly linked list. | |
3783689a | 229 | * For every zspage, zspage->freeobj gives head of this list. |
0959c63f SJ |
230 | * |
231 | * This must be power of 2 and less than or equal to ZS_ALIGN | |
232 | */ | |
233 | struct link_free { | |
2e40e163 MK |
234 | union { |
235 | /* | |
bfd093f5 | 236 | * Free object index; |
2e40e163 MK |
237 | * It's valid for non-allocated object |
238 | */ | |
bfd093f5 | 239 | unsigned long next; |
2e40e163 MK |
240 | /* |
241 | * Handle of allocated object. | |
242 | */ | |
243 | unsigned long handle; | |
244 | }; | |
0959c63f SJ |
245 | }; |
246 | ||
247 | struct zs_pool { | |
6f3526d6 | 248 | const char *name; |
0f050d99 | 249 | |
40f9fb8c | 250 | struct size_class **size_class; |
2e40e163 | 251 | struct kmem_cache *handle_cachep; |
3783689a | 252 | struct kmem_cache *zspage_cachep; |
0959c63f | 253 | |
13de8933 | 254 | atomic_long_t pages_allocated; |
0f050d99 | 255 | |
7d3f3938 | 256 | struct zs_pool_stats stats; |
ab9d306d SS |
257 | |
258 | /* Compact classes */ | |
259 | struct shrinker shrinker; | |
260 | /* | |
261 | * To signify that register_shrinker() was successful | |
262 | * and unregister_shrinker() will not Oops. | |
263 | */ | |
264 | bool shrinker_enabled; | |
0f050d99 GM |
265 | #ifdef CONFIG_ZSMALLOC_STAT |
266 | struct dentry *stat_dentry; | |
267 | #endif | |
48b4800a MK |
268 | #ifdef CONFIG_COMPACTION |
269 | struct inode *inode; | |
270 | struct work_struct free_work; | |
271 | #endif | |
0959c63f | 272 | }; |
61989a80 NG |
273 | |
274 | /* | |
275 | * A zspage's class index and fullness group | |
276 | * are encoded in its (first)page->mapping | |
277 | */ | |
3783689a MK |
278 | #define FULLNESS_BITS 2 |
279 | #define CLASS_BITS 8 | |
48b4800a MK |
280 | #define ISOLATED_BITS 3 |
281 | #define MAGIC_VAL_BITS 8 | |
4f42047b | 282 | |
3783689a MK |
283 | struct zspage { |
284 | struct { | |
285 | unsigned int fullness:FULLNESS_BITS; | |
286 | unsigned int class:CLASS_BITS; | |
48b4800a MK |
287 | unsigned int isolated:ISOLATED_BITS; |
288 | unsigned int magic:MAGIC_VAL_BITS; | |
3783689a MK |
289 | }; |
290 | unsigned int inuse; | |
bfd093f5 | 291 | unsigned int freeobj; |
3783689a MK |
292 | struct page *first_page; |
293 | struct list_head list; /* fullness list */ | |
48b4800a MK |
294 | #ifdef CONFIG_COMPACTION |
295 | rwlock_t lock; | |
296 | #endif | |
3783689a | 297 | }; |
61989a80 | 298 | |
f553646a | 299 | struct mapping_area { |
1b945aee | 300 | #ifdef CONFIG_PGTABLE_MAPPING |
f553646a SJ |
301 | struct vm_struct *vm; /* vm area for mapping object that span pages */ |
302 | #else | |
303 | char *vm_buf; /* copy buffer for objects that span pages */ | |
304 | #endif | |
305 | char *vm_addr; /* address of kmap_atomic()'ed pages */ | |
306 | enum zs_mapmode vm_mm; /* mapping mode */ | |
307 | }; | |
308 | ||
48b4800a MK |
309 | #ifdef CONFIG_COMPACTION |
310 | static int zs_register_migration(struct zs_pool *pool); | |
311 | static void zs_unregister_migration(struct zs_pool *pool); | |
312 | static void migrate_lock_init(struct zspage *zspage); | |
313 | static void migrate_read_lock(struct zspage *zspage); | |
314 | static void migrate_read_unlock(struct zspage *zspage); | |
315 | static void kick_deferred_free(struct zs_pool *pool); | |
316 | static void init_deferred_free(struct zs_pool *pool); | |
317 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage); | |
318 | #else | |
319 | static int zsmalloc_mount(void) { return 0; } | |
320 | static void zsmalloc_unmount(void) {} | |
321 | static int zs_register_migration(struct zs_pool *pool) { return 0; } | |
322 | static void zs_unregister_migration(struct zs_pool *pool) {} | |
323 | static void migrate_lock_init(struct zspage *zspage) {} | |
324 | static void migrate_read_lock(struct zspage *zspage) {} | |
325 | static void migrate_read_unlock(struct zspage *zspage) {} | |
326 | static void kick_deferred_free(struct zs_pool *pool) {} | |
327 | static void init_deferred_free(struct zs_pool *pool) {} | |
328 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage) {} | |
329 | #endif | |
330 | ||
3783689a | 331 | static int create_cache(struct zs_pool *pool) |
2e40e163 MK |
332 | { |
333 | pool->handle_cachep = kmem_cache_create("zs_handle", ZS_HANDLE_SIZE, | |
334 | 0, 0, NULL); | |
3783689a MK |
335 | if (!pool->handle_cachep) |
336 | return 1; | |
337 | ||
338 | pool->zspage_cachep = kmem_cache_create("zspage", sizeof(struct zspage), | |
339 | 0, 0, NULL); | |
340 | if (!pool->zspage_cachep) { | |
341 | kmem_cache_destroy(pool->handle_cachep); | |
342 | pool->handle_cachep = NULL; | |
343 | return 1; | |
344 | } | |
345 | ||
346 | return 0; | |
2e40e163 MK |
347 | } |
348 | ||
3783689a | 349 | static void destroy_cache(struct zs_pool *pool) |
2e40e163 | 350 | { |
cd10add0 | 351 | kmem_cache_destroy(pool->handle_cachep); |
3783689a | 352 | kmem_cache_destroy(pool->zspage_cachep); |
2e40e163 MK |
353 | } |
354 | ||
3783689a | 355 | static unsigned long cache_alloc_handle(struct zs_pool *pool, gfp_t gfp) |
2e40e163 MK |
356 | { |
357 | return (unsigned long)kmem_cache_alloc(pool->handle_cachep, | |
48b4800a | 358 | gfp & ~(__GFP_HIGHMEM|__GFP_MOVABLE)); |
2e40e163 MK |
359 | } |
360 | ||
3783689a | 361 | static void cache_free_handle(struct zs_pool *pool, unsigned long handle) |
2e40e163 MK |
362 | { |
363 | kmem_cache_free(pool->handle_cachep, (void *)handle); | |
364 | } | |
365 | ||
3783689a MK |
366 | static struct zspage *cache_alloc_zspage(struct zs_pool *pool, gfp_t flags) |
367 | { | |
48b4800a MK |
368 | return kmem_cache_alloc(pool->zspage_cachep, |
369 | flags & ~(__GFP_HIGHMEM|__GFP_MOVABLE)); | |
3783689a MK |
370 | }; |
371 | ||
372 | static void cache_free_zspage(struct zs_pool *pool, struct zspage *zspage) | |
373 | { | |
374 | kmem_cache_free(pool->zspage_cachep, zspage); | |
375 | } | |
376 | ||
2e40e163 MK |
377 | static void record_obj(unsigned long handle, unsigned long obj) |
378 | { | |
c102f07c JL |
379 | /* |
380 | * lsb of @obj represents handle lock while other bits | |
381 | * represent object value the handle is pointing so | |
382 | * updating shouldn't do store tearing. | |
383 | */ | |
384 | WRITE_ONCE(*(unsigned long *)handle, obj); | |
2e40e163 MK |
385 | } |
386 | ||
c795779d DS |
387 | /* zpool driver */ |
388 | ||
389 | #ifdef CONFIG_ZPOOL | |
390 | ||
6f3526d6 | 391 | static void *zs_zpool_create(const char *name, gfp_t gfp, |
78672779 | 392 | const struct zpool_ops *zpool_ops, |
479305fd | 393 | struct zpool *zpool) |
c795779d | 394 | { |
d0d8da2d SS |
395 | /* |
396 | * Ignore global gfp flags: zs_malloc() may be invoked from | |
397 | * different contexts and its caller must provide a valid | |
398 | * gfp mask. | |
399 | */ | |
400 | return zs_create_pool(name); | |
c795779d DS |
401 | } |
402 | ||
403 | static void zs_zpool_destroy(void *pool) | |
404 | { | |
405 | zs_destroy_pool(pool); | |
406 | } | |
407 | ||
408 | static int zs_zpool_malloc(void *pool, size_t size, gfp_t gfp, | |
409 | unsigned long *handle) | |
410 | { | |
d0d8da2d | 411 | *handle = zs_malloc(pool, size, gfp); |
c795779d DS |
412 | return *handle ? 0 : -1; |
413 | } | |
414 | static void zs_zpool_free(void *pool, unsigned long handle) | |
415 | { | |
416 | zs_free(pool, handle); | |
417 | } | |
418 | ||
419 | static int zs_zpool_shrink(void *pool, unsigned int pages, | |
420 | unsigned int *reclaimed) | |
421 | { | |
422 | return -EINVAL; | |
423 | } | |
424 | ||
425 | static void *zs_zpool_map(void *pool, unsigned long handle, | |
426 | enum zpool_mapmode mm) | |
427 | { | |
428 | enum zs_mapmode zs_mm; | |
429 | ||
430 | switch (mm) { | |
431 | case ZPOOL_MM_RO: | |
432 | zs_mm = ZS_MM_RO; | |
433 | break; | |
434 | case ZPOOL_MM_WO: | |
435 | zs_mm = ZS_MM_WO; | |
436 | break; | |
437 | case ZPOOL_MM_RW: /* fallthru */ | |
438 | default: | |
439 | zs_mm = ZS_MM_RW; | |
440 | break; | |
441 | } | |
442 | ||
443 | return zs_map_object(pool, handle, zs_mm); | |
444 | } | |
445 | static void zs_zpool_unmap(void *pool, unsigned long handle) | |
446 | { | |
447 | zs_unmap_object(pool, handle); | |
448 | } | |
449 | ||
450 | static u64 zs_zpool_total_size(void *pool) | |
451 | { | |
722cdc17 | 452 | return zs_get_total_pages(pool) << PAGE_SHIFT; |
c795779d DS |
453 | } |
454 | ||
455 | static struct zpool_driver zs_zpool_driver = { | |
456 | .type = "zsmalloc", | |
457 | .owner = THIS_MODULE, | |
458 | .create = zs_zpool_create, | |
459 | .destroy = zs_zpool_destroy, | |
460 | .malloc = zs_zpool_malloc, | |
461 | .free = zs_zpool_free, | |
462 | .shrink = zs_zpool_shrink, | |
463 | .map = zs_zpool_map, | |
464 | .unmap = zs_zpool_unmap, | |
465 | .total_size = zs_zpool_total_size, | |
466 | }; | |
467 | ||
137f8cff | 468 | MODULE_ALIAS("zpool-zsmalloc"); |
c795779d DS |
469 | #endif /* CONFIG_ZPOOL */ |
470 | ||
61989a80 NG |
471 | /* per-cpu VM mapping areas for zspage accesses that cross page boundaries */ |
472 | static DEFINE_PER_CPU(struct mapping_area, zs_map_area); | |
473 | ||
48b4800a MK |
474 | static bool is_zspage_isolated(struct zspage *zspage) |
475 | { | |
476 | return zspage->isolated; | |
477 | } | |
478 | ||
61989a80 NG |
479 | static int is_first_page(struct page *page) |
480 | { | |
a27545bf | 481 | return PagePrivate(page); |
61989a80 NG |
482 | } |
483 | ||
48b4800a | 484 | /* Protected by class->lock */ |
3783689a | 485 | static inline int get_zspage_inuse(struct zspage *zspage) |
4f42047b | 486 | { |
3783689a | 487 | return zspage->inuse; |
4f42047b MK |
488 | } |
489 | ||
3783689a | 490 | static inline void set_zspage_inuse(struct zspage *zspage, int val) |
4f42047b | 491 | { |
3783689a | 492 | zspage->inuse = val; |
4f42047b MK |
493 | } |
494 | ||
3783689a | 495 | static inline void mod_zspage_inuse(struct zspage *zspage, int val) |
4f42047b | 496 | { |
3783689a | 497 | zspage->inuse += val; |
4f42047b MK |
498 | } |
499 | ||
48b4800a | 500 | static inline struct page *get_first_page(struct zspage *zspage) |
4f42047b | 501 | { |
48b4800a | 502 | struct page *first_page = zspage->first_page; |
3783689a | 503 | |
48b4800a MK |
504 | VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); |
505 | return first_page; | |
4f42047b MK |
506 | } |
507 | ||
48b4800a | 508 | static inline int get_first_obj_offset(struct page *page) |
4f42047b | 509 | { |
48b4800a MK |
510 | return page->units; |
511 | } | |
3783689a | 512 | |
48b4800a MK |
513 | static inline void set_first_obj_offset(struct page *page, int offset) |
514 | { | |
515 | page->units = offset; | |
4f42047b MK |
516 | } |
517 | ||
bfd093f5 | 518 | static inline unsigned int get_freeobj(struct zspage *zspage) |
4f42047b | 519 | { |
bfd093f5 | 520 | return zspage->freeobj; |
4f42047b MK |
521 | } |
522 | ||
bfd093f5 | 523 | static inline void set_freeobj(struct zspage *zspage, unsigned int obj) |
4f42047b | 524 | { |
bfd093f5 | 525 | zspage->freeobj = obj; |
4f42047b MK |
526 | } |
527 | ||
3783689a | 528 | static void get_zspage_mapping(struct zspage *zspage, |
a4209467 | 529 | unsigned int *class_idx, |
61989a80 NG |
530 | enum fullness_group *fullness) |
531 | { | |
48b4800a MK |
532 | BUG_ON(zspage->magic != ZSPAGE_MAGIC); |
533 | ||
3783689a MK |
534 | *fullness = zspage->fullness; |
535 | *class_idx = zspage->class; | |
61989a80 NG |
536 | } |
537 | ||
3783689a | 538 | static void set_zspage_mapping(struct zspage *zspage, |
a4209467 | 539 | unsigned int class_idx, |
61989a80 NG |
540 | enum fullness_group fullness) |
541 | { | |
3783689a MK |
542 | zspage->class = class_idx; |
543 | zspage->fullness = fullness; | |
61989a80 NG |
544 | } |
545 | ||
c3e3e88a NC |
546 | /* |
547 | * zsmalloc divides the pool into various size classes where each | |
548 | * class maintains a list of zspages where each zspage is divided | |
549 | * into equal sized chunks. Each allocation falls into one of these | |
550 | * classes depending on its size. This function returns index of the | |
551 | * size class which has chunk size big enough to hold the give size. | |
552 | */ | |
61989a80 NG |
553 | static int get_size_class_index(int size) |
554 | { | |
555 | int idx = 0; | |
556 | ||
557 | if (likely(size > ZS_MIN_ALLOC_SIZE)) | |
558 | idx = DIV_ROUND_UP(size - ZS_MIN_ALLOC_SIZE, | |
559 | ZS_SIZE_CLASS_DELTA); | |
560 | ||
7b60a685 | 561 | return min(zs_size_classes - 1, idx); |
61989a80 NG |
562 | } |
563 | ||
248ca1b0 MK |
564 | static inline void zs_stat_inc(struct size_class *class, |
565 | enum zs_stat_type type, unsigned long cnt) | |
566 | { | |
48b4800a | 567 | class->stats.objs[type] += cnt; |
248ca1b0 MK |
568 | } |
569 | ||
570 | static inline void zs_stat_dec(struct size_class *class, | |
571 | enum zs_stat_type type, unsigned long cnt) | |
572 | { | |
48b4800a | 573 | class->stats.objs[type] -= cnt; |
248ca1b0 MK |
574 | } |
575 | ||
576 | static inline unsigned long zs_stat_get(struct size_class *class, | |
577 | enum zs_stat_type type) | |
578 | { | |
48b4800a | 579 | return class->stats.objs[type]; |
248ca1b0 MK |
580 | } |
581 | ||
57244594 SS |
582 | #ifdef CONFIG_ZSMALLOC_STAT |
583 | ||
4abaac9b | 584 | static void __init zs_stat_init(void) |
248ca1b0 | 585 | { |
4abaac9b DS |
586 | if (!debugfs_initialized()) { |
587 | pr_warn("debugfs not available, stat dir not created\n"); | |
588 | return; | |
589 | } | |
248ca1b0 MK |
590 | |
591 | zs_stat_root = debugfs_create_dir("zsmalloc", NULL); | |
592 | if (!zs_stat_root) | |
4abaac9b | 593 | pr_warn("debugfs 'zsmalloc' stat dir creation failed\n"); |
248ca1b0 MK |
594 | } |
595 | ||
596 | static void __exit zs_stat_exit(void) | |
597 | { | |
598 | debugfs_remove_recursive(zs_stat_root); | |
599 | } | |
600 | ||
1120ed54 SS |
601 | static unsigned long zs_can_compact(struct size_class *class); |
602 | ||
248ca1b0 MK |
603 | static int zs_stats_size_show(struct seq_file *s, void *v) |
604 | { | |
605 | int i; | |
606 | struct zs_pool *pool = s->private; | |
607 | struct size_class *class; | |
608 | int objs_per_zspage; | |
609 | unsigned long class_almost_full, class_almost_empty; | |
1120ed54 | 610 | unsigned long obj_allocated, obj_used, pages_used, freeable; |
248ca1b0 MK |
611 | unsigned long total_class_almost_full = 0, total_class_almost_empty = 0; |
612 | unsigned long total_objs = 0, total_used_objs = 0, total_pages = 0; | |
1120ed54 | 613 | unsigned long total_freeable = 0; |
248ca1b0 | 614 | |
1120ed54 | 615 | seq_printf(s, " %5s %5s %11s %12s %13s %10s %10s %16s %8s\n", |
248ca1b0 MK |
616 | "class", "size", "almost_full", "almost_empty", |
617 | "obj_allocated", "obj_used", "pages_used", | |
1120ed54 | 618 | "pages_per_zspage", "freeable"); |
248ca1b0 MK |
619 | |
620 | for (i = 0; i < zs_size_classes; i++) { | |
621 | class = pool->size_class[i]; | |
622 | ||
623 | if (class->index != i) | |
624 | continue; | |
625 | ||
626 | spin_lock(&class->lock); | |
627 | class_almost_full = zs_stat_get(class, CLASS_ALMOST_FULL); | |
628 | class_almost_empty = zs_stat_get(class, CLASS_ALMOST_EMPTY); | |
629 | obj_allocated = zs_stat_get(class, OBJ_ALLOCATED); | |
630 | obj_used = zs_stat_get(class, OBJ_USED); | |
1120ed54 | 631 | freeable = zs_can_compact(class); |
248ca1b0 MK |
632 | spin_unlock(&class->lock); |
633 | ||
b4fd07a0 | 634 | objs_per_zspage = class->objs_per_zspage; |
248ca1b0 MK |
635 | pages_used = obj_allocated / objs_per_zspage * |
636 | class->pages_per_zspage; | |
637 | ||
1120ed54 SS |
638 | seq_printf(s, " %5u %5u %11lu %12lu %13lu" |
639 | " %10lu %10lu %16d %8lu\n", | |
248ca1b0 MK |
640 | i, class->size, class_almost_full, class_almost_empty, |
641 | obj_allocated, obj_used, pages_used, | |
1120ed54 | 642 | class->pages_per_zspage, freeable); |
248ca1b0 MK |
643 | |
644 | total_class_almost_full += class_almost_full; | |
645 | total_class_almost_empty += class_almost_empty; | |
646 | total_objs += obj_allocated; | |
647 | total_used_objs += obj_used; | |
648 | total_pages += pages_used; | |
1120ed54 | 649 | total_freeable += freeable; |
248ca1b0 MK |
650 | } |
651 | ||
652 | seq_puts(s, "\n"); | |
1120ed54 | 653 | seq_printf(s, " %5s %5s %11lu %12lu %13lu %10lu %10lu %16s %8lu\n", |
248ca1b0 MK |
654 | "Total", "", total_class_almost_full, |
655 | total_class_almost_empty, total_objs, | |
1120ed54 | 656 | total_used_objs, total_pages, "", total_freeable); |
248ca1b0 MK |
657 | |
658 | return 0; | |
659 | } | |
660 | ||
661 | static int zs_stats_size_open(struct inode *inode, struct file *file) | |
662 | { | |
663 | return single_open(file, zs_stats_size_show, inode->i_private); | |
664 | } | |
665 | ||
666 | static const struct file_operations zs_stat_size_ops = { | |
667 | .open = zs_stats_size_open, | |
668 | .read = seq_read, | |
669 | .llseek = seq_lseek, | |
670 | .release = single_release, | |
671 | }; | |
672 | ||
d34f6157 | 673 | static void zs_pool_stat_create(struct zs_pool *pool, const char *name) |
248ca1b0 MK |
674 | { |
675 | struct dentry *entry; | |
676 | ||
4abaac9b DS |
677 | if (!zs_stat_root) { |
678 | pr_warn("no root stat dir, not creating <%s> stat dir\n", name); | |
d34f6157 | 679 | return; |
4abaac9b | 680 | } |
248ca1b0 MK |
681 | |
682 | entry = debugfs_create_dir(name, zs_stat_root); | |
683 | if (!entry) { | |
684 | pr_warn("debugfs dir <%s> creation failed\n", name); | |
d34f6157 | 685 | return; |
248ca1b0 MK |
686 | } |
687 | pool->stat_dentry = entry; | |
688 | ||
689 | entry = debugfs_create_file("classes", S_IFREG | S_IRUGO, | |
690 | pool->stat_dentry, pool, &zs_stat_size_ops); | |
691 | if (!entry) { | |
692 | pr_warn("%s: debugfs file entry <%s> creation failed\n", | |
693 | name, "classes"); | |
4abaac9b DS |
694 | debugfs_remove_recursive(pool->stat_dentry); |
695 | pool->stat_dentry = NULL; | |
248ca1b0 | 696 | } |
248ca1b0 MK |
697 | } |
698 | ||
699 | static void zs_pool_stat_destroy(struct zs_pool *pool) | |
700 | { | |
701 | debugfs_remove_recursive(pool->stat_dentry); | |
702 | } | |
703 | ||
704 | #else /* CONFIG_ZSMALLOC_STAT */ | |
4abaac9b | 705 | static void __init zs_stat_init(void) |
248ca1b0 | 706 | { |
248ca1b0 MK |
707 | } |
708 | ||
709 | static void __exit zs_stat_exit(void) | |
710 | { | |
711 | } | |
712 | ||
d34f6157 | 713 | static inline void zs_pool_stat_create(struct zs_pool *pool, const char *name) |
248ca1b0 | 714 | { |
248ca1b0 MK |
715 | } |
716 | ||
717 | static inline void zs_pool_stat_destroy(struct zs_pool *pool) | |
718 | { | |
719 | } | |
248ca1b0 MK |
720 | #endif |
721 | ||
48b4800a | 722 | |
c3e3e88a NC |
723 | /* |
724 | * For each size class, zspages are divided into different groups | |
725 | * depending on how "full" they are. This was done so that we could | |
726 | * easily find empty or nearly empty zspages when we try to shrink | |
727 | * the pool (not yet implemented). This function returns fullness | |
728 | * status of the given page. | |
729 | */ | |
1fc6e27d | 730 | static enum fullness_group get_fullness_group(struct size_class *class, |
3783689a | 731 | struct zspage *zspage) |
61989a80 | 732 | { |
1fc6e27d | 733 | int inuse, objs_per_zspage; |
61989a80 | 734 | enum fullness_group fg; |
830e4bc5 | 735 | |
3783689a | 736 | inuse = get_zspage_inuse(zspage); |
1fc6e27d | 737 | objs_per_zspage = class->objs_per_zspage; |
61989a80 NG |
738 | |
739 | if (inuse == 0) | |
740 | fg = ZS_EMPTY; | |
1fc6e27d | 741 | else if (inuse == objs_per_zspage) |
61989a80 | 742 | fg = ZS_FULL; |
1fc6e27d | 743 | else if (inuse <= 3 * objs_per_zspage / fullness_threshold_frac) |
61989a80 NG |
744 | fg = ZS_ALMOST_EMPTY; |
745 | else | |
746 | fg = ZS_ALMOST_FULL; | |
747 | ||
748 | return fg; | |
749 | } | |
750 | ||
c3e3e88a NC |
751 | /* |
752 | * Each size class maintains various freelists and zspages are assigned | |
753 | * to one of these freelists based on the number of live objects they | |
754 | * have. This functions inserts the given zspage into the freelist | |
755 | * identified by <class, fullness_group>. | |
756 | */ | |
251cbb95 | 757 | static void insert_zspage(struct size_class *class, |
3783689a MK |
758 | struct zspage *zspage, |
759 | enum fullness_group fullness) | |
61989a80 | 760 | { |
3783689a | 761 | struct zspage *head; |
61989a80 | 762 | |
48b4800a | 763 | zs_stat_inc(class, fullness, 1); |
3783689a MK |
764 | head = list_first_entry_or_null(&class->fullness_list[fullness], |
765 | struct zspage, list); | |
58f17117 | 766 | /* |
3783689a MK |
767 | * We want to see more ZS_FULL pages and less almost empty/full. |
768 | * Put pages with higher ->inuse first. | |
58f17117 | 769 | */ |
3783689a MK |
770 | if (head) { |
771 | if (get_zspage_inuse(zspage) < get_zspage_inuse(head)) { | |
772 | list_add(&zspage->list, &head->list); | |
773 | return; | |
774 | } | |
775 | } | |
776 | list_add(&zspage->list, &class->fullness_list[fullness]); | |
61989a80 NG |
777 | } |
778 | ||
c3e3e88a NC |
779 | /* |
780 | * This function removes the given zspage from the freelist identified | |
781 | * by <class, fullness_group>. | |
782 | */ | |
251cbb95 | 783 | static void remove_zspage(struct size_class *class, |
3783689a MK |
784 | struct zspage *zspage, |
785 | enum fullness_group fullness) | |
61989a80 | 786 | { |
3783689a | 787 | VM_BUG_ON(list_empty(&class->fullness_list[fullness])); |
48b4800a | 788 | VM_BUG_ON(is_zspage_isolated(zspage)); |
61989a80 | 789 | |
3783689a | 790 | list_del_init(&zspage->list); |
48b4800a | 791 | zs_stat_dec(class, fullness, 1); |
61989a80 NG |
792 | } |
793 | ||
c3e3e88a NC |
794 | /* |
795 | * Each size class maintains zspages in different fullness groups depending | |
796 | * on the number of live objects they contain. When allocating or freeing | |
797 | * objects, the fullness status of the page can change, say, from ALMOST_FULL | |
798 | * to ALMOST_EMPTY when freeing an object. This function checks if such | |
799 | * a status change has occurred for the given page and accordingly moves the | |
800 | * page from the freelist of the old fullness group to that of the new | |
801 | * fullness group. | |
802 | */ | |
c7806261 | 803 | static enum fullness_group fix_fullness_group(struct size_class *class, |
3783689a | 804 | struct zspage *zspage) |
61989a80 NG |
805 | { |
806 | int class_idx; | |
61989a80 NG |
807 | enum fullness_group currfg, newfg; |
808 | ||
3783689a MK |
809 | get_zspage_mapping(zspage, &class_idx, &currfg); |
810 | newfg = get_fullness_group(class, zspage); | |
61989a80 NG |
811 | if (newfg == currfg) |
812 | goto out; | |
813 | ||
48b4800a MK |
814 | if (!is_zspage_isolated(zspage)) { |
815 | remove_zspage(class, zspage, currfg); | |
816 | insert_zspage(class, zspage, newfg); | |
817 | } | |
818 | ||
3783689a | 819 | set_zspage_mapping(zspage, class_idx, newfg); |
61989a80 NG |
820 | |
821 | out: | |
822 | return newfg; | |
823 | } | |
824 | ||
825 | /* | |
826 | * We have to decide on how many pages to link together | |
827 | * to form a zspage for each size class. This is important | |
828 | * to reduce wastage due to unusable space left at end of | |
829 | * each zspage which is given as: | |
888fa374 YX |
830 | * wastage = Zp % class_size |
831 | * usage = Zp - wastage | |
61989a80 NG |
832 | * where Zp = zspage size = k * PAGE_SIZE where k = 1, 2, ... |
833 | * | |
834 | * For example, for size class of 3/8 * PAGE_SIZE, we should | |
835 | * link together 3 PAGE_SIZE sized pages to form a zspage | |
836 | * since then we can perfectly fit in 8 such objects. | |
837 | */ | |
2e3b6154 | 838 | static int get_pages_per_zspage(int class_size) |
61989a80 NG |
839 | { |
840 | int i, max_usedpc = 0; | |
841 | /* zspage order which gives maximum used size per KB */ | |
842 | int max_usedpc_order = 1; | |
843 | ||
84d4faab | 844 | for (i = 1; i <= ZS_MAX_PAGES_PER_ZSPAGE; i++) { |
61989a80 NG |
845 | int zspage_size; |
846 | int waste, usedpc; | |
847 | ||
848 | zspage_size = i * PAGE_SIZE; | |
849 | waste = zspage_size % class_size; | |
850 | usedpc = (zspage_size - waste) * 100 / zspage_size; | |
851 | ||
852 | if (usedpc > max_usedpc) { | |
853 | max_usedpc = usedpc; | |
854 | max_usedpc_order = i; | |
855 | } | |
856 | } | |
857 | ||
858 | return max_usedpc_order; | |
859 | } | |
860 | ||
3783689a | 861 | static struct zspage *get_zspage(struct page *page) |
61989a80 | 862 | { |
48b4800a MK |
863 | struct zspage *zspage = (struct zspage *)page->private; |
864 | ||
865 | BUG_ON(zspage->magic != ZSPAGE_MAGIC); | |
866 | return zspage; | |
61989a80 NG |
867 | } |
868 | ||
869 | static struct page *get_next_page(struct page *page) | |
870 | { | |
48b4800a MK |
871 | if (unlikely(PageHugeObject(page))) |
872 | return NULL; | |
873 | ||
874 | return page->freelist; | |
61989a80 NG |
875 | } |
876 | ||
bfd093f5 MK |
877 | /** |
878 | * obj_to_location - get (<page>, <obj_idx>) from encoded object value | |
879 | * @page: page object resides in zspage | |
880 | * @obj_idx: object index | |
67296874 | 881 | */ |
bfd093f5 MK |
882 | static void obj_to_location(unsigned long obj, struct page **page, |
883 | unsigned int *obj_idx) | |
61989a80 | 884 | { |
bfd093f5 MK |
885 | obj >>= OBJ_TAG_BITS; |
886 | *page = pfn_to_page(obj >> OBJ_INDEX_BITS); | |
887 | *obj_idx = (obj & OBJ_INDEX_MASK); | |
888 | } | |
61989a80 | 889 | |
bfd093f5 MK |
890 | /** |
891 | * location_to_obj - get obj value encoded from (<page>, <obj_idx>) | |
892 | * @page: page object resides in zspage | |
893 | * @obj_idx: object index | |
894 | */ | |
895 | static unsigned long location_to_obj(struct page *page, unsigned int obj_idx) | |
896 | { | |
897 | unsigned long obj; | |
61989a80 | 898 | |
312fcae2 | 899 | obj = page_to_pfn(page) << OBJ_INDEX_BITS; |
bfd093f5 | 900 | obj |= obj_idx & OBJ_INDEX_MASK; |
312fcae2 | 901 | obj <<= OBJ_TAG_BITS; |
61989a80 | 902 | |
bfd093f5 | 903 | return obj; |
61989a80 NG |
904 | } |
905 | ||
2e40e163 MK |
906 | static unsigned long handle_to_obj(unsigned long handle) |
907 | { | |
908 | return *(unsigned long *)handle; | |
909 | } | |
910 | ||
48b4800a | 911 | static unsigned long obj_to_head(struct page *page, void *obj) |
312fcae2 | 912 | { |
48b4800a | 913 | if (unlikely(PageHugeObject(page))) { |
830e4bc5 | 914 | VM_BUG_ON_PAGE(!is_first_page(page), page); |
3783689a | 915 | return page->index; |
7b60a685 MK |
916 | } else |
917 | return *(unsigned long *)obj; | |
312fcae2 MK |
918 | } |
919 | ||
48b4800a MK |
920 | static inline int testpin_tag(unsigned long handle) |
921 | { | |
922 | return bit_spin_is_locked(HANDLE_PIN_BIT, (unsigned long *)handle); | |
923 | } | |
924 | ||
312fcae2 MK |
925 | static inline int trypin_tag(unsigned long handle) |
926 | { | |
1b8320b6 | 927 | return bit_spin_trylock(HANDLE_PIN_BIT, (unsigned long *)handle); |
312fcae2 MK |
928 | } |
929 | ||
930 | static void pin_tag(unsigned long handle) | |
931 | { | |
1b8320b6 | 932 | bit_spin_lock(HANDLE_PIN_BIT, (unsigned long *)handle); |
312fcae2 MK |
933 | } |
934 | ||
935 | static void unpin_tag(unsigned long handle) | |
936 | { | |
1b8320b6 | 937 | bit_spin_unlock(HANDLE_PIN_BIT, (unsigned long *)handle); |
312fcae2 MK |
938 | } |
939 | ||
f4477e90 NG |
940 | static void reset_page(struct page *page) |
941 | { | |
48b4800a | 942 | __ClearPageMovable(page); |
18fd06bf GM |
943 | ClearPagePrivate(page); |
944 | ClearPagePrivate2(page); | |
f4477e90 | 945 | set_page_private(page, 0); |
48b4800a MK |
946 | page_mapcount_reset(page); |
947 | ClearPageHugeObject(page); | |
948 | page->freelist = NULL; | |
949 | } | |
950 | ||
951 | /* | |
952 | * To prevent zspage destroy during migration, zspage freeing should | |
953 | * hold locks of all pages in the zspage. | |
954 | */ | |
955 | void lock_zspage(struct zspage *zspage) | |
956 | { | |
957 | struct page *page = get_first_page(zspage); | |
958 | ||
959 | do { | |
960 | lock_page(page); | |
961 | } while ((page = get_next_page(page)) != NULL); | |
962 | } | |
963 | ||
964 | int trylock_zspage(struct zspage *zspage) | |
965 | { | |
966 | struct page *cursor, *fail; | |
967 | ||
968 | for (cursor = get_first_page(zspage); cursor != NULL; cursor = | |
969 | get_next_page(cursor)) { | |
970 | if (!trylock_page(cursor)) { | |
971 | fail = cursor; | |
972 | goto unlock; | |
973 | } | |
974 | } | |
975 | ||
976 | return 1; | |
977 | unlock: | |
978 | for (cursor = get_first_page(zspage); cursor != fail; cursor = | |
979 | get_next_page(cursor)) | |
980 | unlock_page(cursor); | |
981 | ||
982 | return 0; | |
f4477e90 NG |
983 | } |
984 | ||
48b4800a MK |
985 | static void __free_zspage(struct zs_pool *pool, struct size_class *class, |
986 | struct zspage *zspage) | |
61989a80 | 987 | { |
3783689a | 988 | struct page *page, *next; |
48b4800a MK |
989 | enum fullness_group fg; |
990 | unsigned int class_idx; | |
991 | ||
992 | get_zspage_mapping(zspage, &class_idx, &fg); | |
993 | ||
994 | assert_spin_locked(&class->lock); | |
61989a80 | 995 | |
3783689a | 996 | VM_BUG_ON(get_zspage_inuse(zspage)); |
48b4800a | 997 | VM_BUG_ON(fg != ZS_EMPTY); |
61989a80 | 998 | |
48b4800a | 999 | next = page = get_first_page(zspage); |
3783689a | 1000 | do { |
48b4800a MK |
1001 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1002 | next = get_next_page(page); | |
3783689a | 1003 | reset_page(page); |
48b4800a | 1004 | unlock_page(page); |
91537fee | 1005 | dec_zone_page_state(page, NR_ZSPAGES); |
3783689a MK |
1006 | put_page(page); |
1007 | page = next; | |
1008 | } while (page != NULL); | |
61989a80 | 1009 | |
3783689a | 1010 | cache_free_zspage(pool, zspage); |
48b4800a | 1011 | |
b4fd07a0 | 1012 | zs_stat_dec(class, OBJ_ALLOCATED, class->objs_per_zspage); |
48b4800a MK |
1013 | atomic_long_sub(class->pages_per_zspage, |
1014 | &pool->pages_allocated); | |
1015 | } | |
1016 | ||
1017 | static void free_zspage(struct zs_pool *pool, struct size_class *class, | |
1018 | struct zspage *zspage) | |
1019 | { | |
1020 | VM_BUG_ON(get_zspage_inuse(zspage)); | |
1021 | VM_BUG_ON(list_empty(&zspage->list)); | |
1022 | ||
1023 | if (!trylock_zspage(zspage)) { | |
1024 | kick_deferred_free(pool); | |
1025 | return; | |
1026 | } | |
1027 | ||
1028 | remove_zspage(class, zspage, ZS_EMPTY); | |
1029 | __free_zspage(pool, class, zspage); | |
61989a80 NG |
1030 | } |
1031 | ||
1032 | /* Initialize a newly allocated zspage */ | |
3783689a | 1033 | static void init_zspage(struct size_class *class, struct zspage *zspage) |
61989a80 | 1034 | { |
bfd093f5 | 1035 | unsigned int freeobj = 1; |
61989a80 | 1036 | unsigned long off = 0; |
48b4800a | 1037 | struct page *page = get_first_page(zspage); |
830e4bc5 | 1038 | |
61989a80 NG |
1039 | while (page) { |
1040 | struct page *next_page; | |
1041 | struct link_free *link; | |
af4ee5e9 | 1042 | void *vaddr; |
61989a80 | 1043 | |
3783689a | 1044 | set_first_obj_offset(page, off); |
61989a80 | 1045 | |
af4ee5e9 MK |
1046 | vaddr = kmap_atomic(page); |
1047 | link = (struct link_free *)vaddr + off / sizeof(*link); | |
5538c562 DS |
1048 | |
1049 | while ((off += class->size) < PAGE_SIZE) { | |
3b1d9ca6 | 1050 | link->next = freeobj++ << OBJ_TAG_BITS; |
5538c562 | 1051 | link += class->size / sizeof(*link); |
61989a80 NG |
1052 | } |
1053 | ||
1054 | /* | |
1055 | * We now come to the last (full or partial) object on this | |
1056 | * page, which must point to the first object on the next | |
1057 | * page (if present) | |
1058 | */ | |
1059 | next_page = get_next_page(page); | |
bfd093f5 | 1060 | if (next_page) { |
3b1d9ca6 | 1061 | link->next = freeobj++ << OBJ_TAG_BITS; |
bfd093f5 MK |
1062 | } else { |
1063 | /* | |
3b1d9ca6 | 1064 | * Reset OBJ_TAG_BITS bit to last link to tell |
bfd093f5 MK |
1065 | * whether it's allocated object or not. |
1066 | */ | |
3b1d9ca6 | 1067 | link->next = -1 << OBJ_TAG_BITS; |
bfd093f5 | 1068 | } |
af4ee5e9 | 1069 | kunmap_atomic(vaddr); |
61989a80 | 1070 | page = next_page; |
5538c562 | 1071 | off %= PAGE_SIZE; |
61989a80 | 1072 | } |
bdb0af7c | 1073 | |
bfd093f5 | 1074 | set_freeobj(zspage, 0); |
61989a80 NG |
1075 | } |
1076 | ||
48b4800a MK |
1077 | static void create_page_chain(struct size_class *class, struct zspage *zspage, |
1078 | struct page *pages[]) | |
61989a80 | 1079 | { |
bdb0af7c MK |
1080 | int i; |
1081 | struct page *page; | |
1082 | struct page *prev_page = NULL; | |
48b4800a | 1083 | int nr_pages = class->pages_per_zspage; |
61989a80 NG |
1084 | |
1085 | /* | |
1086 | * Allocate individual pages and link them together as: | |
48b4800a | 1087 | * 1. all pages are linked together using page->freelist |
3783689a | 1088 | * 2. each sub-page point to zspage using page->private |
61989a80 | 1089 | * |
3783689a MK |
1090 | * we set PG_private to identify the first page (i.e. no other sub-page |
1091 | * has this flag set) and PG_private_2 to identify the last page. | |
61989a80 | 1092 | */ |
bdb0af7c MK |
1093 | for (i = 0; i < nr_pages; i++) { |
1094 | page = pages[i]; | |
3783689a | 1095 | set_page_private(page, (unsigned long)zspage); |
48b4800a | 1096 | page->freelist = NULL; |
bdb0af7c | 1097 | if (i == 0) { |
3783689a | 1098 | zspage->first_page = page; |
a27545bf | 1099 | SetPagePrivate(page); |
48b4800a MK |
1100 | if (unlikely(class->objs_per_zspage == 1 && |
1101 | class->pages_per_zspage == 1)) | |
1102 | SetPageHugeObject(page); | |
3783689a | 1103 | } else { |
48b4800a | 1104 | prev_page->freelist = page; |
61989a80 | 1105 | } |
48b4800a | 1106 | if (i == nr_pages - 1) |
a27545bf | 1107 | SetPagePrivate2(page); |
61989a80 NG |
1108 | prev_page = page; |
1109 | } | |
bdb0af7c | 1110 | } |
61989a80 | 1111 | |
bdb0af7c MK |
1112 | /* |
1113 | * Allocate a zspage for the given size class | |
1114 | */ | |
3783689a MK |
1115 | static struct zspage *alloc_zspage(struct zs_pool *pool, |
1116 | struct size_class *class, | |
1117 | gfp_t gfp) | |
bdb0af7c MK |
1118 | { |
1119 | int i; | |
bdb0af7c | 1120 | struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE]; |
3783689a MK |
1121 | struct zspage *zspage = cache_alloc_zspage(pool, gfp); |
1122 | ||
1123 | if (!zspage) | |
1124 | return NULL; | |
1125 | ||
1126 | memset(zspage, 0, sizeof(struct zspage)); | |
48b4800a MK |
1127 | zspage->magic = ZSPAGE_MAGIC; |
1128 | migrate_lock_init(zspage); | |
61989a80 | 1129 | |
bdb0af7c MK |
1130 | for (i = 0; i < class->pages_per_zspage; i++) { |
1131 | struct page *page; | |
61989a80 | 1132 | |
3783689a | 1133 | page = alloc_page(gfp); |
bdb0af7c | 1134 | if (!page) { |
91537fee MK |
1135 | while (--i >= 0) { |
1136 | dec_zone_page_state(pages[i], NR_ZSPAGES); | |
bdb0af7c | 1137 | __free_page(pages[i]); |
91537fee | 1138 | } |
3783689a | 1139 | cache_free_zspage(pool, zspage); |
bdb0af7c MK |
1140 | return NULL; |
1141 | } | |
91537fee MK |
1142 | |
1143 | inc_zone_page_state(page, NR_ZSPAGES); | |
bdb0af7c | 1144 | pages[i] = page; |
61989a80 NG |
1145 | } |
1146 | ||
48b4800a | 1147 | create_page_chain(class, zspage, pages); |
3783689a | 1148 | init_zspage(class, zspage); |
bdb0af7c | 1149 | |
3783689a | 1150 | return zspage; |
61989a80 NG |
1151 | } |
1152 | ||
3783689a | 1153 | static struct zspage *find_get_zspage(struct size_class *class) |
61989a80 NG |
1154 | { |
1155 | int i; | |
3783689a | 1156 | struct zspage *zspage; |
61989a80 | 1157 | |
48b4800a | 1158 | for (i = ZS_ALMOST_FULL; i >= ZS_EMPTY; i--) { |
3783689a MK |
1159 | zspage = list_first_entry_or_null(&class->fullness_list[i], |
1160 | struct zspage, list); | |
1161 | if (zspage) | |
61989a80 NG |
1162 | break; |
1163 | } | |
1164 | ||
3783689a | 1165 | return zspage; |
61989a80 NG |
1166 | } |
1167 | ||
1b945aee | 1168 | #ifdef CONFIG_PGTABLE_MAPPING |
f553646a SJ |
1169 | static inline int __zs_cpu_up(struct mapping_area *area) |
1170 | { | |
1171 | /* | |
1172 | * Make sure we don't leak memory if a cpu UP notification | |
1173 | * and zs_init() race and both call zs_cpu_up() on the same cpu | |
1174 | */ | |
1175 | if (area->vm) | |
1176 | return 0; | |
1177 | area->vm = alloc_vm_area(PAGE_SIZE * 2, NULL); | |
1178 | if (!area->vm) | |
1179 | return -ENOMEM; | |
1180 | return 0; | |
1181 | } | |
1182 | ||
1183 | static inline void __zs_cpu_down(struct mapping_area *area) | |
1184 | { | |
1185 | if (area->vm) | |
1186 | free_vm_area(area->vm); | |
1187 | area->vm = NULL; | |
1188 | } | |
1189 | ||
1190 | static inline void *__zs_map_object(struct mapping_area *area, | |
1191 | struct page *pages[2], int off, int size) | |
1192 | { | |
f6f8ed47 | 1193 | BUG_ON(map_vm_area(area->vm, PAGE_KERNEL, pages)); |
f553646a SJ |
1194 | area->vm_addr = area->vm->addr; |
1195 | return area->vm_addr + off; | |
1196 | } | |
1197 | ||
1198 | static inline void __zs_unmap_object(struct mapping_area *area, | |
1199 | struct page *pages[2], int off, int size) | |
1200 | { | |
1201 | unsigned long addr = (unsigned long)area->vm_addr; | |
f553646a | 1202 | |
d95abbbb | 1203 | unmap_kernel_range(addr, PAGE_SIZE * 2); |
f553646a SJ |
1204 | } |
1205 | ||
1b945aee | 1206 | #else /* CONFIG_PGTABLE_MAPPING */ |
f553646a SJ |
1207 | |
1208 | static inline int __zs_cpu_up(struct mapping_area *area) | |
1209 | { | |
1210 | /* | |
1211 | * Make sure we don't leak memory if a cpu UP notification | |
1212 | * and zs_init() race and both call zs_cpu_up() on the same cpu | |
1213 | */ | |
1214 | if (area->vm_buf) | |
1215 | return 0; | |
40f9fb8c | 1216 | area->vm_buf = kmalloc(ZS_MAX_ALLOC_SIZE, GFP_KERNEL); |
f553646a SJ |
1217 | if (!area->vm_buf) |
1218 | return -ENOMEM; | |
1219 | return 0; | |
1220 | } | |
1221 | ||
1222 | static inline void __zs_cpu_down(struct mapping_area *area) | |
1223 | { | |
40f9fb8c | 1224 | kfree(area->vm_buf); |
f553646a SJ |
1225 | area->vm_buf = NULL; |
1226 | } | |
1227 | ||
1228 | static void *__zs_map_object(struct mapping_area *area, | |
1229 | struct page *pages[2], int off, int size) | |
5f601902 | 1230 | { |
5f601902 SJ |
1231 | int sizes[2]; |
1232 | void *addr; | |
f553646a | 1233 | char *buf = area->vm_buf; |
5f601902 | 1234 | |
f553646a SJ |
1235 | /* disable page faults to match kmap_atomic() return conditions */ |
1236 | pagefault_disable(); | |
1237 | ||
1238 | /* no read fastpath */ | |
1239 | if (area->vm_mm == ZS_MM_WO) | |
1240 | goto out; | |
5f601902 SJ |
1241 | |
1242 | sizes[0] = PAGE_SIZE - off; | |
1243 | sizes[1] = size - sizes[0]; | |
1244 | ||
5f601902 SJ |
1245 | /* copy object to per-cpu buffer */ |
1246 | addr = kmap_atomic(pages[0]); | |
1247 | memcpy(buf, addr + off, sizes[0]); | |
1248 | kunmap_atomic(addr); | |
1249 | addr = kmap_atomic(pages[1]); | |
1250 | memcpy(buf + sizes[0], addr, sizes[1]); | |
1251 | kunmap_atomic(addr); | |
f553646a SJ |
1252 | out: |
1253 | return area->vm_buf; | |
5f601902 SJ |
1254 | } |
1255 | ||
f553646a SJ |
1256 | static void __zs_unmap_object(struct mapping_area *area, |
1257 | struct page *pages[2], int off, int size) | |
5f601902 | 1258 | { |
5f601902 SJ |
1259 | int sizes[2]; |
1260 | void *addr; | |
2e40e163 | 1261 | char *buf; |
5f601902 | 1262 | |
f553646a SJ |
1263 | /* no write fastpath */ |
1264 | if (area->vm_mm == ZS_MM_RO) | |
1265 | goto out; | |
5f601902 | 1266 | |
7b60a685 | 1267 | buf = area->vm_buf; |
a82cbf07 YX |
1268 | buf = buf + ZS_HANDLE_SIZE; |
1269 | size -= ZS_HANDLE_SIZE; | |
1270 | off += ZS_HANDLE_SIZE; | |
2e40e163 | 1271 | |
5f601902 SJ |
1272 | sizes[0] = PAGE_SIZE - off; |
1273 | sizes[1] = size - sizes[0]; | |
1274 | ||
1275 | /* copy per-cpu buffer to object */ | |
1276 | addr = kmap_atomic(pages[0]); | |
1277 | memcpy(addr + off, buf, sizes[0]); | |
1278 | kunmap_atomic(addr); | |
1279 | addr = kmap_atomic(pages[1]); | |
1280 | memcpy(addr, buf + sizes[0], sizes[1]); | |
1281 | kunmap_atomic(addr); | |
f553646a SJ |
1282 | |
1283 | out: | |
1284 | /* enable page faults to match kunmap_atomic() return conditions */ | |
1285 | pagefault_enable(); | |
5f601902 | 1286 | } |
61989a80 | 1287 | |
1b945aee | 1288 | #endif /* CONFIG_PGTABLE_MAPPING */ |
f553646a | 1289 | |
61989a80 NG |
1290 | static int zs_cpu_notifier(struct notifier_block *nb, unsigned long action, |
1291 | void *pcpu) | |
1292 | { | |
f553646a | 1293 | int ret, cpu = (long)pcpu; |
61989a80 NG |
1294 | struct mapping_area *area; |
1295 | ||
1296 | switch (action) { | |
1297 | case CPU_UP_PREPARE: | |
1298 | area = &per_cpu(zs_map_area, cpu); | |
f553646a SJ |
1299 | ret = __zs_cpu_up(area); |
1300 | if (ret) | |
1301 | return notifier_from_errno(ret); | |
61989a80 NG |
1302 | break; |
1303 | case CPU_DEAD: | |
1304 | case CPU_UP_CANCELED: | |
1305 | area = &per_cpu(zs_map_area, cpu); | |
f553646a | 1306 | __zs_cpu_down(area); |
61989a80 NG |
1307 | break; |
1308 | } | |
1309 | ||
1310 | return NOTIFY_OK; | |
1311 | } | |
1312 | ||
1313 | static struct notifier_block zs_cpu_nb = { | |
1314 | .notifier_call = zs_cpu_notifier | |
1315 | }; | |
1316 | ||
b1b00a5b | 1317 | static int zs_register_cpu_notifier(void) |
61989a80 | 1318 | { |
b1b00a5b | 1319 | int cpu, uninitialized_var(ret); |
61989a80 | 1320 | |
f0e71fcd SB |
1321 | cpu_notifier_register_begin(); |
1322 | ||
1323 | __register_cpu_notifier(&zs_cpu_nb); | |
61989a80 NG |
1324 | for_each_online_cpu(cpu) { |
1325 | ret = zs_cpu_notifier(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
b1b00a5b SS |
1326 | if (notifier_to_errno(ret)) |
1327 | break; | |
61989a80 | 1328 | } |
f0e71fcd SB |
1329 | |
1330 | cpu_notifier_register_done(); | |
b1b00a5b SS |
1331 | return notifier_to_errno(ret); |
1332 | } | |
f0e71fcd | 1333 | |
66cdef66 | 1334 | static void zs_unregister_cpu_notifier(void) |
40f9fb8c | 1335 | { |
66cdef66 | 1336 | int cpu; |
40f9fb8c | 1337 | |
66cdef66 | 1338 | cpu_notifier_register_begin(); |
40f9fb8c | 1339 | |
66cdef66 GM |
1340 | for_each_online_cpu(cpu) |
1341 | zs_cpu_notifier(NULL, CPU_DEAD, (void *)(long)cpu); | |
1342 | __unregister_cpu_notifier(&zs_cpu_nb); | |
40f9fb8c | 1343 | |
66cdef66 | 1344 | cpu_notifier_register_done(); |
b1b00a5b SS |
1345 | } |
1346 | ||
35b3445e | 1347 | static void __init init_zs_size_classes(void) |
b1b00a5b | 1348 | { |
66cdef66 | 1349 | int nr; |
c795779d | 1350 | |
66cdef66 GM |
1351 | nr = (ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) / ZS_SIZE_CLASS_DELTA + 1; |
1352 | if ((ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) % ZS_SIZE_CLASS_DELTA) | |
1353 | nr += 1; | |
40f9fb8c | 1354 | |
66cdef66 | 1355 | zs_size_classes = nr; |
61989a80 NG |
1356 | } |
1357 | ||
64d90465 GM |
1358 | static bool can_merge(struct size_class *prev, int pages_per_zspage, |
1359 | int objs_per_zspage) | |
9eec4cd5 | 1360 | { |
64d90465 GM |
1361 | if (prev->pages_per_zspage == pages_per_zspage && |
1362 | prev->objs_per_zspage == objs_per_zspage) | |
1363 | return true; | |
9eec4cd5 | 1364 | |
64d90465 | 1365 | return false; |
9eec4cd5 JK |
1366 | } |
1367 | ||
3783689a | 1368 | static bool zspage_full(struct size_class *class, struct zspage *zspage) |
312fcae2 | 1369 | { |
3783689a | 1370 | return get_zspage_inuse(zspage) == class->objs_per_zspage; |
312fcae2 MK |
1371 | } |
1372 | ||
66cdef66 GM |
1373 | unsigned long zs_get_total_pages(struct zs_pool *pool) |
1374 | { | |
1375 | return atomic_long_read(&pool->pages_allocated); | |
1376 | } | |
1377 | EXPORT_SYMBOL_GPL(zs_get_total_pages); | |
1378 | ||
4bbc0bc0 | 1379 | /** |
66cdef66 GM |
1380 | * zs_map_object - get address of allocated object from handle. |
1381 | * @pool: pool from which the object was allocated | |
1382 | * @handle: handle returned from zs_malloc | |
4bbc0bc0 | 1383 | * |
66cdef66 GM |
1384 | * Before using an object allocated from zs_malloc, it must be mapped using |
1385 | * this function. When done with the object, it must be unmapped using | |
1386 | * zs_unmap_object. | |
4bbc0bc0 | 1387 | * |
66cdef66 GM |
1388 | * Only one object can be mapped per cpu at a time. There is no protection |
1389 | * against nested mappings. | |
1390 | * | |
1391 | * This function returns with preemption and page faults disabled. | |
4bbc0bc0 | 1392 | */ |
66cdef66 GM |
1393 | void *zs_map_object(struct zs_pool *pool, unsigned long handle, |
1394 | enum zs_mapmode mm) | |
61989a80 | 1395 | { |
3783689a | 1396 | struct zspage *zspage; |
66cdef66 | 1397 | struct page *page; |
bfd093f5 MK |
1398 | unsigned long obj, off; |
1399 | unsigned int obj_idx; | |
61989a80 | 1400 | |
66cdef66 GM |
1401 | unsigned int class_idx; |
1402 | enum fullness_group fg; | |
1403 | struct size_class *class; | |
1404 | struct mapping_area *area; | |
1405 | struct page *pages[2]; | |
2e40e163 | 1406 | void *ret; |
61989a80 | 1407 | |
9eec4cd5 | 1408 | /* |
66cdef66 GM |
1409 | * Because we use per-cpu mapping areas shared among the |
1410 | * pools/users, we can't allow mapping in interrupt context | |
1411 | * because it can corrupt another users mappings. | |
9eec4cd5 | 1412 | */ |
830e4bc5 | 1413 | WARN_ON_ONCE(in_interrupt()); |
61989a80 | 1414 | |
312fcae2 MK |
1415 | /* From now on, migration cannot move the object */ |
1416 | pin_tag(handle); | |
1417 | ||
2e40e163 MK |
1418 | obj = handle_to_obj(handle); |
1419 | obj_to_location(obj, &page, &obj_idx); | |
3783689a | 1420 | zspage = get_zspage(page); |
48b4800a MK |
1421 | |
1422 | /* migration cannot move any subpage in this zspage */ | |
1423 | migrate_read_lock(zspage); | |
1424 | ||
3783689a | 1425 | get_zspage_mapping(zspage, &class_idx, &fg); |
66cdef66 | 1426 | class = pool->size_class[class_idx]; |
bfd093f5 | 1427 | off = (class->size * obj_idx) & ~PAGE_MASK; |
df8b5bb9 | 1428 | |
66cdef66 GM |
1429 | area = &get_cpu_var(zs_map_area); |
1430 | area->vm_mm = mm; | |
1431 | if (off + class->size <= PAGE_SIZE) { | |
1432 | /* this object is contained entirely within a page */ | |
1433 | area->vm_addr = kmap_atomic(page); | |
2e40e163 MK |
1434 | ret = area->vm_addr + off; |
1435 | goto out; | |
61989a80 NG |
1436 | } |
1437 | ||
66cdef66 GM |
1438 | /* this object spans two pages */ |
1439 | pages[0] = page; | |
1440 | pages[1] = get_next_page(page); | |
1441 | BUG_ON(!pages[1]); | |
9eec4cd5 | 1442 | |
2e40e163 MK |
1443 | ret = __zs_map_object(area, pages, off, class->size); |
1444 | out: | |
48b4800a | 1445 | if (likely(!PageHugeObject(page))) |
7b60a685 MK |
1446 | ret += ZS_HANDLE_SIZE; |
1447 | ||
1448 | return ret; | |
61989a80 | 1449 | } |
66cdef66 | 1450 | EXPORT_SYMBOL_GPL(zs_map_object); |
61989a80 | 1451 | |
66cdef66 | 1452 | void zs_unmap_object(struct zs_pool *pool, unsigned long handle) |
61989a80 | 1453 | { |
3783689a | 1454 | struct zspage *zspage; |
66cdef66 | 1455 | struct page *page; |
bfd093f5 MK |
1456 | unsigned long obj, off; |
1457 | unsigned int obj_idx; | |
61989a80 | 1458 | |
66cdef66 GM |
1459 | unsigned int class_idx; |
1460 | enum fullness_group fg; | |
1461 | struct size_class *class; | |
1462 | struct mapping_area *area; | |
9eec4cd5 | 1463 | |
2e40e163 MK |
1464 | obj = handle_to_obj(handle); |
1465 | obj_to_location(obj, &page, &obj_idx); | |
3783689a MK |
1466 | zspage = get_zspage(page); |
1467 | get_zspage_mapping(zspage, &class_idx, &fg); | |
66cdef66 | 1468 | class = pool->size_class[class_idx]; |
bfd093f5 | 1469 | off = (class->size * obj_idx) & ~PAGE_MASK; |
61989a80 | 1470 | |
66cdef66 GM |
1471 | area = this_cpu_ptr(&zs_map_area); |
1472 | if (off + class->size <= PAGE_SIZE) | |
1473 | kunmap_atomic(area->vm_addr); | |
1474 | else { | |
1475 | struct page *pages[2]; | |
40f9fb8c | 1476 | |
66cdef66 GM |
1477 | pages[0] = page; |
1478 | pages[1] = get_next_page(page); | |
1479 | BUG_ON(!pages[1]); | |
1480 | ||
1481 | __zs_unmap_object(area, pages, off, class->size); | |
1482 | } | |
1483 | put_cpu_var(zs_map_area); | |
48b4800a MK |
1484 | |
1485 | migrate_read_unlock(zspage); | |
312fcae2 | 1486 | unpin_tag(handle); |
61989a80 | 1487 | } |
66cdef66 | 1488 | EXPORT_SYMBOL_GPL(zs_unmap_object); |
61989a80 | 1489 | |
251cbb95 | 1490 | static unsigned long obj_malloc(struct size_class *class, |
3783689a | 1491 | struct zspage *zspage, unsigned long handle) |
c7806261 | 1492 | { |
bfd093f5 | 1493 | int i, nr_page, offset; |
c7806261 MK |
1494 | unsigned long obj; |
1495 | struct link_free *link; | |
1496 | ||
1497 | struct page *m_page; | |
bfd093f5 | 1498 | unsigned long m_offset; |
c7806261 MK |
1499 | void *vaddr; |
1500 | ||
312fcae2 | 1501 | handle |= OBJ_ALLOCATED_TAG; |
3783689a | 1502 | obj = get_freeobj(zspage); |
bfd093f5 MK |
1503 | |
1504 | offset = obj * class->size; | |
1505 | nr_page = offset >> PAGE_SHIFT; | |
1506 | m_offset = offset & ~PAGE_MASK; | |
1507 | m_page = get_first_page(zspage); | |
1508 | ||
1509 | for (i = 0; i < nr_page; i++) | |
1510 | m_page = get_next_page(m_page); | |
c7806261 MK |
1511 | |
1512 | vaddr = kmap_atomic(m_page); | |
1513 | link = (struct link_free *)vaddr + m_offset / sizeof(*link); | |
3b1d9ca6 | 1514 | set_freeobj(zspage, link->next >> OBJ_TAG_BITS); |
48b4800a | 1515 | if (likely(!PageHugeObject(m_page))) |
7b60a685 MK |
1516 | /* record handle in the header of allocated chunk */ |
1517 | link->handle = handle; | |
1518 | else | |
3783689a MK |
1519 | /* record handle to page->index */ |
1520 | zspage->first_page->index = handle; | |
1521 | ||
c7806261 | 1522 | kunmap_atomic(vaddr); |
3783689a | 1523 | mod_zspage_inuse(zspage, 1); |
c7806261 MK |
1524 | zs_stat_inc(class, OBJ_USED, 1); |
1525 | ||
bfd093f5 MK |
1526 | obj = location_to_obj(m_page, obj); |
1527 | ||
c7806261 MK |
1528 | return obj; |
1529 | } | |
1530 | ||
1531 | ||
61989a80 NG |
1532 | /** |
1533 | * zs_malloc - Allocate block of given size from pool. | |
1534 | * @pool: pool to allocate from | |
1535 | * @size: size of block to allocate | |
fd854463 | 1536 | * @gfp: gfp flags when allocating object |
61989a80 | 1537 | * |
00a61d86 | 1538 | * On success, handle to the allocated object is returned, |
c2344348 | 1539 | * otherwise 0. |
61989a80 NG |
1540 | * Allocation requests with size > ZS_MAX_ALLOC_SIZE will fail. |
1541 | */ | |
d0d8da2d | 1542 | unsigned long zs_malloc(struct zs_pool *pool, size_t size, gfp_t gfp) |
61989a80 | 1543 | { |
2e40e163 | 1544 | unsigned long handle, obj; |
61989a80 | 1545 | struct size_class *class; |
48b4800a | 1546 | enum fullness_group newfg; |
3783689a | 1547 | struct zspage *zspage; |
61989a80 | 1548 | |
7b60a685 | 1549 | if (unlikely(!size || size > ZS_MAX_ALLOC_SIZE)) |
2e40e163 MK |
1550 | return 0; |
1551 | ||
3783689a | 1552 | handle = cache_alloc_handle(pool, gfp); |
2e40e163 | 1553 | if (!handle) |
c2344348 | 1554 | return 0; |
61989a80 | 1555 | |
2e40e163 MK |
1556 | /* extra space in chunk to keep the handle */ |
1557 | size += ZS_HANDLE_SIZE; | |
9eec4cd5 | 1558 | class = pool->size_class[get_size_class_index(size)]; |
61989a80 NG |
1559 | |
1560 | spin_lock(&class->lock); | |
3783689a | 1561 | zspage = find_get_zspage(class); |
48b4800a MK |
1562 | if (likely(zspage)) { |
1563 | obj = obj_malloc(class, zspage, handle); | |
1564 | /* Now move the zspage to another fullness group, if required */ | |
1565 | fix_fullness_group(class, zspage); | |
1566 | record_obj(handle, obj); | |
61989a80 | 1567 | spin_unlock(&class->lock); |
61989a80 | 1568 | |
48b4800a MK |
1569 | return handle; |
1570 | } | |
0f050d99 | 1571 | |
48b4800a MK |
1572 | spin_unlock(&class->lock); |
1573 | ||
1574 | zspage = alloc_zspage(pool, class, gfp); | |
1575 | if (!zspage) { | |
1576 | cache_free_handle(pool, handle); | |
1577 | return 0; | |
61989a80 NG |
1578 | } |
1579 | ||
48b4800a | 1580 | spin_lock(&class->lock); |
3783689a | 1581 | obj = obj_malloc(class, zspage, handle); |
48b4800a MK |
1582 | newfg = get_fullness_group(class, zspage); |
1583 | insert_zspage(class, zspage, newfg); | |
1584 | set_zspage_mapping(zspage, class->index, newfg); | |
2e40e163 | 1585 | record_obj(handle, obj); |
48b4800a MK |
1586 | atomic_long_add(class->pages_per_zspage, |
1587 | &pool->pages_allocated); | |
b4fd07a0 | 1588 | zs_stat_inc(class, OBJ_ALLOCATED, class->objs_per_zspage); |
48b4800a MK |
1589 | |
1590 | /* We completely set up zspage so mark them as movable */ | |
1591 | SetZsPageMovable(pool, zspage); | |
61989a80 NG |
1592 | spin_unlock(&class->lock); |
1593 | ||
2e40e163 | 1594 | return handle; |
61989a80 NG |
1595 | } |
1596 | EXPORT_SYMBOL_GPL(zs_malloc); | |
1597 | ||
1ee47165 | 1598 | static void obj_free(struct size_class *class, unsigned long obj) |
61989a80 NG |
1599 | { |
1600 | struct link_free *link; | |
3783689a MK |
1601 | struct zspage *zspage; |
1602 | struct page *f_page; | |
bfd093f5 MK |
1603 | unsigned long f_offset; |
1604 | unsigned int f_objidx; | |
af4ee5e9 | 1605 | void *vaddr; |
61989a80 | 1606 | |
312fcae2 | 1607 | obj &= ~OBJ_ALLOCATED_TAG; |
2e40e163 | 1608 | obj_to_location(obj, &f_page, &f_objidx); |
bfd093f5 | 1609 | f_offset = (class->size * f_objidx) & ~PAGE_MASK; |
3783689a | 1610 | zspage = get_zspage(f_page); |
61989a80 | 1611 | |
c7806261 | 1612 | vaddr = kmap_atomic(f_page); |
61989a80 NG |
1613 | |
1614 | /* Insert this object in containing zspage's freelist */ | |
af4ee5e9 | 1615 | link = (struct link_free *)(vaddr + f_offset); |
3b1d9ca6 | 1616 | link->next = get_freeobj(zspage) << OBJ_TAG_BITS; |
af4ee5e9 | 1617 | kunmap_atomic(vaddr); |
bfd093f5 | 1618 | set_freeobj(zspage, f_objidx); |
3783689a | 1619 | mod_zspage_inuse(zspage, -1); |
0f050d99 | 1620 | zs_stat_dec(class, OBJ_USED, 1); |
c7806261 MK |
1621 | } |
1622 | ||
1623 | void zs_free(struct zs_pool *pool, unsigned long handle) | |
1624 | { | |
3783689a MK |
1625 | struct zspage *zspage; |
1626 | struct page *f_page; | |
bfd093f5 MK |
1627 | unsigned long obj; |
1628 | unsigned int f_objidx; | |
c7806261 MK |
1629 | int class_idx; |
1630 | struct size_class *class; | |
1631 | enum fullness_group fullness; | |
48b4800a | 1632 | bool isolated; |
c7806261 MK |
1633 | |
1634 | if (unlikely(!handle)) | |
1635 | return; | |
1636 | ||
312fcae2 | 1637 | pin_tag(handle); |
c7806261 | 1638 | obj = handle_to_obj(handle); |
c7806261 | 1639 | obj_to_location(obj, &f_page, &f_objidx); |
3783689a | 1640 | zspage = get_zspage(f_page); |
c7806261 | 1641 | |
48b4800a MK |
1642 | migrate_read_lock(zspage); |
1643 | ||
3783689a | 1644 | get_zspage_mapping(zspage, &class_idx, &fullness); |
c7806261 MK |
1645 | class = pool->size_class[class_idx]; |
1646 | ||
1647 | spin_lock(&class->lock); | |
1ee47165 | 1648 | obj_free(class, obj); |
3783689a | 1649 | fullness = fix_fullness_group(class, zspage); |
48b4800a MK |
1650 | if (fullness != ZS_EMPTY) { |
1651 | migrate_read_unlock(zspage); | |
1652 | goto out; | |
312fcae2 | 1653 | } |
48b4800a MK |
1654 | |
1655 | isolated = is_zspage_isolated(zspage); | |
1656 | migrate_read_unlock(zspage); | |
1657 | /* If zspage is isolated, zs_page_putback will free the zspage */ | |
1658 | if (likely(!isolated)) | |
1659 | free_zspage(pool, class, zspage); | |
1660 | out: | |
1661 | ||
61989a80 | 1662 | spin_unlock(&class->lock); |
312fcae2 | 1663 | unpin_tag(handle); |
3783689a | 1664 | cache_free_handle(pool, handle); |
312fcae2 MK |
1665 | } |
1666 | EXPORT_SYMBOL_GPL(zs_free); | |
1667 | ||
251cbb95 MK |
1668 | static void zs_object_copy(struct size_class *class, unsigned long dst, |
1669 | unsigned long src) | |
312fcae2 MK |
1670 | { |
1671 | struct page *s_page, *d_page; | |
bfd093f5 | 1672 | unsigned int s_objidx, d_objidx; |
312fcae2 MK |
1673 | unsigned long s_off, d_off; |
1674 | void *s_addr, *d_addr; | |
1675 | int s_size, d_size, size; | |
1676 | int written = 0; | |
1677 | ||
1678 | s_size = d_size = class->size; | |
1679 | ||
1680 | obj_to_location(src, &s_page, &s_objidx); | |
1681 | obj_to_location(dst, &d_page, &d_objidx); | |
1682 | ||
bfd093f5 MK |
1683 | s_off = (class->size * s_objidx) & ~PAGE_MASK; |
1684 | d_off = (class->size * d_objidx) & ~PAGE_MASK; | |
312fcae2 MK |
1685 | |
1686 | if (s_off + class->size > PAGE_SIZE) | |
1687 | s_size = PAGE_SIZE - s_off; | |
1688 | ||
1689 | if (d_off + class->size > PAGE_SIZE) | |
1690 | d_size = PAGE_SIZE - d_off; | |
1691 | ||
1692 | s_addr = kmap_atomic(s_page); | |
1693 | d_addr = kmap_atomic(d_page); | |
1694 | ||
1695 | while (1) { | |
1696 | size = min(s_size, d_size); | |
1697 | memcpy(d_addr + d_off, s_addr + s_off, size); | |
1698 | written += size; | |
1699 | ||
1700 | if (written == class->size) | |
1701 | break; | |
1702 | ||
495819ea SS |
1703 | s_off += size; |
1704 | s_size -= size; | |
1705 | d_off += size; | |
1706 | d_size -= size; | |
1707 | ||
1708 | if (s_off >= PAGE_SIZE) { | |
312fcae2 MK |
1709 | kunmap_atomic(d_addr); |
1710 | kunmap_atomic(s_addr); | |
1711 | s_page = get_next_page(s_page); | |
312fcae2 MK |
1712 | s_addr = kmap_atomic(s_page); |
1713 | d_addr = kmap_atomic(d_page); | |
1714 | s_size = class->size - written; | |
1715 | s_off = 0; | |
312fcae2 MK |
1716 | } |
1717 | ||
495819ea | 1718 | if (d_off >= PAGE_SIZE) { |
312fcae2 MK |
1719 | kunmap_atomic(d_addr); |
1720 | d_page = get_next_page(d_page); | |
312fcae2 MK |
1721 | d_addr = kmap_atomic(d_page); |
1722 | d_size = class->size - written; | |
1723 | d_off = 0; | |
312fcae2 MK |
1724 | } |
1725 | } | |
1726 | ||
1727 | kunmap_atomic(d_addr); | |
1728 | kunmap_atomic(s_addr); | |
1729 | } | |
1730 | ||
1731 | /* | |
1732 | * Find alloced object in zspage from index object and | |
1733 | * return handle. | |
1734 | */ | |
251cbb95 | 1735 | static unsigned long find_alloced_obj(struct size_class *class, |
cf675acb | 1736 | struct page *page, int *obj_idx) |
312fcae2 MK |
1737 | { |
1738 | unsigned long head; | |
1739 | int offset = 0; | |
cf675acb | 1740 | int index = *obj_idx; |
312fcae2 MK |
1741 | unsigned long handle = 0; |
1742 | void *addr = kmap_atomic(page); | |
1743 | ||
3783689a | 1744 | offset = get_first_obj_offset(page); |
312fcae2 MK |
1745 | offset += class->size * index; |
1746 | ||
1747 | while (offset < PAGE_SIZE) { | |
48b4800a | 1748 | head = obj_to_head(page, addr + offset); |
312fcae2 MK |
1749 | if (head & OBJ_ALLOCATED_TAG) { |
1750 | handle = head & ~OBJ_ALLOCATED_TAG; | |
1751 | if (trypin_tag(handle)) | |
1752 | break; | |
1753 | handle = 0; | |
1754 | } | |
1755 | ||
1756 | offset += class->size; | |
1757 | index++; | |
1758 | } | |
1759 | ||
1760 | kunmap_atomic(addr); | |
cf675acb GM |
1761 | |
1762 | *obj_idx = index; | |
1763 | ||
312fcae2 MK |
1764 | return handle; |
1765 | } | |
1766 | ||
1767 | struct zs_compact_control { | |
3783689a | 1768 | /* Source spage for migration which could be a subpage of zspage */ |
312fcae2 MK |
1769 | struct page *s_page; |
1770 | /* Destination page for migration which should be a first page | |
1771 | * of zspage. */ | |
1772 | struct page *d_page; | |
c27cbaa3 GM |
1773 | /* Starting object index within @s_page which used for live object |
1774 | * in the subpage. */ | |
41b88e14 | 1775 | int obj_idx; |
c27cbaa3 GM |
1776 | |
1777 | unsigned long nr_migrated_obj; | |
1778 | unsigned long nr_freed_pages; | |
312fcae2 MK |
1779 | }; |
1780 | ||
1781 | static int migrate_zspage(struct zs_pool *pool, struct size_class *class, | |
1782 | struct zs_compact_control *cc) | |
1783 | { | |
1784 | unsigned long used_obj, free_obj; | |
1785 | unsigned long handle; | |
1786 | struct page *s_page = cc->s_page; | |
1787 | struct page *d_page = cc->d_page; | |
41b88e14 | 1788 | int obj_idx = cc->obj_idx; |
312fcae2 MK |
1789 | int ret = 0; |
1790 | ||
1791 | while (1) { | |
cf675acb | 1792 | handle = find_alloced_obj(class, s_page, &obj_idx); |
312fcae2 MK |
1793 | if (!handle) { |
1794 | s_page = get_next_page(s_page); | |
1795 | if (!s_page) | |
1796 | break; | |
41b88e14 | 1797 | obj_idx = 0; |
312fcae2 MK |
1798 | continue; |
1799 | } | |
1800 | ||
1801 | /* Stop if there is no more space */ | |
3783689a | 1802 | if (zspage_full(class, get_zspage(d_page))) { |
312fcae2 MK |
1803 | unpin_tag(handle); |
1804 | ret = -ENOMEM; | |
1805 | break; | |
1806 | } | |
1807 | ||
1808 | used_obj = handle_to_obj(handle); | |
3783689a | 1809 | free_obj = obj_malloc(class, get_zspage(d_page), handle); |
251cbb95 | 1810 | zs_object_copy(class, free_obj, used_obj); |
41b88e14 | 1811 | obj_idx++; |
c27cbaa3 | 1812 | cc->nr_migrated_obj++; |
c102f07c JL |
1813 | /* |
1814 | * record_obj updates handle's value to free_obj and it will | |
1815 | * invalidate lock bit(ie, HANDLE_PIN_BIT) of handle, which | |
1816 | * breaks synchronization using pin_tag(e,g, zs_free) so | |
1817 | * let's keep the lock bit. | |
1818 | */ | |
1819 | free_obj |= BIT(HANDLE_PIN_BIT); | |
312fcae2 MK |
1820 | record_obj(handle, free_obj); |
1821 | unpin_tag(handle); | |
1ee47165 | 1822 | obj_free(class, used_obj); |
312fcae2 MK |
1823 | } |
1824 | ||
1825 | /* Remember last position in this iteration */ | |
1826 | cc->s_page = s_page; | |
41b88e14 | 1827 | cc->obj_idx = obj_idx; |
312fcae2 MK |
1828 | |
1829 | return ret; | |
1830 | } | |
1831 | ||
3783689a | 1832 | static struct zspage *isolate_zspage(struct size_class *class, bool source) |
312fcae2 MK |
1833 | { |
1834 | int i; | |
3783689a MK |
1835 | struct zspage *zspage; |
1836 | enum fullness_group fg[2] = {ZS_ALMOST_EMPTY, ZS_ALMOST_FULL}; | |
312fcae2 | 1837 | |
3783689a MK |
1838 | if (!source) { |
1839 | fg[0] = ZS_ALMOST_FULL; | |
1840 | fg[1] = ZS_ALMOST_EMPTY; | |
1841 | } | |
1842 | ||
1843 | for (i = 0; i < 2; i++) { | |
1844 | zspage = list_first_entry_or_null(&class->fullness_list[fg[i]], | |
1845 | struct zspage, list); | |
1846 | if (zspage) { | |
48b4800a | 1847 | VM_BUG_ON(is_zspage_isolated(zspage)); |
3783689a MK |
1848 | remove_zspage(class, zspage, fg[i]); |
1849 | return zspage; | |
312fcae2 MK |
1850 | } |
1851 | } | |
1852 | ||
3783689a | 1853 | return zspage; |
312fcae2 MK |
1854 | } |
1855 | ||
860c707d | 1856 | /* |
3783689a | 1857 | * putback_zspage - add @zspage into right class's fullness list |
860c707d | 1858 | * @class: destination class |
3783689a | 1859 | * @zspage: target page |
860c707d | 1860 | * |
3783689a | 1861 | * Return @zspage's fullness_group |
860c707d | 1862 | */ |
4aa409ca | 1863 | static enum fullness_group putback_zspage(struct size_class *class, |
3783689a | 1864 | struct zspage *zspage) |
312fcae2 | 1865 | { |
312fcae2 MK |
1866 | enum fullness_group fullness; |
1867 | ||
48b4800a MK |
1868 | VM_BUG_ON(is_zspage_isolated(zspage)); |
1869 | ||
3783689a MK |
1870 | fullness = get_fullness_group(class, zspage); |
1871 | insert_zspage(class, zspage, fullness); | |
1872 | set_zspage_mapping(zspage, class->index, fullness); | |
839373e6 | 1873 | |
860c707d | 1874 | return fullness; |
61989a80 | 1875 | } |
312fcae2 | 1876 | |
48b4800a MK |
1877 | #ifdef CONFIG_COMPACTION |
1878 | static struct dentry *zs_mount(struct file_system_type *fs_type, | |
1879 | int flags, const char *dev_name, void *data) | |
1880 | { | |
1881 | static const struct dentry_operations ops = { | |
1882 | .d_dname = simple_dname, | |
1883 | }; | |
1884 | ||
1885 | return mount_pseudo(fs_type, "zsmalloc:", NULL, &ops, ZSMALLOC_MAGIC); | |
1886 | } | |
1887 | ||
1888 | static struct file_system_type zsmalloc_fs = { | |
1889 | .name = "zsmalloc", | |
1890 | .mount = zs_mount, | |
1891 | .kill_sb = kill_anon_super, | |
1892 | }; | |
1893 | ||
1894 | static int zsmalloc_mount(void) | |
1895 | { | |
1896 | int ret = 0; | |
1897 | ||
1898 | zsmalloc_mnt = kern_mount(&zsmalloc_fs); | |
1899 | if (IS_ERR(zsmalloc_mnt)) | |
1900 | ret = PTR_ERR(zsmalloc_mnt); | |
1901 | ||
1902 | return ret; | |
1903 | } | |
1904 | ||
1905 | static void zsmalloc_unmount(void) | |
1906 | { | |
1907 | kern_unmount(zsmalloc_mnt); | |
1908 | } | |
1909 | ||
1910 | static void migrate_lock_init(struct zspage *zspage) | |
1911 | { | |
1912 | rwlock_init(&zspage->lock); | |
1913 | } | |
1914 | ||
1915 | static void migrate_read_lock(struct zspage *zspage) | |
1916 | { | |
1917 | read_lock(&zspage->lock); | |
1918 | } | |
1919 | ||
1920 | static void migrate_read_unlock(struct zspage *zspage) | |
1921 | { | |
1922 | read_unlock(&zspage->lock); | |
1923 | } | |
1924 | ||
1925 | static void migrate_write_lock(struct zspage *zspage) | |
1926 | { | |
1927 | write_lock(&zspage->lock); | |
1928 | } | |
1929 | ||
1930 | static void migrate_write_unlock(struct zspage *zspage) | |
1931 | { | |
1932 | write_unlock(&zspage->lock); | |
1933 | } | |
1934 | ||
1935 | /* Number of isolated subpage for *page migration* in this zspage */ | |
1936 | static void inc_zspage_isolation(struct zspage *zspage) | |
1937 | { | |
1938 | zspage->isolated++; | |
1939 | } | |
1940 | ||
1941 | static void dec_zspage_isolation(struct zspage *zspage) | |
1942 | { | |
1943 | zspage->isolated--; | |
1944 | } | |
1945 | ||
1946 | static void replace_sub_page(struct size_class *class, struct zspage *zspage, | |
1947 | struct page *newpage, struct page *oldpage) | |
1948 | { | |
1949 | struct page *page; | |
1950 | struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE] = {NULL, }; | |
1951 | int idx = 0; | |
1952 | ||
1953 | page = get_first_page(zspage); | |
1954 | do { | |
1955 | if (page == oldpage) | |
1956 | pages[idx] = newpage; | |
1957 | else | |
1958 | pages[idx] = page; | |
1959 | idx++; | |
1960 | } while ((page = get_next_page(page)) != NULL); | |
1961 | ||
1962 | create_page_chain(class, zspage, pages); | |
1963 | set_first_obj_offset(newpage, get_first_obj_offset(oldpage)); | |
1964 | if (unlikely(PageHugeObject(oldpage))) | |
1965 | newpage->index = oldpage->index; | |
1966 | __SetPageMovable(newpage, page_mapping(oldpage)); | |
1967 | } | |
1968 | ||
1969 | bool zs_page_isolate(struct page *page, isolate_mode_t mode) | |
1970 | { | |
1971 | struct zs_pool *pool; | |
1972 | struct size_class *class; | |
1973 | int class_idx; | |
1974 | enum fullness_group fullness; | |
1975 | struct zspage *zspage; | |
1976 | struct address_space *mapping; | |
1977 | ||
1978 | /* | |
1979 | * Page is locked so zspage couldn't be destroyed. For detail, look at | |
1980 | * lock_zspage in free_zspage. | |
1981 | */ | |
1982 | VM_BUG_ON_PAGE(!PageMovable(page), page); | |
1983 | VM_BUG_ON_PAGE(PageIsolated(page), page); | |
1984 | ||
1985 | zspage = get_zspage(page); | |
1986 | ||
1987 | /* | |
1988 | * Without class lock, fullness could be stale while class_idx is okay | |
1989 | * because class_idx is constant unless page is freed so we should get | |
1990 | * fullness again under class lock. | |
1991 | */ | |
1992 | get_zspage_mapping(zspage, &class_idx, &fullness); | |
1993 | mapping = page_mapping(page); | |
1994 | pool = mapping->private_data; | |
1995 | class = pool->size_class[class_idx]; | |
1996 | ||
1997 | spin_lock(&class->lock); | |
1998 | if (get_zspage_inuse(zspage) == 0) { | |
1999 | spin_unlock(&class->lock); | |
2000 | return false; | |
2001 | } | |
2002 | ||
2003 | /* zspage is isolated for object migration */ | |
2004 | if (list_empty(&zspage->list) && !is_zspage_isolated(zspage)) { | |
2005 | spin_unlock(&class->lock); | |
2006 | return false; | |
2007 | } | |
2008 | ||
2009 | /* | |
2010 | * If this is first time isolation for the zspage, isolate zspage from | |
2011 | * size_class to prevent further object allocation from the zspage. | |
2012 | */ | |
2013 | if (!list_empty(&zspage->list) && !is_zspage_isolated(zspage)) { | |
2014 | get_zspage_mapping(zspage, &class_idx, &fullness); | |
2015 | remove_zspage(class, zspage, fullness); | |
2016 | } | |
2017 | ||
2018 | inc_zspage_isolation(zspage); | |
2019 | spin_unlock(&class->lock); | |
2020 | ||
2021 | return true; | |
2022 | } | |
2023 | ||
2024 | int zs_page_migrate(struct address_space *mapping, struct page *newpage, | |
2025 | struct page *page, enum migrate_mode mode) | |
2026 | { | |
2027 | struct zs_pool *pool; | |
2028 | struct size_class *class; | |
2029 | int class_idx; | |
2030 | enum fullness_group fullness; | |
2031 | struct zspage *zspage; | |
2032 | struct page *dummy; | |
2033 | void *s_addr, *d_addr, *addr; | |
2034 | int offset, pos; | |
2035 | unsigned long handle, head; | |
2036 | unsigned long old_obj, new_obj; | |
2037 | unsigned int obj_idx; | |
2038 | int ret = -EAGAIN; | |
2039 | ||
2040 | VM_BUG_ON_PAGE(!PageMovable(page), page); | |
2041 | VM_BUG_ON_PAGE(!PageIsolated(page), page); | |
2042 | ||
2043 | zspage = get_zspage(page); | |
2044 | ||
2045 | /* Concurrent compactor cannot migrate any subpage in zspage */ | |
2046 | migrate_write_lock(zspage); | |
2047 | get_zspage_mapping(zspage, &class_idx, &fullness); | |
2048 | pool = mapping->private_data; | |
2049 | class = pool->size_class[class_idx]; | |
2050 | offset = get_first_obj_offset(page); | |
2051 | ||
2052 | spin_lock(&class->lock); | |
2053 | if (!get_zspage_inuse(zspage)) { | |
2054 | ret = -EBUSY; | |
2055 | goto unlock_class; | |
2056 | } | |
2057 | ||
2058 | pos = offset; | |
2059 | s_addr = kmap_atomic(page); | |
2060 | while (pos < PAGE_SIZE) { | |
2061 | head = obj_to_head(page, s_addr + pos); | |
2062 | if (head & OBJ_ALLOCATED_TAG) { | |
2063 | handle = head & ~OBJ_ALLOCATED_TAG; | |
2064 | if (!trypin_tag(handle)) | |
2065 | goto unpin_objects; | |
2066 | } | |
2067 | pos += class->size; | |
2068 | } | |
2069 | ||
2070 | /* | |
2071 | * Here, any user cannot access all objects in the zspage so let's move. | |
2072 | */ | |
2073 | d_addr = kmap_atomic(newpage); | |
2074 | memcpy(d_addr, s_addr, PAGE_SIZE); | |
2075 | kunmap_atomic(d_addr); | |
2076 | ||
2077 | for (addr = s_addr + offset; addr < s_addr + pos; | |
2078 | addr += class->size) { | |
2079 | head = obj_to_head(page, addr); | |
2080 | if (head & OBJ_ALLOCATED_TAG) { | |
2081 | handle = head & ~OBJ_ALLOCATED_TAG; | |
2082 | if (!testpin_tag(handle)) | |
2083 | BUG(); | |
2084 | ||
2085 | old_obj = handle_to_obj(handle); | |
2086 | obj_to_location(old_obj, &dummy, &obj_idx); | |
2087 | new_obj = (unsigned long)location_to_obj(newpage, | |
2088 | obj_idx); | |
2089 | new_obj |= BIT(HANDLE_PIN_BIT); | |
2090 | record_obj(handle, new_obj); | |
2091 | } | |
2092 | } | |
2093 | ||
2094 | replace_sub_page(class, zspage, newpage, page); | |
2095 | get_page(newpage); | |
2096 | ||
2097 | dec_zspage_isolation(zspage); | |
2098 | ||
2099 | /* | |
2100 | * Page migration is done so let's putback isolated zspage to | |
2101 | * the list if @page is final isolated subpage in the zspage. | |
2102 | */ | |
2103 | if (!is_zspage_isolated(zspage)) | |
2104 | putback_zspage(class, zspage); | |
2105 | ||
2106 | reset_page(page); | |
2107 | put_page(page); | |
2108 | page = newpage; | |
2109 | ||
dd4123f3 | 2110 | ret = MIGRATEPAGE_SUCCESS; |
48b4800a MK |
2111 | unpin_objects: |
2112 | for (addr = s_addr + offset; addr < s_addr + pos; | |
2113 | addr += class->size) { | |
2114 | head = obj_to_head(page, addr); | |
2115 | if (head & OBJ_ALLOCATED_TAG) { | |
2116 | handle = head & ~OBJ_ALLOCATED_TAG; | |
2117 | if (!testpin_tag(handle)) | |
2118 | BUG(); | |
2119 | unpin_tag(handle); | |
2120 | } | |
2121 | } | |
2122 | kunmap_atomic(s_addr); | |
2123 | unlock_class: | |
2124 | spin_unlock(&class->lock); | |
2125 | migrate_write_unlock(zspage); | |
2126 | ||
2127 | return ret; | |
2128 | } | |
2129 | ||
2130 | void zs_page_putback(struct page *page) | |
2131 | { | |
2132 | struct zs_pool *pool; | |
2133 | struct size_class *class; | |
2134 | int class_idx; | |
2135 | enum fullness_group fg; | |
2136 | struct address_space *mapping; | |
2137 | struct zspage *zspage; | |
2138 | ||
2139 | VM_BUG_ON_PAGE(!PageMovable(page), page); | |
2140 | VM_BUG_ON_PAGE(!PageIsolated(page), page); | |
2141 | ||
2142 | zspage = get_zspage(page); | |
2143 | get_zspage_mapping(zspage, &class_idx, &fg); | |
2144 | mapping = page_mapping(page); | |
2145 | pool = mapping->private_data; | |
2146 | class = pool->size_class[class_idx]; | |
2147 | ||
2148 | spin_lock(&class->lock); | |
2149 | dec_zspage_isolation(zspage); | |
2150 | if (!is_zspage_isolated(zspage)) { | |
2151 | fg = putback_zspage(class, zspage); | |
2152 | /* | |
2153 | * Due to page_lock, we cannot free zspage immediately | |
2154 | * so let's defer. | |
2155 | */ | |
2156 | if (fg == ZS_EMPTY) | |
2157 | schedule_work(&pool->free_work); | |
2158 | } | |
2159 | spin_unlock(&class->lock); | |
2160 | } | |
2161 | ||
2162 | const struct address_space_operations zsmalloc_aops = { | |
2163 | .isolate_page = zs_page_isolate, | |
2164 | .migratepage = zs_page_migrate, | |
2165 | .putback_page = zs_page_putback, | |
2166 | }; | |
2167 | ||
2168 | static int zs_register_migration(struct zs_pool *pool) | |
2169 | { | |
2170 | pool->inode = alloc_anon_inode(zsmalloc_mnt->mnt_sb); | |
2171 | if (IS_ERR(pool->inode)) { | |
2172 | pool->inode = NULL; | |
2173 | return 1; | |
2174 | } | |
2175 | ||
2176 | pool->inode->i_mapping->private_data = pool; | |
2177 | pool->inode->i_mapping->a_ops = &zsmalloc_aops; | |
2178 | return 0; | |
2179 | } | |
2180 | ||
2181 | static void zs_unregister_migration(struct zs_pool *pool) | |
2182 | { | |
2183 | flush_work(&pool->free_work); | |
c3491eca | 2184 | iput(pool->inode); |
48b4800a MK |
2185 | } |
2186 | ||
2187 | /* | |
2188 | * Caller should hold page_lock of all pages in the zspage | |
2189 | * In here, we cannot use zspage meta data. | |
2190 | */ | |
2191 | static void async_free_zspage(struct work_struct *work) | |
2192 | { | |
2193 | int i; | |
2194 | struct size_class *class; | |
2195 | unsigned int class_idx; | |
2196 | enum fullness_group fullness; | |
2197 | struct zspage *zspage, *tmp; | |
2198 | LIST_HEAD(free_pages); | |
2199 | struct zs_pool *pool = container_of(work, struct zs_pool, | |
2200 | free_work); | |
2201 | ||
2202 | for (i = 0; i < zs_size_classes; i++) { | |
2203 | class = pool->size_class[i]; | |
2204 | if (class->index != i) | |
2205 | continue; | |
2206 | ||
2207 | spin_lock(&class->lock); | |
2208 | list_splice_init(&class->fullness_list[ZS_EMPTY], &free_pages); | |
2209 | spin_unlock(&class->lock); | |
2210 | } | |
2211 | ||
2212 | ||
2213 | list_for_each_entry_safe(zspage, tmp, &free_pages, list) { | |
2214 | list_del(&zspage->list); | |
2215 | lock_zspage(zspage); | |
2216 | ||
2217 | get_zspage_mapping(zspage, &class_idx, &fullness); | |
2218 | VM_BUG_ON(fullness != ZS_EMPTY); | |
2219 | class = pool->size_class[class_idx]; | |
2220 | spin_lock(&class->lock); | |
2221 | __free_zspage(pool, pool->size_class[class_idx], zspage); | |
2222 | spin_unlock(&class->lock); | |
2223 | } | |
2224 | }; | |
2225 | ||
2226 | static void kick_deferred_free(struct zs_pool *pool) | |
2227 | { | |
2228 | schedule_work(&pool->free_work); | |
2229 | } | |
2230 | ||
2231 | static void init_deferred_free(struct zs_pool *pool) | |
2232 | { | |
2233 | INIT_WORK(&pool->free_work, async_free_zspage); | |
2234 | } | |
2235 | ||
2236 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage) | |
2237 | { | |
2238 | struct page *page = get_first_page(zspage); | |
2239 | ||
2240 | do { | |
2241 | WARN_ON(!trylock_page(page)); | |
2242 | __SetPageMovable(page, pool->inode->i_mapping); | |
2243 | unlock_page(page); | |
2244 | } while ((page = get_next_page(page)) != NULL); | |
2245 | } | |
2246 | #endif | |
2247 | ||
04f05909 SS |
2248 | /* |
2249 | * | |
2250 | * Based on the number of unused allocated objects calculate | |
2251 | * and return the number of pages that we can free. | |
04f05909 SS |
2252 | */ |
2253 | static unsigned long zs_can_compact(struct size_class *class) | |
2254 | { | |
2255 | unsigned long obj_wasted; | |
44f43e99 SS |
2256 | unsigned long obj_allocated = zs_stat_get(class, OBJ_ALLOCATED); |
2257 | unsigned long obj_used = zs_stat_get(class, OBJ_USED); | |
04f05909 | 2258 | |
44f43e99 SS |
2259 | if (obj_allocated <= obj_used) |
2260 | return 0; | |
04f05909 | 2261 | |
44f43e99 | 2262 | obj_wasted = obj_allocated - obj_used; |
b4fd07a0 | 2263 | obj_wasted /= class->objs_per_zspage; |
04f05909 | 2264 | |
6cbf16b3 | 2265 | return obj_wasted * class->pages_per_zspage; |
04f05909 SS |
2266 | } |
2267 | ||
7d3f3938 | 2268 | static void __zs_compact(struct zs_pool *pool, struct size_class *class) |
312fcae2 | 2269 | { |
c27cbaa3 GM |
2270 | struct zs_compact_control cc = { |
2271 | .nr_migrated_obj = 0, | |
2272 | .nr_freed_pages = 0, | |
2273 | }; | |
3783689a MK |
2274 | struct zspage *src_zspage; |
2275 | struct zspage *dst_zspage = NULL; | |
312fcae2 | 2276 | |
312fcae2 | 2277 | spin_lock(&class->lock); |
3783689a | 2278 | while ((src_zspage = isolate_zspage(class, true))) { |
312fcae2 | 2279 | |
04f05909 SS |
2280 | if (!zs_can_compact(class)) |
2281 | break; | |
2282 | ||
41b88e14 | 2283 | cc.obj_idx = 0; |
48b4800a | 2284 | cc.s_page = get_first_page(src_zspage); |
312fcae2 | 2285 | |
3783689a | 2286 | while ((dst_zspage = isolate_zspage(class, false))) { |
48b4800a | 2287 | cc.d_page = get_first_page(dst_zspage); |
312fcae2 | 2288 | /* |
0dc63d48 SS |
2289 | * If there is no more space in dst_page, resched |
2290 | * and see if anyone had allocated another zspage. | |
312fcae2 MK |
2291 | */ |
2292 | if (!migrate_zspage(pool, class, &cc)) | |
2293 | break; | |
2294 | ||
4aa409ca | 2295 | putback_zspage(class, dst_zspage); |
312fcae2 MK |
2296 | } |
2297 | ||
2298 | /* Stop if we couldn't find slot */ | |
3783689a | 2299 | if (dst_zspage == NULL) |
312fcae2 MK |
2300 | break; |
2301 | ||
4aa409ca MK |
2302 | putback_zspage(class, dst_zspage); |
2303 | if (putback_zspage(class, src_zspage) == ZS_EMPTY) { | |
48b4800a | 2304 | free_zspage(pool, class, src_zspage); |
c27cbaa3 | 2305 | cc.nr_freed_pages += class->pages_per_zspage; |
4aa409ca | 2306 | } |
312fcae2 | 2307 | spin_unlock(&class->lock); |
312fcae2 MK |
2308 | cond_resched(); |
2309 | spin_lock(&class->lock); | |
2310 | } | |
2311 | ||
3783689a | 2312 | if (src_zspage) |
4aa409ca | 2313 | putback_zspage(class, src_zspage); |
312fcae2 | 2314 | |
7d3f3938 | 2315 | spin_unlock(&class->lock); |
c27cbaa3 GM |
2316 | |
2317 | pool->stats.pages_compacted += cc.nr_freed_pages; | |
2318 | trace_zs_compact(class->index, cc.nr_migrated_obj, cc.nr_freed_pages); | |
312fcae2 MK |
2319 | } |
2320 | ||
2321 | unsigned long zs_compact(struct zs_pool *pool) | |
2322 | { | |
2323 | int i; | |
312fcae2 | 2324 | struct size_class *class; |
40bf5a9d GM |
2325 | unsigned long pages_compacted_before = pool->stats.pages_compacted; |
2326 | ||
c27cbaa3 | 2327 | trace_zs_compact_start(pool->name); |
312fcae2 MK |
2328 | |
2329 | for (i = zs_size_classes - 1; i >= 0; i--) { | |
2330 | class = pool->size_class[i]; | |
2331 | if (!class) | |
2332 | continue; | |
2333 | if (class->index != i) | |
2334 | continue; | |
7d3f3938 | 2335 | __zs_compact(pool, class); |
312fcae2 MK |
2336 | } |
2337 | ||
c27cbaa3 GM |
2338 | trace_zs_compact_end(pool->name, |
2339 | pool->stats.pages_compacted - pages_compacted_before); | |
40bf5a9d | 2340 | |
860c707d | 2341 | return pool->stats.pages_compacted; |
312fcae2 MK |
2342 | } |
2343 | EXPORT_SYMBOL_GPL(zs_compact); | |
61989a80 | 2344 | |
7d3f3938 SS |
2345 | void zs_pool_stats(struct zs_pool *pool, struct zs_pool_stats *stats) |
2346 | { | |
2347 | memcpy(stats, &pool->stats, sizeof(struct zs_pool_stats)); | |
2348 | } | |
2349 | EXPORT_SYMBOL_GPL(zs_pool_stats); | |
2350 | ||
ab9d306d SS |
2351 | static unsigned long zs_shrinker_scan(struct shrinker *shrinker, |
2352 | struct shrink_control *sc) | |
2353 | { | |
2354 | unsigned long pages_freed; | |
2355 | struct zs_pool *pool = container_of(shrinker, struct zs_pool, | |
2356 | shrinker); | |
2357 | ||
2358 | pages_freed = pool->stats.pages_compacted; | |
2359 | /* | |
2360 | * Compact classes and calculate compaction delta. | |
2361 | * Can run concurrently with a manually triggered | |
2362 | * (by user) compaction. | |
2363 | */ | |
2364 | pages_freed = zs_compact(pool) - pages_freed; | |
2365 | ||
2366 | return pages_freed ? pages_freed : SHRINK_STOP; | |
2367 | } | |
2368 | ||
2369 | static unsigned long zs_shrinker_count(struct shrinker *shrinker, | |
2370 | struct shrink_control *sc) | |
2371 | { | |
2372 | int i; | |
2373 | struct size_class *class; | |
2374 | unsigned long pages_to_free = 0; | |
2375 | struct zs_pool *pool = container_of(shrinker, struct zs_pool, | |
2376 | shrinker); | |
2377 | ||
ab9d306d SS |
2378 | for (i = zs_size_classes - 1; i >= 0; i--) { |
2379 | class = pool->size_class[i]; | |
2380 | if (!class) | |
2381 | continue; | |
2382 | if (class->index != i) | |
2383 | continue; | |
2384 | ||
ab9d306d | 2385 | pages_to_free += zs_can_compact(class); |
ab9d306d SS |
2386 | } |
2387 | ||
2388 | return pages_to_free; | |
2389 | } | |
2390 | ||
2391 | static void zs_unregister_shrinker(struct zs_pool *pool) | |
2392 | { | |
2393 | if (pool->shrinker_enabled) { | |
2394 | unregister_shrinker(&pool->shrinker); | |
2395 | pool->shrinker_enabled = false; | |
2396 | } | |
2397 | } | |
2398 | ||
2399 | static int zs_register_shrinker(struct zs_pool *pool) | |
2400 | { | |
2401 | pool->shrinker.scan_objects = zs_shrinker_scan; | |
2402 | pool->shrinker.count_objects = zs_shrinker_count; | |
2403 | pool->shrinker.batch = 0; | |
2404 | pool->shrinker.seeks = DEFAULT_SEEKS; | |
2405 | ||
2406 | return register_shrinker(&pool->shrinker); | |
2407 | } | |
2408 | ||
00a61d86 | 2409 | /** |
66cdef66 | 2410 | * zs_create_pool - Creates an allocation pool to work from. |
fd854463 | 2411 | * @name: pool name to be created |
166cfda7 | 2412 | * |
66cdef66 GM |
2413 | * This function must be called before anything when using |
2414 | * the zsmalloc allocator. | |
166cfda7 | 2415 | * |
66cdef66 GM |
2416 | * On success, a pointer to the newly created pool is returned, |
2417 | * otherwise NULL. | |
396b7fd6 | 2418 | */ |
d0d8da2d | 2419 | struct zs_pool *zs_create_pool(const char *name) |
61989a80 | 2420 | { |
66cdef66 GM |
2421 | int i; |
2422 | struct zs_pool *pool; | |
2423 | struct size_class *prev_class = NULL; | |
61989a80 | 2424 | |
66cdef66 GM |
2425 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
2426 | if (!pool) | |
2427 | return NULL; | |
61989a80 | 2428 | |
48b4800a | 2429 | init_deferred_free(pool); |
66cdef66 GM |
2430 | pool->size_class = kcalloc(zs_size_classes, sizeof(struct size_class *), |
2431 | GFP_KERNEL); | |
2432 | if (!pool->size_class) { | |
2433 | kfree(pool); | |
2434 | return NULL; | |
2435 | } | |
61989a80 | 2436 | |
2e40e163 MK |
2437 | pool->name = kstrdup(name, GFP_KERNEL); |
2438 | if (!pool->name) | |
2439 | goto err; | |
2440 | ||
3783689a | 2441 | if (create_cache(pool)) |
2e40e163 MK |
2442 | goto err; |
2443 | ||
c60369f0 | 2444 | /* |
66cdef66 GM |
2445 | * Iterate reversly, because, size of size_class that we want to use |
2446 | * for merging should be larger or equal to current size. | |
c60369f0 | 2447 | */ |
66cdef66 GM |
2448 | for (i = zs_size_classes - 1; i >= 0; i--) { |
2449 | int size; | |
2450 | int pages_per_zspage; | |
64d90465 | 2451 | int objs_per_zspage; |
66cdef66 | 2452 | struct size_class *class; |
3783689a | 2453 | int fullness = 0; |
c60369f0 | 2454 | |
66cdef66 GM |
2455 | size = ZS_MIN_ALLOC_SIZE + i * ZS_SIZE_CLASS_DELTA; |
2456 | if (size > ZS_MAX_ALLOC_SIZE) | |
2457 | size = ZS_MAX_ALLOC_SIZE; | |
2458 | pages_per_zspage = get_pages_per_zspage(size); | |
64d90465 | 2459 | objs_per_zspage = pages_per_zspage * PAGE_SIZE / size; |
61989a80 | 2460 | |
66cdef66 GM |
2461 | /* |
2462 | * size_class is used for normal zsmalloc operation such | |
2463 | * as alloc/free for that size. Although it is natural that we | |
2464 | * have one size_class for each size, there is a chance that we | |
2465 | * can get more memory utilization if we use one size_class for | |
2466 | * many different sizes whose size_class have same | |
2467 | * characteristics. So, we makes size_class point to | |
2468 | * previous size_class if possible. | |
2469 | */ | |
2470 | if (prev_class) { | |
64d90465 | 2471 | if (can_merge(prev_class, pages_per_zspage, objs_per_zspage)) { |
66cdef66 GM |
2472 | pool->size_class[i] = prev_class; |
2473 | continue; | |
2474 | } | |
2475 | } | |
2476 | ||
2477 | class = kzalloc(sizeof(struct size_class), GFP_KERNEL); | |
2478 | if (!class) | |
2479 | goto err; | |
2480 | ||
2481 | class->size = size; | |
2482 | class->index = i; | |
2483 | class->pages_per_zspage = pages_per_zspage; | |
64d90465 | 2484 | class->objs_per_zspage = objs_per_zspage; |
66cdef66 GM |
2485 | spin_lock_init(&class->lock); |
2486 | pool->size_class[i] = class; | |
48b4800a MK |
2487 | for (fullness = ZS_EMPTY; fullness < NR_ZS_FULLNESS; |
2488 | fullness++) | |
3783689a | 2489 | INIT_LIST_HEAD(&class->fullness_list[fullness]); |
66cdef66 GM |
2490 | |
2491 | prev_class = class; | |
61989a80 NG |
2492 | } |
2493 | ||
d34f6157 DS |
2494 | /* debug only, don't abort if it fails */ |
2495 | zs_pool_stat_create(pool, name); | |
0f050d99 | 2496 | |
48b4800a MK |
2497 | if (zs_register_migration(pool)) |
2498 | goto err; | |
2499 | ||
ab9d306d SS |
2500 | /* |
2501 | * Not critical, we still can use the pool | |
2502 | * and user can trigger compaction manually. | |
2503 | */ | |
2504 | if (zs_register_shrinker(pool) == 0) | |
2505 | pool->shrinker_enabled = true; | |
66cdef66 GM |
2506 | return pool; |
2507 | ||
2508 | err: | |
2509 | zs_destroy_pool(pool); | |
2510 | return NULL; | |
61989a80 | 2511 | } |
66cdef66 | 2512 | EXPORT_SYMBOL_GPL(zs_create_pool); |
61989a80 | 2513 | |
66cdef66 | 2514 | void zs_destroy_pool(struct zs_pool *pool) |
61989a80 | 2515 | { |
66cdef66 | 2516 | int i; |
61989a80 | 2517 | |
ab9d306d | 2518 | zs_unregister_shrinker(pool); |
48b4800a | 2519 | zs_unregister_migration(pool); |
0f050d99 GM |
2520 | zs_pool_stat_destroy(pool); |
2521 | ||
66cdef66 GM |
2522 | for (i = 0; i < zs_size_classes; i++) { |
2523 | int fg; | |
2524 | struct size_class *class = pool->size_class[i]; | |
61989a80 | 2525 | |
66cdef66 GM |
2526 | if (!class) |
2527 | continue; | |
61989a80 | 2528 | |
66cdef66 GM |
2529 | if (class->index != i) |
2530 | continue; | |
61989a80 | 2531 | |
48b4800a | 2532 | for (fg = ZS_EMPTY; fg < NR_ZS_FULLNESS; fg++) { |
3783689a | 2533 | if (!list_empty(&class->fullness_list[fg])) { |
66cdef66 GM |
2534 | pr_info("Freeing non-empty class with size %db, fullness group %d\n", |
2535 | class->size, fg); | |
2536 | } | |
2537 | } | |
2538 | kfree(class); | |
2539 | } | |
f553646a | 2540 | |
3783689a | 2541 | destroy_cache(pool); |
66cdef66 | 2542 | kfree(pool->size_class); |
0f050d99 | 2543 | kfree(pool->name); |
66cdef66 GM |
2544 | kfree(pool); |
2545 | } | |
2546 | EXPORT_SYMBOL_GPL(zs_destroy_pool); | |
b7418510 | 2547 | |
66cdef66 GM |
2548 | static int __init zs_init(void) |
2549 | { | |
48b4800a MK |
2550 | int ret; |
2551 | ||
2552 | ret = zsmalloc_mount(); | |
2553 | if (ret) | |
2554 | goto out; | |
2555 | ||
2556 | ret = zs_register_cpu_notifier(); | |
66cdef66 | 2557 | |
0f050d99 GM |
2558 | if (ret) |
2559 | goto notifier_fail; | |
66cdef66 GM |
2560 | |
2561 | init_zs_size_classes(); | |
2562 | ||
2563 | #ifdef CONFIG_ZPOOL | |
2564 | zpool_register_driver(&zs_zpool_driver); | |
2565 | #endif | |
0f050d99 | 2566 | |
4abaac9b DS |
2567 | zs_stat_init(); |
2568 | ||
66cdef66 | 2569 | return 0; |
0f050d99 | 2570 | |
0f050d99 GM |
2571 | notifier_fail: |
2572 | zs_unregister_cpu_notifier(); | |
48b4800a MK |
2573 | zsmalloc_unmount(); |
2574 | out: | |
0f050d99 | 2575 | return ret; |
61989a80 | 2576 | } |
61989a80 | 2577 | |
66cdef66 | 2578 | static void __exit zs_exit(void) |
61989a80 | 2579 | { |
66cdef66 GM |
2580 | #ifdef CONFIG_ZPOOL |
2581 | zpool_unregister_driver(&zs_zpool_driver); | |
2582 | #endif | |
48b4800a | 2583 | zsmalloc_unmount(); |
66cdef66 | 2584 | zs_unregister_cpu_notifier(); |
0f050d99 GM |
2585 | |
2586 | zs_stat_exit(); | |
61989a80 | 2587 | } |
069f101f BH |
2588 | |
2589 | module_init(zs_init); | |
2590 | module_exit(zs_exit); | |
2591 | ||
2592 | MODULE_LICENSE("Dual BSD/GPL"); | |
2593 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); |