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 |
ffedd09f | 20 | * page->index: links together all component pages of a zspage |
48b4800a MK |
21 | * For the huge page, this is always 0, so we use this field |
22 | * to store handle. | |
ffedd09f | 23 | * page->page_type: 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 | |
399d8eeb | 27 | * PG_owner_priv_1: identifies the huge component page |
2db51dae NG |
28 | * |
29 | */ | |
30 | ||
4abaac9b DS |
31 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
32 | ||
b475d42d MK |
33 | /* |
34 | * lock ordering: | |
35 | * page_lock | |
c0547d0b | 36 | * pool->lock |
b475d42d MK |
37 | * zspage->lock |
38 | */ | |
39 | ||
61989a80 NG |
40 | #include <linux/module.h> |
41 | #include <linux/kernel.h> | |
312fcae2 | 42 | #include <linux/sched.h> |
61989a80 NG |
43 | #include <linux/bitops.h> |
44 | #include <linux/errno.h> | |
45 | #include <linux/highmem.h> | |
61989a80 NG |
46 | #include <linux/string.h> |
47 | #include <linux/slab.h> | |
ca5999fd | 48 | #include <linux/pgtable.h> |
65fddcfc | 49 | #include <asm/tlbflush.h> |
61989a80 NG |
50 | #include <linux/cpumask.h> |
51 | #include <linux/cpu.h> | |
0cbb613f | 52 | #include <linux/vmalloc.h> |
759b26b2 | 53 | #include <linux/preempt.h> |
0959c63f | 54 | #include <linux/spinlock.h> |
93144ca3 | 55 | #include <linux/shrinker.h> |
0959c63f | 56 | #include <linux/types.h> |
0f050d99 | 57 | #include <linux/debugfs.h> |
bcf1647d | 58 | #include <linux/zsmalloc.h> |
c795779d | 59 | #include <linux/zpool.h> |
dd4123f3 | 60 | #include <linux/migrate.h> |
701d6785 | 61 | #include <linux/wait.h> |
48b4800a | 62 | #include <linux/pagemap.h> |
cdc346b3 | 63 | #include <linux/fs.h> |
a3726599 | 64 | #include <linux/local_lock.h> |
48b4800a MK |
65 | |
66 | #define ZSPAGE_MAGIC 0x58 | |
0959c63f SJ |
67 | |
68 | /* | |
cb152a1a | 69 | * This must be power of 2 and greater than or equal to sizeof(link_free). |
0959c63f SJ |
70 | * These two conditions ensure that any 'struct link_free' itself doesn't |
71 | * span more than 1 page which avoids complex case of mapping 2 pages simply | |
72 | * to restore link_free pointer values. | |
73 | */ | |
74 | #define ZS_ALIGN 8 | |
75 | ||
2e40e163 MK |
76 | #define ZS_HANDLE_SIZE (sizeof(unsigned long)) |
77 | ||
0959c63f SJ |
78 | /* |
79 | * Object location (<PFN>, <obj_idx>) is encoded as | |
b956b5ac | 80 | * a single (unsigned long) handle value. |
0959c63f | 81 | * |
bfd093f5 | 82 | * Note that object index <obj_idx> starts from 0. |
0959c63f SJ |
83 | * |
84 | * This is made more complicated by various memory models and PAE. | |
85 | */ | |
86 | ||
02390b87 KS |
87 | #ifndef MAX_POSSIBLE_PHYSMEM_BITS |
88 | #ifdef MAX_PHYSMEM_BITS | |
89 | #define MAX_POSSIBLE_PHYSMEM_BITS MAX_PHYSMEM_BITS | |
90 | #else | |
0959c63f SJ |
91 | /* |
92 | * If this definition of MAX_PHYSMEM_BITS is used, OBJ_INDEX_BITS will just | |
93 | * be PAGE_SHIFT | |
94 | */ | |
02390b87 | 95 | #define MAX_POSSIBLE_PHYSMEM_BITS BITS_PER_LONG |
0959c63f SJ |
96 | #endif |
97 | #endif | |
02390b87 KS |
98 | |
99 | #define _PFN_BITS (MAX_POSSIBLE_PHYSMEM_BITS - PAGE_SHIFT) | |
312fcae2 | 100 | |
312fcae2 MK |
101 | /* |
102 | * Head in allocated object should have OBJ_ALLOCATED_TAG | |
103 | * to identify the object was allocated or not. | |
104 | * It's okay to add the status bit in the least bit because | |
105 | * header keeps handle which is 4byte-aligned address so we | |
106 | * have room for two bit at least. | |
107 | */ | |
108 | #define OBJ_ALLOCATED_TAG 1 | |
85b32581 | 109 | |
85b32581 NP |
110 | #define OBJ_TAG_BITS 1 |
111 | #define OBJ_TAG_MASK OBJ_ALLOCATED_TAG | |
85b32581 | 112 | |
26e93839 | 113 | #define OBJ_INDEX_BITS (BITS_PER_LONG - _PFN_BITS) |
0959c63f SJ |
114 | #define OBJ_INDEX_MASK ((_AC(1, UL) << OBJ_INDEX_BITS) - 1) |
115 | ||
a41ec880 | 116 | #define HUGE_BITS 1 |
4c7ac972 | 117 | #define FULLNESS_BITS 4 |
cf8e0fed | 118 | #define CLASS_BITS 8 |
cf8e0fed JM |
119 | #define MAGIC_VAL_BITS 8 |
120 | ||
0959c63f | 121 | #define MAX(a, b) ((a) >= (b) ? (a) : (b)) |
4ff93b29 SS |
122 | |
123 | #define ZS_MAX_PAGES_PER_ZSPAGE (_AC(CONFIG_ZSMALLOC_CHAIN_SIZE, UL)) | |
124 | ||
0959c63f SJ |
125 | /* ZS_MIN_ALLOC_SIZE must be multiple of ZS_ALIGN */ |
126 | #define ZS_MIN_ALLOC_SIZE \ | |
127 | MAX(32, (ZS_MAX_PAGES_PER_ZSPAGE << PAGE_SHIFT >> OBJ_INDEX_BITS)) | |
2e40e163 | 128 | /* each chunk includes extra space to keep handle */ |
7b60a685 | 129 | #define ZS_MAX_ALLOC_SIZE PAGE_SIZE |
0959c63f SJ |
130 | |
131 | /* | |
7eb52512 | 132 | * On systems with 4K page size, this gives 255 size classes! There is a |
0959c63f SJ |
133 | * trader-off here: |
134 | * - Large number of size classes is potentially wasteful as free page are | |
135 | * spread across these classes | |
136 | * - Small number of size classes causes large internal fragmentation | |
137 | * - Probably its better to use specific size classes (empirically | |
138 | * determined). NOTE: all those class sizes must be set as multiple of | |
139 | * ZS_ALIGN to make sure link_free itself never has to span 2 pages. | |
140 | * | |
141 | * ZS_MIN_ALLOC_SIZE and ZS_SIZE_CLASS_DELTA must be multiple of ZS_ALIGN | |
142 | * (reason above) | |
143 | */ | |
3783689a | 144 | #define ZS_SIZE_CLASS_DELTA (PAGE_SIZE >> CLASS_BITS) |
cf8e0fed JM |
145 | #define ZS_SIZE_CLASSES (DIV_ROUND_UP(ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE, \ |
146 | ZS_SIZE_CLASS_DELTA) + 1) | |
0959c63f | 147 | |
4c7ac972 SS |
148 | /* |
149 | * Pages are distinguished by the ratio of used memory (that is the ratio | |
150 | * of ->inuse objects to all objects that page can store). For example, | |
151 | * INUSE_RATIO_10 means that the ratio of used objects is > 0% and <= 10%. | |
152 | * | |
153 | * The number of fullness groups is not random. It allows us to keep | |
154 | * difference between the least busy page in the group (minimum permitted | |
155 | * number of ->inuse objects) and the most busy page (maximum permitted | |
156 | * number of ->inuse objects) at a reasonable value. | |
157 | */ | |
0959c63f | 158 | enum fullness_group { |
4c7ac972 SS |
159 | ZS_INUSE_RATIO_0, |
160 | ZS_INUSE_RATIO_10, | |
e1807d5d | 161 | /* NOTE: 8 more fullness groups here */ |
4c7ac972 SS |
162 | ZS_INUSE_RATIO_99 = 10, |
163 | ZS_INUSE_RATIO_100, | |
164 | NR_FULLNESS_GROUPS, | |
0959c63f SJ |
165 | }; |
166 | ||
3828a764 | 167 | enum class_stat_type { |
4c7ac972 SS |
168 | /* NOTE: stats for 12 fullness groups here: from inuse 0 to 100 */ |
169 | ZS_OBJS_ALLOCATED = NR_FULLNESS_GROUPS, | |
170 | ZS_OBJS_INUSE, | |
171 | NR_CLASS_STAT_TYPES, | |
0f050d99 GM |
172 | }; |
173 | ||
0f050d99 | 174 | struct zs_size_stat { |
4c7ac972 | 175 | unsigned long objs[NR_CLASS_STAT_TYPES]; |
0f050d99 GM |
176 | }; |
177 | ||
57244594 SS |
178 | #ifdef CONFIG_ZSMALLOC_STAT |
179 | static struct dentry *zs_stat_root; | |
0f050d99 GM |
180 | #endif |
181 | ||
010b495e | 182 | static size_t huge_class_size; |
0959c63f SJ |
183 | |
184 | struct size_class { | |
4c7ac972 | 185 | struct list_head fullness_list[NR_FULLNESS_GROUPS]; |
0959c63f SJ |
186 | /* |
187 | * Size of objects stored in this class. Must be multiple | |
188 | * of ZS_ALIGN. | |
189 | */ | |
190 | int size; | |
1fc6e27d | 191 | int objs_per_zspage; |
7dfa4612 WY |
192 | /* Number of PAGE_SIZE sized pages to combine to form a 'zspage' */ |
193 | int pages_per_zspage; | |
48b4800a MK |
194 | |
195 | unsigned int index; | |
196 | struct zs_size_stat stats; | |
0959c63f SJ |
197 | }; |
198 | ||
199 | /* | |
200 | * Placed within free objects to form a singly linked list. | |
3783689a | 201 | * For every zspage, zspage->freeobj gives head of this list. |
0959c63f SJ |
202 | * |
203 | * This must be power of 2 and less than or equal to ZS_ALIGN | |
204 | */ | |
205 | struct link_free { | |
2e40e163 MK |
206 | union { |
207 | /* | |
bfd093f5 | 208 | * Free object index; |
2e40e163 MK |
209 | * It's valid for non-allocated object |
210 | */ | |
bfd093f5 | 211 | unsigned long next; |
2e40e163 MK |
212 | /* |
213 | * Handle of allocated object. | |
214 | */ | |
215 | unsigned long handle; | |
216 | }; | |
0959c63f SJ |
217 | }; |
218 | ||
219 | struct zs_pool { | |
6f3526d6 | 220 | const char *name; |
0f050d99 | 221 | |
cf8e0fed | 222 | struct size_class *size_class[ZS_SIZE_CLASSES]; |
2e40e163 | 223 | struct kmem_cache *handle_cachep; |
3783689a | 224 | struct kmem_cache *zspage_cachep; |
0959c63f | 225 | |
13de8933 | 226 | atomic_long_t pages_allocated; |
0f050d99 | 227 | |
7d3f3938 | 228 | struct zs_pool_stats stats; |
ab9d306d SS |
229 | |
230 | /* Compact classes */ | |
c19b548b | 231 | struct shrinker *shrinker; |
93144ca3 | 232 | |
0f050d99 GM |
233 | #ifdef CONFIG_ZSMALLOC_STAT |
234 | struct dentry *stat_dentry; | |
235 | #endif | |
48b4800a | 236 | #ifdef CONFIG_COMPACTION |
48b4800a MK |
237 | struct work_struct free_work; |
238 | #endif | |
c0547d0b | 239 | spinlock_t lock; |
d2658f20 | 240 | atomic_t compaction_in_progress; |
0959c63f | 241 | }; |
61989a80 | 242 | |
3783689a MK |
243 | struct zspage { |
244 | struct { | |
a41ec880 | 245 | unsigned int huge:HUGE_BITS; |
3783689a | 246 | unsigned int fullness:FULLNESS_BITS; |
85d492f2 | 247 | unsigned int class:CLASS_BITS + 1; |
48b4800a | 248 | unsigned int magic:MAGIC_VAL_BITS; |
3783689a MK |
249 | }; |
250 | unsigned int inuse; | |
bfd093f5 | 251 | unsigned int freeobj; |
3783689a MK |
252 | struct page *first_page; |
253 | struct list_head list; /* fullness list */ | |
68f2736a | 254 | struct zs_pool *pool; |
48b4800a | 255 | rwlock_t lock; |
3783689a | 256 | }; |
61989a80 | 257 | |
f553646a | 258 | struct mapping_area { |
a3726599 | 259 | local_lock_t lock; |
f553646a | 260 | char *vm_buf; /* copy buffer for objects that span pages */ |
f553646a SJ |
261 | char *vm_addr; /* address of kmap_atomic()'ed pages */ |
262 | enum zs_mapmode vm_mm; /* mapping mode */ | |
263 | }; | |
264 | ||
a41ec880 MK |
265 | /* huge object: pages_per_zspage == 1 && maxobj_per_zspage == 1 */ |
266 | static void SetZsHugePage(struct zspage *zspage) | |
267 | { | |
268 | zspage->huge = 1; | |
269 | } | |
270 | ||
271 | static bool ZsHugePage(struct zspage *zspage) | |
272 | { | |
273 | return zspage->huge; | |
274 | } | |
275 | ||
48b4800a MK |
276 | static void migrate_lock_init(struct zspage *zspage); |
277 | static void migrate_read_lock(struct zspage *zspage); | |
278 | static void migrate_read_unlock(struct zspage *zspage); | |
b475d42d | 279 | static void migrate_write_lock(struct zspage *zspage); |
b475d42d | 280 | static void migrate_write_unlock(struct zspage *zspage); |
568b567f CZ |
281 | |
282 | #ifdef CONFIG_COMPACTION | |
48b4800a MK |
283 | static void kick_deferred_free(struct zs_pool *pool); |
284 | static void init_deferred_free(struct zs_pool *pool); | |
285 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage); | |
286 | #else | |
48b4800a MK |
287 | static void kick_deferred_free(struct zs_pool *pool) {} |
288 | static void init_deferred_free(struct zs_pool *pool) {} | |
289 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage) {} | |
290 | #endif | |
291 | ||
3783689a | 292 | static int create_cache(struct zs_pool *pool) |
2e40e163 MK |
293 | { |
294 | pool->handle_cachep = kmem_cache_create("zs_handle", ZS_HANDLE_SIZE, | |
295 | 0, 0, NULL); | |
3783689a MK |
296 | if (!pool->handle_cachep) |
297 | return 1; | |
298 | ||
299 | pool->zspage_cachep = kmem_cache_create("zspage", sizeof(struct zspage), | |
300 | 0, 0, NULL); | |
301 | if (!pool->zspage_cachep) { | |
302 | kmem_cache_destroy(pool->handle_cachep); | |
303 | pool->handle_cachep = NULL; | |
304 | return 1; | |
305 | } | |
306 | ||
307 | return 0; | |
2e40e163 MK |
308 | } |
309 | ||
3783689a | 310 | static void destroy_cache(struct zs_pool *pool) |
2e40e163 | 311 | { |
cd10add0 | 312 | kmem_cache_destroy(pool->handle_cachep); |
3783689a | 313 | kmem_cache_destroy(pool->zspage_cachep); |
2e40e163 MK |
314 | } |
315 | ||
3783689a | 316 | static unsigned long cache_alloc_handle(struct zs_pool *pool, gfp_t gfp) |
2e40e163 MK |
317 | { |
318 | return (unsigned long)kmem_cache_alloc(pool->handle_cachep, | |
48b4800a | 319 | gfp & ~(__GFP_HIGHMEM|__GFP_MOVABLE)); |
2e40e163 MK |
320 | } |
321 | ||
3783689a | 322 | static void cache_free_handle(struct zs_pool *pool, unsigned long handle) |
2e40e163 MK |
323 | { |
324 | kmem_cache_free(pool->handle_cachep, (void *)handle); | |
325 | } | |
326 | ||
3783689a MK |
327 | static struct zspage *cache_alloc_zspage(struct zs_pool *pool, gfp_t flags) |
328 | { | |
f0231305 | 329 | return kmem_cache_zalloc(pool->zspage_cachep, |
48b4800a | 330 | flags & ~(__GFP_HIGHMEM|__GFP_MOVABLE)); |
399d8eeb | 331 | } |
3783689a MK |
332 | |
333 | static void cache_free_zspage(struct zs_pool *pool, struct zspage *zspage) | |
334 | { | |
335 | kmem_cache_free(pool->zspage_cachep, zspage); | |
336 | } | |
337 | ||
c0547d0b | 338 | /* pool->lock(which owns the handle) synchronizes races */ |
2e40e163 MK |
339 | static void record_obj(unsigned long handle, unsigned long obj) |
340 | { | |
b475d42d | 341 | *(unsigned long *)handle = obj; |
2e40e163 MK |
342 | } |
343 | ||
c795779d DS |
344 | /* zpool driver */ |
345 | ||
346 | #ifdef CONFIG_ZPOOL | |
347 | ||
35499e2b | 348 | static void *zs_zpool_create(const char *name, gfp_t gfp) |
c795779d | 349 | { |
d0d8da2d SS |
350 | /* |
351 | * Ignore global gfp flags: zs_malloc() may be invoked from | |
352 | * different contexts and its caller must provide a valid | |
353 | * gfp mask. | |
354 | */ | |
b3067742 | 355 | return zs_create_pool(name); |
c795779d DS |
356 | } |
357 | ||
358 | static void zs_zpool_destroy(void *pool) | |
359 | { | |
360 | zs_destroy_pool(pool); | |
361 | } | |
362 | ||
363 | static int zs_zpool_malloc(void *pool, size_t size, gfp_t gfp, | |
364 | unsigned long *handle) | |
365 | { | |
d0d8da2d | 366 | *handle = zs_malloc(pool, size, gfp); |
c7e6f17b | 367 | |
65917b53 | 368 | if (IS_ERR_VALUE(*handle)) |
c7e6f17b HZ |
369 | return PTR_ERR((void *)*handle); |
370 | return 0; | |
c795779d DS |
371 | } |
372 | static void zs_zpool_free(void *pool, unsigned long handle) | |
373 | { | |
374 | zs_free(pool, handle); | |
375 | } | |
376 | ||
c795779d DS |
377 | static void *zs_zpool_map(void *pool, unsigned long handle, |
378 | enum zpool_mapmode mm) | |
379 | { | |
380 | enum zs_mapmode zs_mm; | |
381 | ||
382 | switch (mm) { | |
383 | case ZPOOL_MM_RO: | |
384 | zs_mm = ZS_MM_RO; | |
385 | break; | |
386 | case ZPOOL_MM_WO: | |
387 | zs_mm = ZS_MM_WO; | |
388 | break; | |
e4a9bc58 | 389 | case ZPOOL_MM_RW: |
c795779d DS |
390 | default: |
391 | zs_mm = ZS_MM_RW; | |
392 | break; | |
393 | } | |
394 | ||
395 | return zs_map_object(pool, handle, zs_mm); | |
396 | } | |
397 | static void zs_zpool_unmap(void *pool, unsigned long handle) | |
398 | { | |
399 | zs_unmap_object(pool, handle); | |
400 | } | |
401 | ||
402 | static u64 zs_zpool_total_size(void *pool) | |
403 | { | |
722cdc17 | 404 | return zs_get_total_pages(pool) << PAGE_SHIFT; |
c795779d DS |
405 | } |
406 | ||
407 | static struct zpool_driver zs_zpool_driver = { | |
c165f25d HZ |
408 | .type = "zsmalloc", |
409 | .owner = THIS_MODULE, | |
410 | .create = zs_zpool_create, | |
411 | .destroy = zs_zpool_destroy, | |
412 | .malloc_support_movable = true, | |
413 | .malloc = zs_zpool_malloc, | |
414 | .free = zs_zpool_free, | |
415 | .map = zs_zpool_map, | |
416 | .unmap = zs_zpool_unmap, | |
417 | .total_size = zs_zpool_total_size, | |
c795779d DS |
418 | }; |
419 | ||
137f8cff | 420 | MODULE_ALIAS("zpool-zsmalloc"); |
c795779d DS |
421 | #endif /* CONFIG_ZPOOL */ |
422 | ||
61989a80 | 423 | /* per-cpu VM mapping areas for zspage accesses that cross page boundaries */ |
a3726599 MG |
424 | static DEFINE_PER_CPU(struct mapping_area, zs_map_area) = { |
425 | .lock = INIT_LOCAL_LOCK(lock), | |
426 | }; | |
61989a80 | 427 | |
3457f414 | 428 | static __maybe_unused int is_first_page(struct page *page) |
61989a80 | 429 | { |
a27545bf | 430 | return PagePrivate(page); |
61989a80 NG |
431 | } |
432 | ||
c0547d0b | 433 | /* Protected by pool->lock */ |
3783689a | 434 | static inline int get_zspage_inuse(struct zspage *zspage) |
4f42047b | 435 | { |
3783689a | 436 | return zspage->inuse; |
4f42047b MK |
437 | } |
438 | ||
4f42047b | 439 | |
3783689a | 440 | static inline void mod_zspage_inuse(struct zspage *zspage, int val) |
4f42047b | 441 | { |
3783689a | 442 | zspage->inuse += val; |
4f42047b MK |
443 | } |
444 | ||
48b4800a | 445 | static inline struct page *get_first_page(struct zspage *zspage) |
4f42047b | 446 | { |
48b4800a | 447 | struct page *first_page = zspage->first_page; |
3783689a | 448 | |
48b4800a MK |
449 | VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); |
450 | return first_page; | |
4f42047b MK |
451 | } |
452 | ||
671f2fa8 | 453 | static inline unsigned int get_first_obj_offset(struct page *page) |
4f42047b | 454 | { |
ffedd09f | 455 | return page->page_type; |
48b4800a | 456 | } |
3783689a | 457 | |
671f2fa8 | 458 | static inline void set_first_obj_offset(struct page *page, unsigned int offset) |
48b4800a | 459 | { |
ffedd09f | 460 | page->page_type = offset; |
4f42047b MK |
461 | } |
462 | ||
bfd093f5 | 463 | static inline unsigned int get_freeobj(struct zspage *zspage) |
4f42047b | 464 | { |
bfd093f5 | 465 | return zspage->freeobj; |
4f42047b MK |
466 | } |
467 | ||
bfd093f5 | 468 | static inline void set_freeobj(struct zspage *zspage, unsigned int obj) |
4f42047b | 469 | { |
bfd093f5 | 470 | zspage->freeobj = obj; |
4f42047b MK |
471 | } |
472 | ||
67f1c9cd | 473 | static struct size_class *zspage_class(struct zs_pool *pool, |
4c7ac972 | 474 | struct zspage *zspage) |
67f1c9cd MK |
475 | { |
476 | return pool->size_class[zspage->class]; | |
477 | } | |
478 | ||
c3e3e88a NC |
479 | /* |
480 | * zsmalloc divides the pool into various size classes where each | |
481 | * class maintains a list of zspages where each zspage is divided | |
482 | * into equal sized chunks. Each allocation falls into one of these | |
483 | * classes depending on its size. This function returns index of the | |
cb152a1a | 484 | * size class which has chunk size big enough to hold the given size. |
c3e3e88a | 485 | */ |
61989a80 NG |
486 | static int get_size_class_index(int size) |
487 | { | |
488 | int idx = 0; | |
489 | ||
490 | if (likely(size > ZS_MIN_ALLOC_SIZE)) | |
491 | idx = DIV_ROUND_UP(size - ZS_MIN_ALLOC_SIZE, | |
492 | ZS_SIZE_CLASS_DELTA); | |
493 | ||
cf8e0fed | 494 | return min_t(int, ZS_SIZE_CLASSES - 1, idx); |
61989a80 NG |
495 | } |
496 | ||
3828a764 | 497 | static inline void class_stat_inc(struct size_class *class, |
3eb95fea | 498 | int type, unsigned long cnt) |
248ca1b0 | 499 | { |
48b4800a | 500 | class->stats.objs[type] += cnt; |
248ca1b0 MK |
501 | } |
502 | ||
3828a764 | 503 | static inline void class_stat_dec(struct size_class *class, |
3eb95fea | 504 | int type, unsigned long cnt) |
248ca1b0 | 505 | { |
48b4800a | 506 | class->stats.objs[type] -= cnt; |
248ca1b0 MK |
507 | } |
508 | ||
4c7ac972 | 509 | static inline unsigned long zs_stat_get(struct size_class *class, int type) |
248ca1b0 | 510 | { |
48b4800a | 511 | return class->stats.objs[type]; |
248ca1b0 MK |
512 | } |
513 | ||
57244594 SS |
514 | #ifdef CONFIG_ZSMALLOC_STAT |
515 | ||
4abaac9b | 516 | static void __init zs_stat_init(void) |
248ca1b0 | 517 | { |
4abaac9b DS |
518 | if (!debugfs_initialized()) { |
519 | pr_warn("debugfs not available, stat dir not created\n"); | |
520 | return; | |
521 | } | |
248ca1b0 MK |
522 | |
523 | zs_stat_root = debugfs_create_dir("zsmalloc", NULL); | |
248ca1b0 MK |
524 | } |
525 | ||
526 | static void __exit zs_stat_exit(void) | |
527 | { | |
528 | debugfs_remove_recursive(zs_stat_root); | |
529 | } | |
530 | ||
1120ed54 SS |
531 | static unsigned long zs_can_compact(struct size_class *class); |
532 | ||
248ca1b0 MK |
533 | static int zs_stats_size_show(struct seq_file *s, void *v) |
534 | { | |
e1807d5d | 535 | int i, fg; |
248ca1b0 MK |
536 | struct zs_pool *pool = s->private; |
537 | struct size_class *class; | |
538 | int objs_per_zspage; | |
1120ed54 | 539 | unsigned long obj_allocated, obj_used, pages_used, freeable; |
248ca1b0 | 540 | unsigned long total_objs = 0, total_used_objs = 0, total_pages = 0; |
1120ed54 | 541 | unsigned long total_freeable = 0; |
e1807d5d | 542 | unsigned long inuse_totals[NR_FULLNESS_GROUPS] = {0, }; |
248ca1b0 | 543 | |
e1807d5d SS |
544 | seq_printf(s, " %5s %5s %9s %9s %9s %9s %9s %9s %9s %9s %9s %9s %9s %13s %10s %10s %16s %8s\n", |
545 | "class", "size", "10%", "20%", "30%", "40%", | |
546 | "50%", "60%", "70%", "80%", "90%", "99%", "100%", | |
248ca1b0 | 547 | "obj_allocated", "obj_used", "pages_used", |
1120ed54 | 548 | "pages_per_zspage", "freeable"); |
248ca1b0 | 549 | |
cf8e0fed | 550 | for (i = 0; i < ZS_SIZE_CLASSES; i++) { |
4c7ac972 | 551 | |
248ca1b0 MK |
552 | class = pool->size_class[i]; |
553 | ||
554 | if (class->index != i) | |
555 | continue; | |
556 | ||
c0547d0b | 557 | spin_lock(&pool->lock); |
e1807d5d SS |
558 | |
559 | seq_printf(s, " %5u %5u ", i, class->size); | |
560 | for (fg = ZS_INUSE_RATIO_10; fg < NR_FULLNESS_GROUPS; fg++) { | |
561 | inuse_totals[fg] += zs_stat_get(class, fg); | |
562 | seq_printf(s, "%9lu ", zs_stat_get(class, fg)); | |
563 | } | |
4c7ac972 SS |
564 | |
565 | obj_allocated = zs_stat_get(class, ZS_OBJS_ALLOCATED); | |
566 | obj_used = zs_stat_get(class, ZS_OBJS_INUSE); | |
1120ed54 | 567 | freeable = zs_can_compact(class); |
c0547d0b | 568 | spin_unlock(&pool->lock); |
248ca1b0 | 569 | |
b4fd07a0 | 570 | objs_per_zspage = class->objs_per_zspage; |
248ca1b0 MK |
571 | pages_used = obj_allocated / objs_per_zspage * |
572 | class->pages_per_zspage; | |
573 | ||
e1807d5d SS |
574 | seq_printf(s, "%13lu %10lu %10lu %16d %8lu\n", |
575 | obj_allocated, obj_used, pages_used, | |
576 | class->pages_per_zspage, freeable); | |
248ca1b0 | 577 | |
248ca1b0 MK |
578 | total_objs += obj_allocated; |
579 | total_used_objs += obj_used; | |
580 | total_pages += pages_used; | |
1120ed54 | 581 | total_freeable += freeable; |
248ca1b0 MK |
582 | } |
583 | ||
584 | seq_puts(s, "\n"); | |
e1807d5d SS |
585 | seq_printf(s, " %5s %5s ", "Total", ""); |
586 | ||
587 | for (fg = ZS_INUSE_RATIO_10; fg < NR_FULLNESS_GROUPS; fg++) | |
588 | seq_printf(s, "%9lu ", inuse_totals[fg]); | |
589 | ||
590 | seq_printf(s, "%13lu %10lu %10lu %16s %8lu\n", | |
591 | total_objs, total_used_objs, total_pages, "", | |
592 | total_freeable); | |
248ca1b0 MK |
593 | |
594 | return 0; | |
595 | } | |
5ad35093 | 596 | DEFINE_SHOW_ATTRIBUTE(zs_stats_size); |
248ca1b0 | 597 | |
d34f6157 | 598 | static void zs_pool_stat_create(struct zs_pool *pool, const char *name) |
248ca1b0 | 599 | { |
4abaac9b DS |
600 | if (!zs_stat_root) { |
601 | pr_warn("no root stat dir, not creating <%s> stat dir\n", name); | |
d34f6157 | 602 | return; |
4abaac9b | 603 | } |
248ca1b0 | 604 | |
4268509a GKH |
605 | pool->stat_dentry = debugfs_create_dir(name, zs_stat_root); |
606 | ||
607 | debugfs_create_file("classes", S_IFREG | 0444, pool->stat_dentry, pool, | |
608 | &zs_stats_size_fops); | |
248ca1b0 MK |
609 | } |
610 | ||
611 | static void zs_pool_stat_destroy(struct zs_pool *pool) | |
612 | { | |
613 | debugfs_remove_recursive(pool->stat_dentry); | |
614 | } | |
615 | ||
616 | #else /* CONFIG_ZSMALLOC_STAT */ | |
4abaac9b | 617 | static void __init zs_stat_init(void) |
248ca1b0 | 618 | { |
248ca1b0 MK |
619 | } |
620 | ||
621 | static void __exit zs_stat_exit(void) | |
622 | { | |
623 | } | |
624 | ||
d34f6157 | 625 | static inline void zs_pool_stat_create(struct zs_pool *pool, const char *name) |
248ca1b0 | 626 | { |
248ca1b0 MK |
627 | } |
628 | ||
629 | static inline void zs_pool_stat_destroy(struct zs_pool *pool) | |
630 | { | |
631 | } | |
248ca1b0 MK |
632 | #endif |
633 | ||
48b4800a | 634 | |
c3e3e88a NC |
635 | /* |
636 | * For each size class, zspages are divided into different groups | |
4c7ac972 | 637 | * depending on their usage ratio. This function returns fullness |
c3e3e88a NC |
638 | * status of the given page. |
639 | */ | |
4c7ac972 | 640 | static int get_fullness_group(struct size_class *class, struct zspage *zspage) |
61989a80 | 641 | { |
4c7ac972 | 642 | int inuse, objs_per_zspage, ratio; |
830e4bc5 | 643 | |
3783689a | 644 | inuse = get_zspage_inuse(zspage); |
1fc6e27d | 645 | objs_per_zspage = class->objs_per_zspage; |
61989a80 NG |
646 | |
647 | if (inuse == 0) | |
4c7ac972 SS |
648 | return ZS_INUSE_RATIO_0; |
649 | if (inuse == objs_per_zspage) | |
650 | return ZS_INUSE_RATIO_100; | |
61989a80 | 651 | |
4c7ac972 SS |
652 | ratio = 100 * inuse / objs_per_zspage; |
653 | /* | |
654 | * Take integer division into consideration: a page with one inuse | |
655 | * object out of 127 possible, will end up having 0 usage ratio, | |
656 | * which is wrong as it belongs in ZS_INUSE_RATIO_10 fullness group. | |
657 | */ | |
658 | return ratio / 10 + 1; | |
61989a80 NG |
659 | } |
660 | ||
c3e3e88a NC |
661 | /* |
662 | * Each size class maintains various freelists and zspages are assigned | |
663 | * to one of these freelists based on the number of live objects they | |
664 | * have. This functions inserts the given zspage into the freelist | |
665 | * identified by <class, fullness_group>. | |
666 | */ | |
251cbb95 | 667 | static void insert_zspage(struct size_class *class, |
3783689a | 668 | struct zspage *zspage, |
4c7ac972 | 669 | int fullness) |
61989a80 | 670 | { |
3828a764 | 671 | class_stat_inc(class, fullness, 1); |
a40a71e8 | 672 | list_add(&zspage->list, &class->fullness_list[fullness]); |
a6a8cdfd | 673 | zspage->fullness = fullness; |
61989a80 NG |
674 | } |
675 | ||
c3e3e88a NC |
676 | /* |
677 | * This function removes the given zspage from the freelist identified | |
678 | * by <class, fullness_group>. | |
679 | */ | |
67eaedc1 | 680 | static void remove_zspage(struct size_class *class, struct zspage *zspage) |
61989a80 | 681 | { |
67eaedc1 CZ |
682 | int fullness = zspage->fullness; |
683 | ||
3783689a | 684 | VM_BUG_ON(list_empty(&class->fullness_list[fullness])); |
61989a80 | 685 | |
3783689a | 686 | list_del_init(&zspage->list); |
3828a764 | 687 | class_stat_dec(class, fullness, 1); |
61989a80 NG |
688 | } |
689 | ||
c3e3e88a NC |
690 | /* |
691 | * Each size class maintains zspages in different fullness groups depending | |
692 | * on the number of live objects they contain. When allocating or freeing | |
4c7ac972 SS |
693 | * objects, the fullness status of the page can change, for instance, from |
694 | * INUSE_RATIO_80 to INUSE_RATIO_70 when freeing an object. This function | |
695 | * checks if such a status change has occurred for the given page and | |
696 | * accordingly moves the page from the list of the old fullness group to that | |
697 | * of the new fullness group. | |
c3e3e88a | 698 | */ |
4c7ac972 | 699 | static int fix_fullness_group(struct size_class *class, struct zspage *zspage) |
61989a80 | 700 | { |
ce335e07 | 701 | int newfg; |
61989a80 | 702 | |
3783689a | 703 | newfg = get_fullness_group(class, zspage); |
ce335e07 | 704 | if (newfg == zspage->fullness) |
61989a80 NG |
705 | goto out; |
706 | ||
67eaedc1 | 707 | remove_zspage(class, zspage); |
c4549b87 | 708 | insert_zspage(class, zspage, newfg); |
61989a80 NG |
709 | out: |
710 | return newfg; | |
711 | } | |
712 | ||
3783689a | 713 | static struct zspage *get_zspage(struct page *page) |
61989a80 | 714 | { |
a6c5e0f7 | 715 | struct zspage *zspage = (struct zspage *)page_private(page); |
48b4800a MK |
716 | |
717 | BUG_ON(zspage->magic != ZSPAGE_MAGIC); | |
718 | return zspage; | |
61989a80 NG |
719 | } |
720 | ||
721 | static struct page *get_next_page(struct page *page) | |
722 | { | |
a41ec880 MK |
723 | struct zspage *zspage = get_zspage(page); |
724 | ||
725 | if (unlikely(ZsHugePage(zspage))) | |
48b4800a MK |
726 | return NULL; |
727 | ||
ffedd09f | 728 | return (struct page *)page->index; |
61989a80 NG |
729 | } |
730 | ||
bfd093f5 MK |
731 | /** |
732 | * obj_to_location - get (<page>, <obj_idx>) from encoded object value | |
e8b098fc | 733 | * @obj: the encoded object value |
bfd093f5 MK |
734 | * @page: page object resides in zspage |
735 | * @obj_idx: object index | |
67296874 | 736 | */ |
bfd093f5 MK |
737 | static void obj_to_location(unsigned long obj, struct page **page, |
738 | unsigned int *obj_idx) | |
61989a80 | 739 | { |
bfd093f5 MK |
740 | *page = pfn_to_page(obj >> OBJ_INDEX_BITS); |
741 | *obj_idx = (obj & OBJ_INDEX_MASK); | |
742 | } | |
61989a80 | 743 | |
67f1c9cd MK |
744 | static void obj_to_page(unsigned long obj, struct page **page) |
745 | { | |
67f1c9cd MK |
746 | *page = pfn_to_page(obj >> OBJ_INDEX_BITS); |
747 | } | |
748 | ||
bfd093f5 MK |
749 | /** |
750 | * location_to_obj - get obj value encoded from (<page>, <obj_idx>) | |
751 | * @page: page object resides in zspage | |
752 | * @obj_idx: object index | |
753 | */ | |
754 | static unsigned long location_to_obj(struct page *page, unsigned int obj_idx) | |
755 | { | |
756 | unsigned long obj; | |
61989a80 | 757 | |
312fcae2 | 758 | obj = page_to_pfn(page) << OBJ_INDEX_BITS; |
bfd093f5 | 759 | obj |= obj_idx & OBJ_INDEX_MASK; |
61989a80 | 760 | |
bfd093f5 | 761 | return obj; |
61989a80 NG |
762 | } |
763 | ||
2e40e163 MK |
764 | static unsigned long handle_to_obj(unsigned long handle) |
765 | { | |
766 | return *(unsigned long *)handle; | |
767 | } | |
768 | ||
f9044f17 SS |
769 | static inline bool obj_allocated(struct page *page, void *obj, |
770 | unsigned long *phandle) | |
312fcae2 | 771 | { |
3ae92ac2 | 772 | unsigned long handle; |
a41ec880 | 773 | struct zspage *zspage = get_zspage(page); |
3ae92ac2 | 774 | |
a41ec880 | 775 | if (unlikely(ZsHugePage(zspage))) { |
830e4bc5 | 776 | VM_BUG_ON_PAGE(!is_first_page(page), page); |
3ae92ac2 | 777 | handle = page->index; |
7b60a685 | 778 | } else |
3ae92ac2 MK |
779 | handle = *(unsigned long *)obj; |
780 | ||
f9044f17 | 781 | if (!(handle & OBJ_ALLOCATED_TAG)) |
3ae92ac2 MK |
782 | return false; |
783 | ||
85b32581 NP |
784 | /* Clear all tags before returning the handle */ |
785 | *phandle = handle & ~OBJ_TAG_MASK; | |
3ae92ac2 | 786 | return true; |
312fcae2 MK |
787 | } |
788 | ||
f4477e90 NG |
789 | static void reset_page(struct page *page) |
790 | { | |
48b4800a | 791 | __ClearPageMovable(page); |
18fd06bf | 792 | ClearPagePrivate(page); |
f4477e90 | 793 | set_page_private(page, 0); |
48b4800a | 794 | page_mapcount_reset(page); |
ffedd09f | 795 | page->index = 0; |
48b4800a MK |
796 | } |
797 | ||
4d0a5402 | 798 | static int trylock_zspage(struct zspage *zspage) |
48b4800a MK |
799 | { |
800 | struct page *cursor, *fail; | |
801 | ||
802 | for (cursor = get_first_page(zspage); cursor != NULL; cursor = | |
803 | get_next_page(cursor)) { | |
804 | if (!trylock_page(cursor)) { | |
805 | fail = cursor; | |
806 | goto unlock; | |
807 | } | |
808 | } | |
809 | ||
810 | return 1; | |
811 | unlock: | |
812 | for (cursor = get_first_page(zspage); cursor != fail; cursor = | |
813 | get_next_page(cursor)) | |
814 | unlock_page(cursor); | |
815 | ||
816 | return 0; | |
f4477e90 NG |
817 | } |
818 | ||
48b4800a MK |
819 | static void __free_zspage(struct zs_pool *pool, struct size_class *class, |
820 | struct zspage *zspage) | |
61989a80 | 821 | { |
3783689a | 822 | struct page *page, *next; |
48b4800a | 823 | |
c0547d0b | 824 | assert_spin_locked(&pool->lock); |
61989a80 | 825 | |
3783689a | 826 | VM_BUG_ON(get_zspage_inuse(zspage)); |
ce335e07 | 827 | VM_BUG_ON(zspage->fullness != ZS_INUSE_RATIO_0); |
61989a80 | 828 | |
48b4800a | 829 | next = page = get_first_page(zspage); |
3783689a | 830 | do { |
48b4800a MK |
831 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
832 | next = get_next_page(page); | |
3783689a | 833 | reset_page(page); |
48b4800a | 834 | unlock_page(page); |
91537fee | 835 | dec_zone_page_state(page, NR_ZSPAGES); |
3783689a MK |
836 | put_page(page); |
837 | page = next; | |
838 | } while (page != NULL); | |
61989a80 | 839 | |
3783689a | 840 | cache_free_zspage(pool, zspage); |
48b4800a | 841 | |
4c7ac972 SS |
842 | class_stat_dec(class, ZS_OBJS_ALLOCATED, class->objs_per_zspage); |
843 | atomic_long_sub(class->pages_per_zspage, &pool->pages_allocated); | |
48b4800a MK |
844 | } |
845 | ||
846 | static void free_zspage(struct zs_pool *pool, struct size_class *class, | |
847 | struct zspage *zspage) | |
848 | { | |
849 | VM_BUG_ON(get_zspage_inuse(zspage)); | |
850 | VM_BUG_ON(list_empty(&zspage->list)); | |
851 | ||
b475d42d MK |
852 | /* |
853 | * Since zs_free couldn't be sleepable, this function cannot call | |
854 | * lock_page. The page locks trylock_zspage got will be released | |
855 | * by __free_zspage. | |
856 | */ | |
48b4800a MK |
857 | if (!trylock_zspage(zspage)) { |
858 | kick_deferred_free(pool); | |
859 | return; | |
860 | } | |
861 | ||
67eaedc1 | 862 | remove_zspage(class, zspage); |
48b4800a | 863 | __free_zspage(pool, class, zspage); |
61989a80 NG |
864 | } |
865 | ||
866 | /* Initialize a newly allocated zspage */ | |
3783689a | 867 | static void init_zspage(struct size_class *class, struct zspage *zspage) |
61989a80 | 868 | { |
bfd093f5 | 869 | unsigned int freeobj = 1; |
61989a80 | 870 | unsigned long off = 0; |
48b4800a | 871 | struct page *page = get_first_page(zspage); |
830e4bc5 | 872 | |
61989a80 NG |
873 | while (page) { |
874 | struct page *next_page; | |
875 | struct link_free *link; | |
af4ee5e9 | 876 | void *vaddr; |
61989a80 | 877 | |
3783689a | 878 | set_first_obj_offset(page, off); |
61989a80 | 879 | |
af4ee5e9 MK |
880 | vaddr = kmap_atomic(page); |
881 | link = (struct link_free *)vaddr + off / sizeof(*link); | |
5538c562 DS |
882 | |
883 | while ((off += class->size) < PAGE_SIZE) { | |
3b1d9ca6 | 884 | link->next = freeobj++ << OBJ_TAG_BITS; |
5538c562 | 885 | link += class->size / sizeof(*link); |
61989a80 NG |
886 | } |
887 | ||
888 | /* | |
889 | * We now come to the last (full or partial) object on this | |
890 | * page, which must point to the first object on the next | |
891 | * page (if present) | |
892 | */ | |
893 | next_page = get_next_page(page); | |
bfd093f5 | 894 | if (next_page) { |
3b1d9ca6 | 895 | link->next = freeobj++ << OBJ_TAG_BITS; |
bfd093f5 MK |
896 | } else { |
897 | /* | |
3b1d9ca6 | 898 | * Reset OBJ_TAG_BITS bit to last link to tell |
bfd093f5 MK |
899 | * whether it's allocated object or not. |
900 | */ | |
01a6ad9a | 901 | link->next = -1UL << OBJ_TAG_BITS; |
bfd093f5 | 902 | } |
af4ee5e9 | 903 | kunmap_atomic(vaddr); |
61989a80 | 904 | page = next_page; |
5538c562 | 905 | off %= PAGE_SIZE; |
61989a80 | 906 | } |
bdb0af7c | 907 | |
bfd093f5 | 908 | set_freeobj(zspage, 0); |
61989a80 NG |
909 | } |
910 | ||
48b4800a MK |
911 | static void create_page_chain(struct size_class *class, struct zspage *zspage, |
912 | struct page *pages[]) | |
61989a80 | 913 | { |
bdb0af7c MK |
914 | int i; |
915 | struct page *page; | |
916 | struct page *prev_page = NULL; | |
48b4800a | 917 | int nr_pages = class->pages_per_zspage; |
61989a80 NG |
918 | |
919 | /* | |
920 | * Allocate individual pages and link them together as: | |
ffedd09f | 921 | * 1. all pages are linked together using page->index |
3783689a | 922 | * 2. each sub-page point to zspage using page->private |
61989a80 | 923 | * |
3783689a | 924 | * we set PG_private to identify the first page (i.e. no other sub-page |
22c5cef1 | 925 | * has this flag set). |
61989a80 | 926 | */ |
bdb0af7c MK |
927 | for (i = 0; i < nr_pages; i++) { |
928 | page = pages[i]; | |
3783689a | 929 | set_page_private(page, (unsigned long)zspage); |
ffedd09f | 930 | page->index = 0; |
bdb0af7c | 931 | if (i == 0) { |
3783689a | 932 | zspage->first_page = page; |
a27545bf | 933 | SetPagePrivate(page); |
48b4800a MK |
934 | if (unlikely(class->objs_per_zspage == 1 && |
935 | class->pages_per_zspage == 1)) | |
a41ec880 | 936 | SetZsHugePage(zspage); |
3783689a | 937 | } else { |
ffedd09f | 938 | prev_page->index = (unsigned long)page; |
61989a80 | 939 | } |
61989a80 NG |
940 | prev_page = page; |
941 | } | |
bdb0af7c | 942 | } |
61989a80 | 943 | |
bdb0af7c MK |
944 | /* |
945 | * Allocate a zspage for the given size class | |
946 | */ | |
3783689a MK |
947 | static struct zspage *alloc_zspage(struct zs_pool *pool, |
948 | struct size_class *class, | |
949 | gfp_t gfp) | |
bdb0af7c MK |
950 | { |
951 | int i; | |
bdb0af7c | 952 | struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE]; |
3783689a MK |
953 | struct zspage *zspage = cache_alloc_zspage(pool, gfp); |
954 | ||
955 | if (!zspage) | |
956 | return NULL; | |
957 | ||
48b4800a MK |
958 | zspage->magic = ZSPAGE_MAGIC; |
959 | migrate_lock_init(zspage); | |
61989a80 | 960 | |
bdb0af7c MK |
961 | for (i = 0; i < class->pages_per_zspage; i++) { |
962 | struct page *page; | |
61989a80 | 963 | |
3783689a | 964 | page = alloc_page(gfp); |
bdb0af7c | 965 | if (!page) { |
91537fee MK |
966 | while (--i >= 0) { |
967 | dec_zone_page_state(pages[i], NR_ZSPAGES); | |
bdb0af7c | 968 | __free_page(pages[i]); |
91537fee | 969 | } |
3783689a | 970 | cache_free_zspage(pool, zspage); |
bdb0af7c MK |
971 | return NULL; |
972 | } | |
91537fee MK |
973 | |
974 | inc_zone_page_state(page, NR_ZSPAGES); | |
bdb0af7c | 975 | pages[i] = page; |
61989a80 NG |
976 | } |
977 | ||
48b4800a | 978 | create_page_chain(class, zspage, pages); |
3783689a | 979 | init_zspage(class, zspage); |
68f2736a | 980 | zspage->pool = pool; |
a6a8cdfd | 981 | zspage->class = class->index; |
bdb0af7c | 982 | |
3783689a | 983 | return zspage; |
61989a80 NG |
984 | } |
985 | ||
3783689a | 986 | static struct zspage *find_get_zspage(struct size_class *class) |
61989a80 NG |
987 | { |
988 | int i; | |
3783689a | 989 | struct zspage *zspage; |
61989a80 | 990 | |
4c7ac972 | 991 | for (i = ZS_INUSE_RATIO_99; i >= ZS_INUSE_RATIO_0; i--) { |
3783689a | 992 | zspage = list_first_entry_or_null(&class->fullness_list[i], |
4c7ac972 | 993 | struct zspage, list); |
3783689a | 994 | if (zspage) |
61989a80 NG |
995 | break; |
996 | } | |
997 | ||
3783689a | 998 | return zspage; |
61989a80 NG |
999 | } |
1000 | ||
f553646a SJ |
1001 | static inline int __zs_cpu_up(struct mapping_area *area) |
1002 | { | |
1003 | /* | |
1004 | * Make sure we don't leak memory if a cpu UP notification | |
1005 | * and zs_init() race and both call zs_cpu_up() on the same cpu | |
1006 | */ | |
1007 | if (area->vm_buf) | |
1008 | return 0; | |
40f9fb8c | 1009 | area->vm_buf = kmalloc(ZS_MAX_ALLOC_SIZE, GFP_KERNEL); |
f553646a SJ |
1010 | if (!area->vm_buf) |
1011 | return -ENOMEM; | |
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | static inline void __zs_cpu_down(struct mapping_area *area) | |
1016 | { | |
40f9fb8c | 1017 | kfree(area->vm_buf); |
f553646a SJ |
1018 | area->vm_buf = NULL; |
1019 | } | |
1020 | ||
1021 | static void *__zs_map_object(struct mapping_area *area, | |
1022 | struct page *pages[2], int off, int size) | |
5f601902 | 1023 | { |
5f601902 SJ |
1024 | int sizes[2]; |
1025 | void *addr; | |
f553646a | 1026 | char *buf = area->vm_buf; |
5f601902 | 1027 | |
f553646a SJ |
1028 | /* disable page faults to match kmap_atomic() return conditions */ |
1029 | pagefault_disable(); | |
1030 | ||
1031 | /* no read fastpath */ | |
1032 | if (area->vm_mm == ZS_MM_WO) | |
1033 | goto out; | |
5f601902 SJ |
1034 | |
1035 | sizes[0] = PAGE_SIZE - off; | |
1036 | sizes[1] = size - sizes[0]; | |
1037 | ||
5f601902 SJ |
1038 | /* copy object to per-cpu buffer */ |
1039 | addr = kmap_atomic(pages[0]); | |
1040 | memcpy(buf, addr + off, sizes[0]); | |
1041 | kunmap_atomic(addr); | |
1042 | addr = kmap_atomic(pages[1]); | |
1043 | memcpy(buf + sizes[0], addr, sizes[1]); | |
1044 | kunmap_atomic(addr); | |
f553646a SJ |
1045 | out: |
1046 | return area->vm_buf; | |
5f601902 SJ |
1047 | } |
1048 | ||
f553646a SJ |
1049 | static void __zs_unmap_object(struct mapping_area *area, |
1050 | struct page *pages[2], int off, int size) | |
5f601902 | 1051 | { |
5f601902 SJ |
1052 | int sizes[2]; |
1053 | void *addr; | |
2e40e163 | 1054 | char *buf; |
5f601902 | 1055 | |
f553646a SJ |
1056 | /* no write fastpath */ |
1057 | if (area->vm_mm == ZS_MM_RO) | |
1058 | goto out; | |
5f601902 | 1059 | |
7b60a685 | 1060 | buf = area->vm_buf; |
a82cbf07 YX |
1061 | buf = buf + ZS_HANDLE_SIZE; |
1062 | size -= ZS_HANDLE_SIZE; | |
1063 | off += ZS_HANDLE_SIZE; | |
2e40e163 | 1064 | |
5f601902 SJ |
1065 | sizes[0] = PAGE_SIZE - off; |
1066 | sizes[1] = size - sizes[0]; | |
1067 | ||
1068 | /* copy per-cpu buffer to object */ | |
1069 | addr = kmap_atomic(pages[0]); | |
1070 | memcpy(addr + off, buf, sizes[0]); | |
1071 | kunmap_atomic(addr); | |
1072 | addr = kmap_atomic(pages[1]); | |
1073 | memcpy(addr, buf + sizes[0], sizes[1]); | |
1074 | kunmap_atomic(addr); | |
f553646a SJ |
1075 | |
1076 | out: | |
1077 | /* enable page faults to match kunmap_atomic() return conditions */ | |
1078 | pagefault_enable(); | |
5f601902 | 1079 | } |
61989a80 | 1080 | |
215c89d0 | 1081 | static int zs_cpu_prepare(unsigned int cpu) |
61989a80 | 1082 | { |
61989a80 NG |
1083 | struct mapping_area *area; |
1084 | ||
215c89d0 SAS |
1085 | area = &per_cpu(zs_map_area, cpu); |
1086 | return __zs_cpu_up(area); | |
61989a80 NG |
1087 | } |
1088 | ||
215c89d0 | 1089 | static int zs_cpu_dead(unsigned int cpu) |
61989a80 | 1090 | { |
215c89d0 | 1091 | struct mapping_area *area; |
40f9fb8c | 1092 | |
215c89d0 SAS |
1093 | area = &per_cpu(zs_map_area, cpu); |
1094 | __zs_cpu_down(area); | |
1095 | return 0; | |
b1b00a5b SS |
1096 | } |
1097 | ||
64d90465 GM |
1098 | static bool can_merge(struct size_class *prev, int pages_per_zspage, |
1099 | int objs_per_zspage) | |
9eec4cd5 | 1100 | { |
64d90465 GM |
1101 | if (prev->pages_per_zspage == pages_per_zspage && |
1102 | prev->objs_per_zspage == objs_per_zspage) | |
1103 | return true; | |
9eec4cd5 | 1104 | |
64d90465 | 1105 | return false; |
9eec4cd5 JK |
1106 | } |
1107 | ||
3783689a | 1108 | static bool zspage_full(struct size_class *class, struct zspage *zspage) |
312fcae2 | 1109 | { |
3783689a | 1110 | return get_zspage_inuse(zspage) == class->objs_per_zspage; |
312fcae2 | 1111 | } |
7c2af309 | 1112 | |
df9cd3cb SS |
1113 | static bool zspage_empty(struct zspage *zspage) |
1114 | { | |
1115 | return get_zspage_inuse(zspage) == 0; | |
1116 | } | |
1117 | ||
7c2af309 AR |
1118 | /** |
1119 | * zs_lookup_class_index() - Returns index of the zsmalloc &size_class | |
1120 | * that hold objects of the provided size. | |
1121 | * @pool: zsmalloc pool to use | |
1122 | * @size: object size | |
1123 | * | |
1124 | * Context: Any context. | |
1125 | * | |
1126 | * Return: the index of the zsmalloc &size_class that hold objects of the | |
1127 | * provided size. | |
1128 | */ | |
1129 | unsigned int zs_lookup_class_index(struct zs_pool *pool, unsigned int size) | |
1130 | { | |
1131 | struct size_class *class; | |
1132 | ||
1133 | class = pool->size_class[get_size_class_index(size)]; | |
1134 | ||
1135 | return class->index; | |
1136 | } | |
1137 | EXPORT_SYMBOL_GPL(zs_lookup_class_index); | |
312fcae2 | 1138 | |
66cdef66 GM |
1139 | unsigned long zs_get_total_pages(struct zs_pool *pool) |
1140 | { | |
1141 | return atomic_long_read(&pool->pages_allocated); | |
1142 | } | |
1143 | EXPORT_SYMBOL_GPL(zs_get_total_pages); | |
1144 | ||
4bbc0bc0 | 1145 | /** |
66cdef66 GM |
1146 | * zs_map_object - get address of allocated object from handle. |
1147 | * @pool: pool from which the object was allocated | |
1148 | * @handle: handle returned from zs_malloc | |
f0953a1b | 1149 | * @mm: mapping mode to use |
4bbc0bc0 | 1150 | * |
66cdef66 GM |
1151 | * Before using an object allocated from zs_malloc, it must be mapped using |
1152 | * this function. When done with the object, it must be unmapped using | |
1153 | * zs_unmap_object. | |
4bbc0bc0 | 1154 | * |
66cdef66 GM |
1155 | * Only one object can be mapped per cpu at a time. There is no protection |
1156 | * against nested mappings. | |
1157 | * | |
1158 | * This function returns with preemption and page faults disabled. | |
4bbc0bc0 | 1159 | */ |
66cdef66 GM |
1160 | void *zs_map_object(struct zs_pool *pool, unsigned long handle, |
1161 | enum zs_mapmode mm) | |
61989a80 | 1162 | { |
3783689a | 1163 | struct zspage *zspage; |
66cdef66 | 1164 | struct page *page; |
bfd093f5 MK |
1165 | unsigned long obj, off; |
1166 | unsigned int obj_idx; | |
61989a80 | 1167 | |
66cdef66 GM |
1168 | struct size_class *class; |
1169 | struct mapping_area *area; | |
1170 | struct page *pages[2]; | |
2e40e163 | 1171 | void *ret; |
61989a80 | 1172 | |
9eec4cd5 | 1173 | /* |
66cdef66 GM |
1174 | * Because we use per-cpu mapping areas shared among the |
1175 | * pools/users, we can't allow mapping in interrupt context | |
1176 | * because it can corrupt another users mappings. | |
9eec4cd5 | 1177 | */ |
1aedcafb | 1178 | BUG_ON(in_interrupt()); |
61989a80 | 1179 | |
b475d42d | 1180 | /* It guarantees it can get zspage from handle safely */ |
c0547d0b | 1181 | spin_lock(&pool->lock); |
2e40e163 MK |
1182 | obj = handle_to_obj(handle); |
1183 | obj_to_location(obj, &page, &obj_idx); | |
3783689a | 1184 | zspage = get_zspage(page); |
48b4800a | 1185 | |
b475d42d | 1186 | /* |
c0547d0b | 1187 | * migration cannot move any zpages in this zspage. Here, pool->lock |
b475d42d MK |
1188 | * is too heavy since callers would take some time until they calls |
1189 | * zs_unmap_object API so delegate the locking from class to zspage | |
1190 | * which is smaller granularity. | |
1191 | */ | |
48b4800a | 1192 | migrate_read_lock(zspage); |
c0547d0b | 1193 | spin_unlock(&pool->lock); |
48b4800a | 1194 | |
67f1c9cd | 1195 | class = zspage_class(pool, zspage); |
f24f66ee | 1196 | off = offset_in_page(class->size * obj_idx); |
df8b5bb9 | 1197 | |
a3726599 MG |
1198 | local_lock(&zs_map_area.lock); |
1199 | area = this_cpu_ptr(&zs_map_area); | |
66cdef66 GM |
1200 | area->vm_mm = mm; |
1201 | if (off + class->size <= PAGE_SIZE) { | |
1202 | /* this object is contained entirely within a page */ | |
1203 | area->vm_addr = kmap_atomic(page); | |
2e40e163 MK |
1204 | ret = area->vm_addr + off; |
1205 | goto out; | |
61989a80 NG |
1206 | } |
1207 | ||
66cdef66 GM |
1208 | /* this object spans two pages */ |
1209 | pages[0] = page; | |
1210 | pages[1] = get_next_page(page); | |
1211 | BUG_ON(!pages[1]); | |
9eec4cd5 | 1212 | |
2e40e163 MK |
1213 | ret = __zs_map_object(area, pages, off, class->size); |
1214 | out: | |
a41ec880 | 1215 | if (likely(!ZsHugePage(zspage))) |
7b60a685 MK |
1216 | ret += ZS_HANDLE_SIZE; |
1217 | ||
1218 | return ret; | |
61989a80 | 1219 | } |
66cdef66 | 1220 | EXPORT_SYMBOL_GPL(zs_map_object); |
61989a80 | 1221 | |
66cdef66 | 1222 | void zs_unmap_object(struct zs_pool *pool, unsigned long handle) |
61989a80 | 1223 | { |
3783689a | 1224 | struct zspage *zspage; |
66cdef66 | 1225 | struct page *page; |
bfd093f5 MK |
1226 | unsigned long obj, off; |
1227 | unsigned int obj_idx; | |
61989a80 | 1228 | |
66cdef66 GM |
1229 | struct size_class *class; |
1230 | struct mapping_area *area; | |
9eec4cd5 | 1231 | |
2e40e163 MK |
1232 | obj = handle_to_obj(handle); |
1233 | obj_to_location(obj, &page, &obj_idx); | |
3783689a | 1234 | zspage = get_zspage(page); |
67f1c9cd | 1235 | class = zspage_class(pool, zspage); |
f24f66ee | 1236 | off = offset_in_page(class->size * obj_idx); |
61989a80 | 1237 | |
66cdef66 GM |
1238 | area = this_cpu_ptr(&zs_map_area); |
1239 | if (off + class->size <= PAGE_SIZE) | |
1240 | kunmap_atomic(area->vm_addr); | |
1241 | else { | |
1242 | struct page *pages[2]; | |
40f9fb8c | 1243 | |
66cdef66 GM |
1244 | pages[0] = page; |
1245 | pages[1] = get_next_page(page); | |
1246 | BUG_ON(!pages[1]); | |
1247 | ||
1248 | __zs_unmap_object(area, pages, off, class->size); | |
1249 | } | |
a3726599 | 1250 | local_unlock(&zs_map_area.lock); |
48b4800a MK |
1251 | |
1252 | migrate_read_unlock(zspage); | |
61989a80 | 1253 | } |
66cdef66 | 1254 | EXPORT_SYMBOL_GPL(zs_unmap_object); |
61989a80 | 1255 | |
010b495e SS |
1256 | /** |
1257 | * zs_huge_class_size() - Returns the size (in bytes) of the first huge | |
1258 | * zsmalloc &size_class. | |
1259 | * @pool: zsmalloc pool to use | |
1260 | * | |
1261 | * The function returns the size of the first huge class - any object of equal | |
1262 | * or bigger size will be stored in zspage consisting of a single physical | |
1263 | * page. | |
1264 | * | |
1265 | * Context: Any context. | |
1266 | * | |
1267 | * Return: the size (in bytes) of the first huge zsmalloc &size_class. | |
1268 | */ | |
1269 | size_t zs_huge_class_size(struct zs_pool *pool) | |
1270 | { | |
1271 | return huge_class_size; | |
1272 | } | |
1273 | EXPORT_SYMBOL_GPL(zs_huge_class_size); | |
1274 | ||
0a5f079b | 1275 | static unsigned long obj_malloc(struct zs_pool *pool, |
3783689a | 1276 | struct zspage *zspage, unsigned long handle) |
c7806261 | 1277 | { |
bfd093f5 | 1278 | int i, nr_page, offset; |
c7806261 MK |
1279 | unsigned long obj; |
1280 | struct link_free *link; | |
0a5f079b | 1281 | struct size_class *class; |
c7806261 MK |
1282 | |
1283 | struct page *m_page; | |
bfd093f5 | 1284 | unsigned long m_offset; |
c7806261 MK |
1285 | void *vaddr; |
1286 | ||
0a5f079b | 1287 | class = pool->size_class[zspage->class]; |
312fcae2 | 1288 | handle |= OBJ_ALLOCATED_TAG; |
3783689a | 1289 | obj = get_freeobj(zspage); |
bfd093f5 MK |
1290 | |
1291 | offset = obj * class->size; | |
1292 | nr_page = offset >> PAGE_SHIFT; | |
f24f66ee | 1293 | m_offset = offset_in_page(offset); |
bfd093f5 MK |
1294 | m_page = get_first_page(zspage); |
1295 | ||
1296 | for (i = 0; i < nr_page; i++) | |
1297 | m_page = get_next_page(m_page); | |
c7806261 MK |
1298 | |
1299 | vaddr = kmap_atomic(m_page); | |
1300 | link = (struct link_free *)vaddr + m_offset / sizeof(*link); | |
3b1d9ca6 | 1301 | set_freeobj(zspage, link->next >> OBJ_TAG_BITS); |
a41ec880 | 1302 | if (likely(!ZsHugePage(zspage))) |
7b60a685 MK |
1303 | /* record handle in the header of allocated chunk */ |
1304 | link->handle = handle; | |
1305 | else | |
3783689a MK |
1306 | /* record handle to page->index */ |
1307 | zspage->first_page->index = handle; | |
1308 | ||
c7806261 | 1309 | kunmap_atomic(vaddr); |
3783689a | 1310 | mod_zspage_inuse(zspage, 1); |
c7806261 | 1311 | |
bfd093f5 MK |
1312 | obj = location_to_obj(m_page, obj); |
1313 | ||
c7806261 MK |
1314 | return obj; |
1315 | } | |
1316 | ||
1317 | ||
61989a80 NG |
1318 | /** |
1319 | * zs_malloc - Allocate block of given size from pool. | |
1320 | * @pool: pool to allocate from | |
1321 | * @size: size of block to allocate | |
fd854463 | 1322 | * @gfp: gfp flags when allocating object |
61989a80 | 1323 | * |
00a61d86 | 1324 | * On success, handle to the allocated object is returned, |
c7e6f17b | 1325 | * otherwise an ERR_PTR(). |
61989a80 NG |
1326 | * Allocation requests with size > ZS_MAX_ALLOC_SIZE will fail. |
1327 | */ | |
d0d8da2d | 1328 | unsigned long zs_malloc(struct zs_pool *pool, size_t size, gfp_t gfp) |
61989a80 | 1329 | { |
2e40e163 | 1330 | unsigned long handle, obj; |
61989a80 | 1331 | struct size_class *class; |
4c7ac972 | 1332 | int newfg; |
3783689a | 1333 | struct zspage *zspage; |
61989a80 | 1334 | |
fc8580ed | 1335 | if (unlikely(!size)) |
c7e6f17b | 1336 | return (unsigned long)ERR_PTR(-EINVAL); |
2e40e163 | 1337 | |
fc8580ed BS |
1338 | if (unlikely(size > ZS_MAX_ALLOC_SIZE)) |
1339 | return (unsigned long)ERR_PTR(-ENOSPC); | |
1340 | ||
3783689a | 1341 | handle = cache_alloc_handle(pool, gfp); |
2e40e163 | 1342 | if (!handle) |
c7e6f17b | 1343 | return (unsigned long)ERR_PTR(-ENOMEM); |
61989a80 | 1344 | |
2e40e163 MK |
1345 | /* extra space in chunk to keep the handle */ |
1346 | size += ZS_HANDLE_SIZE; | |
9eec4cd5 | 1347 | class = pool->size_class[get_size_class_index(size)]; |
61989a80 | 1348 | |
c0547d0b NP |
1349 | /* pool->lock effectively protects the zpage migration */ |
1350 | spin_lock(&pool->lock); | |
3783689a | 1351 | zspage = find_get_zspage(class); |
48b4800a | 1352 | if (likely(zspage)) { |
0a5f079b | 1353 | obj = obj_malloc(pool, zspage, handle); |
48b4800a MK |
1354 | /* Now move the zspage to another fullness group, if required */ |
1355 | fix_fullness_group(class, zspage); | |
1356 | record_obj(handle, obj); | |
4c7ac972 | 1357 | class_stat_inc(class, ZS_OBJS_INUSE, 1); |
61989a80 | 1358 | |
d461aac9 | 1359 | goto out; |
48b4800a | 1360 | } |
0f050d99 | 1361 | |
c0547d0b | 1362 | spin_unlock(&pool->lock); |
48b4800a MK |
1363 | |
1364 | zspage = alloc_zspage(pool, class, gfp); | |
1365 | if (!zspage) { | |
1366 | cache_free_handle(pool, handle); | |
c7e6f17b | 1367 | return (unsigned long)ERR_PTR(-ENOMEM); |
61989a80 NG |
1368 | } |
1369 | ||
c0547d0b | 1370 | spin_lock(&pool->lock); |
0a5f079b | 1371 | obj = obj_malloc(pool, zspage, handle); |
48b4800a MK |
1372 | newfg = get_fullness_group(class, zspage); |
1373 | insert_zspage(class, zspage, newfg); | |
2e40e163 | 1374 | record_obj(handle, obj); |
4c7ac972 SS |
1375 | atomic_long_add(class->pages_per_zspage, &pool->pages_allocated); |
1376 | class_stat_inc(class, ZS_OBJS_ALLOCATED, class->objs_per_zspage); | |
1377 | class_stat_inc(class, ZS_OBJS_INUSE, 1); | |
48b4800a MK |
1378 | |
1379 | /* We completely set up zspage so mark them as movable */ | |
1380 | SetZsPageMovable(pool, zspage); | |
d461aac9 | 1381 | out: |
c0547d0b | 1382 | spin_unlock(&pool->lock); |
61989a80 | 1383 | |
2e40e163 | 1384 | return handle; |
61989a80 NG |
1385 | } |
1386 | EXPORT_SYMBOL_GPL(zs_malloc); | |
1387 | ||
b3067742 | 1388 | static void obj_free(int class_size, unsigned long obj) |
61989a80 NG |
1389 | { |
1390 | struct link_free *link; | |
3783689a MK |
1391 | struct zspage *zspage; |
1392 | struct page *f_page; | |
bfd093f5 MK |
1393 | unsigned long f_offset; |
1394 | unsigned int f_objidx; | |
af4ee5e9 | 1395 | void *vaddr; |
61989a80 | 1396 | |
2e40e163 | 1397 | obj_to_location(obj, &f_page, &f_objidx); |
f24f66ee | 1398 | f_offset = offset_in_page(class_size * f_objidx); |
3783689a | 1399 | zspage = get_zspage(f_page); |
61989a80 | 1400 | |
c7806261 | 1401 | vaddr = kmap_atomic(f_page); |
af4ee5e9 | 1402 | link = (struct link_free *)(vaddr + f_offset); |
85b32581 | 1403 | |
b3067742 DC |
1404 | /* Insert this object in containing zspage's freelist */ |
1405 | if (likely(!ZsHugePage(zspage))) | |
1406 | link->next = get_freeobj(zspage) << OBJ_TAG_BITS; | |
1407 | else | |
1408 | f_page->index = 0; | |
1409 | set_freeobj(zspage, f_objidx); | |
85b32581 | 1410 | |
af4ee5e9 | 1411 | kunmap_atomic(vaddr); |
3783689a | 1412 | mod_zspage_inuse(zspage, -1); |
c7806261 MK |
1413 | } |
1414 | ||
1415 | void zs_free(struct zs_pool *pool, unsigned long handle) | |
1416 | { | |
3783689a MK |
1417 | struct zspage *zspage; |
1418 | struct page *f_page; | |
bfd093f5 | 1419 | unsigned long obj; |
c7806261 | 1420 | struct size_class *class; |
4c7ac972 | 1421 | int fullness; |
c7806261 | 1422 | |
a5d21721 | 1423 | if (IS_ERR_OR_NULL((void *)handle)) |
c7806261 MK |
1424 | return; |
1425 | ||
b475d42d | 1426 | /* |
c0547d0b | 1427 | * The pool->lock protects the race with zpage's migration |
b475d42d MK |
1428 | * so it's safe to get the page from handle. |
1429 | */ | |
c0547d0b | 1430 | spin_lock(&pool->lock); |
c7806261 | 1431 | obj = handle_to_obj(handle); |
67f1c9cd | 1432 | obj_to_page(obj, &f_page); |
3783689a | 1433 | zspage = get_zspage(f_page); |
67f1c9cd | 1434 | class = zspage_class(pool, zspage); |
b475d42d | 1435 | |
4c7ac972 | 1436 | class_stat_dec(class, ZS_OBJS_INUSE, 1); |
b3067742 | 1437 | obj_free(class->size, obj); |
85b32581 | 1438 | |
3783689a | 1439 | fullness = fix_fullness_group(class, zspage); |
4c7ac972 | 1440 | if (fullness == ZS_INUSE_RATIO_0) |
9997bc01 | 1441 | free_zspage(pool, class, zspage); |
48b4800a | 1442 | |
c0547d0b | 1443 | spin_unlock(&pool->lock); |
3783689a | 1444 | cache_free_handle(pool, handle); |
312fcae2 MK |
1445 | } |
1446 | EXPORT_SYMBOL_GPL(zs_free); | |
1447 | ||
251cbb95 MK |
1448 | static void zs_object_copy(struct size_class *class, unsigned long dst, |
1449 | unsigned long src) | |
312fcae2 MK |
1450 | { |
1451 | struct page *s_page, *d_page; | |
bfd093f5 | 1452 | unsigned int s_objidx, d_objidx; |
312fcae2 MK |
1453 | unsigned long s_off, d_off; |
1454 | void *s_addr, *d_addr; | |
1455 | int s_size, d_size, size; | |
1456 | int written = 0; | |
1457 | ||
1458 | s_size = d_size = class->size; | |
1459 | ||
1460 | obj_to_location(src, &s_page, &s_objidx); | |
1461 | obj_to_location(dst, &d_page, &d_objidx); | |
1462 | ||
f24f66ee AR |
1463 | s_off = offset_in_page(class->size * s_objidx); |
1464 | d_off = offset_in_page(class->size * d_objidx); | |
312fcae2 MK |
1465 | |
1466 | if (s_off + class->size > PAGE_SIZE) | |
1467 | s_size = PAGE_SIZE - s_off; | |
1468 | ||
1469 | if (d_off + class->size > PAGE_SIZE) | |
1470 | d_size = PAGE_SIZE - d_off; | |
1471 | ||
1472 | s_addr = kmap_atomic(s_page); | |
1473 | d_addr = kmap_atomic(d_page); | |
1474 | ||
1475 | while (1) { | |
1476 | size = min(s_size, d_size); | |
1477 | memcpy(d_addr + d_off, s_addr + s_off, size); | |
1478 | written += size; | |
1479 | ||
1480 | if (written == class->size) | |
1481 | break; | |
1482 | ||
495819ea SS |
1483 | s_off += size; |
1484 | s_size -= size; | |
1485 | d_off += size; | |
1486 | d_size -= size; | |
1487 | ||
050a388b AR |
1488 | /* |
1489 | * Calling kunmap_atomic(d_addr) is necessary. kunmap_atomic() | |
1490 | * calls must occurs in reverse order of calls to kmap_atomic(). | |
1491 | * So, to call kunmap_atomic(s_addr) we should first call | |
46e87152 AR |
1492 | * kunmap_atomic(d_addr). For more details see |
1493 | * Documentation/mm/highmem.rst. | |
050a388b | 1494 | */ |
495819ea | 1495 | if (s_off >= PAGE_SIZE) { |
312fcae2 MK |
1496 | kunmap_atomic(d_addr); |
1497 | kunmap_atomic(s_addr); | |
1498 | s_page = get_next_page(s_page); | |
312fcae2 MK |
1499 | s_addr = kmap_atomic(s_page); |
1500 | d_addr = kmap_atomic(d_page); | |
1501 | s_size = class->size - written; | |
1502 | s_off = 0; | |
312fcae2 MK |
1503 | } |
1504 | ||
495819ea | 1505 | if (d_off >= PAGE_SIZE) { |
312fcae2 MK |
1506 | kunmap_atomic(d_addr); |
1507 | d_page = get_next_page(d_page); | |
312fcae2 MK |
1508 | d_addr = kmap_atomic(d_page); |
1509 | d_size = class->size - written; | |
1510 | d_off = 0; | |
312fcae2 MK |
1511 | } |
1512 | } | |
1513 | ||
1514 | kunmap_atomic(d_addr); | |
1515 | kunmap_atomic(s_addr); | |
1516 | } | |
1517 | ||
1518 | /* | |
f9044f17 | 1519 | * Find alloced object in zspage from index object and |
312fcae2 MK |
1520 | * return handle. |
1521 | */ | |
f9044f17 SS |
1522 | static unsigned long find_alloced_obj(struct size_class *class, |
1523 | struct page *page, int *obj_idx) | |
312fcae2 | 1524 | { |
671f2fa8 | 1525 | unsigned int offset; |
cf675acb | 1526 | int index = *obj_idx; |
312fcae2 MK |
1527 | unsigned long handle = 0; |
1528 | void *addr = kmap_atomic(page); | |
1529 | ||
3783689a | 1530 | offset = get_first_obj_offset(page); |
312fcae2 MK |
1531 | offset += class->size * index; |
1532 | ||
1533 | while (offset < PAGE_SIZE) { | |
f9044f17 | 1534 | if (obj_allocated(page, addr + offset, &handle)) |
b475d42d | 1535 | break; |
312fcae2 MK |
1536 | |
1537 | offset += class->size; | |
1538 | index++; | |
1539 | } | |
1540 | ||
1541 | kunmap_atomic(addr); | |
cf675acb GM |
1542 | |
1543 | *obj_idx = index; | |
1544 | ||
312fcae2 MK |
1545 | return handle; |
1546 | } | |
1547 | ||
ada5caed MK |
1548 | static void migrate_zspage(struct zs_pool *pool, struct zspage *src_zspage, |
1549 | struct zspage *dst_zspage) | |
312fcae2 MK |
1550 | { |
1551 | unsigned long used_obj, free_obj; | |
1552 | unsigned long handle; | |
ada5caed MK |
1553 | int obj_idx = 0; |
1554 | struct page *s_page = get_first_page(src_zspage); | |
1555 | struct size_class *class = pool->size_class[src_zspage->class]; | |
312fcae2 MK |
1556 | |
1557 | while (1) { | |
cf675acb | 1558 | handle = find_alloced_obj(class, s_page, &obj_idx); |
312fcae2 MK |
1559 | if (!handle) { |
1560 | s_page = get_next_page(s_page); | |
1561 | if (!s_page) | |
1562 | break; | |
41b88e14 | 1563 | obj_idx = 0; |
312fcae2 MK |
1564 | continue; |
1565 | } | |
1566 | ||
312fcae2 | 1567 | used_obj = handle_to_obj(handle); |
ada5caed | 1568 | free_obj = obj_malloc(pool, dst_zspage, handle); |
251cbb95 | 1569 | zs_object_copy(class, free_obj, used_obj); |
41b88e14 | 1570 | obj_idx++; |
312fcae2 | 1571 | record_obj(handle, free_obj); |
b3067742 | 1572 | obj_free(class->size, used_obj); |
df9cd3cb | 1573 | |
4ce36584 | 1574 | /* Stop if there is no more space */ |
ada5caed | 1575 | if (zspage_full(class, dst_zspage)) |
4ce36584 SS |
1576 | break; |
1577 | ||
df9cd3cb | 1578 | /* Stop if there are no more objects to migrate */ |
ada5caed | 1579 | if (zspage_empty(src_zspage)) |
df9cd3cb | 1580 | break; |
312fcae2 | 1581 | } |
312fcae2 MK |
1582 | } |
1583 | ||
4c7ac972 | 1584 | static struct zspage *isolate_src_zspage(struct size_class *class) |
312fcae2 | 1585 | { |
3783689a | 1586 | struct zspage *zspage; |
4c7ac972 | 1587 | int fg; |
312fcae2 | 1588 | |
4c7ac972 SS |
1589 | for (fg = ZS_INUSE_RATIO_10; fg <= ZS_INUSE_RATIO_99; fg++) { |
1590 | zspage = list_first_entry_or_null(&class->fullness_list[fg], | |
1591 | struct zspage, list); | |
1592 | if (zspage) { | |
67eaedc1 | 1593 | remove_zspage(class, zspage); |
4c7ac972 SS |
1594 | return zspage; |
1595 | } | |
3783689a MK |
1596 | } |
1597 | ||
4c7ac972 SS |
1598 | return zspage; |
1599 | } | |
1600 | ||
1601 | static struct zspage *isolate_dst_zspage(struct size_class *class) | |
1602 | { | |
1603 | struct zspage *zspage; | |
1604 | int fg; | |
1605 | ||
1606 | for (fg = ZS_INUSE_RATIO_99; fg >= ZS_INUSE_RATIO_10; fg--) { | |
1607 | zspage = list_first_entry_or_null(&class->fullness_list[fg], | |
1608 | struct zspage, list); | |
3783689a | 1609 | if (zspage) { |
67eaedc1 | 1610 | remove_zspage(class, zspage); |
3783689a | 1611 | return zspage; |
312fcae2 MK |
1612 | } |
1613 | } | |
1614 | ||
3783689a | 1615 | return zspage; |
312fcae2 MK |
1616 | } |
1617 | ||
860c707d | 1618 | /* |
3783689a | 1619 | * putback_zspage - add @zspage into right class's fullness list |
860c707d | 1620 | * @class: destination class |
3783689a | 1621 | * @zspage: target page |
860c707d | 1622 | * |
4c7ac972 | 1623 | * Return @zspage's fullness status |
860c707d | 1624 | */ |
4c7ac972 | 1625 | static int putback_zspage(struct size_class *class, struct zspage *zspage) |
312fcae2 | 1626 | { |
4c7ac972 | 1627 | int fullness; |
312fcae2 | 1628 | |
3783689a MK |
1629 | fullness = get_fullness_group(class, zspage); |
1630 | insert_zspage(class, zspage, fullness); | |
839373e6 | 1631 | |
860c707d | 1632 | return fullness; |
61989a80 | 1633 | } |
312fcae2 | 1634 | |
b3067742 | 1635 | #ifdef CONFIG_COMPACTION |
4d0a5402 CIK |
1636 | /* |
1637 | * To prevent zspage destroy during migration, zspage freeing should | |
1638 | * hold locks of all pages in the zspage. | |
1639 | */ | |
1640 | static void lock_zspage(struct zspage *zspage) | |
1641 | { | |
2505a981 | 1642 | struct page *curr_page, *page; |
4d0a5402 | 1643 | |
2505a981 SA |
1644 | /* |
1645 | * Pages we haven't locked yet can be migrated off the list while we're | |
1646 | * trying to lock them, so we need to be careful and only attempt to | |
1647 | * lock each page under migrate_read_lock(). Otherwise, the page we lock | |
1648 | * may no longer belong to the zspage. This means that we may wait for | |
1649 | * the wrong page to unlock, so we must take a reference to the page | |
1650 | * prior to waiting for it to unlock outside migrate_read_lock(). | |
1651 | */ | |
1652 | while (1) { | |
1653 | migrate_read_lock(zspage); | |
1654 | page = get_first_page(zspage); | |
1655 | if (trylock_page(page)) | |
1656 | break; | |
1657 | get_page(page); | |
1658 | migrate_read_unlock(zspage); | |
1659 | wait_on_page_locked(page); | |
1660 | put_page(page); | |
1661 | } | |
1662 | ||
1663 | curr_page = page; | |
1664 | while ((page = get_next_page(curr_page))) { | |
1665 | if (trylock_page(page)) { | |
1666 | curr_page = page; | |
1667 | } else { | |
1668 | get_page(page); | |
1669 | migrate_read_unlock(zspage); | |
1670 | wait_on_page_locked(page); | |
1671 | put_page(page); | |
1672 | migrate_read_lock(zspage); | |
1673 | } | |
1674 | } | |
1675 | migrate_read_unlock(zspage); | |
4d0a5402 | 1676 | } |
b3067742 | 1677 | #endif /* CONFIG_COMPACTION */ |
4d0a5402 | 1678 | |
48b4800a MK |
1679 | static void migrate_lock_init(struct zspage *zspage) |
1680 | { | |
1681 | rwlock_init(&zspage->lock); | |
1682 | } | |
1683 | ||
cfc451cf | 1684 | static void migrate_read_lock(struct zspage *zspage) __acquires(&zspage->lock) |
48b4800a MK |
1685 | { |
1686 | read_lock(&zspage->lock); | |
1687 | } | |
1688 | ||
8a374ccc | 1689 | static void migrate_read_unlock(struct zspage *zspage) __releases(&zspage->lock) |
48b4800a MK |
1690 | { |
1691 | read_unlock(&zspage->lock); | |
1692 | } | |
1693 | ||
1694 | static void migrate_write_lock(struct zspage *zspage) | |
1695 | { | |
1696 | write_lock(&zspage->lock); | |
1697 | } | |
1698 | ||
1699 | static void migrate_write_unlock(struct zspage *zspage) | |
1700 | { | |
1701 | write_unlock(&zspage->lock); | |
1702 | } | |
1703 | ||
568b567f | 1704 | #ifdef CONFIG_COMPACTION |
48b4800a | 1705 | |
68f2736a MWO |
1706 | static const struct movable_operations zsmalloc_mops; |
1707 | ||
48b4800a MK |
1708 | static void replace_sub_page(struct size_class *class, struct zspage *zspage, |
1709 | struct page *newpage, struct page *oldpage) | |
1710 | { | |
1711 | struct page *page; | |
1712 | struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE] = {NULL, }; | |
1713 | int idx = 0; | |
1714 | ||
1715 | page = get_first_page(zspage); | |
1716 | do { | |
1717 | if (page == oldpage) | |
1718 | pages[idx] = newpage; | |
1719 | else | |
1720 | pages[idx] = page; | |
1721 | idx++; | |
1722 | } while ((page = get_next_page(page)) != NULL); | |
1723 | ||
1724 | create_page_chain(class, zspage, pages); | |
1725 | set_first_obj_offset(newpage, get_first_obj_offset(oldpage)); | |
a41ec880 | 1726 | if (unlikely(ZsHugePage(zspage))) |
48b4800a | 1727 | newpage->index = oldpage->index; |
68f2736a | 1728 | __SetPageMovable(newpage, &zsmalloc_mops); |
48b4800a MK |
1729 | } |
1730 | ||
4d0a5402 | 1731 | static bool zs_page_isolate(struct page *page, isolate_mode_t mode) |
48b4800a | 1732 | { |
48b4800a MK |
1733 | /* |
1734 | * Page is locked so zspage couldn't be destroyed. For detail, look at | |
1735 | * lock_zspage in free_zspage. | |
1736 | */ | |
48b4800a MK |
1737 | VM_BUG_ON_PAGE(PageIsolated(page), page); |
1738 | ||
48b4800a MK |
1739 | return true; |
1740 | } | |
1741 | ||
68f2736a MWO |
1742 | static int zs_page_migrate(struct page *newpage, struct page *page, |
1743 | enum migrate_mode mode) | |
48b4800a MK |
1744 | { |
1745 | struct zs_pool *pool; | |
1746 | struct size_class *class; | |
48b4800a MK |
1747 | struct zspage *zspage; |
1748 | struct page *dummy; | |
1749 | void *s_addr, *d_addr, *addr; | |
671f2fa8 | 1750 | unsigned int offset; |
3ae92ac2 | 1751 | unsigned long handle; |
48b4800a MK |
1752 | unsigned long old_obj, new_obj; |
1753 | unsigned int obj_idx; | |
48b4800a | 1754 | |
2916ecc0 JG |
1755 | /* |
1756 | * We cannot support the _NO_COPY case here, because copy needs to | |
1757 | * happen under the zs lock, which does not work with | |
1758 | * MIGRATE_SYNC_NO_COPY workflow. | |
1759 | */ | |
1760 | if (mode == MIGRATE_SYNC_NO_COPY) | |
1761 | return -EINVAL; | |
1762 | ||
48b4800a MK |
1763 | VM_BUG_ON_PAGE(!PageIsolated(page), page); |
1764 | ||
68f2736a MWO |
1765 | /* The page is locked, so this pointer must remain valid */ |
1766 | zspage = get_zspage(page); | |
1767 | pool = zspage->pool; | |
b475d42d MK |
1768 | |
1769 | /* | |
c0547d0b | 1770 | * The pool's lock protects the race between zpage migration |
b475d42d MK |
1771 | * and zs_free. |
1772 | */ | |
c0547d0b | 1773 | spin_lock(&pool->lock); |
67f1c9cd | 1774 | class = zspage_class(pool, zspage); |
48b4800a | 1775 | |
b475d42d MK |
1776 | /* the migrate_write_lock protects zpage access via zs_map_object */ |
1777 | migrate_write_lock(zspage); | |
48b4800a | 1778 | |
b475d42d | 1779 | offset = get_first_obj_offset(page); |
48b4800a | 1780 | s_addr = kmap_atomic(page); |
48b4800a MK |
1781 | |
1782 | /* | |
1783 | * Here, any user cannot access all objects in the zspage so let's move. | |
1784 | */ | |
1785 | d_addr = kmap_atomic(newpage); | |
afb2d666 | 1786 | copy_page(d_addr, s_addr); |
48b4800a MK |
1787 | kunmap_atomic(d_addr); |
1788 | ||
b475d42d | 1789 | for (addr = s_addr + offset; addr < s_addr + PAGE_SIZE; |
48b4800a | 1790 | addr += class->size) { |
3ae92ac2 | 1791 | if (obj_allocated(page, addr, &handle)) { |
48b4800a MK |
1792 | |
1793 | old_obj = handle_to_obj(handle); | |
1794 | obj_to_location(old_obj, &dummy, &obj_idx); | |
1795 | new_obj = (unsigned long)location_to_obj(newpage, | |
1796 | obj_idx); | |
48b4800a MK |
1797 | record_obj(handle, new_obj); |
1798 | } | |
1799 | } | |
b475d42d | 1800 | kunmap_atomic(s_addr); |
48b4800a MK |
1801 | |
1802 | replace_sub_page(class, zspage, newpage, page); | |
b475d42d MK |
1803 | /* |
1804 | * Since we complete the data copy and set up new zspage structure, | |
c0547d0b | 1805 | * it's okay to release the pool's lock. |
b475d42d | 1806 | */ |
c0547d0b | 1807 | spin_unlock(&pool->lock); |
b475d42d | 1808 | migrate_write_unlock(zspage); |
48b4800a | 1809 | |
b475d42d | 1810 | get_page(newpage); |
ac8f05da CM |
1811 | if (page_zone(newpage) != page_zone(page)) { |
1812 | dec_zone_page_state(page, NR_ZSPAGES); | |
1813 | inc_zone_page_state(newpage, NR_ZSPAGES); | |
1814 | } | |
1815 | ||
48b4800a MK |
1816 | reset_page(page); |
1817 | put_page(page); | |
48b4800a | 1818 | |
b475d42d | 1819 | return MIGRATEPAGE_SUCCESS; |
48b4800a MK |
1820 | } |
1821 | ||
4d0a5402 | 1822 | static void zs_page_putback(struct page *page) |
48b4800a | 1823 | { |
48b4800a | 1824 | VM_BUG_ON_PAGE(!PageIsolated(page), page); |
48b4800a MK |
1825 | } |
1826 | ||
68f2736a | 1827 | static const struct movable_operations zsmalloc_mops = { |
48b4800a | 1828 | .isolate_page = zs_page_isolate, |
68f2736a | 1829 | .migrate_page = zs_page_migrate, |
48b4800a MK |
1830 | .putback_page = zs_page_putback, |
1831 | }; | |
1832 | ||
48b4800a MK |
1833 | /* |
1834 | * Caller should hold page_lock of all pages in the zspage | |
1835 | * In here, we cannot use zspage meta data. | |
1836 | */ | |
1837 | static void async_free_zspage(struct work_struct *work) | |
1838 | { | |
1839 | int i; | |
1840 | struct size_class *class; | |
48b4800a MK |
1841 | struct zspage *zspage, *tmp; |
1842 | LIST_HEAD(free_pages); | |
1843 | struct zs_pool *pool = container_of(work, struct zs_pool, | |
1844 | free_work); | |
1845 | ||
cf8e0fed | 1846 | for (i = 0; i < ZS_SIZE_CLASSES; i++) { |
48b4800a MK |
1847 | class = pool->size_class[i]; |
1848 | if (class->index != i) | |
1849 | continue; | |
1850 | ||
c0547d0b | 1851 | spin_lock(&pool->lock); |
4c7ac972 SS |
1852 | list_splice_init(&class->fullness_list[ZS_INUSE_RATIO_0], |
1853 | &free_pages); | |
c0547d0b | 1854 | spin_unlock(&pool->lock); |
48b4800a MK |
1855 | } |
1856 | ||
48b4800a MK |
1857 | list_for_each_entry_safe(zspage, tmp, &free_pages, list) { |
1858 | list_del(&zspage->list); | |
1859 | lock_zspage(zspage); | |
1860 | ||
c0547d0b | 1861 | spin_lock(&pool->lock); |
ce335e07 | 1862 | class = zspage_class(pool, zspage); |
33848337 | 1863 | __free_zspage(pool, class, zspage); |
c0547d0b | 1864 | spin_unlock(&pool->lock); |
48b4800a MK |
1865 | } |
1866 | }; | |
1867 | ||
1868 | static void kick_deferred_free(struct zs_pool *pool) | |
1869 | { | |
1870 | schedule_work(&pool->free_work); | |
1871 | } | |
1872 | ||
68f2736a MWO |
1873 | static void zs_flush_migration(struct zs_pool *pool) |
1874 | { | |
1875 | flush_work(&pool->free_work); | |
1876 | } | |
1877 | ||
48b4800a MK |
1878 | static void init_deferred_free(struct zs_pool *pool) |
1879 | { | |
1880 | INIT_WORK(&pool->free_work, async_free_zspage); | |
1881 | } | |
1882 | ||
1883 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage) | |
1884 | { | |
1885 | struct page *page = get_first_page(zspage); | |
1886 | ||
1887 | do { | |
1888 | WARN_ON(!trylock_page(page)); | |
68f2736a | 1889 | __SetPageMovable(page, &zsmalloc_mops); |
48b4800a MK |
1890 | unlock_page(page); |
1891 | } while ((page = get_next_page(page)) != NULL); | |
1892 | } | |
68f2736a MWO |
1893 | #else |
1894 | static inline void zs_flush_migration(struct zs_pool *pool) { } | |
48b4800a MK |
1895 | #endif |
1896 | ||
04f05909 SS |
1897 | /* |
1898 | * | |
1899 | * Based on the number of unused allocated objects calculate | |
1900 | * and return the number of pages that we can free. | |
04f05909 SS |
1901 | */ |
1902 | static unsigned long zs_can_compact(struct size_class *class) | |
1903 | { | |
1904 | unsigned long obj_wasted; | |
4c7ac972 SS |
1905 | unsigned long obj_allocated = zs_stat_get(class, ZS_OBJS_ALLOCATED); |
1906 | unsigned long obj_used = zs_stat_get(class, ZS_OBJS_INUSE); | |
04f05909 | 1907 | |
44f43e99 SS |
1908 | if (obj_allocated <= obj_used) |
1909 | return 0; | |
04f05909 | 1910 | |
44f43e99 | 1911 | obj_wasted = obj_allocated - obj_used; |
b4fd07a0 | 1912 | obj_wasted /= class->objs_per_zspage; |
04f05909 | 1913 | |
6cbf16b3 | 1914 | return obj_wasted * class->pages_per_zspage; |
04f05909 SS |
1915 | } |
1916 | ||
23959281 RY |
1917 | static unsigned long __zs_compact(struct zs_pool *pool, |
1918 | struct size_class *class) | |
312fcae2 | 1919 | { |
5a845e9f | 1920 | struct zspage *src_zspage = NULL; |
3783689a | 1921 | struct zspage *dst_zspage = NULL; |
23959281 | 1922 | unsigned long pages_freed = 0; |
312fcae2 | 1923 | |
c0547d0b NP |
1924 | /* |
1925 | * protect the race between zpage migration and zs_free | |
1926 | * as well as zpage allocation/free | |
1927 | */ | |
1928 | spin_lock(&pool->lock); | |
5a845e9f SS |
1929 | while (zs_can_compact(class)) { |
1930 | int fg; | |
312fcae2 | 1931 | |
5a845e9f SS |
1932 | if (!dst_zspage) { |
1933 | dst_zspage = isolate_dst_zspage(class); | |
1934 | if (!dst_zspage) | |
1935 | break; | |
5a845e9f SS |
1936 | } |
1937 | ||
1938 | src_zspage = isolate_src_zspage(class); | |
1939 | if (!src_zspage) | |
04f05909 SS |
1940 | break; |
1941 | ||
59def443 | 1942 | migrate_write_lock(src_zspage); |
ada5caed | 1943 | migrate_zspage(pool, src_zspage, dst_zspage); |
5a845e9f | 1944 | migrate_write_unlock(src_zspage); |
312fcae2 | 1945 | |
59def443 | 1946 | fg = putback_zspage(class, src_zspage); |
5a845e9f SS |
1947 | if (fg == ZS_INUSE_RATIO_0) { |
1948 | free_zspage(pool, class, src_zspage); | |
1949 | pages_freed += class->pages_per_zspage; | |
5a845e9f | 1950 | } |
f7ddb612 | 1951 | src_zspage = NULL; |
312fcae2 | 1952 | |
5a845e9f SS |
1953 | if (get_fullness_group(class, dst_zspage) == ZS_INUSE_RATIO_100 |
1954 | || spin_is_contended(&pool->lock)) { | |
4aa409ca | 1955 | putback_zspage(class, dst_zspage); |
b475d42d | 1956 | dst_zspage = NULL; |
b475d42d | 1957 | |
5a845e9f SS |
1958 | spin_unlock(&pool->lock); |
1959 | cond_resched(); | |
1960 | spin_lock(&pool->lock); | |
1961 | } | |
312fcae2 MK |
1962 | } |
1963 | ||
59def443 | 1964 | if (src_zspage) |
4aa409ca | 1965 | putback_zspage(class, src_zspage); |
312fcae2 | 1966 | |
59def443 | 1967 | if (dst_zspage) |
5a845e9f | 1968 | putback_zspage(class, dst_zspage); |
59def443 | 1969 | |
c0547d0b | 1970 | spin_unlock(&pool->lock); |
23959281 RY |
1971 | |
1972 | return pages_freed; | |
312fcae2 MK |
1973 | } |
1974 | ||
1975 | unsigned long zs_compact(struct zs_pool *pool) | |
1976 | { | |
1977 | int i; | |
312fcae2 | 1978 | struct size_class *class; |
23959281 | 1979 | unsigned long pages_freed = 0; |
312fcae2 | 1980 | |
d2658f20 SS |
1981 | /* |
1982 | * Pool compaction is performed under pool->lock so it is basically | |
1983 | * single-threaded. Having more than one thread in __zs_compact() | |
1984 | * will increase pool->lock contention, which will impact other | |
1985 | * zsmalloc operations that need pool->lock. | |
1986 | */ | |
1987 | if (atomic_xchg(&pool->compaction_in_progress, 1)) | |
1988 | return 0; | |
1989 | ||
cf8e0fed | 1990 | for (i = ZS_SIZE_CLASSES - 1; i >= 0; i--) { |
312fcae2 | 1991 | class = pool->size_class[i]; |
312fcae2 MK |
1992 | if (class->index != i) |
1993 | continue; | |
23959281 | 1994 | pages_freed += __zs_compact(pool, class); |
312fcae2 | 1995 | } |
23959281 | 1996 | atomic_long_add(pages_freed, &pool->stats.pages_compacted); |
d2658f20 | 1997 | atomic_set(&pool->compaction_in_progress, 0); |
312fcae2 | 1998 | |
23959281 | 1999 | return pages_freed; |
312fcae2 MK |
2000 | } |
2001 | EXPORT_SYMBOL_GPL(zs_compact); | |
61989a80 | 2002 | |
7d3f3938 SS |
2003 | void zs_pool_stats(struct zs_pool *pool, struct zs_pool_stats *stats) |
2004 | { | |
2005 | memcpy(stats, &pool->stats, sizeof(struct zs_pool_stats)); | |
2006 | } | |
2007 | EXPORT_SYMBOL_GPL(zs_pool_stats); | |
2008 | ||
ab9d306d SS |
2009 | static unsigned long zs_shrinker_scan(struct shrinker *shrinker, |
2010 | struct shrink_control *sc) | |
2011 | { | |
2012 | unsigned long pages_freed; | |
c19b548b | 2013 | struct zs_pool *pool = shrinker->private_data; |
ab9d306d | 2014 | |
ab9d306d SS |
2015 | /* |
2016 | * Compact classes and calculate compaction delta. | |
2017 | * Can run concurrently with a manually triggered | |
2018 | * (by user) compaction. | |
2019 | */ | |
23959281 | 2020 | pages_freed = zs_compact(pool); |
ab9d306d SS |
2021 | |
2022 | return pages_freed ? pages_freed : SHRINK_STOP; | |
2023 | } | |
2024 | ||
2025 | static unsigned long zs_shrinker_count(struct shrinker *shrinker, | |
2026 | struct shrink_control *sc) | |
2027 | { | |
2028 | int i; | |
2029 | struct size_class *class; | |
2030 | unsigned long pages_to_free = 0; | |
c19b548b | 2031 | struct zs_pool *pool = shrinker->private_data; |
ab9d306d | 2032 | |
cf8e0fed | 2033 | for (i = ZS_SIZE_CLASSES - 1; i >= 0; i--) { |
ab9d306d | 2034 | class = pool->size_class[i]; |
ab9d306d SS |
2035 | if (class->index != i) |
2036 | continue; | |
2037 | ||
ab9d306d | 2038 | pages_to_free += zs_can_compact(class); |
ab9d306d SS |
2039 | } |
2040 | ||
2041 | return pages_to_free; | |
2042 | } | |
2043 | ||
2044 | static void zs_unregister_shrinker(struct zs_pool *pool) | |
2045 | { | |
c19b548b | 2046 | shrinker_free(pool->shrinker); |
ab9d306d SS |
2047 | } |
2048 | ||
2049 | static int zs_register_shrinker(struct zs_pool *pool) | |
2050 | { | |
c19b548b QZ |
2051 | pool->shrinker = shrinker_alloc(0, "mm-zspool:%s", pool->name); |
2052 | if (!pool->shrinker) | |
2053 | return -ENOMEM; | |
2054 | ||
2055 | pool->shrinker->scan_objects = zs_shrinker_scan; | |
2056 | pool->shrinker->count_objects = zs_shrinker_count; | |
2057 | pool->shrinker->batch = 0; | |
2058 | pool->shrinker->private_data = pool; | |
ab9d306d | 2059 | |
c19b548b QZ |
2060 | shrinker_register(pool->shrinker); |
2061 | ||
2062 | return 0; | |
ab9d306d SS |
2063 | } |
2064 | ||
6260ae35 SS |
2065 | static int calculate_zspage_chain_size(int class_size) |
2066 | { | |
2067 | int i, min_waste = INT_MAX; | |
2068 | int chain_size = 1; | |
2069 | ||
e1d1f354 SS |
2070 | if (is_power_of_2(class_size)) |
2071 | return chain_size; | |
2072 | ||
6260ae35 SS |
2073 | for (i = 1; i <= ZS_MAX_PAGES_PER_ZSPAGE; i++) { |
2074 | int waste; | |
2075 | ||
2076 | waste = (i * PAGE_SIZE) % class_size; | |
2077 | if (waste < min_waste) { | |
2078 | min_waste = waste; | |
2079 | chain_size = i; | |
2080 | } | |
2081 | } | |
2082 | ||
2083 | return chain_size; | |
2084 | } | |
2085 | ||
00a61d86 | 2086 | /** |
66cdef66 | 2087 | * zs_create_pool - Creates an allocation pool to work from. |
fd854463 | 2088 | * @name: pool name to be created |
166cfda7 | 2089 | * |
66cdef66 GM |
2090 | * This function must be called before anything when using |
2091 | * the zsmalloc allocator. | |
166cfda7 | 2092 | * |
66cdef66 GM |
2093 | * On success, a pointer to the newly created pool is returned, |
2094 | * otherwise NULL. | |
396b7fd6 | 2095 | */ |
d0d8da2d | 2096 | struct zs_pool *zs_create_pool(const char *name) |
61989a80 | 2097 | { |
66cdef66 GM |
2098 | int i; |
2099 | struct zs_pool *pool; | |
2100 | struct size_class *prev_class = NULL; | |
61989a80 | 2101 | |
66cdef66 GM |
2102 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
2103 | if (!pool) | |
2104 | return NULL; | |
61989a80 | 2105 | |
48b4800a | 2106 | init_deferred_free(pool); |
c0547d0b | 2107 | spin_lock_init(&pool->lock); |
d2658f20 | 2108 | atomic_set(&pool->compaction_in_progress, 0); |
61989a80 | 2109 | |
2e40e163 MK |
2110 | pool->name = kstrdup(name, GFP_KERNEL); |
2111 | if (!pool->name) | |
2112 | goto err; | |
2113 | ||
3783689a | 2114 | if (create_cache(pool)) |
2e40e163 MK |
2115 | goto err; |
2116 | ||
c60369f0 | 2117 | /* |
399d8eeb | 2118 | * Iterate reversely, because, size of size_class that we want to use |
66cdef66 | 2119 | * for merging should be larger or equal to current size. |
c60369f0 | 2120 | */ |
cf8e0fed | 2121 | for (i = ZS_SIZE_CLASSES - 1; i >= 0; i--) { |
66cdef66 GM |
2122 | int size; |
2123 | int pages_per_zspage; | |
64d90465 | 2124 | int objs_per_zspage; |
66cdef66 | 2125 | struct size_class *class; |
4c7ac972 | 2126 | int fullness; |
c60369f0 | 2127 | |
66cdef66 GM |
2128 | size = ZS_MIN_ALLOC_SIZE + i * ZS_SIZE_CLASS_DELTA; |
2129 | if (size > ZS_MAX_ALLOC_SIZE) | |
2130 | size = ZS_MAX_ALLOC_SIZE; | |
6260ae35 | 2131 | pages_per_zspage = calculate_zspage_chain_size(size); |
64d90465 | 2132 | objs_per_zspage = pages_per_zspage * PAGE_SIZE / size; |
61989a80 | 2133 | |
010b495e SS |
2134 | /* |
2135 | * We iterate from biggest down to smallest classes, | |
2136 | * so huge_class_size holds the size of the first huge | |
2137 | * class. Any object bigger than or equal to that will | |
2138 | * endup in the huge class. | |
2139 | */ | |
2140 | if (pages_per_zspage != 1 && objs_per_zspage != 1 && | |
2141 | !huge_class_size) { | |
2142 | huge_class_size = size; | |
2143 | /* | |
2144 | * The object uses ZS_HANDLE_SIZE bytes to store the | |
2145 | * handle. We need to subtract it, because zs_malloc() | |
2146 | * unconditionally adds handle size before it performs | |
2147 | * size class search - so object may be smaller than | |
2148 | * huge class size, yet it still can end up in the huge | |
2149 | * class because it grows by ZS_HANDLE_SIZE extra bytes | |
2150 | * right before class lookup. | |
2151 | */ | |
2152 | huge_class_size -= (ZS_HANDLE_SIZE - 1); | |
2153 | } | |
2154 | ||
66cdef66 GM |
2155 | /* |
2156 | * size_class is used for normal zsmalloc operation such | |
2157 | * as alloc/free for that size. Although it is natural that we | |
2158 | * have one size_class for each size, there is a chance that we | |
2159 | * can get more memory utilization if we use one size_class for | |
2160 | * many different sizes whose size_class have same | |
2161 | * characteristics. So, we makes size_class point to | |
2162 | * previous size_class if possible. | |
2163 | */ | |
2164 | if (prev_class) { | |
64d90465 | 2165 | if (can_merge(prev_class, pages_per_zspage, objs_per_zspage)) { |
66cdef66 GM |
2166 | pool->size_class[i] = prev_class; |
2167 | continue; | |
2168 | } | |
2169 | } | |
2170 | ||
2171 | class = kzalloc(sizeof(struct size_class), GFP_KERNEL); | |
2172 | if (!class) | |
2173 | goto err; | |
2174 | ||
2175 | class->size = size; | |
2176 | class->index = i; | |
2177 | class->pages_per_zspage = pages_per_zspage; | |
64d90465 | 2178 | class->objs_per_zspage = objs_per_zspage; |
66cdef66 | 2179 | pool->size_class[i] = class; |
4c7ac972 SS |
2180 | |
2181 | fullness = ZS_INUSE_RATIO_0; | |
2182 | while (fullness < NR_FULLNESS_GROUPS) { | |
3783689a | 2183 | INIT_LIST_HEAD(&class->fullness_list[fullness]); |
4c7ac972 SS |
2184 | fullness++; |
2185 | } | |
66cdef66 GM |
2186 | |
2187 | prev_class = class; | |
61989a80 NG |
2188 | } |
2189 | ||
d34f6157 DS |
2190 | /* debug only, don't abort if it fails */ |
2191 | zs_pool_stat_create(pool, name); | |
0f050d99 | 2192 | |
ab9d306d | 2193 | /* |
93144ca3 AK |
2194 | * Not critical since shrinker is only used to trigger internal |
2195 | * defragmentation of the pool which is pretty optional thing. If | |
2196 | * registration fails we still can use the pool normally and user can | |
2197 | * trigger compaction manually. Thus, ignore return code. | |
ab9d306d | 2198 | */ |
93144ca3 AK |
2199 | zs_register_shrinker(pool); |
2200 | ||
66cdef66 GM |
2201 | return pool; |
2202 | ||
2203 | err: | |
2204 | zs_destroy_pool(pool); | |
2205 | return NULL; | |
61989a80 | 2206 | } |
66cdef66 | 2207 | EXPORT_SYMBOL_GPL(zs_create_pool); |
61989a80 | 2208 | |
66cdef66 | 2209 | void zs_destroy_pool(struct zs_pool *pool) |
61989a80 | 2210 | { |
66cdef66 | 2211 | int i; |
61989a80 | 2212 | |
ab9d306d | 2213 | zs_unregister_shrinker(pool); |
68f2736a | 2214 | zs_flush_migration(pool); |
0f050d99 GM |
2215 | zs_pool_stat_destroy(pool); |
2216 | ||
cf8e0fed | 2217 | for (i = 0; i < ZS_SIZE_CLASSES; i++) { |
66cdef66 GM |
2218 | int fg; |
2219 | struct size_class *class = pool->size_class[i]; | |
61989a80 | 2220 | |
4249a05f AR |
2221 | if (!class) |
2222 | continue; | |
2223 | ||
66cdef66 GM |
2224 | if (class->index != i) |
2225 | continue; | |
61989a80 | 2226 | |
4c7ac972 SS |
2227 | for (fg = ZS_INUSE_RATIO_0; fg < NR_FULLNESS_GROUPS; fg++) { |
2228 | if (list_empty(&class->fullness_list[fg])) | |
2229 | continue; | |
2230 | ||
2231 | pr_err("Class-%d fullness group %d is not empty\n", | |
2232 | class->size, fg); | |
66cdef66 GM |
2233 | } |
2234 | kfree(class); | |
2235 | } | |
f553646a | 2236 | |
3783689a | 2237 | destroy_cache(pool); |
0f050d99 | 2238 | kfree(pool->name); |
66cdef66 GM |
2239 | kfree(pool); |
2240 | } | |
2241 | EXPORT_SYMBOL_GPL(zs_destroy_pool); | |
b7418510 | 2242 | |
66cdef66 GM |
2243 | static int __init zs_init(void) |
2244 | { | |
48b4800a MK |
2245 | int ret; |
2246 | ||
215c89d0 SAS |
2247 | ret = cpuhp_setup_state(CPUHP_MM_ZS_PREPARE, "mm/zsmalloc:prepare", |
2248 | zs_cpu_prepare, zs_cpu_dead); | |
0f050d99 | 2249 | if (ret) |
68f2736a | 2250 | goto out; |
66cdef66 | 2251 | |
66cdef66 GM |
2252 | #ifdef CONFIG_ZPOOL |
2253 | zpool_register_driver(&zs_zpool_driver); | |
2254 | #endif | |
0f050d99 | 2255 | |
4abaac9b DS |
2256 | zs_stat_init(); |
2257 | ||
66cdef66 | 2258 | return 0; |
0f050d99 | 2259 | |
48b4800a | 2260 | out: |
0f050d99 | 2261 | return ret; |
61989a80 | 2262 | } |
61989a80 | 2263 | |
66cdef66 | 2264 | static void __exit zs_exit(void) |
61989a80 | 2265 | { |
66cdef66 GM |
2266 | #ifdef CONFIG_ZPOOL |
2267 | zpool_unregister_driver(&zs_zpool_driver); | |
2268 | #endif | |
215c89d0 | 2269 | cpuhp_remove_state(CPUHP_MM_ZS_PREPARE); |
0f050d99 GM |
2270 | |
2271 | zs_stat_exit(); | |
61989a80 | 2272 | } |
069f101f BH |
2273 | |
2274 | module_init(zs_init); | |
2275 | module_exit(zs_exit); | |
2276 | ||
2277 | MODULE_LICENSE("Dual BSD/GPL"); | |
2278 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); |