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 | ||
76 | /* | |
77 | * A single 'zspage' is composed of up to 2^N discontiguous 0-order (single) | |
78 | * pages. ZS_MAX_ZSPAGE_ORDER defines upper limit on N. | |
79 | */ | |
80 | #define ZS_MAX_ZSPAGE_ORDER 2 | |
81 | #define ZS_MAX_PAGES_PER_ZSPAGE (_AC(1, UL) << ZS_MAX_ZSPAGE_ORDER) | |
82 | ||
2e40e163 MK |
83 | #define ZS_HANDLE_SIZE (sizeof(unsigned long)) |
84 | ||
0959c63f SJ |
85 | /* |
86 | * Object location (<PFN>, <obj_idx>) is encoded as | |
b956b5ac | 87 | * a single (unsigned long) handle value. |
0959c63f | 88 | * |
bfd093f5 | 89 | * Note that object index <obj_idx> starts from 0. |
0959c63f SJ |
90 | * |
91 | * This is made more complicated by various memory models and PAE. | |
92 | */ | |
93 | ||
02390b87 KS |
94 | #ifndef MAX_POSSIBLE_PHYSMEM_BITS |
95 | #ifdef MAX_PHYSMEM_BITS | |
96 | #define MAX_POSSIBLE_PHYSMEM_BITS MAX_PHYSMEM_BITS | |
97 | #else | |
0959c63f SJ |
98 | /* |
99 | * If this definition of MAX_PHYSMEM_BITS is used, OBJ_INDEX_BITS will just | |
100 | * be PAGE_SHIFT | |
101 | */ | |
02390b87 | 102 | #define MAX_POSSIBLE_PHYSMEM_BITS BITS_PER_LONG |
0959c63f SJ |
103 | #endif |
104 | #endif | |
02390b87 KS |
105 | |
106 | #define _PFN_BITS (MAX_POSSIBLE_PHYSMEM_BITS - PAGE_SHIFT) | |
312fcae2 | 107 | |
312fcae2 MK |
108 | /* |
109 | * Head in allocated object should have OBJ_ALLOCATED_TAG | |
110 | * to identify the object was allocated or not. | |
111 | * It's okay to add the status bit in the least bit because | |
112 | * header keeps handle which is 4byte-aligned address so we | |
113 | * have room for two bit at least. | |
114 | */ | |
115 | #define OBJ_ALLOCATED_TAG 1 | |
85b32581 NP |
116 | |
117 | #ifdef CONFIG_ZPOOL | |
118 | /* | |
119 | * The second least-significant bit in the object's header identifies if the | |
120 | * value stored at the header is a deferred handle from the last reclaim | |
121 | * attempt. | |
122 | * | |
123 | * As noted above, this is valid because we have room for two bits. | |
124 | */ | |
125 | #define OBJ_DEFERRED_HANDLE_TAG 2 | |
126 | #define OBJ_TAG_BITS 2 | |
127 | #define OBJ_TAG_MASK (OBJ_ALLOCATED_TAG | OBJ_DEFERRED_HANDLE_TAG) | |
128 | #else | |
129 | #define OBJ_TAG_BITS 1 | |
130 | #define OBJ_TAG_MASK OBJ_ALLOCATED_TAG | |
131 | #endif /* CONFIG_ZPOOL */ | |
132 | ||
312fcae2 | 133 | #define OBJ_INDEX_BITS (BITS_PER_LONG - _PFN_BITS - OBJ_TAG_BITS) |
0959c63f SJ |
134 | #define OBJ_INDEX_MASK ((_AC(1, UL) << OBJ_INDEX_BITS) - 1) |
135 | ||
a41ec880 | 136 | #define HUGE_BITS 1 |
cf8e0fed JM |
137 | #define FULLNESS_BITS 2 |
138 | #define CLASS_BITS 8 | |
139 | #define ISOLATED_BITS 3 | |
140 | #define MAGIC_VAL_BITS 8 | |
141 | ||
0959c63f SJ |
142 | #define MAX(a, b) ((a) >= (b) ? (a) : (b)) |
143 | /* ZS_MIN_ALLOC_SIZE must be multiple of ZS_ALIGN */ | |
144 | #define ZS_MIN_ALLOC_SIZE \ | |
145 | MAX(32, (ZS_MAX_PAGES_PER_ZSPAGE << PAGE_SHIFT >> OBJ_INDEX_BITS)) | |
2e40e163 | 146 | /* each chunk includes extra space to keep handle */ |
7b60a685 | 147 | #define ZS_MAX_ALLOC_SIZE PAGE_SIZE |
0959c63f SJ |
148 | |
149 | /* | |
7eb52512 | 150 | * On systems with 4K page size, this gives 255 size classes! There is a |
0959c63f SJ |
151 | * trader-off here: |
152 | * - Large number of size classes is potentially wasteful as free page are | |
153 | * spread across these classes | |
154 | * - Small number of size classes causes large internal fragmentation | |
155 | * - Probably its better to use specific size classes (empirically | |
156 | * determined). NOTE: all those class sizes must be set as multiple of | |
157 | * ZS_ALIGN to make sure link_free itself never has to span 2 pages. | |
158 | * | |
159 | * ZS_MIN_ALLOC_SIZE and ZS_SIZE_CLASS_DELTA must be multiple of ZS_ALIGN | |
160 | * (reason above) | |
161 | */ | |
3783689a | 162 | #define ZS_SIZE_CLASS_DELTA (PAGE_SIZE >> CLASS_BITS) |
cf8e0fed JM |
163 | #define ZS_SIZE_CLASSES (DIV_ROUND_UP(ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE, \ |
164 | ZS_SIZE_CLASS_DELTA) + 1) | |
0959c63f | 165 | |
0959c63f | 166 | enum fullness_group { |
0959c63f | 167 | ZS_EMPTY, |
48b4800a MK |
168 | ZS_ALMOST_EMPTY, |
169 | ZS_ALMOST_FULL, | |
170 | ZS_FULL, | |
171 | NR_ZS_FULLNESS, | |
0959c63f SJ |
172 | }; |
173 | ||
3828a764 | 174 | enum class_stat_type { |
48b4800a MK |
175 | CLASS_EMPTY, |
176 | CLASS_ALMOST_EMPTY, | |
177 | CLASS_ALMOST_FULL, | |
178 | CLASS_FULL, | |
0f050d99 GM |
179 | OBJ_ALLOCATED, |
180 | OBJ_USED, | |
48b4800a | 181 | NR_ZS_STAT_TYPE, |
0f050d99 GM |
182 | }; |
183 | ||
0f050d99 GM |
184 | struct zs_size_stat { |
185 | unsigned long objs[NR_ZS_STAT_TYPE]; | |
186 | }; | |
187 | ||
57244594 SS |
188 | #ifdef CONFIG_ZSMALLOC_STAT |
189 | static struct dentry *zs_stat_root; | |
0f050d99 GM |
190 | #endif |
191 | ||
0959c63f SJ |
192 | /* |
193 | * We assign a page to ZS_ALMOST_EMPTY fullness group when: | |
194 | * n <= N / f, where | |
195 | * n = number of allocated objects | |
196 | * N = total number of objects zspage can store | |
6dd9737e | 197 | * f = fullness_threshold_frac |
0959c63f SJ |
198 | * |
199 | * Similarly, we assign zspage to: | |
200 | * ZS_ALMOST_FULL when n > N / f | |
201 | * ZS_EMPTY when n == 0 | |
202 | * ZS_FULL when n == N | |
203 | * | |
204 | * (see: fix_fullness_group()) | |
205 | */ | |
206 | static const int fullness_threshold_frac = 4; | |
010b495e | 207 | static size_t huge_class_size; |
0959c63f SJ |
208 | |
209 | struct size_class { | |
48b4800a | 210 | struct list_head fullness_list[NR_ZS_FULLNESS]; |
0959c63f SJ |
211 | /* |
212 | * Size of objects stored in this class. Must be multiple | |
213 | * of ZS_ALIGN. | |
214 | */ | |
215 | int size; | |
1fc6e27d | 216 | int objs_per_zspage; |
7dfa4612 WY |
217 | /* Number of PAGE_SIZE sized pages to combine to form a 'zspage' */ |
218 | int pages_per_zspage; | |
48b4800a MK |
219 | |
220 | unsigned int index; | |
221 | struct zs_size_stat stats; | |
0959c63f SJ |
222 | }; |
223 | ||
224 | /* | |
225 | * Placed within free objects to form a singly linked list. | |
3783689a | 226 | * For every zspage, zspage->freeobj gives head of this list. |
0959c63f SJ |
227 | * |
228 | * This must be power of 2 and less than or equal to ZS_ALIGN | |
229 | */ | |
230 | struct link_free { | |
2e40e163 MK |
231 | union { |
232 | /* | |
bfd093f5 | 233 | * Free object index; |
2e40e163 MK |
234 | * It's valid for non-allocated object |
235 | */ | |
bfd093f5 | 236 | unsigned long next; |
2e40e163 MK |
237 | /* |
238 | * Handle of allocated object. | |
239 | */ | |
240 | unsigned long handle; | |
85b32581 NP |
241 | #ifdef CONFIG_ZPOOL |
242 | /* | |
243 | * Deferred handle of a reclaimed object. | |
244 | */ | |
245 | unsigned long deferred_handle; | |
246 | #endif | |
2e40e163 | 247 | }; |
0959c63f SJ |
248 | }; |
249 | ||
250 | struct zs_pool { | |
6f3526d6 | 251 | const char *name; |
0f050d99 | 252 | |
cf8e0fed | 253 | struct size_class *size_class[ZS_SIZE_CLASSES]; |
2e40e163 | 254 | struct kmem_cache *handle_cachep; |
3783689a | 255 | struct kmem_cache *zspage_cachep; |
0959c63f | 256 | |
13de8933 | 257 | atomic_long_t pages_allocated; |
0f050d99 | 258 | |
7d3f3938 | 259 | struct zs_pool_stats stats; |
ab9d306d SS |
260 | |
261 | /* Compact classes */ | |
262 | struct shrinker shrinker; | |
93144ca3 | 263 | |
64f768c6 NP |
264 | #ifdef CONFIG_ZPOOL |
265 | /* List tracking the zspages in LRU order by most recently added object */ | |
266 | struct list_head lru; | |
bd0fded2 NP |
267 | struct zpool *zpool; |
268 | const struct zpool_ops *zpool_ops; | |
64f768c6 NP |
269 | #endif |
270 | ||
0f050d99 GM |
271 | #ifdef CONFIG_ZSMALLOC_STAT |
272 | struct dentry *stat_dentry; | |
273 | #endif | |
48b4800a | 274 | #ifdef CONFIG_COMPACTION |
48b4800a MK |
275 | struct work_struct free_work; |
276 | #endif | |
c0547d0b | 277 | spinlock_t lock; |
0959c63f | 278 | }; |
61989a80 | 279 | |
3783689a MK |
280 | struct zspage { |
281 | struct { | |
a41ec880 | 282 | unsigned int huge:HUGE_BITS; |
3783689a | 283 | unsigned int fullness:FULLNESS_BITS; |
85d492f2 | 284 | unsigned int class:CLASS_BITS + 1; |
48b4800a MK |
285 | unsigned int isolated:ISOLATED_BITS; |
286 | unsigned int magic:MAGIC_VAL_BITS; | |
3783689a MK |
287 | }; |
288 | unsigned int inuse; | |
bfd093f5 | 289 | unsigned int freeobj; |
3783689a MK |
290 | struct page *first_page; |
291 | struct list_head list; /* fullness list */ | |
64f768c6 NP |
292 | |
293 | #ifdef CONFIG_ZPOOL | |
294 | /* links the zspage to the lru list in the pool */ | |
295 | struct list_head lru; | |
9997bc01 | 296 | bool under_reclaim; |
64f768c6 NP |
297 | #endif |
298 | ||
68f2736a | 299 | struct zs_pool *pool; |
48b4800a | 300 | rwlock_t lock; |
3783689a | 301 | }; |
61989a80 | 302 | |
f553646a | 303 | struct mapping_area { |
a3726599 | 304 | local_lock_t lock; |
f553646a | 305 | char *vm_buf; /* copy buffer for objects that span pages */ |
f553646a SJ |
306 | char *vm_addr; /* address of kmap_atomic()'ed pages */ |
307 | enum zs_mapmode vm_mm; /* mapping mode */ | |
308 | }; | |
309 | ||
a41ec880 MK |
310 | /* huge object: pages_per_zspage == 1 && maxobj_per_zspage == 1 */ |
311 | static void SetZsHugePage(struct zspage *zspage) | |
312 | { | |
313 | zspage->huge = 1; | |
314 | } | |
315 | ||
316 | static bool ZsHugePage(struct zspage *zspage) | |
317 | { | |
318 | return zspage->huge; | |
319 | } | |
320 | ||
48b4800a MK |
321 | static void migrate_lock_init(struct zspage *zspage); |
322 | static void migrate_read_lock(struct zspage *zspage); | |
323 | static void migrate_read_unlock(struct zspage *zspage); | |
9997bc01 NP |
324 | |
325 | #ifdef CONFIG_COMPACTION | |
b475d42d MK |
326 | static void migrate_write_lock(struct zspage *zspage); |
327 | static void migrate_write_lock_nested(struct zspage *zspage); | |
328 | static void migrate_write_unlock(struct zspage *zspage); | |
48b4800a MK |
329 | static void kick_deferred_free(struct zs_pool *pool); |
330 | static void init_deferred_free(struct zs_pool *pool); | |
331 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage); | |
332 | #else | |
b475d42d MK |
333 | static void migrate_write_lock(struct zspage *zspage) {} |
334 | static void migrate_write_lock_nested(struct zspage *zspage) {} | |
335 | static void migrate_write_unlock(struct zspage *zspage) {} | |
48b4800a MK |
336 | static void kick_deferred_free(struct zs_pool *pool) {} |
337 | static void init_deferred_free(struct zs_pool *pool) {} | |
338 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage) {} | |
339 | #endif | |
340 | ||
3783689a | 341 | static int create_cache(struct zs_pool *pool) |
2e40e163 MK |
342 | { |
343 | pool->handle_cachep = kmem_cache_create("zs_handle", ZS_HANDLE_SIZE, | |
344 | 0, 0, NULL); | |
3783689a MK |
345 | if (!pool->handle_cachep) |
346 | return 1; | |
347 | ||
348 | pool->zspage_cachep = kmem_cache_create("zspage", sizeof(struct zspage), | |
349 | 0, 0, NULL); | |
350 | if (!pool->zspage_cachep) { | |
351 | kmem_cache_destroy(pool->handle_cachep); | |
352 | pool->handle_cachep = NULL; | |
353 | return 1; | |
354 | } | |
355 | ||
356 | return 0; | |
2e40e163 MK |
357 | } |
358 | ||
3783689a | 359 | static void destroy_cache(struct zs_pool *pool) |
2e40e163 | 360 | { |
cd10add0 | 361 | kmem_cache_destroy(pool->handle_cachep); |
3783689a | 362 | kmem_cache_destroy(pool->zspage_cachep); |
2e40e163 MK |
363 | } |
364 | ||
3783689a | 365 | static unsigned long cache_alloc_handle(struct zs_pool *pool, gfp_t gfp) |
2e40e163 MK |
366 | { |
367 | return (unsigned long)kmem_cache_alloc(pool->handle_cachep, | |
48b4800a | 368 | gfp & ~(__GFP_HIGHMEM|__GFP_MOVABLE)); |
2e40e163 MK |
369 | } |
370 | ||
3783689a | 371 | static void cache_free_handle(struct zs_pool *pool, unsigned long handle) |
2e40e163 MK |
372 | { |
373 | kmem_cache_free(pool->handle_cachep, (void *)handle); | |
374 | } | |
375 | ||
3783689a MK |
376 | static struct zspage *cache_alloc_zspage(struct zs_pool *pool, gfp_t flags) |
377 | { | |
f0231305 | 378 | return kmem_cache_zalloc(pool->zspage_cachep, |
48b4800a | 379 | flags & ~(__GFP_HIGHMEM|__GFP_MOVABLE)); |
399d8eeb | 380 | } |
3783689a MK |
381 | |
382 | static void cache_free_zspage(struct zs_pool *pool, struct zspage *zspage) | |
383 | { | |
384 | kmem_cache_free(pool->zspage_cachep, zspage); | |
385 | } | |
386 | ||
c0547d0b | 387 | /* pool->lock(which owns the handle) synchronizes races */ |
2e40e163 MK |
388 | static void record_obj(unsigned long handle, unsigned long obj) |
389 | { | |
b475d42d | 390 | *(unsigned long *)handle = obj; |
2e40e163 MK |
391 | } |
392 | ||
c795779d DS |
393 | /* zpool driver */ |
394 | ||
395 | #ifdef CONFIG_ZPOOL | |
396 | ||
6f3526d6 | 397 | static void *zs_zpool_create(const char *name, gfp_t gfp, |
78672779 | 398 | const struct zpool_ops *zpool_ops, |
479305fd | 399 | struct zpool *zpool) |
c795779d | 400 | { |
d0d8da2d SS |
401 | /* |
402 | * Ignore global gfp flags: zs_malloc() may be invoked from | |
403 | * different contexts and its caller must provide a valid | |
404 | * gfp mask. | |
405 | */ | |
bd0fded2 NP |
406 | struct zs_pool *pool = zs_create_pool(name); |
407 | ||
408 | if (pool) { | |
409 | pool->zpool = zpool; | |
410 | pool->zpool_ops = zpool_ops; | |
411 | } | |
412 | ||
413 | return pool; | |
c795779d DS |
414 | } |
415 | ||
416 | static void zs_zpool_destroy(void *pool) | |
417 | { | |
418 | zs_destroy_pool(pool); | |
419 | } | |
420 | ||
421 | static int zs_zpool_malloc(void *pool, size_t size, gfp_t gfp, | |
422 | unsigned long *handle) | |
423 | { | |
d0d8da2d | 424 | *handle = zs_malloc(pool, size, gfp); |
c7e6f17b | 425 | |
65917b53 | 426 | if (IS_ERR_VALUE(*handle)) |
c7e6f17b HZ |
427 | return PTR_ERR((void *)*handle); |
428 | return 0; | |
c795779d DS |
429 | } |
430 | static void zs_zpool_free(void *pool, unsigned long handle) | |
431 | { | |
432 | zs_free(pool, handle); | |
433 | } | |
434 | ||
9997bc01 NP |
435 | static int zs_reclaim_page(struct zs_pool *pool, unsigned int retries); |
436 | ||
437 | static int zs_zpool_shrink(void *pool, unsigned int pages, | |
438 | unsigned int *reclaimed) | |
439 | { | |
440 | unsigned int total = 0; | |
441 | int ret = -EINVAL; | |
442 | ||
443 | while (total < pages) { | |
444 | ret = zs_reclaim_page(pool, 8); | |
445 | if (ret < 0) | |
446 | break; | |
447 | total++; | |
448 | } | |
449 | ||
450 | if (reclaimed) | |
451 | *reclaimed = total; | |
452 | ||
453 | return ret; | |
454 | } | |
455 | ||
c795779d DS |
456 | static void *zs_zpool_map(void *pool, unsigned long handle, |
457 | enum zpool_mapmode mm) | |
458 | { | |
459 | enum zs_mapmode zs_mm; | |
460 | ||
461 | switch (mm) { | |
462 | case ZPOOL_MM_RO: | |
463 | zs_mm = ZS_MM_RO; | |
464 | break; | |
465 | case ZPOOL_MM_WO: | |
466 | zs_mm = ZS_MM_WO; | |
467 | break; | |
e4a9bc58 | 468 | case ZPOOL_MM_RW: |
c795779d DS |
469 | default: |
470 | zs_mm = ZS_MM_RW; | |
471 | break; | |
472 | } | |
473 | ||
474 | return zs_map_object(pool, handle, zs_mm); | |
475 | } | |
476 | static void zs_zpool_unmap(void *pool, unsigned long handle) | |
477 | { | |
478 | zs_unmap_object(pool, handle); | |
479 | } | |
480 | ||
481 | static u64 zs_zpool_total_size(void *pool) | |
482 | { | |
722cdc17 | 483 | return zs_get_total_pages(pool) << PAGE_SHIFT; |
c795779d DS |
484 | } |
485 | ||
486 | static struct zpool_driver zs_zpool_driver = { | |
c165f25d HZ |
487 | .type = "zsmalloc", |
488 | .owner = THIS_MODULE, | |
489 | .create = zs_zpool_create, | |
490 | .destroy = zs_zpool_destroy, | |
491 | .malloc_support_movable = true, | |
492 | .malloc = zs_zpool_malloc, | |
493 | .free = zs_zpool_free, | |
9997bc01 | 494 | .shrink = zs_zpool_shrink, |
c165f25d HZ |
495 | .map = zs_zpool_map, |
496 | .unmap = zs_zpool_unmap, | |
497 | .total_size = zs_zpool_total_size, | |
c795779d DS |
498 | }; |
499 | ||
137f8cff | 500 | MODULE_ALIAS("zpool-zsmalloc"); |
c795779d DS |
501 | #endif /* CONFIG_ZPOOL */ |
502 | ||
61989a80 | 503 | /* per-cpu VM mapping areas for zspage accesses that cross page boundaries */ |
a3726599 MG |
504 | static DEFINE_PER_CPU(struct mapping_area, zs_map_area) = { |
505 | .lock = INIT_LOCAL_LOCK(lock), | |
506 | }; | |
61989a80 | 507 | |
3457f414 | 508 | static __maybe_unused int is_first_page(struct page *page) |
61989a80 | 509 | { |
a27545bf | 510 | return PagePrivate(page); |
61989a80 NG |
511 | } |
512 | ||
c0547d0b | 513 | /* Protected by pool->lock */ |
3783689a | 514 | static inline int get_zspage_inuse(struct zspage *zspage) |
4f42047b | 515 | { |
3783689a | 516 | return zspage->inuse; |
4f42047b MK |
517 | } |
518 | ||
4f42047b | 519 | |
3783689a | 520 | static inline void mod_zspage_inuse(struct zspage *zspage, int val) |
4f42047b | 521 | { |
3783689a | 522 | zspage->inuse += val; |
4f42047b MK |
523 | } |
524 | ||
48b4800a | 525 | static inline struct page *get_first_page(struct zspage *zspage) |
4f42047b | 526 | { |
48b4800a | 527 | struct page *first_page = zspage->first_page; |
3783689a | 528 | |
48b4800a MK |
529 | VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); |
530 | return first_page; | |
4f42047b MK |
531 | } |
532 | ||
671f2fa8 | 533 | static inline unsigned int get_first_obj_offset(struct page *page) |
4f42047b | 534 | { |
ffedd09f | 535 | return page->page_type; |
48b4800a | 536 | } |
3783689a | 537 | |
671f2fa8 | 538 | static inline void set_first_obj_offset(struct page *page, unsigned int offset) |
48b4800a | 539 | { |
ffedd09f | 540 | page->page_type = offset; |
4f42047b MK |
541 | } |
542 | ||
bfd093f5 | 543 | static inline unsigned int get_freeobj(struct zspage *zspage) |
4f42047b | 544 | { |
bfd093f5 | 545 | return zspage->freeobj; |
4f42047b MK |
546 | } |
547 | ||
bfd093f5 | 548 | static inline void set_freeobj(struct zspage *zspage, unsigned int obj) |
4f42047b | 549 | { |
bfd093f5 | 550 | zspage->freeobj = obj; |
4f42047b MK |
551 | } |
552 | ||
3783689a | 553 | static void get_zspage_mapping(struct zspage *zspage, |
a4209467 | 554 | unsigned int *class_idx, |
61989a80 NG |
555 | enum fullness_group *fullness) |
556 | { | |
48b4800a MK |
557 | BUG_ON(zspage->magic != ZSPAGE_MAGIC); |
558 | ||
3783689a MK |
559 | *fullness = zspage->fullness; |
560 | *class_idx = zspage->class; | |
61989a80 NG |
561 | } |
562 | ||
67f1c9cd MK |
563 | static struct size_class *zspage_class(struct zs_pool *pool, |
564 | struct zspage *zspage) | |
565 | { | |
566 | return pool->size_class[zspage->class]; | |
567 | } | |
568 | ||
3783689a | 569 | static void set_zspage_mapping(struct zspage *zspage, |
a4209467 | 570 | unsigned int class_idx, |
61989a80 NG |
571 | enum fullness_group fullness) |
572 | { | |
3783689a MK |
573 | zspage->class = class_idx; |
574 | zspage->fullness = fullness; | |
61989a80 NG |
575 | } |
576 | ||
c3e3e88a NC |
577 | /* |
578 | * zsmalloc divides the pool into various size classes where each | |
579 | * class maintains a list of zspages where each zspage is divided | |
580 | * into equal sized chunks. Each allocation falls into one of these | |
581 | * classes depending on its size. This function returns index of the | |
cb152a1a | 582 | * size class which has chunk size big enough to hold the given size. |
c3e3e88a | 583 | */ |
61989a80 NG |
584 | static int get_size_class_index(int size) |
585 | { | |
586 | int idx = 0; | |
587 | ||
588 | if (likely(size > ZS_MIN_ALLOC_SIZE)) | |
589 | idx = DIV_ROUND_UP(size - ZS_MIN_ALLOC_SIZE, | |
590 | ZS_SIZE_CLASS_DELTA); | |
591 | ||
cf8e0fed | 592 | return min_t(int, ZS_SIZE_CLASSES - 1, idx); |
61989a80 NG |
593 | } |
594 | ||
3828a764 MK |
595 | /* type can be of enum type class_stat_type or fullness_group */ |
596 | static inline void class_stat_inc(struct size_class *class, | |
3eb95fea | 597 | int type, unsigned long cnt) |
248ca1b0 | 598 | { |
48b4800a | 599 | class->stats.objs[type] += cnt; |
248ca1b0 MK |
600 | } |
601 | ||
3828a764 MK |
602 | /* type can be of enum type class_stat_type or fullness_group */ |
603 | static inline void class_stat_dec(struct size_class *class, | |
3eb95fea | 604 | int type, unsigned long cnt) |
248ca1b0 | 605 | { |
48b4800a | 606 | class->stats.objs[type] -= cnt; |
248ca1b0 MK |
607 | } |
608 | ||
3828a764 | 609 | /* type can be of enum type class_stat_type or fullness_group */ |
248ca1b0 | 610 | static inline unsigned long zs_stat_get(struct size_class *class, |
3eb95fea | 611 | int type) |
248ca1b0 | 612 | { |
48b4800a | 613 | return class->stats.objs[type]; |
248ca1b0 MK |
614 | } |
615 | ||
57244594 SS |
616 | #ifdef CONFIG_ZSMALLOC_STAT |
617 | ||
4abaac9b | 618 | static void __init zs_stat_init(void) |
248ca1b0 | 619 | { |
4abaac9b DS |
620 | if (!debugfs_initialized()) { |
621 | pr_warn("debugfs not available, stat dir not created\n"); | |
622 | return; | |
623 | } | |
248ca1b0 MK |
624 | |
625 | zs_stat_root = debugfs_create_dir("zsmalloc", NULL); | |
248ca1b0 MK |
626 | } |
627 | ||
628 | static void __exit zs_stat_exit(void) | |
629 | { | |
630 | debugfs_remove_recursive(zs_stat_root); | |
631 | } | |
632 | ||
1120ed54 SS |
633 | static unsigned long zs_can_compact(struct size_class *class); |
634 | ||
248ca1b0 MK |
635 | static int zs_stats_size_show(struct seq_file *s, void *v) |
636 | { | |
637 | int i; | |
638 | struct zs_pool *pool = s->private; | |
639 | struct size_class *class; | |
640 | int objs_per_zspage; | |
641 | unsigned long class_almost_full, class_almost_empty; | |
1120ed54 | 642 | unsigned long obj_allocated, obj_used, pages_used, freeable; |
248ca1b0 MK |
643 | unsigned long total_class_almost_full = 0, total_class_almost_empty = 0; |
644 | unsigned long total_objs = 0, total_used_objs = 0, total_pages = 0; | |
1120ed54 | 645 | unsigned long total_freeable = 0; |
248ca1b0 | 646 | |
1120ed54 | 647 | seq_printf(s, " %5s %5s %11s %12s %13s %10s %10s %16s %8s\n", |
248ca1b0 MK |
648 | "class", "size", "almost_full", "almost_empty", |
649 | "obj_allocated", "obj_used", "pages_used", | |
1120ed54 | 650 | "pages_per_zspage", "freeable"); |
248ca1b0 | 651 | |
cf8e0fed | 652 | for (i = 0; i < ZS_SIZE_CLASSES; i++) { |
248ca1b0 MK |
653 | class = pool->size_class[i]; |
654 | ||
655 | if (class->index != i) | |
656 | continue; | |
657 | ||
c0547d0b | 658 | spin_lock(&pool->lock); |
248ca1b0 MK |
659 | class_almost_full = zs_stat_get(class, CLASS_ALMOST_FULL); |
660 | class_almost_empty = zs_stat_get(class, CLASS_ALMOST_EMPTY); | |
661 | obj_allocated = zs_stat_get(class, OBJ_ALLOCATED); | |
662 | obj_used = zs_stat_get(class, OBJ_USED); | |
1120ed54 | 663 | freeable = zs_can_compact(class); |
c0547d0b | 664 | spin_unlock(&pool->lock); |
248ca1b0 | 665 | |
b4fd07a0 | 666 | objs_per_zspage = class->objs_per_zspage; |
248ca1b0 MK |
667 | pages_used = obj_allocated / objs_per_zspage * |
668 | class->pages_per_zspage; | |
669 | ||
1120ed54 SS |
670 | seq_printf(s, " %5u %5u %11lu %12lu %13lu" |
671 | " %10lu %10lu %16d %8lu\n", | |
248ca1b0 MK |
672 | i, class->size, class_almost_full, class_almost_empty, |
673 | obj_allocated, obj_used, pages_used, | |
1120ed54 | 674 | class->pages_per_zspage, freeable); |
248ca1b0 MK |
675 | |
676 | total_class_almost_full += class_almost_full; | |
677 | total_class_almost_empty += class_almost_empty; | |
678 | total_objs += obj_allocated; | |
679 | total_used_objs += obj_used; | |
680 | total_pages += pages_used; | |
1120ed54 | 681 | total_freeable += freeable; |
248ca1b0 MK |
682 | } |
683 | ||
684 | seq_puts(s, "\n"); | |
1120ed54 | 685 | seq_printf(s, " %5s %5s %11lu %12lu %13lu %10lu %10lu %16s %8lu\n", |
248ca1b0 MK |
686 | "Total", "", total_class_almost_full, |
687 | total_class_almost_empty, total_objs, | |
1120ed54 | 688 | total_used_objs, total_pages, "", total_freeable); |
248ca1b0 MK |
689 | |
690 | return 0; | |
691 | } | |
5ad35093 | 692 | DEFINE_SHOW_ATTRIBUTE(zs_stats_size); |
248ca1b0 | 693 | |
d34f6157 | 694 | static void zs_pool_stat_create(struct zs_pool *pool, const char *name) |
248ca1b0 | 695 | { |
4abaac9b DS |
696 | if (!zs_stat_root) { |
697 | pr_warn("no root stat dir, not creating <%s> stat dir\n", name); | |
d34f6157 | 698 | return; |
4abaac9b | 699 | } |
248ca1b0 | 700 | |
4268509a GKH |
701 | pool->stat_dentry = debugfs_create_dir(name, zs_stat_root); |
702 | ||
703 | debugfs_create_file("classes", S_IFREG | 0444, pool->stat_dentry, pool, | |
704 | &zs_stats_size_fops); | |
248ca1b0 MK |
705 | } |
706 | ||
707 | static void zs_pool_stat_destroy(struct zs_pool *pool) | |
708 | { | |
709 | debugfs_remove_recursive(pool->stat_dentry); | |
710 | } | |
711 | ||
712 | #else /* CONFIG_ZSMALLOC_STAT */ | |
4abaac9b | 713 | static void __init zs_stat_init(void) |
248ca1b0 | 714 | { |
248ca1b0 MK |
715 | } |
716 | ||
717 | static void __exit zs_stat_exit(void) | |
718 | { | |
719 | } | |
720 | ||
d34f6157 | 721 | static inline void zs_pool_stat_create(struct zs_pool *pool, const char *name) |
248ca1b0 | 722 | { |
248ca1b0 MK |
723 | } |
724 | ||
725 | static inline void zs_pool_stat_destroy(struct zs_pool *pool) | |
726 | { | |
727 | } | |
248ca1b0 MK |
728 | #endif |
729 | ||
48b4800a | 730 | |
c3e3e88a NC |
731 | /* |
732 | * For each size class, zspages are divided into different groups | |
733 | * depending on how "full" they are. This was done so that we could | |
734 | * easily find empty or nearly empty zspages when we try to shrink | |
735 | * the pool (not yet implemented). This function returns fullness | |
736 | * status of the given page. | |
737 | */ | |
1fc6e27d | 738 | static enum fullness_group get_fullness_group(struct size_class *class, |
3783689a | 739 | struct zspage *zspage) |
61989a80 | 740 | { |
1fc6e27d | 741 | int inuse, objs_per_zspage; |
61989a80 | 742 | enum fullness_group fg; |
830e4bc5 | 743 | |
3783689a | 744 | inuse = get_zspage_inuse(zspage); |
1fc6e27d | 745 | objs_per_zspage = class->objs_per_zspage; |
61989a80 NG |
746 | |
747 | if (inuse == 0) | |
748 | fg = ZS_EMPTY; | |
1fc6e27d | 749 | else if (inuse == objs_per_zspage) |
61989a80 | 750 | fg = ZS_FULL; |
1fc6e27d | 751 | else if (inuse <= 3 * objs_per_zspage / fullness_threshold_frac) |
61989a80 NG |
752 | fg = ZS_ALMOST_EMPTY; |
753 | else | |
754 | fg = ZS_ALMOST_FULL; | |
755 | ||
756 | return fg; | |
757 | } | |
758 | ||
c3e3e88a NC |
759 | /* |
760 | * Each size class maintains various freelists and zspages are assigned | |
761 | * to one of these freelists based on the number of live objects they | |
762 | * have. This functions inserts the given zspage into the freelist | |
763 | * identified by <class, fullness_group>. | |
764 | */ | |
251cbb95 | 765 | static void insert_zspage(struct size_class *class, |
3783689a MK |
766 | struct zspage *zspage, |
767 | enum fullness_group fullness) | |
61989a80 | 768 | { |
3783689a | 769 | struct zspage *head; |
61989a80 | 770 | |
3828a764 | 771 | class_stat_inc(class, fullness, 1); |
3783689a MK |
772 | head = list_first_entry_or_null(&class->fullness_list[fullness], |
773 | struct zspage, list); | |
58f17117 | 774 | /* |
3783689a MK |
775 | * We want to see more ZS_FULL pages and less almost empty/full. |
776 | * Put pages with higher ->inuse first. | |
58f17117 | 777 | */ |
110ceb82 ML |
778 | if (head && get_zspage_inuse(zspage) < get_zspage_inuse(head)) |
779 | list_add(&zspage->list, &head->list); | |
780 | else | |
781 | list_add(&zspage->list, &class->fullness_list[fullness]); | |
61989a80 NG |
782 | } |
783 | ||
c3e3e88a NC |
784 | /* |
785 | * This function removes the given zspage from the freelist identified | |
786 | * by <class, fullness_group>. | |
787 | */ | |
251cbb95 | 788 | static void remove_zspage(struct size_class *class, |
3783689a MK |
789 | struct zspage *zspage, |
790 | enum fullness_group fullness) | |
61989a80 | 791 | { |
3783689a | 792 | VM_BUG_ON(list_empty(&class->fullness_list[fullness])); |
61989a80 | 793 | |
3783689a | 794 | list_del_init(&zspage->list); |
3828a764 | 795 | class_stat_dec(class, fullness, 1); |
61989a80 NG |
796 | } |
797 | ||
c3e3e88a NC |
798 | /* |
799 | * Each size class maintains zspages in different fullness groups depending | |
800 | * on the number of live objects they contain. When allocating or freeing | |
801 | * objects, the fullness status of the page can change, say, from ALMOST_FULL | |
802 | * to ALMOST_EMPTY when freeing an object. This function checks if such | |
803 | * a status change has occurred for the given page and accordingly moves the | |
804 | * page from the freelist of the old fullness group to that of the new | |
805 | * fullness group. | |
806 | */ | |
c7806261 | 807 | static enum fullness_group fix_fullness_group(struct size_class *class, |
3783689a | 808 | struct zspage *zspage) |
61989a80 NG |
809 | { |
810 | int class_idx; | |
61989a80 NG |
811 | enum fullness_group currfg, newfg; |
812 | ||
3783689a MK |
813 | get_zspage_mapping(zspage, &class_idx, &currfg); |
814 | newfg = get_fullness_group(class, zspage); | |
61989a80 NG |
815 | if (newfg == currfg) |
816 | goto out; | |
817 | ||
c4549b87 MK |
818 | remove_zspage(class, zspage, currfg); |
819 | insert_zspage(class, zspage, newfg); | |
3783689a | 820 | set_zspage_mapping(zspage, class_idx, newfg); |
61989a80 NG |
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 | { |
a6c5e0f7 | 863 | struct zspage *zspage = (struct zspage *)page_private(page); |
48b4800a MK |
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 | { | |
a41ec880 MK |
871 | struct zspage *zspage = get_zspage(page); |
872 | ||
873 | if (unlikely(ZsHugePage(zspage))) | |
48b4800a MK |
874 | return NULL; |
875 | ||
ffedd09f | 876 | return (struct page *)page->index; |
61989a80 NG |
877 | } |
878 | ||
bfd093f5 MK |
879 | /** |
880 | * obj_to_location - get (<page>, <obj_idx>) from encoded object value | |
e8b098fc | 881 | * @obj: the encoded object value |
bfd093f5 MK |
882 | * @page: page object resides in zspage |
883 | * @obj_idx: object index | |
67296874 | 884 | */ |
bfd093f5 MK |
885 | static void obj_to_location(unsigned long obj, struct page **page, |
886 | unsigned int *obj_idx) | |
61989a80 | 887 | { |
bfd093f5 MK |
888 | obj >>= OBJ_TAG_BITS; |
889 | *page = pfn_to_page(obj >> OBJ_INDEX_BITS); | |
890 | *obj_idx = (obj & OBJ_INDEX_MASK); | |
891 | } | |
61989a80 | 892 | |
67f1c9cd MK |
893 | static void obj_to_page(unsigned long obj, struct page **page) |
894 | { | |
895 | obj >>= OBJ_TAG_BITS; | |
896 | *page = pfn_to_page(obj >> OBJ_INDEX_BITS); | |
897 | } | |
898 | ||
bfd093f5 MK |
899 | /** |
900 | * location_to_obj - get obj value encoded from (<page>, <obj_idx>) | |
901 | * @page: page object resides in zspage | |
902 | * @obj_idx: object index | |
903 | */ | |
904 | static unsigned long location_to_obj(struct page *page, unsigned int obj_idx) | |
905 | { | |
906 | unsigned long obj; | |
61989a80 | 907 | |
312fcae2 | 908 | obj = page_to_pfn(page) << OBJ_INDEX_BITS; |
bfd093f5 | 909 | obj |= obj_idx & OBJ_INDEX_MASK; |
312fcae2 | 910 | obj <<= OBJ_TAG_BITS; |
61989a80 | 911 | |
bfd093f5 | 912 | return obj; |
61989a80 NG |
913 | } |
914 | ||
2e40e163 MK |
915 | static unsigned long handle_to_obj(unsigned long handle) |
916 | { | |
917 | return *(unsigned long *)handle; | |
918 | } | |
919 | ||
85b32581 NP |
920 | static bool obj_tagged(struct page *page, void *obj, unsigned long *phandle, |
921 | int tag) | |
312fcae2 | 922 | { |
3ae92ac2 | 923 | unsigned long handle; |
a41ec880 | 924 | struct zspage *zspage = get_zspage(page); |
3ae92ac2 | 925 | |
a41ec880 | 926 | if (unlikely(ZsHugePage(zspage))) { |
830e4bc5 | 927 | VM_BUG_ON_PAGE(!is_first_page(page), page); |
3ae92ac2 | 928 | handle = page->index; |
7b60a685 | 929 | } else |
3ae92ac2 MK |
930 | handle = *(unsigned long *)obj; |
931 | ||
85b32581 | 932 | if (!(handle & tag)) |
3ae92ac2 MK |
933 | return false; |
934 | ||
85b32581 NP |
935 | /* Clear all tags before returning the handle */ |
936 | *phandle = handle & ~OBJ_TAG_MASK; | |
3ae92ac2 | 937 | return true; |
312fcae2 MK |
938 | } |
939 | ||
85b32581 NP |
940 | static inline bool obj_allocated(struct page *page, void *obj, unsigned long *phandle) |
941 | { | |
942 | return obj_tagged(page, obj, phandle, OBJ_ALLOCATED_TAG); | |
943 | } | |
944 | ||
945 | #ifdef CONFIG_ZPOOL | |
946 | static bool obj_stores_deferred_handle(struct page *page, void *obj, | |
947 | unsigned long *phandle) | |
948 | { | |
949 | return obj_tagged(page, obj, phandle, OBJ_DEFERRED_HANDLE_TAG); | |
950 | } | |
951 | #endif | |
952 | ||
f4477e90 NG |
953 | static void reset_page(struct page *page) |
954 | { | |
48b4800a | 955 | __ClearPageMovable(page); |
18fd06bf | 956 | ClearPagePrivate(page); |
f4477e90 | 957 | set_page_private(page, 0); |
48b4800a | 958 | page_mapcount_reset(page); |
ffedd09f | 959 | page->index = 0; |
48b4800a MK |
960 | } |
961 | ||
4d0a5402 | 962 | static int trylock_zspage(struct zspage *zspage) |
48b4800a MK |
963 | { |
964 | struct page *cursor, *fail; | |
965 | ||
966 | for (cursor = get_first_page(zspage); cursor != NULL; cursor = | |
967 | get_next_page(cursor)) { | |
968 | if (!trylock_page(cursor)) { | |
969 | fail = cursor; | |
970 | goto unlock; | |
971 | } | |
972 | } | |
973 | ||
974 | return 1; | |
975 | unlock: | |
976 | for (cursor = get_first_page(zspage); cursor != fail; cursor = | |
977 | get_next_page(cursor)) | |
978 | unlock_page(cursor); | |
979 | ||
980 | return 0; | |
f4477e90 NG |
981 | } |
982 | ||
9997bc01 | 983 | #ifdef CONFIG_ZPOOL |
85b32581 NP |
984 | static unsigned long find_deferred_handle_obj(struct size_class *class, |
985 | struct page *page, int *obj_idx); | |
986 | ||
9997bc01 NP |
987 | /* |
988 | * Free all the deferred handles whose objects are freed in zs_free. | |
989 | */ | |
85b32581 NP |
990 | static void free_handles(struct zs_pool *pool, struct size_class *class, |
991 | struct zspage *zspage) | |
9997bc01 | 992 | { |
85b32581 NP |
993 | int obj_idx = 0; |
994 | struct page *page = get_first_page(zspage); | |
995 | unsigned long handle; | |
9997bc01 | 996 | |
85b32581 NP |
997 | while (1) { |
998 | handle = find_deferred_handle_obj(class, page, &obj_idx); | |
999 | if (!handle) { | |
1000 | page = get_next_page(page); | |
1001 | if (!page) | |
1002 | break; | |
1003 | obj_idx = 0; | |
1004 | continue; | |
1005 | } | |
9997bc01 NP |
1006 | |
1007 | cache_free_handle(pool, handle); | |
85b32581 | 1008 | obj_idx++; |
9997bc01 NP |
1009 | } |
1010 | } | |
1011 | #else | |
85b32581 NP |
1012 | static inline void free_handles(struct zs_pool *pool, struct size_class *class, |
1013 | struct zspage *zspage) {} | |
9997bc01 NP |
1014 | #endif |
1015 | ||
48b4800a MK |
1016 | static void __free_zspage(struct zs_pool *pool, struct size_class *class, |
1017 | struct zspage *zspage) | |
61989a80 | 1018 | { |
3783689a | 1019 | struct page *page, *next; |
48b4800a MK |
1020 | enum fullness_group fg; |
1021 | unsigned int class_idx; | |
1022 | ||
1023 | get_zspage_mapping(zspage, &class_idx, &fg); | |
1024 | ||
c0547d0b | 1025 | assert_spin_locked(&pool->lock); |
61989a80 | 1026 | |
3783689a | 1027 | VM_BUG_ON(get_zspage_inuse(zspage)); |
48b4800a | 1028 | VM_BUG_ON(fg != ZS_EMPTY); |
61989a80 | 1029 | |
9997bc01 | 1030 | /* Free all deferred handles from zs_free */ |
85b32581 | 1031 | free_handles(pool, class, zspage); |
9997bc01 | 1032 | |
48b4800a | 1033 | next = page = get_first_page(zspage); |
3783689a | 1034 | do { |
48b4800a MK |
1035 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1036 | next = get_next_page(page); | |
3783689a | 1037 | reset_page(page); |
48b4800a | 1038 | unlock_page(page); |
91537fee | 1039 | dec_zone_page_state(page, NR_ZSPAGES); |
3783689a MK |
1040 | put_page(page); |
1041 | page = next; | |
1042 | } while (page != NULL); | |
61989a80 | 1043 | |
3783689a | 1044 | cache_free_zspage(pool, zspage); |
48b4800a | 1045 | |
3828a764 | 1046 | class_stat_dec(class, OBJ_ALLOCATED, class->objs_per_zspage); |
48b4800a MK |
1047 | atomic_long_sub(class->pages_per_zspage, |
1048 | &pool->pages_allocated); | |
1049 | } | |
1050 | ||
1051 | static void free_zspage(struct zs_pool *pool, struct size_class *class, | |
1052 | struct zspage *zspage) | |
1053 | { | |
1054 | VM_BUG_ON(get_zspage_inuse(zspage)); | |
1055 | VM_BUG_ON(list_empty(&zspage->list)); | |
1056 | ||
b475d42d MK |
1057 | /* |
1058 | * Since zs_free couldn't be sleepable, this function cannot call | |
1059 | * lock_page. The page locks trylock_zspage got will be released | |
1060 | * by __free_zspage. | |
1061 | */ | |
48b4800a MK |
1062 | if (!trylock_zspage(zspage)) { |
1063 | kick_deferred_free(pool); | |
1064 | return; | |
1065 | } | |
1066 | ||
1067 | remove_zspage(class, zspage, ZS_EMPTY); | |
64f768c6 NP |
1068 | #ifdef CONFIG_ZPOOL |
1069 | list_del(&zspage->lru); | |
1070 | #endif | |
48b4800a | 1071 | __free_zspage(pool, class, zspage); |
61989a80 NG |
1072 | } |
1073 | ||
1074 | /* Initialize a newly allocated zspage */ | |
3783689a | 1075 | static void init_zspage(struct size_class *class, struct zspage *zspage) |
61989a80 | 1076 | { |
bfd093f5 | 1077 | unsigned int freeobj = 1; |
61989a80 | 1078 | unsigned long off = 0; |
48b4800a | 1079 | struct page *page = get_first_page(zspage); |
830e4bc5 | 1080 | |
61989a80 NG |
1081 | while (page) { |
1082 | struct page *next_page; | |
1083 | struct link_free *link; | |
af4ee5e9 | 1084 | void *vaddr; |
61989a80 | 1085 | |
3783689a | 1086 | set_first_obj_offset(page, off); |
61989a80 | 1087 | |
af4ee5e9 MK |
1088 | vaddr = kmap_atomic(page); |
1089 | link = (struct link_free *)vaddr + off / sizeof(*link); | |
5538c562 DS |
1090 | |
1091 | while ((off += class->size) < PAGE_SIZE) { | |
3b1d9ca6 | 1092 | link->next = freeobj++ << OBJ_TAG_BITS; |
5538c562 | 1093 | link += class->size / sizeof(*link); |
61989a80 NG |
1094 | } |
1095 | ||
1096 | /* | |
1097 | * We now come to the last (full or partial) object on this | |
1098 | * page, which must point to the first object on the next | |
1099 | * page (if present) | |
1100 | */ | |
1101 | next_page = get_next_page(page); | |
bfd093f5 | 1102 | if (next_page) { |
3b1d9ca6 | 1103 | link->next = freeobj++ << OBJ_TAG_BITS; |
bfd093f5 MK |
1104 | } else { |
1105 | /* | |
3b1d9ca6 | 1106 | * Reset OBJ_TAG_BITS bit to last link to tell |
bfd093f5 MK |
1107 | * whether it's allocated object or not. |
1108 | */ | |
01a6ad9a | 1109 | link->next = -1UL << OBJ_TAG_BITS; |
bfd093f5 | 1110 | } |
af4ee5e9 | 1111 | kunmap_atomic(vaddr); |
61989a80 | 1112 | page = next_page; |
5538c562 | 1113 | off %= PAGE_SIZE; |
61989a80 | 1114 | } |
bdb0af7c | 1115 | |
64f768c6 NP |
1116 | #ifdef CONFIG_ZPOOL |
1117 | INIT_LIST_HEAD(&zspage->lru); | |
9997bc01 | 1118 | zspage->under_reclaim = false; |
64f768c6 NP |
1119 | #endif |
1120 | ||
bfd093f5 | 1121 | set_freeobj(zspage, 0); |
61989a80 NG |
1122 | } |
1123 | ||
48b4800a MK |
1124 | static void create_page_chain(struct size_class *class, struct zspage *zspage, |
1125 | struct page *pages[]) | |
61989a80 | 1126 | { |
bdb0af7c MK |
1127 | int i; |
1128 | struct page *page; | |
1129 | struct page *prev_page = NULL; | |
48b4800a | 1130 | int nr_pages = class->pages_per_zspage; |
61989a80 NG |
1131 | |
1132 | /* | |
1133 | * Allocate individual pages and link them together as: | |
ffedd09f | 1134 | * 1. all pages are linked together using page->index |
3783689a | 1135 | * 2. each sub-page point to zspage using page->private |
61989a80 | 1136 | * |
3783689a | 1137 | * we set PG_private to identify the first page (i.e. no other sub-page |
22c5cef1 | 1138 | * has this flag set). |
61989a80 | 1139 | */ |
bdb0af7c MK |
1140 | for (i = 0; i < nr_pages; i++) { |
1141 | page = pages[i]; | |
3783689a | 1142 | set_page_private(page, (unsigned long)zspage); |
ffedd09f | 1143 | page->index = 0; |
bdb0af7c | 1144 | if (i == 0) { |
3783689a | 1145 | zspage->first_page = page; |
a27545bf | 1146 | SetPagePrivate(page); |
48b4800a MK |
1147 | if (unlikely(class->objs_per_zspage == 1 && |
1148 | class->pages_per_zspage == 1)) | |
a41ec880 | 1149 | SetZsHugePage(zspage); |
3783689a | 1150 | } else { |
ffedd09f | 1151 | prev_page->index = (unsigned long)page; |
61989a80 | 1152 | } |
61989a80 NG |
1153 | prev_page = page; |
1154 | } | |
bdb0af7c | 1155 | } |
61989a80 | 1156 | |
bdb0af7c MK |
1157 | /* |
1158 | * Allocate a zspage for the given size class | |
1159 | */ | |
3783689a MK |
1160 | static struct zspage *alloc_zspage(struct zs_pool *pool, |
1161 | struct size_class *class, | |
1162 | gfp_t gfp) | |
bdb0af7c MK |
1163 | { |
1164 | int i; | |
bdb0af7c | 1165 | struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE]; |
3783689a MK |
1166 | struct zspage *zspage = cache_alloc_zspage(pool, gfp); |
1167 | ||
1168 | if (!zspage) | |
1169 | return NULL; | |
1170 | ||
48b4800a MK |
1171 | zspage->magic = ZSPAGE_MAGIC; |
1172 | migrate_lock_init(zspage); | |
61989a80 | 1173 | |
bdb0af7c MK |
1174 | for (i = 0; i < class->pages_per_zspage; i++) { |
1175 | struct page *page; | |
61989a80 | 1176 | |
3783689a | 1177 | page = alloc_page(gfp); |
bdb0af7c | 1178 | if (!page) { |
91537fee MK |
1179 | while (--i >= 0) { |
1180 | dec_zone_page_state(pages[i], NR_ZSPAGES); | |
bdb0af7c | 1181 | __free_page(pages[i]); |
91537fee | 1182 | } |
3783689a | 1183 | cache_free_zspage(pool, zspage); |
bdb0af7c MK |
1184 | return NULL; |
1185 | } | |
91537fee MK |
1186 | |
1187 | inc_zone_page_state(page, NR_ZSPAGES); | |
bdb0af7c | 1188 | pages[i] = page; |
61989a80 NG |
1189 | } |
1190 | ||
48b4800a | 1191 | create_page_chain(class, zspage, pages); |
3783689a | 1192 | init_zspage(class, zspage); |
68f2736a | 1193 | zspage->pool = pool; |
bdb0af7c | 1194 | |
3783689a | 1195 | return zspage; |
61989a80 NG |
1196 | } |
1197 | ||
3783689a | 1198 | static struct zspage *find_get_zspage(struct size_class *class) |
61989a80 NG |
1199 | { |
1200 | int i; | |
3783689a | 1201 | struct zspage *zspage; |
61989a80 | 1202 | |
48b4800a | 1203 | for (i = ZS_ALMOST_FULL; i >= ZS_EMPTY; i--) { |
3783689a MK |
1204 | zspage = list_first_entry_or_null(&class->fullness_list[i], |
1205 | struct zspage, list); | |
1206 | if (zspage) | |
61989a80 NG |
1207 | break; |
1208 | } | |
1209 | ||
3783689a | 1210 | return zspage; |
61989a80 NG |
1211 | } |
1212 | ||
f553646a SJ |
1213 | static inline int __zs_cpu_up(struct mapping_area *area) |
1214 | { | |
1215 | /* | |
1216 | * Make sure we don't leak memory if a cpu UP notification | |
1217 | * and zs_init() race and both call zs_cpu_up() on the same cpu | |
1218 | */ | |
1219 | if (area->vm_buf) | |
1220 | return 0; | |
40f9fb8c | 1221 | area->vm_buf = kmalloc(ZS_MAX_ALLOC_SIZE, GFP_KERNEL); |
f553646a SJ |
1222 | if (!area->vm_buf) |
1223 | return -ENOMEM; | |
1224 | return 0; | |
1225 | } | |
1226 | ||
1227 | static inline void __zs_cpu_down(struct mapping_area *area) | |
1228 | { | |
40f9fb8c | 1229 | kfree(area->vm_buf); |
f553646a SJ |
1230 | area->vm_buf = NULL; |
1231 | } | |
1232 | ||
1233 | static void *__zs_map_object(struct mapping_area *area, | |
1234 | struct page *pages[2], int off, int size) | |
5f601902 | 1235 | { |
5f601902 SJ |
1236 | int sizes[2]; |
1237 | void *addr; | |
f553646a | 1238 | char *buf = area->vm_buf; |
5f601902 | 1239 | |
f553646a SJ |
1240 | /* disable page faults to match kmap_atomic() return conditions */ |
1241 | pagefault_disable(); | |
1242 | ||
1243 | /* no read fastpath */ | |
1244 | if (area->vm_mm == ZS_MM_WO) | |
1245 | goto out; | |
5f601902 SJ |
1246 | |
1247 | sizes[0] = PAGE_SIZE - off; | |
1248 | sizes[1] = size - sizes[0]; | |
1249 | ||
5f601902 SJ |
1250 | /* copy object to per-cpu buffer */ |
1251 | addr = kmap_atomic(pages[0]); | |
1252 | memcpy(buf, addr + off, sizes[0]); | |
1253 | kunmap_atomic(addr); | |
1254 | addr = kmap_atomic(pages[1]); | |
1255 | memcpy(buf + sizes[0], addr, sizes[1]); | |
1256 | kunmap_atomic(addr); | |
f553646a SJ |
1257 | out: |
1258 | return area->vm_buf; | |
5f601902 SJ |
1259 | } |
1260 | ||
f553646a SJ |
1261 | static void __zs_unmap_object(struct mapping_area *area, |
1262 | struct page *pages[2], int off, int size) | |
5f601902 | 1263 | { |
5f601902 SJ |
1264 | int sizes[2]; |
1265 | void *addr; | |
2e40e163 | 1266 | char *buf; |
5f601902 | 1267 | |
f553646a SJ |
1268 | /* no write fastpath */ |
1269 | if (area->vm_mm == ZS_MM_RO) | |
1270 | goto out; | |
5f601902 | 1271 | |
7b60a685 | 1272 | buf = area->vm_buf; |
a82cbf07 YX |
1273 | buf = buf + ZS_HANDLE_SIZE; |
1274 | size -= ZS_HANDLE_SIZE; | |
1275 | off += ZS_HANDLE_SIZE; | |
2e40e163 | 1276 | |
5f601902 SJ |
1277 | sizes[0] = PAGE_SIZE - off; |
1278 | sizes[1] = size - sizes[0]; | |
1279 | ||
1280 | /* copy per-cpu buffer to object */ | |
1281 | addr = kmap_atomic(pages[0]); | |
1282 | memcpy(addr + off, buf, sizes[0]); | |
1283 | kunmap_atomic(addr); | |
1284 | addr = kmap_atomic(pages[1]); | |
1285 | memcpy(addr, buf + sizes[0], sizes[1]); | |
1286 | kunmap_atomic(addr); | |
f553646a SJ |
1287 | |
1288 | out: | |
1289 | /* enable page faults to match kunmap_atomic() return conditions */ | |
1290 | pagefault_enable(); | |
5f601902 | 1291 | } |
61989a80 | 1292 | |
215c89d0 | 1293 | static int zs_cpu_prepare(unsigned int cpu) |
61989a80 | 1294 | { |
61989a80 NG |
1295 | struct mapping_area *area; |
1296 | ||
215c89d0 SAS |
1297 | area = &per_cpu(zs_map_area, cpu); |
1298 | return __zs_cpu_up(area); | |
61989a80 NG |
1299 | } |
1300 | ||
215c89d0 | 1301 | static int zs_cpu_dead(unsigned int cpu) |
61989a80 | 1302 | { |
215c89d0 | 1303 | struct mapping_area *area; |
40f9fb8c | 1304 | |
215c89d0 SAS |
1305 | area = &per_cpu(zs_map_area, cpu); |
1306 | __zs_cpu_down(area); | |
1307 | return 0; | |
b1b00a5b SS |
1308 | } |
1309 | ||
64d90465 GM |
1310 | static bool can_merge(struct size_class *prev, int pages_per_zspage, |
1311 | int objs_per_zspage) | |
9eec4cd5 | 1312 | { |
64d90465 GM |
1313 | if (prev->pages_per_zspage == pages_per_zspage && |
1314 | prev->objs_per_zspage == objs_per_zspage) | |
1315 | return true; | |
9eec4cd5 | 1316 | |
64d90465 | 1317 | return false; |
9eec4cd5 JK |
1318 | } |
1319 | ||
3783689a | 1320 | static bool zspage_full(struct size_class *class, struct zspage *zspage) |
312fcae2 | 1321 | { |
3783689a | 1322 | return get_zspage_inuse(zspage) == class->objs_per_zspage; |
312fcae2 | 1323 | } |
7c2af309 AR |
1324 | |
1325 | /** | |
1326 | * zs_lookup_class_index() - Returns index of the zsmalloc &size_class | |
1327 | * that hold objects of the provided size. | |
1328 | * @pool: zsmalloc pool to use | |
1329 | * @size: object size | |
1330 | * | |
1331 | * Context: Any context. | |
1332 | * | |
1333 | * Return: the index of the zsmalloc &size_class that hold objects of the | |
1334 | * provided size. | |
1335 | */ | |
1336 | unsigned int zs_lookup_class_index(struct zs_pool *pool, unsigned int size) | |
1337 | { | |
1338 | struct size_class *class; | |
1339 | ||
1340 | class = pool->size_class[get_size_class_index(size)]; | |
1341 | ||
1342 | return class->index; | |
1343 | } | |
1344 | EXPORT_SYMBOL_GPL(zs_lookup_class_index); | |
312fcae2 | 1345 | |
66cdef66 GM |
1346 | unsigned long zs_get_total_pages(struct zs_pool *pool) |
1347 | { | |
1348 | return atomic_long_read(&pool->pages_allocated); | |
1349 | } | |
1350 | EXPORT_SYMBOL_GPL(zs_get_total_pages); | |
1351 | ||
4bbc0bc0 | 1352 | /** |
66cdef66 GM |
1353 | * zs_map_object - get address of allocated object from handle. |
1354 | * @pool: pool from which the object was allocated | |
1355 | * @handle: handle returned from zs_malloc | |
f0953a1b | 1356 | * @mm: mapping mode to use |
4bbc0bc0 | 1357 | * |
66cdef66 GM |
1358 | * Before using an object allocated from zs_malloc, it must be mapped using |
1359 | * this function. When done with the object, it must be unmapped using | |
1360 | * zs_unmap_object. | |
4bbc0bc0 | 1361 | * |
66cdef66 GM |
1362 | * Only one object can be mapped per cpu at a time. There is no protection |
1363 | * against nested mappings. | |
1364 | * | |
1365 | * This function returns with preemption and page faults disabled. | |
4bbc0bc0 | 1366 | */ |
66cdef66 GM |
1367 | void *zs_map_object(struct zs_pool *pool, unsigned long handle, |
1368 | enum zs_mapmode mm) | |
61989a80 | 1369 | { |
3783689a | 1370 | struct zspage *zspage; |
66cdef66 | 1371 | struct page *page; |
bfd093f5 MK |
1372 | unsigned long obj, off; |
1373 | unsigned int obj_idx; | |
61989a80 | 1374 | |
66cdef66 GM |
1375 | struct size_class *class; |
1376 | struct mapping_area *area; | |
1377 | struct page *pages[2]; | |
2e40e163 | 1378 | void *ret; |
61989a80 | 1379 | |
9eec4cd5 | 1380 | /* |
66cdef66 GM |
1381 | * Because we use per-cpu mapping areas shared among the |
1382 | * pools/users, we can't allow mapping in interrupt context | |
1383 | * because it can corrupt another users mappings. | |
9eec4cd5 | 1384 | */ |
1aedcafb | 1385 | BUG_ON(in_interrupt()); |
61989a80 | 1386 | |
b475d42d | 1387 | /* It guarantees it can get zspage from handle safely */ |
c0547d0b | 1388 | spin_lock(&pool->lock); |
2e40e163 MK |
1389 | obj = handle_to_obj(handle); |
1390 | obj_to_location(obj, &page, &obj_idx); | |
3783689a | 1391 | zspage = get_zspage(page); |
48b4800a | 1392 | |
64f768c6 NP |
1393 | #ifdef CONFIG_ZPOOL |
1394 | /* | |
1395 | * Move the zspage to front of pool's LRU. | |
1396 | * | |
1397 | * Note that this is swap-specific, so by definition there are no ongoing | |
1398 | * accesses to the memory while the page is swapped out that would make | |
1399 | * it "hot". A new entry is hot, then ages to the tail until it gets either | |
1400 | * written back or swaps back in. | |
1401 | * | |
1402 | * Furthermore, map is also called during writeback. We must not put an | |
1403 | * isolated page on the LRU mid-reclaim. | |
1404 | * | |
1405 | * As a result, only update the LRU when the page is mapped for write | |
1406 | * when it's first instantiated. | |
1407 | * | |
1408 | * This is a deviation from the other backends, which perform this update | |
1409 | * in the allocation function (zbud_alloc, z3fold_alloc). | |
1410 | */ | |
1411 | if (mm == ZS_MM_WO) { | |
1412 | if (!list_empty(&zspage->lru)) | |
1413 | list_del(&zspage->lru); | |
1414 | list_add(&zspage->lru, &pool->lru); | |
1415 | } | |
1416 | #endif | |
1417 | ||
b475d42d | 1418 | /* |
c0547d0b | 1419 | * migration cannot move any zpages in this zspage. Here, pool->lock |
b475d42d MK |
1420 | * is too heavy since callers would take some time until they calls |
1421 | * zs_unmap_object API so delegate the locking from class to zspage | |
1422 | * which is smaller granularity. | |
1423 | */ | |
48b4800a | 1424 | migrate_read_lock(zspage); |
c0547d0b | 1425 | spin_unlock(&pool->lock); |
48b4800a | 1426 | |
67f1c9cd | 1427 | class = zspage_class(pool, zspage); |
bfd093f5 | 1428 | off = (class->size * obj_idx) & ~PAGE_MASK; |
df8b5bb9 | 1429 | |
a3726599 MG |
1430 | local_lock(&zs_map_area.lock); |
1431 | area = this_cpu_ptr(&zs_map_area); | |
66cdef66 GM |
1432 | area->vm_mm = mm; |
1433 | if (off + class->size <= PAGE_SIZE) { | |
1434 | /* this object is contained entirely within a page */ | |
1435 | area->vm_addr = kmap_atomic(page); | |
2e40e163 MK |
1436 | ret = area->vm_addr + off; |
1437 | goto out; | |
61989a80 NG |
1438 | } |
1439 | ||
66cdef66 GM |
1440 | /* this object spans two pages */ |
1441 | pages[0] = page; | |
1442 | pages[1] = get_next_page(page); | |
1443 | BUG_ON(!pages[1]); | |
9eec4cd5 | 1444 | |
2e40e163 MK |
1445 | ret = __zs_map_object(area, pages, off, class->size); |
1446 | out: | |
a41ec880 | 1447 | if (likely(!ZsHugePage(zspage))) |
7b60a685 MK |
1448 | ret += ZS_HANDLE_SIZE; |
1449 | ||
1450 | return ret; | |
61989a80 | 1451 | } |
66cdef66 | 1452 | EXPORT_SYMBOL_GPL(zs_map_object); |
61989a80 | 1453 | |
66cdef66 | 1454 | void zs_unmap_object(struct zs_pool *pool, unsigned long handle) |
61989a80 | 1455 | { |
3783689a | 1456 | struct zspage *zspage; |
66cdef66 | 1457 | struct page *page; |
bfd093f5 MK |
1458 | unsigned long obj, off; |
1459 | unsigned int obj_idx; | |
61989a80 | 1460 | |
66cdef66 GM |
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 | 1466 | zspage = get_zspage(page); |
67f1c9cd | 1467 | class = zspage_class(pool, zspage); |
bfd093f5 | 1468 | off = (class->size * obj_idx) & ~PAGE_MASK; |
61989a80 | 1469 | |
66cdef66 GM |
1470 | area = this_cpu_ptr(&zs_map_area); |
1471 | if (off + class->size <= PAGE_SIZE) | |
1472 | kunmap_atomic(area->vm_addr); | |
1473 | else { | |
1474 | struct page *pages[2]; | |
40f9fb8c | 1475 | |
66cdef66 GM |
1476 | pages[0] = page; |
1477 | pages[1] = get_next_page(page); | |
1478 | BUG_ON(!pages[1]); | |
1479 | ||
1480 | __zs_unmap_object(area, pages, off, class->size); | |
1481 | } | |
a3726599 | 1482 | local_unlock(&zs_map_area.lock); |
48b4800a MK |
1483 | |
1484 | migrate_read_unlock(zspage); | |
61989a80 | 1485 | } |
66cdef66 | 1486 | EXPORT_SYMBOL_GPL(zs_unmap_object); |
61989a80 | 1487 | |
010b495e SS |
1488 | /** |
1489 | * zs_huge_class_size() - Returns the size (in bytes) of the first huge | |
1490 | * zsmalloc &size_class. | |
1491 | * @pool: zsmalloc pool to use | |
1492 | * | |
1493 | * The function returns the size of the first huge class - any object of equal | |
1494 | * or bigger size will be stored in zspage consisting of a single physical | |
1495 | * page. | |
1496 | * | |
1497 | * Context: Any context. | |
1498 | * | |
1499 | * Return: the size (in bytes) of the first huge zsmalloc &size_class. | |
1500 | */ | |
1501 | size_t zs_huge_class_size(struct zs_pool *pool) | |
1502 | { | |
1503 | return huge_class_size; | |
1504 | } | |
1505 | EXPORT_SYMBOL_GPL(zs_huge_class_size); | |
1506 | ||
0a5f079b | 1507 | static unsigned long obj_malloc(struct zs_pool *pool, |
3783689a | 1508 | struct zspage *zspage, unsigned long handle) |
c7806261 | 1509 | { |
bfd093f5 | 1510 | int i, nr_page, offset; |
c7806261 MK |
1511 | unsigned long obj; |
1512 | struct link_free *link; | |
0a5f079b | 1513 | struct size_class *class; |
c7806261 MK |
1514 | |
1515 | struct page *m_page; | |
bfd093f5 | 1516 | unsigned long m_offset; |
c7806261 MK |
1517 | void *vaddr; |
1518 | ||
0a5f079b | 1519 | class = pool->size_class[zspage->class]; |
312fcae2 | 1520 | handle |= OBJ_ALLOCATED_TAG; |
3783689a | 1521 | obj = get_freeobj(zspage); |
bfd093f5 MK |
1522 | |
1523 | offset = obj * class->size; | |
1524 | nr_page = offset >> PAGE_SHIFT; | |
1525 | m_offset = offset & ~PAGE_MASK; | |
1526 | m_page = get_first_page(zspage); | |
1527 | ||
1528 | for (i = 0; i < nr_page; i++) | |
1529 | m_page = get_next_page(m_page); | |
c7806261 MK |
1530 | |
1531 | vaddr = kmap_atomic(m_page); | |
1532 | link = (struct link_free *)vaddr + m_offset / sizeof(*link); | |
3b1d9ca6 | 1533 | set_freeobj(zspage, link->next >> OBJ_TAG_BITS); |
a41ec880 | 1534 | if (likely(!ZsHugePage(zspage))) |
7b60a685 MK |
1535 | /* record handle in the header of allocated chunk */ |
1536 | link->handle = handle; | |
1537 | else | |
3783689a MK |
1538 | /* record handle to page->index */ |
1539 | zspage->first_page->index = handle; | |
1540 | ||
c7806261 | 1541 | kunmap_atomic(vaddr); |
3783689a | 1542 | mod_zspage_inuse(zspage, 1); |
c7806261 | 1543 | |
bfd093f5 MK |
1544 | obj = location_to_obj(m_page, obj); |
1545 | ||
c7806261 MK |
1546 | return obj; |
1547 | } | |
1548 | ||
1549 | ||
61989a80 NG |
1550 | /** |
1551 | * zs_malloc - Allocate block of given size from pool. | |
1552 | * @pool: pool to allocate from | |
1553 | * @size: size of block to allocate | |
fd854463 | 1554 | * @gfp: gfp flags when allocating object |
61989a80 | 1555 | * |
00a61d86 | 1556 | * On success, handle to the allocated object is returned, |
c7e6f17b | 1557 | * otherwise an ERR_PTR(). |
61989a80 NG |
1558 | * Allocation requests with size > ZS_MAX_ALLOC_SIZE will fail. |
1559 | */ | |
d0d8da2d | 1560 | unsigned long zs_malloc(struct zs_pool *pool, size_t size, gfp_t gfp) |
61989a80 | 1561 | { |
2e40e163 | 1562 | unsigned long handle, obj; |
61989a80 | 1563 | struct size_class *class; |
48b4800a | 1564 | enum fullness_group newfg; |
3783689a | 1565 | struct zspage *zspage; |
61989a80 | 1566 | |
7b60a685 | 1567 | if (unlikely(!size || size > ZS_MAX_ALLOC_SIZE)) |
c7e6f17b | 1568 | return (unsigned long)ERR_PTR(-EINVAL); |
2e40e163 | 1569 | |
3783689a | 1570 | handle = cache_alloc_handle(pool, gfp); |
2e40e163 | 1571 | if (!handle) |
c7e6f17b | 1572 | return (unsigned long)ERR_PTR(-ENOMEM); |
61989a80 | 1573 | |
2e40e163 MK |
1574 | /* extra space in chunk to keep the handle */ |
1575 | size += ZS_HANDLE_SIZE; | |
9eec4cd5 | 1576 | class = pool->size_class[get_size_class_index(size)]; |
61989a80 | 1577 | |
c0547d0b NP |
1578 | /* pool->lock effectively protects the zpage migration */ |
1579 | spin_lock(&pool->lock); | |
3783689a | 1580 | zspage = find_get_zspage(class); |
48b4800a | 1581 | if (likely(zspage)) { |
0a5f079b | 1582 | obj = obj_malloc(pool, zspage, handle); |
48b4800a MK |
1583 | /* Now move the zspage to another fullness group, if required */ |
1584 | fix_fullness_group(class, zspage); | |
1585 | record_obj(handle, obj); | |
0a5f079b | 1586 | class_stat_inc(class, OBJ_USED, 1); |
c0547d0b | 1587 | spin_unlock(&pool->lock); |
61989a80 | 1588 | |
48b4800a MK |
1589 | return handle; |
1590 | } | |
0f050d99 | 1591 | |
c0547d0b | 1592 | spin_unlock(&pool->lock); |
48b4800a MK |
1593 | |
1594 | zspage = alloc_zspage(pool, class, gfp); | |
1595 | if (!zspage) { | |
1596 | cache_free_handle(pool, handle); | |
c7e6f17b | 1597 | return (unsigned long)ERR_PTR(-ENOMEM); |
61989a80 NG |
1598 | } |
1599 | ||
c0547d0b | 1600 | spin_lock(&pool->lock); |
0a5f079b | 1601 | obj = obj_malloc(pool, zspage, handle); |
48b4800a MK |
1602 | newfg = get_fullness_group(class, zspage); |
1603 | insert_zspage(class, zspage, newfg); | |
1604 | set_zspage_mapping(zspage, class->index, newfg); | |
2e40e163 | 1605 | record_obj(handle, obj); |
48b4800a MK |
1606 | atomic_long_add(class->pages_per_zspage, |
1607 | &pool->pages_allocated); | |
3828a764 | 1608 | class_stat_inc(class, OBJ_ALLOCATED, class->objs_per_zspage); |
0a5f079b | 1609 | class_stat_inc(class, OBJ_USED, 1); |
48b4800a MK |
1610 | |
1611 | /* We completely set up zspage so mark them as movable */ | |
1612 | SetZsPageMovable(pool, zspage); | |
c0547d0b | 1613 | spin_unlock(&pool->lock); |
61989a80 | 1614 | |
2e40e163 | 1615 | return handle; |
61989a80 NG |
1616 | } |
1617 | EXPORT_SYMBOL_GPL(zs_malloc); | |
1618 | ||
85b32581 | 1619 | static void obj_free(int class_size, unsigned long obj, unsigned long *handle) |
61989a80 NG |
1620 | { |
1621 | struct link_free *link; | |
3783689a MK |
1622 | struct zspage *zspage; |
1623 | struct page *f_page; | |
bfd093f5 MK |
1624 | unsigned long f_offset; |
1625 | unsigned int f_objidx; | |
af4ee5e9 | 1626 | void *vaddr; |
61989a80 | 1627 | |
2e40e163 | 1628 | obj_to_location(obj, &f_page, &f_objidx); |
0a5f079b | 1629 | f_offset = (class_size * f_objidx) & ~PAGE_MASK; |
3783689a | 1630 | zspage = get_zspage(f_page); |
61989a80 | 1631 | |
c7806261 | 1632 | vaddr = kmap_atomic(f_page); |
af4ee5e9 | 1633 | link = (struct link_free *)(vaddr + f_offset); |
85b32581 NP |
1634 | |
1635 | if (handle) { | |
1636 | #ifdef CONFIG_ZPOOL | |
1637 | /* Stores the (deferred) handle in the object's header */ | |
1638 | *handle |= OBJ_DEFERRED_HANDLE_TAG; | |
1639 | *handle &= ~OBJ_ALLOCATED_TAG; | |
1640 | ||
1641 | if (likely(!ZsHugePage(zspage))) | |
1642 | link->deferred_handle = *handle; | |
1643 | else | |
1644 | f_page->index = *handle; | |
1645 | #endif | |
1646 | } else { | |
1647 | /* Insert this object in containing zspage's freelist */ | |
1648 | if (likely(!ZsHugePage(zspage))) | |
1649 | link->next = get_freeobj(zspage) << OBJ_TAG_BITS; | |
1650 | else | |
1651 | f_page->index = 0; | |
1652 | set_freeobj(zspage, f_objidx); | |
1653 | } | |
1654 | ||
af4ee5e9 | 1655 | kunmap_atomic(vaddr); |
3783689a | 1656 | mod_zspage_inuse(zspage, -1); |
c7806261 MK |
1657 | } |
1658 | ||
1659 | void zs_free(struct zs_pool *pool, unsigned long handle) | |
1660 | { | |
3783689a MK |
1661 | struct zspage *zspage; |
1662 | struct page *f_page; | |
bfd093f5 | 1663 | unsigned long obj; |
c7806261 MK |
1664 | struct size_class *class; |
1665 | enum fullness_group fullness; | |
1666 | ||
a5d21721 | 1667 | if (IS_ERR_OR_NULL((void *)handle)) |
c7806261 MK |
1668 | return; |
1669 | ||
b475d42d | 1670 | /* |
c0547d0b | 1671 | * The pool->lock protects the race with zpage's migration |
b475d42d MK |
1672 | * so it's safe to get the page from handle. |
1673 | */ | |
c0547d0b | 1674 | spin_lock(&pool->lock); |
c7806261 | 1675 | obj = handle_to_obj(handle); |
67f1c9cd | 1676 | obj_to_page(obj, &f_page); |
3783689a | 1677 | zspage = get_zspage(f_page); |
67f1c9cd | 1678 | class = zspage_class(pool, zspage); |
b475d42d | 1679 | |
0a5f079b | 1680 | class_stat_dec(class, OBJ_USED, 1); |
9997bc01 NP |
1681 | |
1682 | #ifdef CONFIG_ZPOOL | |
1683 | if (zspage->under_reclaim) { | |
1684 | /* | |
1685 | * Reclaim needs the handles during writeback. It'll free | |
1686 | * them along with the zspage when it's done with them. | |
1687 | * | |
85b32581 | 1688 | * Record current deferred handle in the object's header. |
9997bc01 | 1689 | */ |
85b32581 | 1690 | obj_free(class->size, obj, &handle); |
9997bc01 NP |
1691 | spin_unlock(&pool->lock); |
1692 | return; | |
1693 | } | |
1694 | #endif | |
85b32581 NP |
1695 | obj_free(class->size, obj, NULL); |
1696 | ||
3783689a | 1697 | fullness = fix_fullness_group(class, zspage); |
9997bc01 NP |
1698 | if (fullness == ZS_EMPTY) |
1699 | free_zspage(pool, class, zspage); | |
48b4800a | 1700 | |
c0547d0b | 1701 | spin_unlock(&pool->lock); |
3783689a | 1702 | cache_free_handle(pool, handle); |
312fcae2 MK |
1703 | } |
1704 | EXPORT_SYMBOL_GPL(zs_free); | |
1705 | ||
251cbb95 MK |
1706 | static void zs_object_copy(struct size_class *class, unsigned long dst, |
1707 | unsigned long src) | |
312fcae2 MK |
1708 | { |
1709 | struct page *s_page, *d_page; | |
bfd093f5 | 1710 | unsigned int s_objidx, d_objidx; |
312fcae2 MK |
1711 | unsigned long s_off, d_off; |
1712 | void *s_addr, *d_addr; | |
1713 | int s_size, d_size, size; | |
1714 | int written = 0; | |
1715 | ||
1716 | s_size = d_size = class->size; | |
1717 | ||
1718 | obj_to_location(src, &s_page, &s_objidx); | |
1719 | obj_to_location(dst, &d_page, &d_objidx); | |
1720 | ||
bfd093f5 MK |
1721 | s_off = (class->size * s_objidx) & ~PAGE_MASK; |
1722 | d_off = (class->size * d_objidx) & ~PAGE_MASK; | |
312fcae2 MK |
1723 | |
1724 | if (s_off + class->size > PAGE_SIZE) | |
1725 | s_size = PAGE_SIZE - s_off; | |
1726 | ||
1727 | if (d_off + class->size > PAGE_SIZE) | |
1728 | d_size = PAGE_SIZE - d_off; | |
1729 | ||
1730 | s_addr = kmap_atomic(s_page); | |
1731 | d_addr = kmap_atomic(d_page); | |
1732 | ||
1733 | while (1) { | |
1734 | size = min(s_size, d_size); | |
1735 | memcpy(d_addr + d_off, s_addr + s_off, size); | |
1736 | written += size; | |
1737 | ||
1738 | if (written == class->size) | |
1739 | break; | |
1740 | ||
495819ea SS |
1741 | s_off += size; |
1742 | s_size -= size; | |
1743 | d_off += size; | |
1744 | d_size -= size; | |
1745 | ||
050a388b AR |
1746 | /* |
1747 | * Calling kunmap_atomic(d_addr) is necessary. kunmap_atomic() | |
1748 | * calls must occurs in reverse order of calls to kmap_atomic(). | |
1749 | * So, to call kunmap_atomic(s_addr) we should first call | |
46e87152 AR |
1750 | * kunmap_atomic(d_addr). For more details see |
1751 | * Documentation/mm/highmem.rst. | |
050a388b | 1752 | */ |
495819ea | 1753 | if (s_off >= PAGE_SIZE) { |
312fcae2 MK |
1754 | kunmap_atomic(d_addr); |
1755 | kunmap_atomic(s_addr); | |
1756 | s_page = get_next_page(s_page); | |
312fcae2 MK |
1757 | s_addr = kmap_atomic(s_page); |
1758 | d_addr = kmap_atomic(d_page); | |
1759 | s_size = class->size - written; | |
1760 | s_off = 0; | |
312fcae2 MK |
1761 | } |
1762 | ||
495819ea | 1763 | if (d_off >= PAGE_SIZE) { |
312fcae2 MK |
1764 | kunmap_atomic(d_addr); |
1765 | d_page = get_next_page(d_page); | |
312fcae2 MK |
1766 | d_addr = kmap_atomic(d_page); |
1767 | d_size = class->size - written; | |
1768 | d_off = 0; | |
312fcae2 MK |
1769 | } |
1770 | } | |
1771 | ||
1772 | kunmap_atomic(d_addr); | |
1773 | kunmap_atomic(s_addr); | |
1774 | } | |
1775 | ||
1776 | /* | |
85b32581 | 1777 | * Find object with a certain tag in zspage from index object and |
312fcae2 MK |
1778 | * return handle. |
1779 | */ | |
85b32581 NP |
1780 | static unsigned long find_tagged_obj(struct size_class *class, |
1781 | struct page *page, int *obj_idx, int tag) | |
312fcae2 | 1782 | { |
671f2fa8 | 1783 | unsigned int offset; |
cf675acb | 1784 | int index = *obj_idx; |
312fcae2 MK |
1785 | unsigned long handle = 0; |
1786 | void *addr = kmap_atomic(page); | |
1787 | ||
3783689a | 1788 | offset = get_first_obj_offset(page); |
312fcae2 MK |
1789 | offset += class->size * index; |
1790 | ||
1791 | while (offset < PAGE_SIZE) { | |
85b32581 | 1792 | if (obj_tagged(page, addr + offset, &handle, tag)) |
b475d42d | 1793 | break; |
312fcae2 MK |
1794 | |
1795 | offset += class->size; | |
1796 | index++; | |
1797 | } | |
1798 | ||
1799 | kunmap_atomic(addr); | |
cf675acb GM |
1800 | |
1801 | *obj_idx = index; | |
1802 | ||
312fcae2 MK |
1803 | return handle; |
1804 | } | |
1805 | ||
85b32581 NP |
1806 | /* |
1807 | * Find alloced object in zspage from index object and | |
1808 | * return handle. | |
1809 | */ | |
1810 | static unsigned long find_alloced_obj(struct size_class *class, | |
1811 | struct page *page, int *obj_idx) | |
1812 | { | |
1813 | return find_tagged_obj(class, page, obj_idx, OBJ_ALLOCATED_TAG); | |
1814 | } | |
1815 | ||
1816 | #ifdef CONFIG_ZPOOL | |
1817 | /* | |
1818 | * Find object storing a deferred handle in header in zspage from index object | |
1819 | * and return handle. | |
1820 | */ | |
1821 | static unsigned long find_deferred_handle_obj(struct size_class *class, | |
1822 | struct page *page, int *obj_idx) | |
1823 | { | |
1824 | return find_tagged_obj(class, page, obj_idx, OBJ_DEFERRED_HANDLE_TAG); | |
1825 | } | |
1826 | #endif | |
1827 | ||
312fcae2 | 1828 | struct zs_compact_control { |
3783689a | 1829 | /* Source spage for migration which could be a subpage of zspage */ |
312fcae2 MK |
1830 | struct page *s_page; |
1831 | /* Destination page for migration which should be a first page | |
1832 | * of zspage. */ | |
1833 | struct page *d_page; | |
1834 | /* Starting object index within @s_page which used for live object | |
1835 | * in the subpage. */ | |
41b88e14 | 1836 | int obj_idx; |
312fcae2 MK |
1837 | }; |
1838 | ||
1839 | static int migrate_zspage(struct zs_pool *pool, struct size_class *class, | |
1840 | struct zs_compact_control *cc) | |
1841 | { | |
1842 | unsigned long used_obj, free_obj; | |
1843 | unsigned long handle; | |
1844 | struct page *s_page = cc->s_page; | |
1845 | struct page *d_page = cc->d_page; | |
41b88e14 | 1846 | int obj_idx = cc->obj_idx; |
312fcae2 MK |
1847 | int ret = 0; |
1848 | ||
1849 | while (1) { | |
cf675acb | 1850 | handle = find_alloced_obj(class, s_page, &obj_idx); |
312fcae2 MK |
1851 | if (!handle) { |
1852 | s_page = get_next_page(s_page); | |
1853 | if (!s_page) | |
1854 | break; | |
41b88e14 | 1855 | obj_idx = 0; |
312fcae2 MK |
1856 | continue; |
1857 | } | |
1858 | ||
1859 | /* Stop if there is no more space */ | |
3783689a | 1860 | if (zspage_full(class, get_zspage(d_page))) { |
312fcae2 MK |
1861 | ret = -ENOMEM; |
1862 | break; | |
1863 | } | |
1864 | ||
1865 | used_obj = handle_to_obj(handle); | |
0a5f079b | 1866 | free_obj = obj_malloc(pool, get_zspage(d_page), handle); |
251cbb95 | 1867 | zs_object_copy(class, free_obj, used_obj); |
41b88e14 | 1868 | obj_idx++; |
312fcae2 | 1869 | record_obj(handle, free_obj); |
85b32581 | 1870 | obj_free(class->size, used_obj, NULL); |
312fcae2 MK |
1871 | } |
1872 | ||
1873 | /* Remember last position in this iteration */ | |
1874 | cc->s_page = s_page; | |
41b88e14 | 1875 | cc->obj_idx = obj_idx; |
312fcae2 MK |
1876 | |
1877 | return ret; | |
1878 | } | |
1879 | ||
3783689a | 1880 | static struct zspage *isolate_zspage(struct size_class *class, bool source) |
312fcae2 MK |
1881 | { |
1882 | int i; | |
3783689a MK |
1883 | struct zspage *zspage; |
1884 | enum fullness_group fg[2] = {ZS_ALMOST_EMPTY, ZS_ALMOST_FULL}; | |
312fcae2 | 1885 | |
3783689a MK |
1886 | if (!source) { |
1887 | fg[0] = ZS_ALMOST_FULL; | |
1888 | fg[1] = ZS_ALMOST_EMPTY; | |
1889 | } | |
1890 | ||
1891 | for (i = 0; i < 2; i++) { | |
1892 | zspage = list_first_entry_or_null(&class->fullness_list[fg[i]], | |
1893 | struct zspage, list); | |
1894 | if (zspage) { | |
1895 | remove_zspage(class, zspage, fg[i]); | |
1896 | return zspage; | |
312fcae2 MK |
1897 | } |
1898 | } | |
1899 | ||
3783689a | 1900 | return zspage; |
312fcae2 MK |
1901 | } |
1902 | ||
860c707d | 1903 | /* |
3783689a | 1904 | * putback_zspage - add @zspage into right class's fullness list |
860c707d | 1905 | * @class: destination class |
3783689a | 1906 | * @zspage: target page |
860c707d | 1907 | * |
3783689a | 1908 | * Return @zspage's fullness_group |
860c707d | 1909 | */ |
4aa409ca | 1910 | static enum fullness_group putback_zspage(struct size_class *class, |
3783689a | 1911 | struct zspage *zspage) |
312fcae2 | 1912 | { |
312fcae2 MK |
1913 | enum fullness_group fullness; |
1914 | ||
3783689a MK |
1915 | fullness = get_fullness_group(class, zspage); |
1916 | insert_zspage(class, zspage, fullness); | |
1917 | set_zspage_mapping(zspage, class->index, fullness); | |
839373e6 | 1918 | |
860c707d | 1919 | return fullness; |
61989a80 | 1920 | } |
312fcae2 | 1921 | |
9997bc01 | 1922 | #if defined(CONFIG_ZPOOL) || defined(CONFIG_COMPACTION) |
4d0a5402 CIK |
1923 | /* |
1924 | * To prevent zspage destroy during migration, zspage freeing should | |
1925 | * hold locks of all pages in the zspage. | |
1926 | */ | |
1927 | static void lock_zspage(struct zspage *zspage) | |
1928 | { | |
2505a981 | 1929 | struct page *curr_page, *page; |
4d0a5402 | 1930 | |
2505a981 SA |
1931 | /* |
1932 | * Pages we haven't locked yet can be migrated off the list while we're | |
1933 | * trying to lock them, so we need to be careful and only attempt to | |
1934 | * lock each page under migrate_read_lock(). Otherwise, the page we lock | |
1935 | * may no longer belong to the zspage. This means that we may wait for | |
1936 | * the wrong page to unlock, so we must take a reference to the page | |
1937 | * prior to waiting for it to unlock outside migrate_read_lock(). | |
1938 | */ | |
1939 | while (1) { | |
1940 | migrate_read_lock(zspage); | |
1941 | page = get_first_page(zspage); | |
1942 | if (trylock_page(page)) | |
1943 | break; | |
1944 | get_page(page); | |
1945 | migrate_read_unlock(zspage); | |
1946 | wait_on_page_locked(page); | |
1947 | put_page(page); | |
1948 | } | |
1949 | ||
1950 | curr_page = page; | |
1951 | while ((page = get_next_page(curr_page))) { | |
1952 | if (trylock_page(page)) { | |
1953 | curr_page = page; | |
1954 | } else { | |
1955 | get_page(page); | |
1956 | migrate_read_unlock(zspage); | |
1957 | wait_on_page_locked(page); | |
1958 | put_page(page); | |
1959 | migrate_read_lock(zspage); | |
1960 | } | |
1961 | } | |
1962 | migrate_read_unlock(zspage); | |
4d0a5402 | 1963 | } |
9997bc01 NP |
1964 | #endif /* defined(CONFIG_ZPOOL) || defined(CONFIG_COMPACTION) */ |
1965 | ||
1966 | #ifdef CONFIG_ZPOOL | |
1967 | /* | |
1968 | * Unlocks all the pages of the zspage. | |
1969 | * | |
1970 | * pool->lock must be held before this function is called | |
1971 | * to prevent the underlying pages from migrating. | |
1972 | */ | |
1973 | static void unlock_zspage(struct zspage *zspage) | |
1974 | { | |
1975 | struct page *page = get_first_page(zspage); | |
1976 | ||
1977 | do { | |
1978 | unlock_page(page); | |
1979 | } while ((page = get_next_page(page)) != NULL); | |
1980 | } | |
1981 | #endif /* CONFIG_ZPOOL */ | |
4d0a5402 | 1982 | |
48b4800a MK |
1983 | static void migrate_lock_init(struct zspage *zspage) |
1984 | { | |
1985 | rwlock_init(&zspage->lock); | |
1986 | } | |
1987 | ||
cfc451cf | 1988 | static void migrate_read_lock(struct zspage *zspage) __acquires(&zspage->lock) |
48b4800a MK |
1989 | { |
1990 | read_lock(&zspage->lock); | |
1991 | } | |
1992 | ||
8a374ccc | 1993 | static void migrate_read_unlock(struct zspage *zspage) __releases(&zspage->lock) |
48b4800a MK |
1994 | { |
1995 | read_unlock(&zspage->lock); | |
1996 | } | |
1997 | ||
9997bc01 | 1998 | #ifdef CONFIG_COMPACTION |
48b4800a MK |
1999 | static void migrate_write_lock(struct zspage *zspage) |
2000 | { | |
2001 | write_lock(&zspage->lock); | |
2002 | } | |
2003 | ||
b475d42d MK |
2004 | static void migrate_write_lock_nested(struct zspage *zspage) |
2005 | { | |
2006 | write_lock_nested(&zspage->lock, SINGLE_DEPTH_NESTING); | |
2007 | } | |
2008 | ||
48b4800a MK |
2009 | static void migrate_write_unlock(struct zspage *zspage) |
2010 | { | |
2011 | write_unlock(&zspage->lock); | |
2012 | } | |
2013 | ||
2014 | /* Number of isolated subpage for *page migration* in this zspage */ | |
2015 | static void inc_zspage_isolation(struct zspage *zspage) | |
2016 | { | |
2017 | zspage->isolated++; | |
2018 | } | |
2019 | ||
2020 | static void dec_zspage_isolation(struct zspage *zspage) | |
2021 | { | |
c4549b87 | 2022 | VM_BUG_ON(zspage->isolated == 0); |
48b4800a MK |
2023 | zspage->isolated--; |
2024 | } | |
2025 | ||
68f2736a MWO |
2026 | static const struct movable_operations zsmalloc_mops; |
2027 | ||
48b4800a MK |
2028 | static void replace_sub_page(struct size_class *class, struct zspage *zspage, |
2029 | struct page *newpage, struct page *oldpage) | |
2030 | { | |
2031 | struct page *page; | |
2032 | struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE] = {NULL, }; | |
2033 | int idx = 0; | |
2034 | ||
2035 | page = get_first_page(zspage); | |
2036 | do { | |
2037 | if (page == oldpage) | |
2038 | pages[idx] = newpage; | |
2039 | else | |
2040 | pages[idx] = page; | |
2041 | idx++; | |
2042 | } while ((page = get_next_page(page)) != NULL); | |
2043 | ||
2044 | create_page_chain(class, zspage, pages); | |
2045 | set_first_obj_offset(newpage, get_first_obj_offset(oldpage)); | |
a41ec880 | 2046 | if (unlikely(ZsHugePage(zspage))) |
48b4800a | 2047 | newpage->index = oldpage->index; |
68f2736a | 2048 | __SetPageMovable(newpage, &zsmalloc_mops); |
48b4800a MK |
2049 | } |
2050 | ||
4d0a5402 | 2051 | static bool zs_page_isolate(struct page *page, isolate_mode_t mode) |
48b4800a | 2052 | { |
48b4800a | 2053 | struct zspage *zspage; |
48b4800a MK |
2054 | |
2055 | /* | |
2056 | * Page is locked so zspage couldn't be destroyed. For detail, look at | |
2057 | * lock_zspage in free_zspage. | |
2058 | */ | |
48b4800a MK |
2059 | VM_BUG_ON_PAGE(PageIsolated(page), page); |
2060 | ||
2061 | zspage = get_zspage(page); | |
c4549b87 | 2062 | migrate_write_lock(zspage); |
48b4800a | 2063 | inc_zspage_isolation(zspage); |
c4549b87 | 2064 | migrate_write_unlock(zspage); |
48b4800a MK |
2065 | |
2066 | return true; | |
2067 | } | |
2068 | ||
68f2736a MWO |
2069 | static int zs_page_migrate(struct page *newpage, struct page *page, |
2070 | enum migrate_mode mode) | |
48b4800a MK |
2071 | { |
2072 | struct zs_pool *pool; | |
2073 | struct size_class *class; | |
48b4800a MK |
2074 | struct zspage *zspage; |
2075 | struct page *dummy; | |
2076 | void *s_addr, *d_addr, *addr; | |
671f2fa8 | 2077 | unsigned int offset; |
3ae92ac2 | 2078 | unsigned long handle; |
48b4800a MK |
2079 | unsigned long old_obj, new_obj; |
2080 | unsigned int obj_idx; | |
48b4800a | 2081 | |
2916ecc0 JG |
2082 | /* |
2083 | * We cannot support the _NO_COPY case here, because copy needs to | |
2084 | * happen under the zs lock, which does not work with | |
2085 | * MIGRATE_SYNC_NO_COPY workflow. | |
2086 | */ | |
2087 | if (mode == MIGRATE_SYNC_NO_COPY) | |
2088 | return -EINVAL; | |
2089 | ||
48b4800a MK |
2090 | VM_BUG_ON_PAGE(!PageIsolated(page), page); |
2091 | ||
68f2736a MWO |
2092 | /* The page is locked, so this pointer must remain valid */ |
2093 | zspage = get_zspage(page); | |
2094 | pool = zspage->pool; | |
b475d42d MK |
2095 | |
2096 | /* | |
c0547d0b | 2097 | * The pool's lock protects the race between zpage migration |
b475d42d MK |
2098 | * and zs_free. |
2099 | */ | |
c0547d0b | 2100 | spin_lock(&pool->lock); |
67f1c9cd | 2101 | class = zspage_class(pool, zspage); |
48b4800a | 2102 | |
b475d42d MK |
2103 | /* the migrate_write_lock protects zpage access via zs_map_object */ |
2104 | migrate_write_lock(zspage); | |
48b4800a | 2105 | |
b475d42d | 2106 | offset = get_first_obj_offset(page); |
48b4800a | 2107 | s_addr = kmap_atomic(page); |
48b4800a MK |
2108 | |
2109 | /* | |
2110 | * Here, any user cannot access all objects in the zspage so let's move. | |
2111 | */ | |
2112 | d_addr = kmap_atomic(newpage); | |
2113 | memcpy(d_addr, s_addr, PAGE_SIZE); | |
2114 | kunmap_atomic(d_addr); | |
2115 | ||
b475d42d | 2116 | for (addr = s_addr + offset; addr < s_addr + PAGE_SIZE; |
48b4800a | 2117 | addr += class->size) { |
3ae92ac2 | 2118 | if (obj_allocated(page, addr, &handle)) { |
48b4800a MK |
2119 | |
2120 | old_obj = handle_to_obj(handle); | |
2121 | obj_to_location(old_obj, &dummy, &obj_idx); | |
2122 | new_obj = (unsigned long)location_to_obj(newpage, | |
2123 | obj_idx); | |
48b4800a MK |
2124 | record_obj(handle, new_obj); |
2125 | } | |
2126 | } | |
b475d42d | 2127 | kunmap_atomic(s_addr); |
48b4800a MK |
2128 | |
2129 | replace_sub_page(class, zspage, newpage, page); | |
b475d42d MK |
2130 | /* |
2131 | * Since we complete the data copy and set up new zspage structure, | |
c0547d0b | 2132 | * it's okay to release the pool's lock. |
b475d42d | 2133 | */ |
c0547d0b | 2134 | spin_unlock(&pool->lock); |
48b4800a | 2135 | dec_zspage_isolation(zspage); |
b475d42d | 2136 | migrate_write_unlock(zspage); |
48b4800a | 2137 | |
b475d42d | 2138 | get_page(newpage); |
ac8f05da CM |
2139 | if (page_zone(newpage) != page_zone(page)) { |
2140 | dec_zone_page_state(page, NR_ZSPAGES); | |
2141 | inc_zone_page_state(newpage, NR_ZSPAGES); | |
2142 | } | |
2143 | ||
48b4800a MK |
2144 | reset_page(page); |
2145 | put_page(page); | |
48b4800a | 2146 | |
b475d42d | 2147 | return MIGRATEPAGE_SUCCESS; |
48b4800a MK |
2148 | } |
2149 | ||
4d0a5402 | 2150 | static void zs_page_putback(struct page *page) |
48b4800a | 2151 | { |
48b4800a MK |
2152 | struct zspage *zspage; |
2153 | ||
48b4800a MK |
2154 | VM_BUG_ON_PAGE(!PageIsolated(page), page); |
2155 | ||
2156 | zspage = get_zspage(page); | |
c4549b87 | 2157 | migrate_write_lock(zspage); |
48b4800a | 2158 | dec_zspage_isolation(zspage); |
c4549b87 | 2159 | migrate_write_unlock(zspage); |
48b4800a MK |
2160 | } |
2161 | ||
68f2736a | 2162 | static const struct movable_operations zsmalloc_mops = { |
48b4800a | 2163 | .isolate_page = zs_page_isolate, |
68f2736a | 2164 | .migrate_page = zs_page_migrate, |
48b4800a MK |
2165 | .putback_page = zs_page_putback, |
2166 | }; | |
2167 | ||
48b4800a MK |
2168 | /* |
2169 | * Caller should hold page_lock of all pages in the zspage | |
2170 | * In here, we cannot use zspage meta data. | |
2171 | */ | |
2172 | static void async_free_zspage(struct work_struct *work) | |
2173 | { | |
2174 | int i; | |
2175 | struct size_class *class; | |
2176 | unsigned int class_idx; | |
2177 | enum fullness_group fullness; | |
2178 | struct zspage *zspage, *tmp; | |
2179 | LIST_HEAD(free_pages); | |
2180 | struct zs_pool *pool = container_of(work, struct zs_pool, | |
2181 | free_work); | |
2182 | ||
cf8e0fed | 2183 | for (i = 0; i < ZS_SIZE_CLASSES; i++) { |
48b4800a MK |
2184 | class = pool->size_class[i]; |
2185 | if (class->index != i) | |
2186 | continue; | |
2187 | ||
c0547d0b | 2188 | spin_lock(&pool->lock); |
48b4800a | 2189 | list_splice_init(&class->fullness_list[ZS_EMPTY], &free_pages); |
c0547d0b | 2190 | spin_unlock(&pool->lock); |
48b4800a MK |
2191 | } |
2192 | ||
48b4800a MK |
2193 | list_for_each_entry_safe(zspage, tmp, &free_pages, list) { |
2194 | list_del(&zspage->list); | |
2195 | lock_zspage(zspage); | |
2196 | ||
2197 | get_zspage_mapping(zspage, &class_idx, &fullness); | |
2198 | VM_BUG_ON(fullness != ZS_EMPTY); | |
2199 | class = pool->size_class[class_idx]; | |
c0547d0b | 2200 | spin_lock(&pool->lock); |
64f768c6 NP |
2201 | #ifdef CONFIG_ZPOOL |
2202 | list_del(&zspage->lru); | |
2203 | #endif | |
33848337 | 2204 | __free_zspage(pool, class, zspage); |
c0547d0b | 2205 | spin_unlock(&pool->lock); |
48b4800a MK |
2206 | } |
2207 | }; | |
2208 | ||
2209 | static void kick_deferred_free(struct zs_pool *pool) | |
2210 | { | |
2211 | schedule_work(&pool->free_work); | |
2212 | } | |
2213 | ||
68f2736a MWO |
2214 | static void zs_flush_migration(struct zs_pool *pool) |
2215 | { | |
2216 | flush_work(&pool->free_work); | |
2217 | } | |
2218 | ||
48b4800a MK |
2219 | static void init_deferred_free(struct zs_pool *pool) |
2220 | { | |
2221 | INIT_WORK(&pool->free_work, async_free_zspage); | |
2222 | } | |
2223 | ||
2224 | static void SetZsPageMovable(struct zs_pool *pool, struct zspage *zspage) | |
2225 | { | |
2226 | struct page *page = get_first_page(zspage); | |
2227 | ||
2228 | do { | |
2229 | WARN_ON(!trylock_page(page)); | |
68f2736a | 2230 | __SetPageMovable(page, &zsmalloc_mops); |
48b4800a MK |
2231 | unlock_page(page); |
2232 | } while ((page = get_next_page(page)) != NULL); | |
2233 | } | |
68f2736a MWO |
2234 | #else |
2235 | static inline void zs_flush_migration(struct zs_pool *pool) { } | |
48b4800a MK |
2236 | #endif |
2237 | ||
04f05909 SS |
2238 | /* |
2239 | * | |
2240 | * Based on the number of unused allocated objects calculate | |
2241 | * and return the number of pages that we can free. | |
04f05909 SS |
2242 | */ |
2243 | static unsigned long zs_can_compact(struct size_class *class) | |
2244 | { | |
2245 | unsigned long obj_wasted; | |
44f43e99 SS |
2246 | unsigned long obj_allocated = zs_stat_get(class, OBJ_ALLOCATED); |
2247 | unsigned long obj_used = zs_stat_get(class, OBJ_USED); | |
04f05909 | 2248 | |
44f43e99 SS |
2249 | if (obj_allocated <= obj_used) |
2250 | return 0; | |
04f05909 | 2251 | |
44f43e99 | 2252 | obj_wasted = obj_allocated - obj_used; |
b4fd07a0 | 2253 | obj_wasted /= class->objs_per_zspage; |
04f05909 | 2254 | |
6cbf16b3 | 2255 | return obj_wasted * class->pages_per_zspage; |
04f05909 SS |
2256 | } |
2257 | ||
23959281 RY |
2258 | static unsigned long __zs_compact(struct zs_pool *pool, |
2259 | struct size_class *class) | |
312fcae2 | 2260 | { |
312fcae2 | 2261 | struct zs_compact_control cc; |
3783689a MK |
2262 | struct zspage *src_zspage; |
2263 | struct zspage *dst_zspage = NULL; | |
23959281 | 2264 | unsigned long pages_freed = 0; |
312fcae2 | 2265 | |
c0547d0b NP |
2266 | /* |
2267 | * protect the race between zpage migration and zs_free | |
2268 | * as well as zpage allocation/free | |
2269 | */ | |
2270 | spin_lock(&pool->lock); | |
3783689a | 2271 | while ((src_zspage = isolate_zspage(class, true))) { |
b475d42d MK |
2272 | /* protect someone accessing the zspage(i.e., zs_map_object) */ |
2273 | migrate_write_lock(src_zspage); | |
312fcae2 | 2274 | |
04f05909 SS |
2275 | if (!zs_can_compact(class)) |
2276 | break; | |
2277 | ||
41b88e14 | 2278 | cc.obj_idx = 0; |
48b4800a | 2279 | cc.s_page = get_first_page(src_zspage); |
312fcae2 | 2280 | |
3783689a | 2281 | while ((dst_zspage = isolate_zspage(class, false))) { |
b475d42d MK |
2282 | migrate_write_lock_nested(dst_zspage); |
2283 | ||
48b4800a | 2284 | cc.d_page = get_first_page(dst_zspage); |
312fcae2 | 2285 | /* |
0dc63d48 SS |
2286 | * If there is no more space in dst_page, resched |
2287 | * and see if anyone had allocated another zspage. | |
312fcae2 MK |
2288 | */ |
2289 | if (!migrate_zspage(pool, class, &cc)) | |
2290 | break; | |
2291 | ||
4aa409ca | 2292 | putback_zspage(class, dst_zspage); |
b475d42d MK |
2293 | migrate_write_unlock(dst_zspage); |
2294 | dst_zspage = NULL; | |
c0547d0b | 2295 | if (spin_is_contended(&pool->lock)) |
b475d42d | 2296 | break; |
312fcae2 MK |
2297 | } |
2298 | ||
2299 | /* Stop if we couldn't find slot */ | |
3783689a | 2300 | if (dst_zspage == NULL) |
312fcae2 MK |
2301 | break; |
2302 | ||
4aa409ca | 2303 | putback_zspage(class, dst_zspage); |
b475d42d MK |
2304 | migrate_write_unlock(dst_zspage); |
2305 | ||
4aa409ca | 2306 | if (putback_zspage(class, src_zspage) == ZS_EMPTY) { |
b475d42d | 2307 | migrate_write_unlock(src_zspage); |
48b4800a | 2308 | free_zspage(pool, class, src_zspage); |
23959281 | 2309 | pages_freed += class->pages_per_zspage; |
b475d42d MK |
2310 | } else |
2311 | migrate_write_unlock(src_zspage); | |
c0547d0b | 2312 | spin_unlock(&pool->lock); |
312fcae2 | 2313 | cond_resched(); |
c0547d0b | 2314 | spin_lock(&pool->lock); |
312fcae2 MK |
2315 | } |
2316 | ||
b475d42d | 2317 | if (src_zspage) { |
4aa409ca | 2318 | putback_zspage(class, src_zspage); |
b475d42d MK |
2319 | migrate_write_unlock(src_zspage); |
2320 | } | |
312fcae2 | 2321 | |
c0547d0b | 2322 | spin_unlock(&pool->lock); |
23959281 RY |
2323 | |
2324 | return pages_freed; | |
312fcae2 MK |
2325 | } |
2326 | ||
2327 | unsigned long zs_compact(struct zs_pool *pool) | |
2328 | { | |
2329 | int i; | |
312fcae2 | 2330 | struct size_class *class; |
23959281 | 2331 | unsigned long pages_freed = 0; |
312fcae2 | 2332 | |
cf8e0fed | 2333 | for (i = ZS_SIZE_CLASSES - 1; i >= 0; i--) { |
312fcae2 | 2334 | class = pool->size_class[i]; |
312fcae2 MK |
2335 | if (class->index != i) |
2336 | continue; | |
23959281 | 2337 | pages_freed += __zs_compact(pool, class); |
312fcae2 | 2338 | } |
23959281 | 2339 | atomic_long_add(pages_freed, &pool->stats.pages_compacted); |
312fcae2 | 2340 | |
23959281 | 2341 | return pages_freed; |
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 | ||
ab9d306d SS |
2358 | /* |
2359 | * Compact classes and calculate compaction delta. | |
2360 | * Can run concurrently with a manually triggered | |
2361 | * (by user) compaction. | |
2362 | */ | |
23959281 | 2363 | pages_freed = zs_compact(pool); |
ab9d306d SS |
2364 | |
2365 | return pages_freed ? pages_freed : SHRINK_STOP; | |
2366 | } | |
2367 | ||
2368 | static unsigned long zs_shrinker_count(struct shrinker *shrinker, | |
2369 | struct shrink_control *sc) | |
2370 | { | |
2371 | int i; | |
2372 | struct size_class *class; | |
2373 | unsigned long pages_to_free = 0; | |
2374 | struct zs_pool *pool = container_of(shrinker, struct zs_pool, | |
2375 | shrinker); | |
2376 | ||
cf8e0fed | 2377 | for (i = ZS_SIZE_CLASSES - 1; i >= 0; i--) { |
ab9d306d | 2378 | class = pool->size_class[i]; |
ab9d306d SS |
2379 | if (class->index != i) |
2380 | continue; | |
2381 | ||
ab9d306d | 2382 | pages_to_free += zs_can_compact(class); |
ab9d306d SS |
2383 | } |
2384 | ||
2385 | return pages_to_free; | |
2386 | } | |
2387 | ||
2388 | static void zs_unregister_shrinker(struct zs_pool *pool) | |
2389 | { | |
93144ca3 | 2390 | unregister_shrinker(&pool->shrinker); |
ab9d306d SS |
2391 | } |
2392 | ||
2393 | static int zs_register_shrinker(struct zs_pool *pool) | |
2394 | { | |
2395 | pool->shrinker.scan_objects = zs_shrinker_scan; | |
2396 | pool->shrinker.count_objects = zs_shrinker_count; | |
2397 | pool->shrinker.batch = 0; | |
2398 | pool->shrinker.seeks = DEFAULT_SEEKS; | |
2399 | ||
e33c267a RG |
2400 | return register_shrinker(&pool->shrinker, "mm-zspool:%s", |
2401 | pool->name); | |
ab9d306d SS |
2402 | } |
2403 | ||
00a61d86 | 2404 | /** |
66cdef66 | 2405 | * zs_create_pool - Creates an allocation pool to work from. |
fd854463 | 2406 | * @name: pool name to be created |
166cfda7 | 2407 | * |
66cdef66 GM |
2408 | * This function must be called before anything when using |
2409 | * the zsmalloc allocator. | |
166cfda7 | 2410 | * |
66cdef66 GM |
2411 | * On success, a pointer to the newly created pool is returned, |
2412 | * otherwise NULL. | |
396b7fd6 | 2413 | */ |
d0d8da2d | 2414 | struct zs_pool *zs_create_pool(const char *name) |
61989a80 | 2415 | { |
66cdef66 GM |
2416 | int i; |
2417 | struct zs_pool *pool; | |
2418 | struct size_class *prev_class = NULL; | |
61989a80 | 2419 | |
66cdef66 GM |
2420 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
2421 | if (!pool) | |
2422 | return NULL; | |
61989a80 | 2423 | |
48b4800a | 2424 | init_deferred_free(pool); |
c0547d0b | 2425 | spin_lock_init(&pool->lock); |
61989a80 | 2426 | |
2e40e163 MK |
2427 | pool->name = kstrdup(name, GFP_KERNEL); |
2428 | if (!pool->name) | |
2429 | goto err; | |
2430 | ||
3783689a | 2431 | if (create_cache(pool)) |
2e40e163 MK |
2432 | goto err; |
2433 | ||
c60369f0 | 2434 | /* |
399d8eeb | 2435 | * Iterate reversely, because, size of size_class that we want to use |
66cdef66 | 2436 | * for merging should be larger or equal to current size. |
c60369f0 | 2437 | */ |
cf8e0fed | 2438 | for (i = ZS_SIZE_CLASSES - 1; i >= 0; i--) { |
66cdef66 GM |
2439 | int size; |
2440 | int pages_per_zspage; | |
64d90465 | 2441 | int objs_per_zspage; |
66cdef66 | 2442 | struct size_class *class; |
3783689a | 2443 | int fullness = 0; |
c60369f0 | 2444 | |
66cdef66 GM |
2445 | size = ZS_MIN_ALLOC_SIZE + i * ZS_SIZE_CLASS_DELTA; |
2446 | if (size > ZS_MAX_ALLOC_SIZE) | |
2447 | size = ZS_MAX_ALLOC_SIZE; | |
2448 | pages_per_zspage = get_pages_per_zspage(size); | |
64d90465 | 2449 | objs_per_zspage = pages_per_zspage * PAGE_SIZE / size; |
61989a80 | 2450 | |
010b495e SS |
2451 | /* |
2452 | * We iterate from biggest down to smallest classes, | |
2453 | * so huge_class_size holds the size of the first huge | |
2454 | * class. Any object bigger than or equal to that will | |
2455 | * endup in the huge class. | |
2456 | */ | |
2457 | if (pages_per_zspage != 1 && objs_per_zspage != 1 && | |
2458 | !huge_class_size) { | |
2459 | huge_class_size = size; | |
2460 | /* | |
2461 | * The object uses ZS_HANDLE_SIZE bytes to store the | |
2462 | * handle. We need to subtract it, because zs_malloc() | |
2463 | * unconditionally adds handle size before it performs | |
2464 | * size class search - so object may be smaller than | |
2465 | * huge class size, yet it still can end up in the huge | |
2466 | * class because it grows by ZS_HANDLE_SIZE extra bytes | |
2467 | * right before class lookup. | |
2468 | */ | |
2469 | huge_class_size -= (ZS_HANDLE_SIZE - 1); | |
2470 | } | |
2471 | ||
66cdef66 GM |
2472 | /* |
2473 | * size_class is used for normal zsmalloc operation such | |
2474 | * as alloc/free for that size. Although it is natural that we | |
2475 | * have one size_class for each size, there is a chance that we | |
2476 | * can get more memory utilization if we use one size_class for | |
2477 | * many different sizes whose size_class have same | |
2478 | * characteristics. So, we makes size_class point to | |
2479 | * previous size_class if possible. | |
2480 | */ | |
2481 | if (prev_class) { | |
64d90465 | 2482 | if (can_merge(prev_class, pages_per_zspage, objs_per_zspage)) { |
66cdef66 GM |
2483 | pool->size_class[i] = prev_class; |
2484 | continue; | |
2485 | } | |
2486 | } | |
2487 | ||
2488 | class = kzalloc(sizeof(struct size_class), GFP_KERNEL); | |
2489 | if (!class) | |
2490 | goto err; | |
2491 | ||
2492 | class->size = size; | |
2493 | class->index = i; | |
2494 | class->pages_per_zspage = pages_per_zspage; | |
64d90465 | 2495 | class->objs_per_zspage = objs_per_zspage; |
66cdef66 | 2496 | pool->size_class[i] = class; |
48b4800a MK |
2497 | for (fullness = ZS_EMPTY; fullness < NR_ZS_FULLNESS; |
2498 | fullness++) | |
3783689a | 2499 | INIT_LIST_HEAD(&class->fullness_list[fullness]); |
66cdef66 GM |
2500 | |
2501 | prev_class = class; | |
61989a80 NG |
2502 | } |
2503 | ||
d34f6157 DS |
2504 | /* debug only, don't abort if it fails */ |
2505 | zs_pool_stat_create(pool, name); | |
0f050d99 | 2506 | |
ab9d306d | 2507 | /* |
93144ca3 AK |
2508 | * Not critical since shrinker is only used to trigger internal |
2509 | * defragmentation of the pool which is pretty optional thing. If | |
2510 | * registration fails we still can use the pool normally and user can | |
2511 | * trigger compaction manually. Thus, ignore return code. | |
ab9d306d | 2512 | */ |
93144ca3 AK |
2513 | zs_register_shrinker(pool); |
2514 | ||
64f768c6 NP |
2515 | #ifdef CONFIG_ZPOOL |
2516 | INIT_LIST_HEAD(&pool->lru); | |
2517 | #endif | |
2518 | ||
66cdef66 GM |
2519 | return pool; |
2520 | ||
2521 | err: | |
2522 | zs_destroy_pool(pool); | |
2523 | return NULL; | |
61989a80 | 2524 | } |
66cdef66 | 2525 | EXPORT_SYMBOL_GPL(zs_create_pool); |
61989a80 | 2526 | |
66cdef66 | 2527 | void zs_destroy_pool(struct zs_pool *pool) |
61989a80 | 2528 | { |
66cdef66 | 2529 | int i; |
61989a80 | 2530 | |
ab9d306d | 2531 | zs_unregister_shrinker(pool); |
68f2736a | 2532 | zs_flush_migration(pool); |
0f050d99 GM |
2533 | zs_pool_stat_destroy(pool); |
2534 | ||
cf8e0fed | 2535 | for (i = 0; i < ZS_SIZE_CLASSES; i++) { |
66cdef66 GM |
2536 | int fg; |
2537 | struct size_class *class = pool->size_class[i]; | |
61989a80 | 2538 | |
4249a05f AR |
2539 | if (!class) |
2540 | continue; | |
2541 | ||
66cdef66 GM |
2542 | if (class->index != i) |
2543 | continue; | |
61989a80 | 2544 | |
48b4800a | 2545 | for (fg = ZS_EMPTY; fg < NR_ZS_FULLNESS; fg++) { |
3783689a | 2546 | if (!list_empty(&class->fullness_list[fg])) { |
66cdef66 GM |
2547 | pr_info("Freeing non-empty class with size %db, fullness group %d\n", |
2548 | class->size, fg); | |
2549 | } | |
2550 | } | |
2551 | kfree(class); | |
2552 | } | |
f553646a | 2553 | |
3783689a | 2554 | destroy_cache(pool); |
0f050d99 | 2555 | kfree(pool->name); |
66cdef66 GM |
2556 | kfree(pool); |
2557 | } | |
2558 | EXPORT_SYMBOL_GPL(zs_destroy_pool); | |
b7418510 | 2559 | |
9997bc01 | 2560 | #ifdef CONFIG_ZPOOL |
85b32581 NP |
2561 | static void restore_freelist(struct zs_pool *pool, struct size_class *class, |
2562 | struct zspage *zspage) | |
2563 | { | |
2564 | unsigned int obj_idx = 0; | |
2565 | unsigned long handle, off = 0; /* off is within-page offset */ | |
2566 | struct page *page = get_first_page(zspage); | |
2567 | struct link_free *prev_free = NULL; | |
2568 | void *prev_page_vaddr = NULL; | |
2569 | ||
2570 | /* in case no free object found */ | |
2571 | set_freeobj(zspage, (unsigned int)(-1UL)); | |
2572 | ||
2573 | while (page) { | |
2574 | void *vaddr = kmap_atomic(page); | |
2575 | struct page *next_page; | |
2576 | ||
2577 | while (off < PAGE_SIZE) { | |
2578 | void *obj_addr = vaddr + off; | |
2579 | ||
2580 | /* skip allocated object */ | |
2581 | if (obj_allocated(page, obj_addr, &handle)) { | |
2582 | obj_idx++; | |
2583 | off += class->size; | |
2584 | continue; | |
2585 | } | |
2586 | ||
2587 | /* free deferred handle from reclaim attempt */ | |
2588 | if (obj_stores_deferred_handle(page, obj_addr, &handle)) | |
2589 | cache_free_handle(pool, handle); | |
2590 | ||
2591 | if (prev_free) | |
2592 | prev_free->next = obj_idx << OBJ_TAG_BITS; | |
2593 | else /* first free object found */ | |
2594 | set_freeobj(zspage, obj_idx); | |
2595 | ||
2596 | prev_free = (struct link_free *)vaddr + off / sizeof(*prev_free); | |
2597 | /* if last free object in a previous page, need to unmap */ | |
2598 | if (prev_page_vaddr) { | |
2599 | kunmap_atomic(prev_page_vaddr); | |
2600 | prev_page_vaddr = NULL; | |
2601 | } | |
2602 | ||
2603 | obj_idx++; | |
2604 | off += class->size; | |
2605 | } | |
2606 | ||
2607 | /* | |
2608 | * Handle the last (full or partial) object on this page. | |
2609 | */ | |
2610 | next_page = get_next_page(page); | |
2611 | if (next_page) { | |
2612 | if (!prev_free || prev_page_vaddr) { | |
2613 | /* | |
2614 | * There is no free object in this page, so we can safely | |
2615 | * unmap it. | |
2616 | */ | |
2617 | kunmap_atomic(vaddr); | |
2618 | } else { | |
2619 | /* update prev_page_vaddr since prev_free is on this page */ | |
2620 | prev_page_vaddr = vaddr; | |
2621 | } | |
2622 | } else { /* this is the last page */ | |
2623 | if (prev_free) { | |
2624 | /* | |
2625 | * Reset OBJ_TAG_BITS bit to last link to tell | |
2626 | * whether it's allocated object or not. | |
2627 | */ | |
2628 | prev_free->next = -1UL << OBJ_TAG_BITS; | |
2629 | } | |
2630 | ||
2631 | /* unmap previous page (if not done yet) */ | |
2632 | if (prev_page_vaddr) { | |
2633 | kunmap_atomic(prev_page_vaddr); | |
2634 | prev_page_vaddr = NULL; | |
2635 | } | |
2636 | ||
2637 | kunmap_atomic(vaddr); | |
2638 | } | |
2639 | ||
2640 | page = next_page; | |
2641 | off %= PAGE_SIZE; | |
2642 | } | |
2643 | } | |
2644 | ||
9997bc01 NP |
2645 | static int zs_reclaim_page(struct zs_pool *pool, unsigned int retries) |
2646 | { | |
2647 | int i, obj_idx, ret = 0; | |
2648 | unsigned long handle; | |
2649 | struct zspage *zspage; | |
2650 | struct page *page; | |
2651 | enum fullness_group fullness; | |
2652 | ||
2653 | /* Lock LRU and fullness list */ | |
2654 | spin_lock(&pool->lock); | |
2655 | if (list_empty(&pool->lru)) { | |
2656 | spin_unlock(&pool->lock); | |
2657 | return -EINVAL; | |
2658 | } | |
2659 | ||
2660 | for (i = 0; i < retries; i++) { | |
2661 | struct size_class *class; | |
2662 | ||
2663 | zspage = list_last_entry(&pool->lru, struct zspage, lru); | |
2664 | list_del(&zspage->lru); | |
2665 | ||
2666 | /* zs_free may free objects, but not the zspage and handles */ | |
2667 | zspage->under_reclaim = true; | |
2668 | ||
2669 | class = zspage_class(pool, zspage); | |
2670 | fullness = get_fullness_group(class, zspage); | |
2671 | ||
2672 | /* Lock out object allocations and object compaction */ | |
2673 | remove_zspage(class, zspage, fullness); | |
2674 | ||
2675 | spin_unlock(&pool->lock); | |
2676 | cond_resched(); | |
2677 | ||
2678 | /* Lock backing pages into place */ | |
2679 | lock_zspage(zspage); | |
2680 | ||
2681 | obj_idx = 0; | |
2682 | page = get_first_page(zspage); | |
2683 | while (1) { | |
2684 | handle = find_alloced_obj(class, page, &obj_idx); | |
2685 | if (!handle) { | |
2686 | page = get_next_page(page); | |
2687 | if (!page) | |
2688 | break; | |
2689 | obj_idx = 0; | |
2690 | continue; | |
2691 | } | |
2692 | ||
2693 | /* | |
2694 | * This will write the object and call zs_free. | |
2695 | * | |
2696 | * zs_free will free the object, but the | |
2697 | * under_reclaim flag prevents it from freeing | |
2698 | * the zspage altogether. This is necessary so | |
2699 | * that we can continue working with the | |
2700 | * zspage potentially after the last object | |
2701 | * has been freed. | |
2702 | */ | |
2703 | ret = pool->zpool_ops->evict(pool->zpool, handle); | |
2704 | if (ret) | |
2705 | goto next; | |
2706 | ||
2707 | obj_idx++; | |
2708 | } | |
2709 | ||
2710 | next: | |
2711 | /* For freeing the zspage, or putting it back in the pool and LRU list. */ | |
2712 | spin_lock(&pool->lock); | |
2713 | zspage->under_reclaim = false; | |
2714 | ||
2715 | if (!get_zspage_inuse(zspage)) { | |
2716 | /* | |
2717 | * Fullness went stale as zs_free() won't touch it | |
2718 | * while the page is removed from the pool. Fix it | |
2719 | * up for the check in __free_zspage(). | |
2720 | */ | |
2721 | zspage->fullness = ZS_EMPTY; | |
2722 | ||
2723 | __free_zspage(pool, class, zspage); | |
2724 | spin_unlock(&pool->lock); | |
2725 | return 0; | |
2726 | } | |
2727 | ||
85b32581 NP |
2728 | /* |
2729 | * Eviction fails on one of the handles, so we need to restore zspage. | |
2730 | * We need to rebuild its freelist (and free stored deferred handles), | |
2731 | * put it back to the correct size class, and add it to the LRU list. | |
2732 | */ | |
2733 | restore_freelist(pool, class, zspage); | |
9997bc01 NP |
2734 | putback_zspage(class, zspage); |
2735 | list_add(&zspage->lru, &pool->lru); | |
2736 | unlock_zspage(zspage); | |
2737 | } | |
2738 | ||
2739 | spin_unlock(&pool->lock); | |
2740 | return -EAGAIN; | |
2741 | } | |
2742 | #endif /* CONFIG_ZPOOL */ | |
2743 | ||
66cdef66 GM |
2744 | static int __init zs_init(void) |
2745 | { | |
48b4800a MK |
2746 | int ret; |
2747 | ||
215c89d0 SAS |
2748 | ret = cpuhp_setup_state(CPUHP_MM_ZS_PREPARE, "mm/zsmalloc:prepare", |
2749 | zs_cpu_prepare, zs_cpu_dead); | |
0f050d99 | 2750 | if (ret) |
68f2736a | 2751 | goto out; |
66cdef66 | 2752 | |
66cdef66 GM |
2753 | #ifdef CONFIG_ZPOOL |
2754 | zpool_register_driver(&zs_zpool_driver); | |
2755 | #endif | |
0f050d99 | 2756 | |
4abaac9b DS |
2757 | zs_stat_init(); |
2758 | ||
66cdef66 | 2759 | return 0; |
0f050d99 | 2760 | |
48b4800a | 2761 | out: |
0f050d99 | 2762 | return ret; |
61989a80 | 2763 | } |
61989a80 | 2764 | |
66cdef66 | 2765 | static void __exit zs_exit(void) |
61989a80 | 2766 | { |
66cdef66 GM |
2767 | #ifdef CONFIG_ZPOOL |
2768 | zpool_unregister_driver(&zs_zpool_driver); | |
2769 | #endif | |
215c89d0 | 2770 | cpuhp_remove_state(CPUHP_MM_ZS_PREPARE); |
0f050d99 GM |
2771 | |
2772 | zs_stat_exit(); | |
61989a80 | 2773 | } |
069f101f BH |
2774 | |
2775 | module_init(zs_init); | |
2776 | module_exit(zs_exit); | |
2777 | ||
2778 | MODULE_LICENSE("Dual BSD/GPL"); | |
2779 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); |