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 | /* |
c3e3e88a NC |
15 | * This allocator is designed for use with zram. Thus, the allocator is |
16 | * supposed to work well under low memory conditions. In particular, it | |
17 | * never attempts higher order page allocation which is very likely to | |
18 | * fail under memory pressure. On the other hand, if we just use single | |
19 | * (0-order) pages, it would suffer from very high fragmentation -- | |
20 | * any object of size PAGE_SIZE/2 or larger would occupy an entire page. | |
21 | * This was one of the major issues with its predecessor (xvmalloc). | |
2db51dae NG |
22 | * |
23 | * To overcome these issues, zsmalloc allocates a bunch of 0-order pages | |
24 | * and links them together using various 'struct page' fields. These linked | |
25 | * pages act as a single higher-order page i.e. an object can span 0-order | |
26 | * page boundaries. The code refers to these linked pages as a single entity | |
27 | * called zspage. | |
28 | * | |
c3e3e88a NC |
29 | * For simplicity, zsmalloc can only allocate objects of size up to PAGE_SIZE |
30 | * since this satisfies the requirements of all its current users (in the | |
31 | * worst case, page is incompressible and is thus stored "as-is" i.e. in | |
32 | * uncompressed form). For allocation requests larger than this size, failure | |
33 | * is returned (see zs_malloc). | |
34 | * | |
35 | * Additionally, zs_malloc() does not return a dereferenceable pointer. | |
36 | * Instead, it returns an opaque handle (unsigned long) which encodes actual | |
37 | * location of the allocated object. The reason for this indirection is that | |
38 | * zsmalloc does not keep zspages permanently mapped since that would cause | |
39 | * issues on 32-bit systems where the VA region for kernel space mappings | |
40 | * is very small. So, before using the allocating memory, the object has to | |
41 | * be mapped using zs_map_object() to get a usable pointer and subsequently | |
42 | * unmapped using zs_unmap_object(). | |
43 | * | |
2db51dae NG |
44 | * Following is how we use various fields and flags of underlying |
45 | * struct page(s) to form a zspage. | |
46 | * | |
47 | * Usage of struct page fields: | |
48 | * page->first_page: points to the first component (0-order) page | |
49 | * page->index (union with page->freelist): offset of the first object | |
50 | * starting in this page. For the first page, this is | |
51 | * always 0, so we use this field (aka freelist) to point | |
52 | * to the first free object in zspage. | |
53 | * page->lru: links together all component pages (except the first page) | |
54 | * of a zspage | |
55 | * | |
56 | * For _first_ page only: | |
57 | * | |
58 | * page->private (union with page->first_page): refers to the | |
59 | * component page after the first page | |
60 | * page->freelist: points to the first free object in zspage. | |
61 | * Free objects are linked together using in-place | |
62 | * metadata. | |
63 | * page->objects: maximum number of objects we can store in this | |
64 | * zspage (class->zspage_order * PAGE_SIZE / class->size) | |
65 | * page->lru: links together first pages of various zspages. | |
66 | * Basically forming list of zspages in a fullness group. | |
67 | * page->mapping: class index and fullness group of the zspage | |
68 | * | |
69 | * Usage of struct page flags: | |
70 | * PG_private: identifies the first component page | |
71 | * PG_private2: identifies the last component page | |
72 | * | |
73 | */ | |
74 | ||
61989a80 NG |
75 | #ifdef CONFIG_ZSMALLOC_DEBUG |
76 | #define DEBUG | |
77 | #endif | |
78 | ||
79 | #include <linux/module.h> | |
80 | #include <linux/kernel.h> | |
81 | #include <linux/bitops.h> | |
82 | #include <linux/errno.h> | |
83 | #include <linux/highmem.h> | |
61989a80 NG |
84 | #include <linux/string.h> |
85 | #include <linux/slab.h> | |
86 | #include <asm/tlbflush.h> | |
87 | #include <asm/pgtable.h> | |
88 | #include <linux/cpumask.h> | |
89 | #include <linux/cpu.h> | |
0cbb613f | 90 | #include <linux/vmalloc.h> |
c60369f0 | 91 | #include <linux/hardirq.h> |
0959c63f SJ |
92 | #include <linux/spinlock.h> |
93 | #include <linux/types.h> | |
bcf1647d | 94 | #include <linux/zsmalloc.h> |
c795779d | 95 | #include <linux/zpool.h> |
0959c63f SJ |
96 | |
97 | /* | |
98 | * This must be power of 2 and greater than of equal to sizeof(link_free). | |
99 | * These two conditions ensure that any 'struct link_free' itself doesn't | |
100 | * span more than 1 page which avoids complex case of mapping 2 pages simply | |
101 | * to restore link_free pointer values. | |
102 | */ | |
103 | #define ZS_ALIGN 8 | |
104 | ||
105 | /* | |
106 | * A single 'zspage' is composed of up to 2^N discontiguous 0-order (single) | |
107 | * pages. ZS_MAX_ZSPAGE_ORDER defines upper limit on N. | |
108 | */ | |
109 | #define ZS_MAX_ZSPAGE_ORDER 2 | |
110 | #define ZS_MAX_PAGES_PER_ZSPAGE (_AC(1, UL) << ZS_MAX_ZSPAGE_ORDER) | |
111 | ||
112 | /* | |
113 | * Object location (<PFN>, <obj_idx>) is encoded as | |
c3e3e88a | 114 | * as single (unsigned long) handle value. |
0959c63f SJ |
115 | * |
116 | * Note that object index <obj_idx> is relative to system | |
117 | * page <PFN> it is stored in, so for each sub-page belonging | |
118 | * to a zspage, obj_idx starts with 0. | |
119 | * | |
120 | * This is made more complicated by various memory models and PAE. | |
121 | */ | |
122 | ||
123 | #ifndef MAX_PHYSMEM_BITS | |
124 | #ifdef CONFIG_HIGHMEM64G | |
125 | #define MAX_PHYSMEM_BITS 36 | |
126 | #else /* !CONFIG_HIGHMEM64G */ | |
127 | /* | |
128 | * If this definition of MAX_PHYSMEM_BITS is used, OBJ_INDEX_BITS will just | |
129 | * be PAGE_SHIFT | |
130 | */ | |
131 | #define MAX_PHYSMEM_BITS BITS_PER_LONG | |
132 | #endif | |
133 | #endif | |
134 | #define _PFN_BITS (MAX_PHYSMEM_BITS - PAGE_SHIFT) | |
135 | #define OBJ_INDEX_BITS (BITS_PER_LONG - _PFN_BITS) | |
136 | #define OBJ_INDEX_MASK ((_AC(1, UL) << OBJ_INDEX_BITS) - 1) | |
137 | ||
138 | #define MAX(a, b) ((a) >= (b) ? (a) : (b)) | |
139 | /* ZS_MIN_ALLOC_SIZE must be multiple of ZS_ALIGN */ | |
140 | #define ZS_MIN_ALLOC_SIZE \ | |
141 | MAX(32, (ZS_MAX_PAGES_PER_ZSPAGE << PAGE_SHIFT >> OBJ_INDEX_BITS)) | |
142 | #define ZS_MAX_ALLOC_SIZE PAGE_SIZE | |
143 | ||
144 | /* | |
7eb52512 | 145 | * On systems with 4K page size, this gives 255 size classes! There is a |
0959c63f SJ |
146 | * trader-off here: |
147 | * - Large number of size classes is potentially wasteful as free page are | |
148 | * spread across these classes | |
149 | * - Small number of size classes causes large internal fragmentation | |
150 | * - Probably its better to use specific size classes (empirically | |
151 | * determined). NOTE: all those class sizes must be set as multiple of | |
152 | * ZS_ALIGN to make sure link_free itself never has to span 2 pages. | |
153 | * | |
154 | * ZS_MIN_ALLOC_SIZE and ZS_SIZE_CLASS_DELTA must be multiple of ZS_ALIGN | |
155 | * (reason above) | |
156 | */ | |
d662b8eb | 157 | #define ZS_SIZE_CLASS_DELTA (PAGE_SIZE >> 8) |
0959c63f SJ |
158 | |
159 | /* | |
160 | * We do not maintain any list for completely empty or full pages | |
161 | */ | |
162 | enum fullness_group { | |
163 | ZS_ALMOST_FULL, | |
164 | ZS_ALMOST_EMPTY, | |
165 | _ZS_NR_FULLNESS_GROUPS, | |
166 | ||
167 | ZS_EMPTY, | |
168 | ZS_FULL | |
169 | }; | |
170 | ||
40f9fb8c MG |
171 | /* |
172 | * number of size_classes | |
173 | */ | |
174 | static int zs_size_classes; | |
175 | ||
0959c63f SJ |
176 | /* |
177 | * We assign a page to ZS_ALMOST_EMPTY fullness group when: | |
178 | * n <= N / f, where | |
179 | * n = number of allocated objects | |
180 | * N = total number of objects zspage can store | |
6dd9737e | 181 | * f = fullness_threshold_frac |
0959c63f SJ |
182 | * |
183 | * Similarly, we assign zspage to: | |
184 | * ZS_ALMOST_FULL when n > N / f | |
185 | * ZS_EMPTY when n == 0 | |
186 | * ZS_FULL when n == N | |
187 | * | |
188 | * (see: fix_fullness_group()) | |
189 | */ | |
190 | static const int fullness_threshold_frac = 4; | |
191 | ||
192 | struct size_class { | |
193 | /* | |
194 | * Size of objects stored in this class. Must be multiple | |
195 | * of ZS_ALIGN. | |
196 | */ | |
197 | int size; | |
198 | unsigned int index; | |
199 | ||
200 | /* Number of PAGE_SIZE sized pages to combine to form a 'zspage' */ | |
201 | int pages_per_zspage; | |
202 | ||
203 | spinlock_t lock; | |
204 | ||
0959c63f SJ |
205 | struct page *fullness_list[_ZS_NR_FULLNESS_GROUPS]; |
206 | }; | |
207 | ||
208 | /* | |
209 | * Placed within free objects to form a singly linked list. | |
210 | * For every zspage, first_page->freelist gives head of this list. | |
211 | * | |
212 | * This must be power of 2 and less than or equal to ZS_ALIGN | |
213 | */ | |
214 | struct link_free { | |
215 | /* Handle of next free chunk (encodes <PFN, obj_idx>) */ | |
216 | void *next; | |
217 | }; | |
218 | ||
219 | struct zs_pool { | |
40f9fb8c | 220 | struct size_class **size_class; |
0959c63f SJ |
221 | |
222 | gfp_t flags; /* allocation flags used when growing pool */ | |
13de8933 | 223 | atomic_long_t pages_allocated; |
0959c63f | 224 | }; |
61989a80 NG |
225 | |
226 | /* | |
227 | * A zspage's class index and fullness group | |
228 | * are encoded in its (first)page->mapping | |
229 | */ | |
230 | #define CLASS_IDX_BITS 28 | |
231 | #define FULLNESS_BITS 4 | |
232 | #define CLASS_IDX_MASK ((1 << CLASS_IDX_BITS) - 1) | |
233 | #define FULLNESS_MASK ((1 << FULLNESS_BITS) - 1) | |
234 | ||
f553646a | 235 | struct mapping_area { |
1b945aee | 236 | #ifdef CONFIG_PGTABLE_MAPPING |
f553646a SJ |
237 | struct vm_struct *vm; /* vm area for mapping object that span pages */ |
238 | #else | |
239 | char *vm_buf; /* copy buffer for objects that span pages */ | |
240 | #endif | |
241 | char *vm_addr; /* address of kmap_atomic()'ed pages */ | |
242 | enum zs_mapmode vm_mm; /* mapping mode */ | |
243 | }; | |
244 | ||
c795779d DS |
245 | /* zpool driver */ |
246 | ||
247 | #ifdef CONFIG_ZPOOL | |
248 | ||
249 | static void *zs_zpool_create(gfp_t gfp, struct zpool_ops *zpool_ops) | |
250 | { | |
251 | return zs_create_pool(gfp); | |
252 | } | |
253 | ||
254 | static void zs_zpool_destroy(void *pool) | |
255 | { | |
256 | zs_destroy_pool(pool); | |
257 | } | |
258 | ||
259 | static int zs_zpool_malloc(void *pool, size_t size, gfp_t gfp, | |
260 | unsigned long *handle) | |
261 | { | |
262 | *handle = zs_malloc(pool, size); | |
263 | return *handle ? 0 : -1; | |
264 | } | |
265 | static void zs_zpool_free(void *pool, unsigned long handle) | |
266 | { | |
267 | zs_free(pool, handle); | |
268 | } | |
269 | ||
270 | static int zs_zpool_shrink(void *pool, unsigned int pages, | |
271 | unsigned int *reclaimed) | |
272 | { | |
273 | return -EINVAL; | |
274 | } | |
275 | ||
276 | static void *zs_zpool_map(void *pool, unsigned long handle, | |
277 | enum zpool_mapmode mm) | |
278 | { | |
279 | enum zs_mapmode zs_mm; | |
280 | ||
281 | switch (mm) { | |
282 | case ZPOOL_MM_RO: | |
283 | zs_mm = ZS_MM_RO; | |
284 | break; | |
285 | case ZPOOL_MM_WO: | |
286 | zs_mm = ZS_MM_WO; | |
287 | break; | |
288 | case ZPOOL_MM_RW: /* fallthru */ | |
289 | default: | |
290 | zs_mm = ZS_MM_RW; | |
291 | break; | |
292 | } | |
293 | ||
294 | return zs_map_object(pool, handle, zs_mm); | |
295 | } | |
296 | static void zs_zpool_unmap(void *pool, unsigned long handle) | |
297 | { | |
298 | zs_unmap_object(pool, handle); | |
299 | } | |
300 | ||
301 | static u64 zs_zpool_total_size(void *pool) | |
302 | { | |
722cdc17 | 303 | return zs_get_total_pages(pool) << PAGE_SHIFT; |
c795779d DS |
304 | } |
305 | ||
306 | static struct zpool_driver zs_zpool_driver = { | |
307 | .type = "zsmalloc", | |
308 | .owner = THIS_MODULE, | |
309 | .create = zs_zpool_create, | |
310 | .destroy = zs_zpool_destroy, | |
311 | .malloc = zs_zpool_malloc, | |
312 | .free = zs_zpool_free, | |
313 | .shrink = zs_zpool_shrink, | |
314 | .map = zs_zpool_map, | |
315 | .unmap = zs_zpool_unmap, | |
316 | .total_size = zs_zpool_total_size, | |
317 | }; | |
318 | ||
137f8cff | 319 | MODULE_ALIAS("zpool-zsmalloc"); |
c795779d DS |
320 | #endif /* CONFIG_ZPOOL */ |
321 | ||
61989a80 NG |
322 | /* per-cpu VM mapping areas for zspage accesses that cross page boundaries */ |
323 | static DEFINE_PER_CPU(struct mapping_area, zs_map_area); | |
324 | ||
325 | static int is_first_page(struct page *page) | |
326 | { | |
a27545bf | 327 | return PagePrivate(page); |
61989a80 NG |
328 | } |
329 | ||
330 | static int is_last_page(struct page *page) | |
331 | { | |
a27545bf | 332 | return PagePrivate2(page); |
61989a80 NG |
333 | } |
334 | ||
335 | static void get_zspage_mapping(struct page *page, unsigned int *class_idx, | |
336 | enum fullness_group *fullness) | |
337 | { | |
338 | unsigned long m; | |
339 | BUG_ON(!is_first_page(page)); | |
340 | ||
341 | m = (unsigned long)page->mapping; | |
342 | *fullness = m & FULLNESS_MASK; | |
343 | *class_idx = (m >> FULLNESS_BITS) & CLASS_IDX_MASK; | |
344 | } | |
345 | ||
346 | static void set_zspage_mapping(struct page *page, unsigned int class_idx, | |
347 | enum fullness_group fullness) | |
348 | { | |
349 | unsigned long m; | |
350 | BUG_ON(!is_first_page(page)); | |
351 | ||
352 | m = ((class_idx & CLASS_IDX_MASK) << FULLNESS_BITS) | | |
353 | (fullness & FULLNESS_MASK); | |
354 | page->mapping = (struct address_space *)m; | |
355 | } | |
356 | ||
c3e3e88a NC |
357 | /* |
358 | * zsmalloc divides the pool into various size classes where each | |
359 | * class maintains a list of zspages where each zspage is divided | |
360 | * into equal sized chunks. Each allocation falls into one of these | |
361 | * classes depending on its size. This function returns index of the | |
362 | * size class which has chunk size big enough to hold the give size. | |
363 | */ | |
61989a80 NG |
364 | static int get_size_class_index(int size) |
365 | { | |
366 | int idx = 0; | |
367 | ||
368 | if (likely(size > ZS_MIN_ALLOC_SIZE)) | |
369 | idx = DIV_ROUND_UP(size - ZS_MIN_ALLOC_SIZE, | |
370 | ZS_SIZE_CLASS_DELTA); | |
371 | ||
372 | return idx; | |
373 | } | |
374 | ||
c3e3e88a NC |
375 | /* |
376 | * For each size class, zspages are divided into different groups | |
377 | * depending on how "full" they are. This was done so that we could | |
378 | * easily find empty or nearly empty zspages when we try to shrink | |
379 | * the pool (not yet implemented). This function returns fullness | |
380 | * status of the given page. | |
381 | */ | |
61989a80 NG |
382 | static enum fullness_group get_fullness_group(struct page *page) |
383 | { | |
384 | int inuse, max_objects; | |
385 | enum fullness_group fg; | |
386 | BUG_ON(!is_first_page(page)); | |
387 | ||
388 | inuse = page->inuse; | |
389 | max_objects = page->objects; | |
390 | ||
391 | if (inuse == 0) | |
392 | fg = ZS_EMPTY; | |
393 | else if (inuse == max_objects) | |
394 | fg = ZS_FULL; | |
395 | else if (inuse <= max_objects / fullness_threshold_frac) | |
396 | fg = ZS_ALMOST_EMPTY; | |
397 | else | |
398 | fg = ZS_ALMOST_FULL; | |
399 | ||
400 | return fg; | |
401 | } | |
402 | ||
c3e3e88a NC |
403 | /* |
404 | * Each size class maintains various freelists and zspages are assigned | |
405 | * to one of these freelists based on the number of live objects they | |
406 | * have. This functions inserts the given zspage into the freelist | |
407 | * identified by <class, fullness_group>. | |
408 | */ | |
61989a80 NG |
409 | static void insert_zspage(struct page *page, struct size_class *class, |
410 | enum fullness_group fullness) | |
411 | { | |
412 | struct page **head; | |
413 | ||
414 | BUG_ON(!is_first_page(page)); | |
415 | ||
416 | if (fullness >= _ZS_NR_FULLNESS_GROUPS) | |
417 | return; | |
418 | ||
419 | head = &class->fullness_list[fullness]; | |
420 | if (*head) | |
421 | list_add_tail(&page->lru, &(*head)->lru); | |
422 | ||
423 | *head = page; | |
424 | } | |
425 | ||
c3e3e88a NC |
426 | /* |
427 | * This function removes the given zspage from the freelist identified | |
428 | * by <class, fullness_group>. | |
429 | */ | |
61989a80 NG |
430 | static void remove_zspage(struct page *page, struct size_class *class, |
431 | enum fullness_group fullness) | |
432 | { | |
433 | struct page **head; | |
434 | ||
435 | BUG_ON(!is_first_page(page)); | |
436 | ||
437 | if (fullness >= _ZS_NR_FULLNESS_GROUPS) | |
438 | return; | |
439 | ||
440 | head = &class->fullness_list[fullness]; | |
441 | BUG_ON(!*head); | |
442 | if (list_empty(&(*head)->lru)) | |
443 | *head = NULL; | |
444 | else if (*head == page) | |
445 | *head = (struct page *)list_entry((*head)->lru.next, | |
446 | struct page, lru); | |
447 | ||
448 | list_del_init(&page->lru); | |
449 | } | |
450 | ||
c3e3e88a NC |
451 | /* |
452 | * Each size class maintains zspages in different fullness groups depending | |
453 | * on the number of live objects they contain. When allocating or freeing | |
454 | * objects, the fullness status of the page can change, say, from ALMOST_FULL | |
455 | * to ALMOST_EMPTY when freeing an object. This function checks if such | |
456 | * a status change has occurred for the given page and accordingly moves the | |
457 | * page from the freelist of the old fullness group to that of the new | |
458 | * fullness group. | |
459 | */ | |
61989a80 NG |
460 | static enum fullness_group fix_fullness_group(struct zs_pool *pool, |
461 | struct page *page) | |
462 | { | |
463 | int class_idx; | |
464 | struct size_class *class; | |
465 | enum fullness_group currfg, newfg; | |
466 | ||
467 | BUG_ON(!is_first_page(page)); | |
468 | ||
469 | get_zspage_mapping(page, &class_idx, &currfg); | |
470 | newfg = get_fullness_group(page); | |
471 | if (newfg == currfg) | |
472 | goto out; | |
473 | ||
9eec4cd5 | 474 | class = pool->size_class[class_idx]; |
61989a80 NG |
475 | remove_zspage(page, class, currfg); |
476 | insert_zspage(page, class, newfg); | |
477 | set_zspage_mapping(page, class_idx, newfg); | |
478 | ||
479 | out: | |
480 | return newfg; | |
481 | } | |
482 | ||
483 | /* | |
484 | * We have to decide on how many pages to link together | |
485 | * to form a zspage for each size class. This is important | |
486 | * to reduce wastage due to unusable space left at end of | |
487 | * each zspage which is given as: | |
488 | * wastage = Zp - Zp % size_class | |
489 | * where Zp = zspage size = k * PAGE_SIZE where k = 1, 2, ... | |
490 | * | |
491 | * For example, for size class of 3/8 * PAGE_SIZE, we should | |
492 | * link together 3 PAGE_SIZE sized pages to form a zspage | |
493 | * since then we can perfectly fit in 8 such objects. | |
494 | */ | |
2e3b6154 | 495 | static int get_pages_per_zspage(int class_size) |
61989a80 NG |
496 | { |
497 | int i, max_usedpc = 0; | |
498 | /* zspage order which gives maximum used size per KB */ | |
499 | int max_usedpc_order = 1; | |
500 | ||
84d4faab | 501 | for (i = 1; i <= ZS_MAX_PAGES_PER_ZSPAGE; i++) { |
61989a80 NG |
502 | int zspage_size; |
503 | int waste, usedpc; | |
504 | ||
505 | zspage_size = i * PAGE_SIZE; | |
506 | waste = zspage_size % class_size; | |
507 | usedpc = (zspage_size - waste) * 100 / zspage_size; | |
508 | ||
509 | if (usedpc > max_usedpc) { | |
510 | max_usedpc = usedpc; | |
511 | max_usedpc_order = i; | |
512 | } | |
513 | } | |
514 | ||
515 | return max_usedpc_order; | |
516 | } | |
517 | ||
518 | /* | |
519 | * A single 'zspage' is composed of many system pages which are | |
520 | * linked together using fields in struct page. This function finds | |
521 | * the first/head page, given any component page of a zspage. | |
522 | */ | |
523 | static struct page *get_first_page(struct page *page) | |
524 | { | |
525 | if (is_first_page(page)) | |
526 | return page; | |
527 | else | |
528 | return page->first_page; | |
529 | } | |
530 | ||
531 | static struct page *get_next_page(struct page *page) | |
532 | { | |
533 | struct page *next; | |
534 | ||
535 | if (is_last_page(page)) | |
536 | next = NULL; | |
537 | else if (is_first_page(page)) | |
e842b976 | 538 | next = (struct page *)page_private(page); |
61989a80 NG |
539 | else |
540 | next = list_entry(page->lru.next, struct page, lru); | |
541 | ||
542 | return next; | |
543 | } | |
544 | ||
67296874 OH |
545 | /* |
546 | * Encode <page, obj_idx> as a single handle value. | |
547 | * On hardware platforms with physical memory starting at 0x0 the pfn | |
548 | * could be 0 so we ensure that the handle will never be 0 by adjusting the | |
549 | * encoded obj_idx value before encoding. | |
550 | */ | |
61989a80 NG |
551 | static void *obj_location_to_handle(struct page *page, unsigned long obj_idx) |
552 | { | |
553 | unsigned long handle; | |
554 | ||
555 | if (!page) { | |
556 | BUG_ON(obj_idx); | |
557 | return NULL; | |
558 | } | |
559 | ||
560 | handle = page_to_pfn(page) << OBJ_INDEX_BITS; | |
67296874 | 561 | handle |= ((obj_idx + 1) & OBJ_INDEX_MASK); |
61989a80 NG |
562 | |
563 | return (void *)handle; | |
564 | } | |
565 | ||
67296874 OH |
566 | /* |
567 | * Decode <page, obj_idx> pair from the given object handle. We adjust the | |
568 | * decoded obj_idx back to its original value since it was adjusted in | |
569 | * obj_location_to_handle(). | |
570 | */ | |
c2344348 | 571 | static void obj_handle_to_location(unsigned long handle, struct page **page, |
61989a80 NG |
572 | unsigned long *obj_idx) |
573 | { | |
c2344348 | 574 | *page = pfn_to_page(handle >> OBJ_INDEX_BITS); |
67296874 | 575 | *obj_idx = (handle & OBJ_INDEX_MASK) - 1; |
61989a80 NG |
576 | } |
577 | ||
578 | static unsigned long obj_idx_to_offset(struct page *page, | |
579 | unsigned long obj_idx, int class_size) | |
580 | { | |
581 | unsigned long off = 0; | |
582 | ||
583 | if (!is_first_page(page)) | |
584 | off = page->index; | |
585 | ||
586 | return off + obj_idx * class_size; | |
587 | } | |
588 | ||
f4477e90 NG |
589 | static void reset_page(struct page *page) |
590 | { | |
591 | clear_bit(PG_private, &page->flags); | |
592 | clear_bit(PG_private_2, &page->flags); | |
593 | set_page_private(page, 0); | |
594 | page->mapping = NULL; | |
595 | page->freelist = NULL; | |
22b751c3 | 596 | page_mapcount_reset(page); |
f4477e90 NG |
597 | } |
598 | ||
61989a80 NG |
599 | static void free_zspage(struct page *first_page) |
600 | { | |
f4477e90 | 601 | struct page *nextp, *tmp, *head_extra; |
61989a80 NG |
602 | |
603 | BUG_ON(!is_first_page(first_page)); | |
604 | BUG_ON(first_page->inuse); | |
605 | ||
f4477e90 | 606 | head_extra = (struct page *)page_private(first_page); |
61989a80 | 607 | |
f4477e90 | 608 | reset_page(first_page); |
61989a80 NG |
609 | __free_page(first_page); |
610 | ||
611 | /* zspage with only 1 system page */ | |
f4477e90 | 612 | if (!head_extra) |
61989a80 NG |
613 | return; |
614 | ||
f4477e90 | 615 | list_for_each_entry_safe(nextp, tmp, &head_extra->lru, lru) { |
61989a80 | 616 | list_del(&nextp->lru); |
f4477e90 | 617 | reset_page(nextp); |
61989a80 NG |
618 | __free_page(nextp); |
619 | } | |
f4477e90 NG |
620 | reset_page(head_extra); |
621 | __free_page(head_extra); | |
61989a80 NG |
622 | } |
623 | ||
624 | /* Initialize a newly allocated zspage */ | |
625 | static void init_zspage(struct page *first_page, struct size_class *class) | |
626 | { | |
627 | unsigned long off = 0; | |
628 | struct page *page = first_page; | |
629 | ||
630 | BUG_ON(!is_first_page(first_page)); | |
631 | while (page) { | |
632 | struct page *next_page; | |
633 | struct link_free *link; | |
5538c562 | 634 | unsigned int i = 1; |
af4ee5e9 | 635 | void *vaddr; |
61989a80 NG |
636 | |
637 | /* | |
638 | * page->index stores offset of first object starting | |
639 | * in the page. For the first page, this is always 0, | |
640 | * so we use first_page->index (aka ->freelist) to store | |
641 | * head of corresponding zspage's freelist. | |
642 | */ | |
643 | if (page != first_page) | |
644 | page->index = off; | |
645 | ||
af4ee5e9 MK |
646 | vaddr = kmap_atomic(page); |
647 | link = (struct link_free *)vaddr + off / sizeof(*link); | |
5538c562 DS |
648 | |
649 | while ((off += class->size) < PAGE_SIZE) { | |
650 | link->next = obj_location_to_handle(page, i++); | |
651 | link += class->size / sizeof(*link); | |
61989a80 NG |
652 | } |
653 | ||
654 | /* | |
655 | * We now come to the last (full or partial) object on this | |
656 | * page, which must point to the first object on the next | |
657 | * page (if present) | |
658 | */ | |
659 | next_page = get_next_page(page); | |
660 | link->next = obj_location_to_handle(next_page, 0); | |
af4ee5e9 | 661 | kunmap_atomic(vaddr); |
61989a80 | 662 | page = next_page; |
5538c562 | 663 | off %= PAGE_SIZE; |
61989a80 NG |
664 | } |
665 | } | |
666 | ||
667 | /* | |
668 | * Allocate a zspage for the given size class | |
669 | */ | |
670 | static struct page *alloc_zspage(struct size_class *class, gfp_t flags) | |
671 | { | |
672 | int i, error; | |
b4b700c5 | 673 | struct page *first_page = NULL, *uninitialized_var(prev_page); |
61989a80 NG |
674 | |
675 | /* | |
676 | * Allocate individual pages and link them together as: | |
677 | * 1. first page->private = first sub-page | |
678 | * 2. all sub-pages are linked together using page->lru | |
679 | * 3. each sub-page is linked to the first page using page->first_page | |
680 | * | |
681 | * For each size class, First/Head pages are linked together using | |
682 | * page->lru. Also, we set PG_private to identify the first page | |
683 | * (i.e. no other sub-page has this flag set) and PG_private_2 to | |
684 | * identify the last page. | |
685 | */ | |
686 | error = -ENOMEM; | |
2e3b6154 | 687 | for (i = 0; i < class->pages_per_zspage; i++) { |
b4b700c5 | 688 | struct page *page; |
61989a80 NG |
689 | |
690 | page = alloc_page(flags); | |
691 | if (!page) | |
692 | goto cleanup; | |
693 | ||
694 | INIT_LIST_HEAD(&page->lru); | |
695 | if (i == 0) { /* first page */ | |
a27545bf | 696 | SetPagePrivate(page); |
61989a80 NG |
697 | set_page_private(page, 0); |
698 | first_page = page; | |
699 | first_page->inuse = 0; | |
700 | } | |
701 | if (i == 1) | |
e842b976 | 702 | set_page_private(first_page, (unsigned long)page); |
61989a80 NG |
703 | if (i >= 1) |
704 | page->first_page = first_page; | |
705 | if (i >= 2) | |
706 | list_add(&page->lru, &prev_page->lru); | |
2e3b6154 | 707 | if (i == class->pages_per_zspage - 1) /* last page */ |
a27545bf | 708 | SetPagePrivate2(page); |
61989a80 NG |
709 | prev_page = page; |
710 | } | |
711 | ||
712 | init_zspage(first_page, class); | |
713 | ||
714 | first_page->freelist = obj_location_to_handle(first_page, 0); | |
715 | /* Maximum number of objects we can store in this zspage */ | |
2e3b6154 | 716 | first_page->objects = class->pages_per_zspage * PAGE_SIZE / class->size; |
61989a80 NG |
717 | |
718 | error = 0; /* Success */ | |
719 | ||
720 | cleanup: | |
721 | if (unlikely(error) && first_page) { | |
722 | free_zspage(first_page); | |
723 | first_page = NULL; | |
724 | } | |
725 | ||
726 | return first_page; | |
727 | } | |
728 | ||
729 | static struct page *find_get_zspage(struct size_class *class) | |
730 | { | |
731 | int i; | |
732 | struct page *page; | |
733 | ||
734 | for (i = 0; i < _ZS_NR_FULLNESS_GROUPS; i++) { | |
735 | page = class->fullness_list[i]; | |
736 | if (page) | |
737 | break; | |
738 | } | |
739 | ||
740 | return page; | |
741 | } | |
742 | ||
1b945aee | 743 | #ifdef CONFIG_PGTABLE_MAPPING |
f553646a SJ |
744 | static inline int __zs_cpu_up(struct mapping_area *area) |
745 | { | |
746 | /* | |
747 | * Make sure we don't leak memory if a cpu UP notification | |
748 | * and zs_init() race and both call zs_cpu_up() on the same cpu | |
749 | */ | |
750 | if (area->vm) | |
751 | return 0; | |
752 | area->vm = alloc_vm_area(PAGE_SIZE * 2, NULL); | |
753 | if (!area->vm) | |
754 | return -ENOMEM; | |
755 | return 0; | |
756 | } | |
757 | ||
758 | static inline void __zs_cpu_down(struct mapping_area *area) | |
759 | { | |
760 | if (area->vm) | |
761 | free_vm_area(area->vm); | |
762 | area->vm = NULL; | |
763 | } | |
764 | ||
765 | static inline void *__zs_map_object(struct mapping_area *area, | |
766 | struct page *pages[2], int off, int size) | |
767 | { | |
f6f8ed47 | 768 | BUG_ON(map_vm_area(area->vm, PAGE_KERNEL, pages)); |
f553646a SJ |
769 | area->vm_addr = area->vm->addr; |
770 | return area->vm_addr + off; | |
771 | } | |
772 | ||
773 | static inline void __zs_unmap_object(struct mapping_area *area, | |
774 | struct page *pages[2], int off, int size) | |
775 | { | |
776 | unsigned long addr = (unsigned long)area->vm_addr; | |
f553646a | 777 | |
d95abbbb | 778 | unmap_kernel_range(addr, PAGE_SIZE * 2); |
f553646a SJ |
779 | } |
780 | ||
1b945aee | 781 | #else /* CONFIG_PGTABLE_MAPPING */ |
f553646a SJ |
782 | |
783 | static inline int __zs_cpu_up(struct mapping_area *area) | |
784 | { | |
785 | /* | |
786 | * Make sure we don't leak memory if a cpu UP notification | |
787 | * and zs_init() race and both call zs_cpu_up() on the same cpu | |
788 | */ | |
789 | if (area->vm_buf) | |
790 | return 0; | |
40f9fb8c | 791 | area->vm_buf = kmalloc(ZS_MAX_ALLOC_SIZE, GFP_KERNEL); |
f553646a SJ |
792 | if (!area->vm_buf) |
793 | return -ENOMEM; | |
794 | return 0; | |
795 | } | |
796 | ||
797 | static inline void __zs_cpu_down(struct mapping_area *area) | |
798 | { | |
40f9fb8c | 799 | kfree(area->vm_buf); |
f553646a SJ |
800 | area->vm_buf = NULL; |
801 | } | |
802 | ||
803 | static void *__zs_map_object(struct mapping_area *area, | |
804 | struct page *pages[2], int off, int size) | |
5f601902 | 805 | { |
5f601902 SJ |
806 | int sizes[2]; |
807 | void *addr; | |
f553646a | 808 | char *buf = area->vm_buf; |
5f601902 | 809 | |
f553646a SJ |
810 | /* disable page faults to match kmap_atomic() return conditions */ |
811 | pagefault_disable(); | |
812 | ||
813 | /* no read fastpath */ | |
814 | if (area->vm_mm == ZS_MM_WO) | |
815 | goto out; | |
5f601902 SJ |
816 | |
817 | sizes[0] = PAGE_SIZE - off; | |
818 | sizes[1] = size - sizes[0]; | |
819 | ||
5f601902 SJ |
820 | /* copy object to per-cpu buffer */ |
821 | addr = kmap_atomic(pages[0]); | |
822 | memcpy(buf, addr + off, sizes[0]); | |
823 | kunmap_atomic(addr); | |
824 | addr = kmap_atomic(pages[1]); | |
825 | memcpy(buf + sizes[0], addr, sizes[1]); | |
826 | kunmap_atomic(addr); | |
f553646a SJ |
827 | out: |
828 | return area->vm_buf; | |
5f601902 SJ |
829 | } |
830 | ||
f553646a SJ |
831 | static void __zs_unmap_object(struct mapping_area *area, |
832 | struct page *pages[2], int off, int size) | |
5f601902 | 833 | { |
5f601902 SJ |
834 | int sizes[2]; |
835 | void *addr; | |
f553646a | 836 | char *buf = area->vm_buf; |
5f601902 | 837 | |
f553646a SJ |
838 | /* no write fastpath */ |
839 | if (area->vm_mm == ZS_MM_RO) | |
840 | goto out; | |
5f601902 SJ |
841 | |
842 | sizes[0] = PAGE_SIZE - off; | |
843 | sizes[1] = size - sizes[0]; | |
844 | ||
845 | /* copy per-cpu buffer to object */ | |
846 | addr = kmap_atomic(pages[0]); | |
847 | memcpy(addr + off, buf, sizes[0]); | |
848 | kunmap_atomic(addr); | |
849 | addr = kmap_atomic(pages[1]); | |
850 | memcpy(addr, buf + sizes[0], sizes[1]); | |
851 | kunmap_atomic(addr); | |
f553646a SJ |
852 | |
853 | out: | |
854 | /* enable page faults to match kunmap_atomic() return conditions */ | |
855 | pagefault_enable(); | |
5f601902 | 856 | } |
61989a80 | 857 | |
1b945aee | 858 | #endif /* CONFIG_PGTABLE_MAPPING */ |
f553646a | 859 | |
61989a80 NG |
860 | static int zs_cpu_notifier(struct notifier_block *nb, unsigned long action, |
861 | void *pcpu) | |
862 | { | |
f553646a | 863 | int ret, cpu = (long)pcpu; |
61989a80 NG |
864 | struct mapping_area *area; |
865 | ||
866 | switch (action) { | |
867 | case CPU_UP_PREPARE: | |
868 | area = &per_cpu(zs_map_area, cpu); | |
f553646a SJ |
869 | ret = __zs_cpu_up(area); |
870 | if (ret) | |
871 | return notifier_from_errno(ret); | |
61989a80 NG |
872 | break; |
873 | case CPU_DEAD: | |
874 | case CPU_UP_CANCELED: | |
875 | area = &per_cpu(zs_map_area, cpu); | |
f553646a | 876 | __zs_cpu_down(area); |
61989a80 NG |
877 | break; |
878 | } | |
879 | ||
880 | return NOTIFY_OK; | |
881 | } | |
882 | ||
883 | static struct notifier_block zs_cpu_nb = { | |
884 | .notifier_call = zs_cpu_notifier | |
885 | }; | |
886 | ||
b1b00a5b | 887 | static void zs_unregister_cpu_notifier(void) |
61989a80 NG |
888 | { |
889 | int cpu; | |
890 | ||
f0e71fcd SB |
891 | cpu_notifier_register_begin(); |
892 | ||
61989a80 NG |
893 | for_each_online_cpu(cpu) |
894 | zs_cpu_notifier(NULL, CPU_DEAD, (void *)(long)cpu); | |
f0e71fcd SB |
895 | __unregister_cpu_notifier(&zs_cpu_nb); |
896 | ||
897 | cpu_notifier_register_done(); | |
61989a80 NG |
898 | } |
899 | ||
b1b00a5b | 900 | static int zs_register_cpu_notifier(void) |
61989a80 | 901 | { |
b1b00a5b | 902 | int cpu, uninitialized_var(ret); |
61989a80 | 903 | |
f0e71fcd SB |
904 | cpu_notifier_register_begin(); |
905 | ||
906 | __register_cpu_notifier(&zs_cpu_nb); | |
61989a80 NG |
907 | for_each_online_cpu(cpu) { |
908 | ret = zs_cpu_notifier(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
b1b00a5b SS |
909 | if (notifier_to_errno(ret)) |
910 | break; | |
61989a80 | 911 | } |
f0e71fcd SB |
912 | |
913 | cpu_notifier_register_done(); | |
b1b00a5b SS |
914 | return notifier_to_errno(ret); |
915 | } | |
f0e71fcd | 916 | |
40f9fb8c MG |
917 | static void init_zs_size_classes(void) |
918 | { | |
919 | int nr; | |
920 | ||
921 | nr = (ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) / ZS_SIZE_CLASS_DELTA + 1; | |
922 | if ((ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) % ZS_SIZE_CLASS_DELTA) | |
923 | nr += 1; | |
924 | ||
925 | zs_size_classes = nr; | |
926 | } | |
927 | ||
b1b00a5b SS |
928 | static void __exit zs_exit(void) |
929 | { | |
c795779d | 930 | #ifdef CONFIG_ZPOOL |
b1b00a5b | 931 | zpool_unregister_driver(&zs_zpool_driver); |
c795779d | 932 | #endif |
b1b00a5b SS |
933 | zs_unregister_cpu_notifier(); |
934 | } | |
935 | ||
936 | static int __init zs_init(void) | |
937 | { | |
938 | int ret = zs_register_cpu_notifier(); | |
939 | ||
940 | if (ret) { | |
941 | zs_unregister_cpu_notifier(); | |
942 | return ret; | |
943 | } | |
c795779d | 944 | |
40f9fb8c MG |
945 | init_zs_size_classes(); |
946 | ||
b1b00a5b SS |
947 | #ifdef CONFIG_ZPOOL |
948 | zpool_register_driver(&zs_zpool_driver); | |
949 | #endif | |
61989a80 | 950 | return 0; |
61989a80 NG |
951 | } |
952 | ||
9eec4cd5 JK |
953 | static unsigned int get_maxobj_per_zspage(int size, int pages_per_zspage) |
954 | { | |
955 | return pages_per_zspage * PAGE_SIZE / size; | |
956 | } | |
957 | ||
958 | static bool can_merge(struct size_class *prev, int size, int pages_per_zspage) | |
959 | { | |
960 | if (prev->pages_per_zspage != pages_per_zspage) | |
961 | return false; | |
962 | ||
963 | if (get_maxobj_per_zspage(prev->size, prev->pages_per_zspage) | |
964 | != get_maxobj_per_zspage(size, pages_per_zspage)) | |
965 | return false; | |
966 | ||
967 | return true; | |
968 | } | |
969 | ||
4bbc0bc0 DB |
970 | /** |
971 | * zs_create_pool - Creates an allocation pool to work from. | |
0d145a50 | 972 | * @flags: allocation flags used to allocate pool metadata |
4bbc0bc0 DB |
973 | * |
974 | * This function must be called before anything when using | |
975 | * the zsmalloc allocator. | |
976 | * | |
977 | * On success, a pointer to the newly created pool is returned, | |
978 | * otherwise NULL. | |
979 | */ | |
0d145a50 | 980 | struct zs_pool *zs_create_pool(gfp_t flags) |
61989a80 | 981 | { |
18136656 | 982 | int i; |
61989a80 | 983 | struct zs_pool *pool; |
df8b5bb9 | 984 | struct size_class *prev_class = NULL; |
61989a80 | 985 | |
18136656 | 986 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
61989a80 NG |
987 | if (!pool) |
988 | return NULL; | |
989 | ||
40f9fb8c MG |
990 | pool->size_class = kcalloc(zs_size_classes, sizeof(struct size_class *), |
991 | GFP_KERNEL); | |
992 | if (!pool->size_class) { | |
993 | kfree(pool); | |
994 | return NULL; | |
995 | } | |
996 | ||
9eec4cd5 JK |
997 | /* |
998 | * Iterate reversly, because, size of size_class that we want to use | |
999 | * for merging should be larger or equal to current size. | |
1000 | */ | |
40f9fb8c | 1001 | for (i = zs_size_classes - 1; i >= 0; i--) { |
61989a80 | 1002 | int size; |
9eec4cd5 | 1003 | int pages_per_zspage; |
61989a80 NG |
1004 | struct size_class *class; |
1005 | ||
1006 | size = ZS_MIN_ALLOC_SIZE + i * ZS_SIZE_CLASS_DELTA; | |
1007 | if (size > ZS_MAX_ALLOC_SIZE) | |
1008 | size = ZS_MAX_ALLOC_SIZE; | |
9eec4cd5 JK |
1009 | pages_per_zspage = get_pages_per_zspage(size); |
1010 | ||
1011 | /* | |
1012 | * size_class is used for normal zsmalloc operation such | |
1013 | * as alloc/free for that size. Although it is natural that we | |
1014 | * have one size_class for each size, there is a chance that we | |
1015 | * can get more memory utilization if we use one size_class for | |
1016 | * many different sizes whose size_class have same | |
1017 | * characteristics. So, we makes size_class point to | |
1018 | * previous size_class if possible. | |
1019 | */ | |
df8b5bb9 | 1020 | if (prev_class) { |
9eec4cd5 JK |
1021 | if (can_merge(prev_class, size, pages_per_zspage)) { |
1022 | pool->size_class[i] = prev_class; | |
1023 | continue; | |
1024 | } | |
1025 | } | |
1026 | ||
1027 | class = kzalloc(sizeof(struct size_class), GFP_KERNEL); | |
1028 | if (!class) | |
1029 | goto err; | |
61989a80 | 1030 | |
61989a80 NG |
1031 | class->size = size; |
1032 | class->index = i; | |
9eec4cd5 | 1033 | class->pages_per_zspage = pages_per_zspage; |
61989a80 | 1034 | spin_lock_init(&class->lock); |
9eec4cd5 | 1035 | pool->size_class[i] = class; |
df8b5bb9 GM |
1036 | |
1037 | prev_class = class; | |
61989a80 NG |
1038 | } |
1039 | ||
61989a80 | 1040 | pool->flags = flags; |
61989a80 | 1041 | |
61989a80 | 1042 | return pool; |
9eec4cd5 JK |
1043 | |
1044 | err: | |
1045 | zs_destroy_pool(pool); | |
1046 | return NULL; | |
61989a80 NG |
1047 | } |
1048 | EXPORT_SYMBOL_GPL(zs_create_pool); | |
1049 | ||
1050 | void zs_destroy_pool(struct zs_pool *pool) | |
1051 | { | |
1052 | int i; | |
1053 | ||
40f9fb8c | 1054 | for (i = 0; i < zs_size_classes; i++) { |
61989a80 | 1055 | int fg; |
9eec4cd5 JK |
1056 | struct size_class *class = pool->size_class[i]; |
1057 | ||
1058 | if (!class) | |
1059 | continue; | |
1060 | ||
1061 | if (class->index != i) | |
1062 | continue; | |
61989a80 NG |
1063 | |
1064 | for (fg = 0; fg < _ZS_NR_FULLNESS_GROUPS; fg++) { | |
1065 | if (class->fullness_list[fg]) { | |
93ad5ab5 | 1066 | pr_info("Freeing non-empty class with size %db, fullness group %d\n", |
61989a80 NG |
1067 | class->size, fg); |
1068 | } | |
1069 | } | |
9eec4cd5 | 1070 | kfree(class); |
61989a80 | 1071 | } |
40f9fb8c MG |
1072 | |
1073 | kfree(pool->size_class); | |
61989a80 NG |
1074 | kfree(pool); |
1075 | } | |
1076 | EXPORT_SYMBOL_GPL(zs_destroy_pool); | |
1077 | ||
1078 | /** | |
1079 | * zs_malloc - Allocate block of given size from pool. | |
1080 | * @pool: pool to allocate from | |
1081 | * @size: size of block to allocate | |
61989a80 | 1082 | * |
00a61d86 | 1083 | * On success, handle to the allocated object is returned, |
c2344348 | 1084 | * otherwise 0. |
61989a80 NG |
1085 | * Allocation requests with size > ZS_MAX_ALLOC_SIZE will fail. |
1086 | */ | |
c2344348 | 1087 | unsigned long zs_malloc(struct zs_pool *pool, size_t size) |
61989a80 | 1088 | { |
c2344348 | 1089 | unsigned long obj; |
61989a80 | 1090 | struct link_free *link; |
61989a80 | 1091 | struct size_class *class; |
af4ee5e9 | 1092 | void *vaddr; |
61989a80 NG |
1093 | |
1094 | struct page *first_page, *m_page; | |
1095 | unsigned long m_objidx, m_offset; | |
1096 | ||
1097 | if (unlikely(!size || size > ZS_MAX_ALLOC_SIZE)) | |
c2344348 | 1098 | return 0; |
61989a80 | 1099 | |
9eec4cd5 | 1100 | class = pool->size_class[get_size_class_index(size)]; |
61989a80 NG |
1101 | |
1102 | spin_lock(&class->lock); | |
1103 | first_page = find_get_zspage(class); | |
1104 | ||
1105 | if (!first_page) { | |
1106 | spin_unlock(&class->lock); | |
1107 | first_page = alloc_zspage(class, pool->flags); | |
1108 | if (unlikely(!first_page)) | |
c2344348 | 1109 | return 0; |
61989a80 NG |
1110 | |
1111 | set_zspage_mapping(first_page, class->index, ZS_EMPTY); | |
13de8933 MK |
1112 | atomic_long_add(class->pages_per_zspage, |
1113 | &pool->pages_allocated); | |
61989a80 | 1114 | spin_lock(&class->lock); |
61989a80 NG |
1115 | } |
1116 | ||
c2344348 | 1117 | obj = (unsigned long)first_page->freelist; |
61989a80 NG |
1118 | obj_handle_to_location(obj, &m_page, &m_objidx); |
1119 | m_offset = obj_idx_to_offset(m_page, m_objidx, class->size); | |
1120 | ||
af4ee5e9 MK |
1121 | vaddr = kmap_atomic(m_page); |
1122 | link = (struct link_free *)vaddr + m_offset / sizeof(*link); | |
61989a80 NG |
1123 | first_page->freelist = link->next; |
1124 | memset(link, POISON_INUSE, sizeof(*link)); | |
af4ee5e9 | 1125 | kunmap_atomic(vaddr); |
61989a80 NG |
1126 | |
1127 | first_page->inuse++; | |
1128 | /* Now move the zspage to another fullness group, if required */ | |
1129 | fix_fullness_group(pool, first_page); | |
1130 | spin_unlock(&class->lock); | |
1131 | ||
1132 | return obj; | |
1133 | } | |
1134 | EXPORT_SYMBOL_GPL(zs_malloc); | |
1135 | ||
c2344348 | 1136 | void zs_free(struct zs_pool *pool, unsigned long obj) |
61989a80 NG |
1137 | { |
1138 | struct link_free *link; | |
1139 | struct page *first_page, *f_page; | |
1140 | unsigned long f_objidx, f_offset; | |
af4ee5e9 | 1141 | void *vaddr; |
61989a80 NG |
1142 | |
1143 | int class_idx; | |
1144 | struct size_class *class; | |
1145 | enum fullness_group fullness; | |
1146 | ||
1147 | if (unlikely(!obj)) | |
1148 | return; | |
1149 | ||
1150 | obj_handle_to_location(obj, &f_page, &f_objidx); | |
1151 | first_page = get_first_page(f_page); | |
1152 | ||
1153 | get_zspage_mapping(first_page, &class_idx, &fullness); | |
9eec4cd5 | 1154 | class = pool->size_class[class_idx]; |
61989a80 NG |
1155 | f_offset = obj_idx_to_offset(f_page, f_objidx, class->size); |
1156 | ||
1157 | spin_lock(&class->lock); | |
1158 | ||
1159 | /* Insert this object in containing zspage's freelist */ | |
af4ee5e9 MK |
1160 | vaddr = kmap_atomic(f_page); |
1161 | link = (struct link_free *)(vaddr + f_offset); | |
61989a80 | 1162 | link->next = first_page->freelist; |
af4ee5e9 | 1163 | kunmap_atomic(vaddr); |
c2344348 | 1164 | first_page->freelist = (void *)obj; |
61989a80 NG |
1165 | |
1166 | first_page->inuse--; | |
1167 | fullness = fix_fullness_group(pool, first_page); | |
61989a80 NG |
1168 | spin_unlock(&class->lock); |
1169 | ||
13de8933 MK |
1170 | if (fullness == ZS_EMPTY) { |
1171 | atomic_long_sub(class->pages_per_zspage, | |
1172 | &pool->pages_allocated); | |
61989a80 | 1173 | free_zspage(first_page); |
13de8933 | 1174 | } |
61989a80 NG |
1175 | } |
1176 | EXPORT_SYMBOL_GPL(zs_free); | |
1177 | ||
00a61d86 MK |
1178 | /** |
1179 | * zs_map_object - get address of allocated object from handle. | |
1180 | * @pool: pool from which the object was allocated | |
1181 | * @handle: handle returned from zs_malloc | |
1182 | * | |
1183 | * Before using an object allocated from zs_malloc, it must be mapped using | |
1184 | * this function. When done with the object, it must be unmapped using | |
166cfda7 SJ |
1185 | * zs_unmap_object. |
1186 | * | |
1187 | * Only one object can be mapped per cpu at a time. There is no protection | |
1188 | * against nested mappings. | |
1189 | * | |
1190 | * This function returns with preemption and page faults disabled. | |
396b7fd6 | 1191 | */ |
b7418510 SJ |
1192 | void *zs_map_object(struct zs_pool *pool, unsigned long handle, |
1193 | enum zs_mapmode mm) | |
61989a80 NG |
1194 | { |
1195 | struct page *page; | |
1196 | unsigned long obj_idx, off; | |
1197 | ||
1198 | unsigned int class_idx; | |
1199 | enum fullness_group fg; | |
1200 | struct size_class *class; | |
1201 | struct mapping_area *area; | |
f553646a | 1202 | struct page *pages[2]; |
61989a80 NG |
1203 | |
1204 | BUG_ON(!handle); | |
1205 | ||
c60369f0 SJ |
1206 | /* |
1207 | * Because we use per-cpu mapping areas shared among the | |
1208 | * pools/users, we can't allow mapping in interrupt context | |
1209 | * because it can corrupt another users mappings. | |
1210 | */ | |
1211 | BUG_ON(in_interrupt()); | |
1212 | ||
61989a80 NG |
1213 | obj_handle_to_location(handle, &page, &obj_idx); |
1214 | get_zspage_mapping(get_first_page(page), &class_idx, &fg); | |
9eec4cd5 | 1215 | class = pool->size_class[class_idx]; |
61989a80 NG |
1216 | off = obj_idx_to_offset(page, obj_idx, class->size); |
1217 | ||
1218 | area = &get_cpu_var(zs_map_area); | |
f553646a | 1219 | area->vm_mm = mm; |
61989a80 NG |
1220 | if (off + class->size <= PAGE_SIZE) { |
1221 | /* this object is contained entirely within a page */ | |
1222 | area->vm_addr = kmap_atomic(page); | |
5f601902 | 1223 | return area->vm_addr + off; |
61989a80 NG |
1224 | } |
1225 | ||
f553646a SJ |
1226 | /* this object spans two pages */ |
1227 | pages[0] = page; | |
1228 | pages[1] = get_next_page(page); | |
1229 | BUG_ON(!pages[1]); | |
b7418510 | 1230 | |
f553646a | 1231 | return __zs_map_object(area, pages, off, class->size); |
61989a80 NG |
1232 | } |
1233 | EXPORT_SYMBOL_GPL(zs_map_object); | |
1234 | ||
c2344348 | 1235 | void zs_unmap_object(struct zs_pool *pool, unsigned long handle) |
61989a80 NG |
1236 | { |
1237 | struct page *page; | |
1238 | unsigned long obj_idx, off; | |
1239 | ||
1240 | unsigned int class_idx; | |
1241 | enum fullness_group fg; | |
1242 | struct size_class *class; | |
1243 | struct mapping_area *area; | |
1244 | ||
1245 | BUG_ON(!handle); | |
1246 | ||
1247 | obj_handle_to_location(handle, &page, &obj_idx); | |
1248 | get_zspage_mapping(get_first_page(page), &class_idx, &fg); | |
9eec4cd5 | 1249 | class = pool->size_class[class_idx]; |
61989a80 NG |
1250 | off = obj_idx_to_offset(page, obj_idx, class->size); |
1251 | ||
7c8e0181 | 1252 | area = this_cpu_ptr(&zs_map_area); |
f553646a SJ |
1253 | if (off + class->size <= PAGE_SIZE) |
1254 | kunmap_atomic(area->vm_addr); | |
1255 | else { | |
1256 | struct page *pages[2]; | |
1257 | ||
1258 | pages[0] = page; | |
1259 | pages[1] = get_next_page(page); | |
1260 | BUG_ON(!pages[1]); | |
b7418510 | 1261 | |
f553646a SJ |
1262 | __zs_unmap_object(area, pages, off, class->size); |
1263 | } | |
61989a80 NG |
1264 | put_cpu_var(zs_map_area); |
1265 | } | |
1266 | EXPORT_SYMBOL_GPL(zs_unmap_object); | |
1267 | ||
722cdc17 | 1268 | unsigned long zs_get_total_pages(struct zs_pool *pool) |
61989a80 | 1269 | { |
722cdc17 | 1270 | return atomic_long_read(&pool->pages_allocated); |
61989a80 | 1271 | } |
722cdc17 | 1272 | EXPORT_SYMBOL_GPL(zs_get_total_pages); |
069f101f BH |
1273 | |
1274 | module_init(zs_init); | |
1275 | module_exit(zs_exit); | |
1276 | ||
1277 | MODULE_LICENSE("Dual BSD/GPL"); | |
1278 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); |