]>
Commit | Line | Data |
---|---|---|
61989a80 NG |
1 | /* |
2 | * zsmalloc memory allocator | |
3 | * | |
4 | * Copyright (C) 2011 Nitin Gupta | |
5 | * | |
6 | * This code is released using a dual license strategy: BSD/GPL | |
7 | * You can choose the license that better fits your requirements. | |
8 | * | |
9 | * Released under the terms of 3-clause BSD License | |
10 | * Released under the terms of GNU General Public License Version 2.0 | |
11 | */ | |
12 | ||
2db51dae NG |
13 | |
14 | /* | |
15 | * This allocator is designed for use with zcache and zram. Thus, the | |
16 | * allocator is supposed to work well under low memory conditions. In | |
17 | * particular, it never attempts higher order page allocation which is | |
18 | * very likely to fail under memory pressure. On the other hand, if we | |
19 | * just use single (0-order) pages, it would suffer from very high | |
20 | * fragmentation -- any object of size PAGE_SIZE/2 or larger would occupy | |
21 | * an entire page. This was one of the major issues with its predecessor | |
22 | * (xvmalloc). | |
23 | * | |
24 | * To overcome these issues, zsmalloc allocates a bunch of 0-order pages | |
25 | * and links them together using various 'struct page' fields. These linked | |
26 | * pages act as a single higher-order page i.e. an object can span 0-order | |
27 | * page boundaries. The code refers to these linked pages as a single entity | |
28 | * called zspage. | |
29 | * | |
30 | * Following is how we use various fields and flags of underlying | |
31 | * struct page(s) to form a zspage. | |
32 | * | |
33 | * Usage of struct page fields: | |
34 | * page->first_page: points to the first component (0-order) page | |
35 | * page->index (union with page->freelist): offset of the first object | |
36 | * starting in this page. For the first page, this is | |
37 | * always 0, so we use this field (aka freelist) to point | |
38 | * to the first free object in zspage. | |
39 | * page->lru: links together all component pages (except the first page) | |
40 | * of a zspage | |
41 | * | |
42 | * For _first_ page only: | |
43 | * | |
44 | * page->private (union with page->first_page): refers to the | |
45 | * component page after the first page | |
46 | * page->freelist: points to the first free object in zspage. | |
47 | * Free objects are linked together using in-place | |
48 | * metadata. | |
49 | * page->objects: maximum number of objects we can store in this | |
50 | * zspage (class->zspage_order * PAGE_SIZE / class->size) | |
51 | * page->lru: links together first pages of various zspages. | |
52 | * Basically forming list of zspages in a fullness group. | |
53 | * page->mapping: class index and fullness group of the zspage | |
54 | * | |
55 | * Usage of struct page flags: | |
56 | * PG_private: identifies the first component page | |
57 | * PG_private2: identifies the last component page | |
58 | * | |
59 | */ | |
60 | ||
61989a80 NG |
61 | #ifdef CONFIG_ZSMALLOC_DEBUG |
62 | #define DEBUG | |
63 | #endif | |
64 | ||
65 | #include <linux/module.h> | |
66 | #include <linux/kernel.h> | |
67 | #include <linux/bitops.h> | |
68 | #include <linux/errno.h> | |
69 | #include <linux/highmem.h> | |
70 | #include <linux/init.h> | |
71 | #include <linux/string.h> | |
72 | #include <linux/slab.h> | |
73 | #include <asm/tlbflush.h> | |
74 | #include <asm/pgtable.h> | |
75 | #include <linux/cpumask.h> | |
76 | #include <linux/cpu.h> | |
0cbb613f | 77 | #include <linux/vmalloc.h> |
61989a80 NG |
78 | |
79 | #include "zsmalloc.h" | |
80 | #include "zsmalloc_int.h" | |
81 | ||
82 | /* | |
83 | * A zspage's class index and fullness group | |
84 | * are encoded in its (first)page->mapping | |
85 | */ | |
86 | #define CLASS_IDX_BITS 28 | |
87 | #define FULLNESS_BITS 4 | |
88 | #define CLASS_IDX_MASK ((1 << CLASS_IDX_BITS) - 1) | |
89 | #define FULLNESS_MASK ((1 << FULLNESS_BITS) - 1) | |
90 | ||
61989a80 NG |
91 | /* per-cpu VM mapping areas for zspage accesses that cross page boundaries */ |
92 | static DEFINE_PER_CPU(struct mapping_area, zs_map_area); | |
93 | ||
94 | static int is_first_page(struct page *page) | |
95 | { | |
a27545bf | 96 | return PagePrivate(page); |
61989a80 NG |
97 | } |
98 | ||
99 | static int is_last_page(struct page *page) | |
100 | { | |
a27545bf | 101 | return PagePrivate2(page); |
61989a80 NG |
102 | } |
103 | ||
104 | static void get_zspage_mapping(struct page *page, unsigned int *class_idx, | |
105 | enum fullness_group *fullness) | |
106 | { | |
107 | unsigned long m; | |
108 | BUG_ON(!is_first_page(page)); | |
109 | ||
110 | m = (unsigned long)page->mapping; | |
111 | *fullness = m & FULLNESS_MASK; | |
112 | *class_idx = (m >> FULLNESS_BITS) & CLASS_IDX_MASK; | |
113 | } | |
114 | ||
115 | static void set_zspage_mapping(struct page *page, unsigned int class_idx, | |
116 | enum fullness_group fullness) | |
117 | { | |
118 | unsigned long m; | |
119 | BUG_ON(!is_first_page(page)); | |
120 | ||
121 | m = ((class_idx & CLASS_IDX_MASK) << FULLNESS_BITS) | | |
122 | (fullness & FULLNESS_MASK); | |
123 | page->mapping = (struct address_space *)m; | |
124 | } | |
125 | ||
126 | static int get_size_class_index(int size) | |
127 | { | |
128 | int idx = 0; | |
129 | ||
130 | if (likely(size > ZS_MIN_ALLOC_SIZE)) | |
131 | idx = DIV_ROUND_UP(size - ZS_MIN_ALLOC_SIZE, | |
132 | ZS_SIZE_CLASS_DELTA); | |
133 | ||
134 | return idx; | |
135 | } | |
136 | ||
137 | static enum fullness_group get_fullness_group(struct page *page) | |
138 | { | |
139 | int inuse, max_objects; | |
140 | enum fullness_group fg; | |
141 | BUG_ON(!is_first_page(page)); | |
142 | ||
143 | inuse = page->inuse; | |
144 | max_objects = page->objects; | |
145 | ||
146 | if (inuse == 0) | |
147 | fg = ZS_EMPTY; | |
148 | else if (inuse == max_objects) | |
149 | fg = ZS_FULL; | |
150 | else if (inuse <= max_objects / fullness_threshold_frac) | |
151 | fg = ZS_ALMOST_EMPTY; | |
152 | else | |
153 | fg = ZS_ALMOST_FULL; | |
154 | ||
155 | return fg; | |
156 | } | |
157 | ||
158 | static void insert_zspage(struct page *page, struct size_class *class, | |
159 | enum fullness_group fullness) | |
160 | { | |
161 | struct page **head; | |
162 | ||
163 | BUG_ON(!is_first_page(page)); | |
164 | ||
165 | if (fullness >= _ZS_NR_FULLNESS_GROUPS) | |
166 | return; | |
167 | ||
168 | head = &class->fullness_list[fullness]; | |
169 | if (*head) | |
170 | list_add_tail(&page->lru, &(*head)->lru); | |
171 | ||
172 | *head = page; | |
173 | } | |
174 | ||
175 | static void remove_zspage(struct page *page, struct size_class *class, | |
176 | enum fullness_group fullness) | |
177 | { | |
178 | struct page **head; | |
179 | ||
180 | BUG_ON(!is_first_page(page)); | |
181 | ||
182 | if (fullness >= _ZS_NR_FULLNESS_GROUPS) | |
183 | return; | |
184 | ||
185 | head = &class->fullness_list[fullness]; | |
186 | BUG_ON(!*head); | |
187 | if (list_empty(&(*head)->lru)) | |
188 | *head = NULL; | |
189 | else if (*head == page) | |
190 | *head = (struct page *)list_entry((*head)->lru.next, | |
191 | struct page, lru); | |
192 | ||
193 | list_del_init(&page->lru); | |
194 | } | |
195 | ||
196 | static enum fullness_group fix_fullness_group(struct zs_pool *pool, | |
197 | struct page *page) | |
198 | { | |
199 | int class_idx; | |
200 | struct size_class *class; | |
201 | enum fullness_group currfg, newfg; | |
202 | ||
203 | BUG_ON(!is_first_page(page)); | |
204 | ||
205 | get_zspage_mapping(page, &class_idx, &currfg); | |
206 | newfg = get_fullness_group(page); | |
207 | if (newfg == currfg) | |
208 | goto out; | |
209 | ||
210 | class = &pool->size_class[class_idx]; | |
211 | remove_zspage(page, class, currfg); | |
212 | insert_zspage(page, class, newfg); | |
213 | set_zspage_mapping(page, class_idx, newfg); | |
214 | ||
215 | out: | |
216 | return newfg; | |
217 | } | |
218 | ||
219 | /* | |
220 | * We have to decide on how many pages to link together | |
221 | * to form a zspage for each size class. This is important | |
222 | * to reduce wastage due to unusable space left at end of | |
223 | * each zspage which is given as: | |
224 | * wastage = Zp - Zp % size_class | |
225 | * where Zp = zspage size = k * PAGE_SIZE where k = 1, 2, ... | |
226 | * | |
227 | * For example, for size class of 3/8 * PAGE_SIZE, we should | |
228 | * link together 3 PAGE_SIZE sized pages to form a zspage | |
229 | * since then we can perfectly fit in 8 such objects. | |
230 | */ | |
2e3b6154 | 231 | static int get_pages_per_zspage(int class_size) |
61989a80 NG |
232 | { |
233 | int i, max_usedpc = 0; | |
234 | /* zspage order which gives maximum used size per KB */ | |
235 | int max_usedpc_order = 1; | |
236 | ||
84d4faab | 237 | for (i = 1; i <= ZS_MAX_PAGES_PER_ZSPAGE; i++) { |
61989a80 NG |
238 | int zspage_size; |
239 | int waste, usedpc; | |
240 | ||
241 | zspage_size = i * PAGE_SIZE; | |
242 | waste = zspage_size % class_size; | |
243 | usedpc = (zspage_size - waste) * 100 / zspage_size; | |
244 | ||
245 | if (usedpc > max_usedpc) { | |
246 | max_usedpc = usedpc; | |
247 | max_usedpc_order = i; | |
248 | } | |
249 | } | |
250 | ||
251 | return max_usedpc_order; | |
252 | } | |
253 | ||
254 | /* | |
255 | * A single 'zspage' is composed of many system pages which are | |
256 | * linked together using fields in struct page. This function finds | |
257 | * the first/head page, given any component page of a zspage. | |
258 | */ | |
259 | static struct page *get_first_page(struct page *page) | |
260 | { | |
261 | if (is_first_page(page)) | |
262 | return page; | |
263 | else | |
264 | return page->first_page; | |
265 | } | |
266 | ||
267 | static struct page *get_next_page(struct page *page) | |
268 | { | |
269 | struct page *next; | |
270 | ||
271 | if (is_last_page(page)) | |
272 | next = NULL; | |
273 | else if (is_first_page(page)) | |
274 | next = (struct page *)page->private; | |
275 | else | |
276 | next = list_entry(page->lru.next, struct page, lru); | |
277 | ||
278 | return next; | |
279 | } | |
280 | ||
281 | /* Encode <page, obj_idx> as a single handle value */ | |
282 | static void *obj_location_to_handle(struct page *page, unsigned long obj_idx) | |
283 | { | |
284 | unsigned long handle; | |
285 | ||
286 | if (!page) { | |
287 | BUG_ON(obj_idx); | |
288 | return NULL; | |
289 | } | |
290 | ||
291 | handle = page_to_pfn(page) << OBJ_INDEX_BITS; | |
292 | handle |= (obj_idx & OBJ_INDEX_MASK); | |
293 | ||
294 | return (void *)handle; | |
295 | } | |
296 | ||
297 | /* Decode <page, obj_idx> pair from the given object handle */ | |
c2344348 | 298 | static void obj_handle_to_location(unsigned long handle, struct page **page, |
61989a80 NG |
299 | unsigned long *obj_idx) |
300 | { | |
c2344348 MK |
301 | *page = pfn_to_page(handle >> OBJ_INDEX_BITS); |
302 | *obj_idx = handle & OBJ_INDEX_MASK; | |
61989a80 NG |
303 | } |
304 | ||
305 | static unsigned long obj_idx_to_offset(struct page *page, | |
306 | unsigned long obj_idx, int class_size) | |
307 | { | |
308 | unsigned long off = 0; | |
309 | ||
310 | if (!is_first_page(page)) | |
311 | off = page->index; | |
312 | ||
313 | return off + obj_idx * class_size; | |
314 | } | |
315 | ||
f4477e90 NG |
316 | static void reset_page(struct page *page) |
317 | { | |
318 | clear_bit(PG_private, &page->flags); | |
319 | clear_bit(PG_private_2, &page->flags); | |
320 | set_page_private(page, 0); | |
321 | page->mapping = NULL; | |
322 | page->freelist = NULL; | |
323 | reset_page_mapcount(page); | |
324 | } | |
325 | ||
61989a80 NG |
326 | static void free_zspage(struct page *first_page) |
327 | { | |
f4477e90 | 328 | struct page *nextp, *tmp, *head_extra; |
61989a80 NG |
329 | |
330 | BUG_ON(!is_first_page(first_page)); | |
331 | BUG_ON(first_page->inuse); | |
332 | ||
f4477e90 | 333 | head_extra = (struct page *)page_private(first_page); |
61989a80 | 334 | |
f4477e90 | 335 | reset_page(first_page); |
61989a80 NG |
336 | __free_page(first_page); |
337 | ||
338 | /* zspage with only 1 system page */ | |
f4477e90 | 339 | if (!head_extra) |
61989a80 NG |
340 | return; |
341 | ||
f4477e90 | 342 | list_for_each_entry_safe(nextp, tmp, &head_extra->lru, lru) { |
61989a80 | 343 | list_del(&nextp->lru); |
f4477e90 | 344 | reset_page(nextp); |
61989a80 NG |
345 | __free_page(nextp); |
346 | } | |
f4477e90 NG |
347 | reset_page(head_extra); |
348 | __free_page(head_extra); | |
61989a80 NG |
349 | } |
350 | ||
351 | /* Initialize a newly allocated zspage */ | |
352 | static void init_zspage(struct page *first_page, struct size_class *class) | |
353 | { | |
354 | unsigned long off = 0; | |
355 | struct page *page = first_page; | |
356 | ||
357 | BUG_ON(!is_first_page(first_page)); | |
358 | while (page) { | |
359 | struct page *next_page; | |
360 | struct link_free *link; | |
361 | unsigned int i, objs_on_page; | |
362 | ||
363 | /* | |
364 | * page->index stores offset of first object starting | |
365 | * in the page. For the first page, this is always 0, | |
366 | * so we use first_page->index (aka ->freelist) to store | |
367 | * head of corresponding zspage's freelist. | |
368 | */ | |
369 | if (page != first_page) | |
370 | page->index = off; | |
371 | ||
372 | link = (struct link_free *)kmap_atomic(page) + | |
373 | off / sizeof(*link); | |
374 | objs_on_page = (PAGE_SIZE - off) / class->size; | |
375 | ||
376 | for (i = 1; i <= objs_on_page; i++) { | |
377 | off += class->size; | |
378 | if (off < PAGE_SIZE) { | |
379 | link->next = obj_location_to_handle(page, i); | |
380 | link += class->size / sizeof(*link); | |
381 | } | |
382 | } | |
383 | ||
384 | /* | |
385 | * We now come to the last (full or partial) object on this | |
386 | * page, which must point to the first object on the next | |
387 | * page (if present) | |
388 | */ | |
389 | next_page = get_next_page(page); | |
390 | link->next = obj_location_to_handle(next_page, 0); | |
391 | kunmap_atomic(link); | |
392 | page = next_page; | |
393 | off = (off + class->size) % PAGE_SIZE; | |
394 | } | |
395 | } | |
396 | ||
397 | /* | |
398 | * Allocate a zspage for the given size class | |
399 | */ | |
400 | static struct page *alloc_zspage(struct size_class *class, gfp_t flags) | |
401 | { | |
402 | int i, error; | |
b4b700c5 | 403 | struct page *first_page = NULL, *uninitialized_var(prev_page); |
61989a80 NG |
404 | |
405 | /* | |
406 | * Allocate individual pages and link them together as: | |
407 | * 1. first page->private = first sub-page | |
408 | * 2. all sub-pages are linked together using page->lru | |
409 | * 3. each sub-page is linked to the first page using page->first_page | |
410 | * | |
411 | * For each size class, First/Head pages are linked together using | |
412 | * page->lru. Also, we set PG_private to identify the first page | |
413 | * (i.e. no other sub-page has this flag set) and PG_private_2 to | |
414 | * identify the last page. | |
415 | */ | |
416 | error = -ENOMEM; | |
2e3b6154 | 417 | for (i = 0; i < class->pages_per_zspage; i++) { |
b4b700c5 | 418 | struct page *page; |
61989a80 NG |
419 | |
420 | page = alloc_page(flags); | |
421 | if (!page) | |
422 | goto cleanup; | |
423 | ||
424 | INIT_LIST_HEAD(&page->lru); | |
425 | if (i == 0) { /* first page */ | |
a27545bf | 426 | SetPagePrivate(page); |
61989a80 NG |
427 | set_page_private(page, 0); |
428 | first_page = page; | |
429 | first_page->inuse = 0; | |
430 | } | |
431 | if (i == 1) | |
432 | first_page->private = (unsigned long)page; | |
433 | if (i >= 1) | |
434 | page->first_page = first_page; | |
435 | if (i >= 2) | |
436 | list_add(&page->lru, &prev_page->lru); | |
2e3b6154 | 437 | if (i == class->pages_per_zspage - 1) /* last page */ |
a27545bf | 438 | SetPagePrivate2(page); |
61989a80 NG |
439 | prev_page = page; |
440 | } | |
441 | ||
442 | init_zspage(first_page, class); | |
443 | ||
444 | first_page->freelist = obj_location_to_handle(first_page, 0); | |
445 | /* Maximum number of objects we can store in this zspage */ | |
2e3b6154 | 446 | first_page->objects = class->pages_per_zspage * PAGE_SIZE / class->size; |
61989a80 NG |
447 | |
448 | error = 0; /* Success */ | |
449 | ||
450 | cleanup: | |
451 | if (unlikely(error) && first_page) { | |
452 | free_zspage(first_page); | |
453 | first_page = NULL; | |
454 | } | |
455 | ||
456 | return first_page; | |
457 | } | |
458 | ||
459 | static struct page *find_get_zspage(struct size_class *class) | |
460 | { | |
461 | int i; | |
462 | struct page *page; | |
463 | ||
464 | for (i = 0; i < _ZS_NR_FULLNESS_GROUPS; i++) { | |
465 | page = class->fullness_list[i]; | |
466 | if (page) | |
467 | break; | |
468 | } | |
469 | ||
470 | return page; | |
471 | } | |
472 | ||
5f601902 SJ |
473 | static void zs_copy_map_object(char *buf, struct page *firstpage, |
474 | int off, int size) | |
475 | { | |
476 | struct page *pages[2]; | |
477 | int sizes[2]; | |
478 | void *addr; | |
479 | ||
480 | pages[0] = firstpage; | |
481 | pages[1] = get_next_page(firstpage); | |
482 | BUG_ON(!pages[1]); | |
483 | ||
484 | sizes[0] = PAGE_SIZE - off; | |
485 | sizes[1] = size - sizes[0]; | |
486 | ||
487 | /* disable page faults to match kmap_atomic() return conditions */ | |
488 | pagefault_disable(); | |
489 | ||
490 | /* copy object to per-cpu buffer */ | |
491 | addr = kmap_atomic(pages[0]); | |
492 | memcpy(buf, addr + off, sizes[0]); | |
493 | kunmap_atomic(addr); | |
494 | addr = kmap_atomic(pages[1]); | |
495 | memcpy(buf + sizes[0], addr, sizes[1]); | |
496 | kunmap_atomic(addr); | |
497 | } | |
498 | ||
499 | static void zs_copy_unmap_object(char *buf, struct page *firstpage, | |
500 | int off, int size) | |
501 | { | |
502 | struct page *pages[2]; | |
503 | int sizes[2]; | |
504 | void *addr; | |
505 | ||
506 | pages[0] = firstpage; | |
507 | pages[1] = get_next_page(firstpage); | |
508 | BUG_ON(!pages[1]); | |
509 | ||
510 | sizes[0] = PAGE_SIZE - off; | |
511 | sizes[1] = size - sizes[0]; | |
512 | ||
513 | /* copy per-cpu buffer to object */ | |
514 | addr = kmap_atomic(pages[0]); | |
515 | memcpy(addr + off, buf, sizes[0]); | |
516 | kunmap_atomic(addr); | |
517 | addr = kmap_atomic(pages[1]); | |
518 | memcpy(addr, buf + sizes[0], sizes[1]); | |
519 | kunmap_atomic(addr); | |
520 | ||
521 | /* enable page faults to match kunmap_atomic() return conditions */ | |
522 | pagefault_enable(); | |
523 | } | |
61989a80 | 524 | |
61989a80 NG |
525 | static int zs_cpu_notifier(struct notifier_block *nb, unsigned long action, |
526 | void *pcpu) | |
527 | { | |
528 | int cpu = (long)pcpu; | |
529 | struct mapping_area *area; | |
530 | ||
531 | switch (action) { | |
532 | case CPU_UP_PREPARE: | |
533 | area = &per_cpu(zs_map_area, cpu); | |
5f601902 SJ |
534 | /* |
535 | * Make sure we don't leak memory if a cpu UP notification | |
536 | * and zs_init() race and both call zs_cpu_up() on the same cpu | |
537 | */ | |
538 | if (area->vm_buf) | |
539 | return 0; | |
540 | area->vm_buf = (char *)__get_free_page(GFP_KERNEL); | |
541 | if (!area->vm_buf) | |
542 | return -ENOMEM; | |
543 | return 0; | |
61989a80 NG |
544 | break; |
545 | case CPU_DEAD: | |
546 | case CPU_UP_CANCELED: | |
547 | area = &per_cpu(zs_map_area, cpu); | |
5f601902 SJ |
548 | if (area->vm_buf) |
549 | free_page((unsigned long)area->vm_buf); | |
550 | area->vm_buf = NULL; | |
61989a80 NG |
551 | break; |
552 | } | |
553 | ||
554 | return NOTIFY_OK; | |
555 | } | |
556 | ||
557 | static struct notifier_block zs_cpu_nb = { | |
558 | .notifier_call = zs_cpu_notifier | |
559 | }; | |
560 | ||
561 | static void zs_exit(void) | |
562 | { | |
563 | int cpu; | |
564 | ||
565 | for_each_online_cpu(cpu) | |
566 | zs_cpu_notifier(NULL, CPU_DEAD, (void *)(long)cpu); | |
567 | unregister_cpu_notifier(&zs_cpu_nb); | |
568 | } | |
569 | ||
570 | static int zs_init(void) | |
571 | { | |
572 | int cpu, ret; | |
573 | ||
574 | register_cpu_notifier(&zs_cpu_nb); | |
575 | for_each_online_cpu(cpu) { | |
576 | ret = zs_cpu_notifier(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
577 | if (notifier_to_errno(ret)) | |
578 | goto fail; | |
579 | } | |
580 | return 0; | |
581 | fail: | |
582 | zs_exit(); | |
583 | return notifier_to_errno(ret); | |
584 | } | |
585 | ||
586 | struct zs_pool *zs_create_pool(const char *name, gfp_t flags) | |
587 | { | |
069f101f | 588 | int i, ovhd_size; |
61989a80 NG |
589 | struct zs_pool *pool; |
590 | ||
591 | if (!name) | |
592 | return NULL; | |
593 | ||
594 | ovhd_size = roundup(sizeof(*pool), PAGE_SIZE); | |
595 | pool = kzalloc(ovhd_size, GFP_KERNEL); | |
596 | if (!pool) | |
597 | return NULL; | |
598 | ||
599 | for (i = 0; i < ZS_SIZE_CLASSES; i++) { | |
600 | int size; | |
601 | struct size_class *class; | |
602 | ||
603 | size = ZS_MIN_ALLOC_SIZE + i * ZS_SIZE_CLASS_DELTA; | |
604 | if (size > ZS_MAX_ALLOC_SIZE) | |
605 | size = ZS_MAX_ALLOC_SIZE; | |
606 | ||
607 | class = &pool->size_class[i]; | |
608 | class->size = size; | |
609 | class->index = i; | |
610 | spin_lock_init(&class->lock); | |
2e3b6154 | 611 | class->pages_per_zspage = get_pages_per_zspage(size); |
61989a80 NG |
612 | |
613 | } | |
614 | ||
61989a80 NG |
615 | pool->flags = flags; |
616 | pool->name = name; | |
617 | ||
61989a80 NG |
618 | return pool; |
619 | } | |
620 | EXPORT_SYMBOL_GPL(zs_create_pool); | |
621 | ||
622 | void zs_destroy_pool(struct zs_pool *pool) | |
623 | { | |
624 | int i; | |
625 | ||
626 | for (i = 0; i < ZS_SIZE_CLASSES; i++) { | |
627 | int fg; | |
628 | struct size_class *class = &pool->size_class[i]; | |
629 | ||
630 | for (fg = 0; fg < _ZS_NR_FULLNESS_GROUPS; fg++) { | |
631 | if (class->fullness_list[fg]) { | |
632 | pr_info("Freeing non-empty class with size " | |
633 | "%db, fullness group %d\n", | |
634 | class->size, fg); | |
635 | } | |
636 | } | |
637 | } | |
638 | kfree(pool); | |
639 | } | |
640 | EXPORT_SYMBOL_GPL(zs_destroy_pool); | |
641 | ||
642 | /** | |
643 | * zs_malloc - Allocate block of given size from pool. | |
644 | * @pool: pool to allocate from | |
645 | * @size: size of block to allocate | |
61989a80 | 646 | * |
00a61d86 | 647 | * On success, handle to the allocated object is returned, |
c2344348 | 648 | * otherwise 0. |
61989a80 NG |
649 | * Allocation requests with size > ZS_MAX_ALLOC_SIZE will fail. |
650 | */ | |
c2344348 | 651 | unsigned long zs_malloc(struct zs_pool *pool, size_t size) |
61989a80 | 652 | { |
c2344348 | 653 | unsigned long obj; |
61989a80 NG |
654 | struct link_free *link; |
655 | int class_idx; | |
656 | struct size_class *class; | |
657 | ||
658 | struct page *first_page, *m_page; | |
659 | unsigned long m_objidx, m_offset; | |
660 | ||
661 | if (unlikely(!size || size > ZS_MAX_ALLOC_SIZE)) | |
c2344348 | 662 | return 0; |
61989a80 NG |
663 | |
664 | class_idx = get_size_class_index(size); | |
665 | class = &pool->size_class[class_idx]; | |
666 | BUG_ON(class_idx != class->index); | |
667 | ||
668 | spin_lock(&class->lock); | |
669 | first_page = find_get_zspage(class); | |
670 | ||
671 | if (!first_page) { | |
672 | spin_unlock(&class->lock); | |
673 | first_page = alloc_zspage(class, pool->flags); | |
674 | if (unlikely(!first_page)) | |
c2344348 | 675 | return 0; |
61989a80 NG |
676 | |
677 | set_zspage_mapping(first_page, class->index, ZS_EMPTY); | |
678 | spin_lock(&class->lock); | |
2e3b6154 | 679 | class->pages_allocated += class->pages_per_zspage; |
61989a80 NG |
680 | } |
681 | ||
c2344348 | 682 | obj = (unsigned long)first_page->freelist; |
61989a80 NG |
683 | obj_handle_to_location(obj, &m_page, &m_objidx); |
684 | m_offset = obj_idx_to_offset(m_page, m_objidx, class->size); | |
685 | ||
686 | link = (struct link_free *)kmap_atomic(m_page) + | |
687 | m_offset / sizeof(*link); | |
688 | first_page->freelist = link->next; | |
689 | memset(link, POISON_INUSE, sizeof(*link)); | |
690 | kunmap_atomic(link); | |
691 | ||
692 | first_page->inuse++; | |
693 | /* Now move the zspage to another fullness group, if required */ | |
694 | fix_fullness_group(pool, first_page); | |
695 | spin_unlock(&class->lock); | |
696 | ||
697 | return obj; | |
698 | } | |
699 | EXPORT_SYMBOL_GPL(zs_malloc); | |
700 | ||
c2344348 | 701 | void zs_free(struct zs_pool *pool, unsigned long obj) |
61989a80 NG |
702 | { |
703 | struct link_free *link; | |
704 | struct page *first_page, *f_page; | |
705 | unsigned long f_objidx, f_offset; | |
706 | ||
707 | int class_idx; | |
708 | struct size_class *class; | |
709 | enum fullness_group fullness; | |
710 | ||
711 | if (unlikely(!obj)) | |
712 | return; | |
713 | ||
714 | obj_handle_to_location(obj, &f_page, &f_objidx); | |
715 | first_page = get_first_page(f_page); | |
716 | ||
717 | get_zspage_mapping(first_page, &class_idx, &fullness); | |
718 | class = &pool->size_class[class_idx]; | |
719 | f_offset = obj_idx_to_offset(f_page, f_objidx, class->size); | |
720 | ||
721 | spin_lock(&class->lock); | |
722 | ||
723 | /* Insert this object in containing zspage's freelist */ | |
724 | link = (struct link_free *)((unsigned char *)kmap_atomic(f_page) | |
725 | + f_offset); | |
726 | link->next = first_page->freelist; | |
727 | kunmap_atomic(link); | |
c2344348 | 728 | first_page->freelist = (void *)obj; |
61989a80 NG |
729 | |
730 | first_page->inuse--; | |
731 | fullness = fix_fullness_group(pool, first_page); | |
732 | ||
733 | if (fullness == ZS_EMPTY) | |
2e3b6154 | 734 | class->pages_allocated -= class->pages_per_zspage; |
61989a80 NG |
735 | |
736 | spin_unlock(&class->lock); | |
737 | ||
738 | if (fullness == ZS_EMPTY) | |
739 | free_zspage(first_page); | |
740 | } | |
741 | EXPORT_SYMBOL_GPL(zs_free); | |
742 | ||
00a61d86 MK |
743 | /** |
744 | * zs_map_object - get address of allocated object from handle. | |
745 | * @pool: pool from which the object was allocated | |
746 | * @handle: handle returned from zs_malloc | |
747 | * | |
748 | * Before using an object allocated from zs_malloc, it must be mapped using | |
749 | * this function. When done with the object, it must be unmapped using | |
166cfda7 SJ |
750 | * zs_unmap_object. |
751 | * | |
752 | * Only one object can be mapped per cpu at a time. There is no protection | |
753 | * against nested mappings. | |
754 | * | |
755 | * This function returns with preemption and page faults disabled. | |
00a61d86 | 756 | */ |
c2344348 | 757 | void *zs_map_object(struct zs_pool *pool, unsigned long handle) |
61989a80 NG |
758 | { |
759 | struct page *page; | |
760 | unsigned long obj_idx, off; | |
761 | ||
762 | unsigned int class_idx; | |
763 | enum fullness_group fg; | |
764 | struct size_class *class; | |
765 | struct mapping_area *area; | |
766 | ||
767 | BUG_ON(!handle); | |
768 | ||
769 | obj_handle_to_location(handle, &page, &obj_idx); | |
770 | get_zspage_mapping(get_first_page(page), &class_idx, &fg); | |
771 | class = &pool->size_class[class_idx]; | |
772 | off = obj_idx_to_offset(page, obj_idx, class->size); | |
773 | ||
774 | area = &get_cpu_var(zs_map_area); | |
775 | if (off + class->size <= PAGE_SIZE) { | |
776 | /* this object is contained entirely within a page */ | |
777 | area->vm_addr = kmap_atomic(page); | |
5f601902 | 778 | return area->vm_addr + off; |
61989a80 NG |
779 | } |
780 | ||
5f601902 | 781 | zs_copy_map_object(area->vm_buf, page, off, class->size); |
10312330 | 782 | area->vm_addr = NULL; |
5f601902 | 783 | return area->vm_buf; |
61989a80 NG |
784 | } |
785 | EXPORT_SYMBOL_GPL(zs_map_object); | |
786 | ||
c2344348 | 787 | void zs_unmap_object(struct zs_pool *pool, unsigned long handle) |
61989a80 NG |
788 | { |
789 | struct page *page; | |
790 | unsigned long obj_idx, off; | |
791 | ||
792 | unsigned int class_idx; | |
793 | enum fullness_group fg; | |
794 | struct size_class *class; | |
795 | struct mapping_area *area; | |
796 | ||
10312330 SJ |
797 | area = &__get_cpu_var(zs_map_area); |
798 | if (area->vm_addr) { | |
799 | /* single-page object fastpath */ | |
800 | kunmap_atomic(area->vm_addr); | |
801 | put_cpu_var(zs_map_area); | |
802 | return; | |
803 | } | |
804 | ||
61989a80 NG |
805 | BUG_ON(!handle); |
806 | ||
807 | obj_handle_to_location(handle, &page, &obj_idx); | |
808 | get_zspage_mapping(get_first_page(page), &class_idx, &fg); | |
809 | class = &pool->size_class[class_idx]; | |
810 | off = obj_idx_to_offset(page, obj_idx, class->size); | |
811 | ||
10312330 | 812 | zs_copy_unmap_object(area->vm_buf, page, off, class->size); |
61989a80 NG |
813 | put_cpu_var(zs_map_area); |
814 | } | |
815 | EXPORT_SYMBOL_GPL(zs_unmap_object); | |
816 | ||
817 | u64 zs_get_total_size_bytes(struct zs_pool *pool) | |
818 | { | |
819 | int i; | |
820 | u64 npages = 0; | |
821 | ||
822 | for (i = 0; i < ZS_SIZE_CLASSES; i++) | |
823 | npages += pool->size_class[i].pages_allocated; | |
824 | ||
825 | return npages << PAGE_SHIFT; | |
826 | } | |
827 | EXPORT_SYMBOL_GPL(zs_get_total_size_bytes); | |
069f101f BH |
828 | |
829 | module_init(zs_init); | |
830 | module_exit(zs_exit); | |
831 | ||
832 | MODULE_LICENSE("Dual BSD/GPL"); | |
833 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); |