~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/zsmalloc.c

Version: ~ [ linux-5.5-rc1 ] ~ [ linux-5.4.2 ] ~ [ linux-5.3.15 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.88 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.158 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.206 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.206 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.78 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * zsmalloc memory allocator
  3  *
  4  * Copyright (C) 2011  Nitin Gupta
  5  * Copyright (C) 2012, 2013 Minchan Kim
  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 
 14 /*
 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).
 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  *
 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  *
 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 
 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>
 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>
 90 #include <linux/vmalloc.h>
 91 #include <linux/hardirq.h>
 92 #include <linux/spinlock.h>
 93 #include <linux/types.h>
 94 #include <linux/debugfs.h>
 95 #include <linux/zsmalloc.h>
 96 #include <linux/zpool.h>
 97 
 98 /*
 99  * This must be power of 2 and greater than of equal to sizeof(link_free).
100  * These two conditions ensure that any 'struct link_free' itself doesn't
101  * span more than 1 page which avoids complex case of mapping 2 pages simply
102  * to restore link_free pointer values.
103  */
104 #define ZS_ALIGN                8
105 
106 /*
107  * A single 'zspage' is composed of up to 2^N discontiguous 0-order (single)
108  * pages. ZS_MAX_ZSPAGE_ORDER defines upper limit on N.
109  */
110 #define ZS_MAX_ZSPAGE_ORDER 2
111 #define ZS_MAX_PAGES_PER_ZSPAGE (_AC(1, UL) << ZS_MAX_ZSPAGE_ORDER)
112 
113 /*
114  * Object location (<PFN>, <obj_idx>) is encoded as
115  * as single (unsigned long) handle value.
116  *
117  * Note that object index <obj_idx> is relative to system
118  * page <PFN> it is stored in, so for each sub-page belonging
119  * to a zspage, obj_idx starts with 0.
120  *
121  * This is made more complicated by various memory models and PAE.
122  */
123 
124 #ifndef MAX_PHYSMEM_BITS
125 #ifdef CONFIG_HIGHMEM64G
126 #define MAX_PHYSMEM_BITS 36
127 #else /* !CONFIG_HIGHMEM64G */
128 /*
129  * If this definition of MAX_PHYSMEM_BITS is used, OBJ_INDEX_BITS will just
130  * be PAGE_SHIFT
131  */
132 #define MAX_PHYSMEM_BITS BITS_PER_LONG
133 #endif
134 #endif
135 #define _PFN_BITS               (MAX_PHYSMEM_BITS - PAGE_SHIFT)
136 #define OBJ_INDEX_BITS  (BITS_PER_LONG - _PFN_BITS)
137 #define OBJ_INDEX_MASK  ((_AC(1, UL) << OBJ_INDEX_BITS) - 1)
138 
139 #define MAX(a, b) ((a) >= (b) ? (a) : (b))
140 /* ZS_MIN_ALLOC_SIZE must be multiple of ZS_ALIGN */
141 #define ZS_MIN_ALLOC_SIZE \
142         MAX(32, (ZS_MAX_PAGES_PER_ZSPAGE << PAGE_SHIFT >> OBJ_INDEX_BITS))
143 #define ZS_MAX_ALLOC_SIZE       PAGE_SIZE
144 
145 /*
146  * On systems with 4K page size, this gives 255 size classes! There is a
147  * trader-off here:
148  *  - Large number of size classes is potentially wasteful as free page are
149  *    spread across these classes
150  *  - Small number of size classes causes large internal fragmentation
151  *  - Probably its better to use specific size classes (empirically
152  *    determined). NOTE: all those class sizes must be set as multiple of
153  *    ZS_ALIGN to make sure link_free itself never has to span 2 pages.
154  *
155  *  ZS_MIN_ALLOC_SIZE and ZS_SIZE_CLASS_DELTA must be multiple of ZS_ALIGN
156  *  (reason above)
157  */
158 #define ZS_SIZE_CLASS_DELTA     (PAGE_SIZE >> 8)
159 
160 /*
161  * We do not maintain any list for completely empty or full pages
162  */
163 enum fullness_group {
164         ZS_ALMOST_FULL,
165         ZS_ALMOST_EMPTY,
166         _ZS_NR_FULLNESS_GROUPS,
167 
168         ZS_EMPTY,
169         ZS_FULL
170 };
171 
172 enum zs_stat_type {
173         OBJ_ALLOCATED,
174         OBJ_USED,
175         NR_ZS_STAT_TYPE,
176 };
177 
178 #ifdef CONFIG_ZSMALLOC_STAT
179 
180 static struct dentry *zs_stat_root;
181 
182 struct zs_size_stat {
183         unsigned long objs[NR_ZS_STAT_TYPE];
184 };
185 
186 #endif
187 
188 /*
189  * number of size_classes
190  */
191 static int zs_size_classes;
192 
193 /*
194  * We assign a page to ZS_ALMOST_EMPTY fullness group when:
195  *      n <= N / f, where
196  * n = number of allocated objects
197  * N = total number of objects zspage can store
198  * f = fullness_threshold_frac
199  *
200  * Similarly, we assign zspage to:
201  *      ZS_ALMOST_FULL  when n > N / f
202  *      ZS_EMPTY        when n == 0
203  *      ZS_FULL         when n == N
204  *
205  * (see: fix_fullness_group())
206  */
207 static const int fullness_threshold_frac = 4;
208 
209 struct size_class {
210         /*
211          * Size of objects stored in this class. Must be multiple
212          * of ZS_ALIGN.
213          */
214         int size;
215         unsigned int index;
216 
217         /* Number of PAGE_SIZE sized pages to combine to form a 'zspage' */
218         int pages_per_zspage;
219 
220 #ifdef CONFIG_ZSMALLOC_STAT
221         struct zs_size_stat stats;
222 #endif
223 
224         spinlock_t lock;
225 
226         struct page *fullness_list[_ZS_NR_FULLNESS_GROUPS];
227 };
228 
229 /*
230  * Placed within free objects to form a singly linked list.
231  * For every zspage, first_page->freelist gives head of this list.
232  *
233  * This must be power of 2 and less than or equal to ZS_ALIGN
234  */
235 struct link_free {
236         /* Handle of next free chunk (encodes <PFN, obj_idx>) */
237         void *next;
238 };
239 
240 struct zs_pool {
241         char *name;
242 
243         struct size_class **size_class;
244 
245         gfp_t flags;    /* allocation flags used when growing pool */
246         atomic_long_t pages_allocated;
247 
248 #ifdef CONFIG_ZSMALLOC_STAT
249         struct dentry *stat_dentry;
250 #endif
251 };
252 
253 /*
254  * A zspage's class index and fullness group
255  * are encoded in its (first)page->mapping
256  */
257 #define CLASS_IDX_BITS  28
258 #define FULLNESS_BITS   4
259 #define CLASS_IDX_MASK  ((1 << CLASS_IDX_BITS) - 1)
260 #define FULLNESS_MASK   ((1 << FULLNESS_BITS) - 1)
261 
262 struct mapping_area {
263 #ifdef CONFIG_PGTABLE_MAPPING
264         struct vm_struct *vm; /* vm area for mapping object that span pages */
265 #else
266         char *vm_buf; /* copy buffer for objects that span pages */
267 #endif
268         char *vm_addr; /* address of kmap_atomic()'ed pages */
269         enum zs_mapmode vm_mm; /* mapping mode */
270 };
271 
272 /* zpool driver */
273 
274 #ifdef CONFIG_ZPOOL
275 
276 static void *zs_zpool_create(char *name, gfp_t gfp, struct zpool_ops *zpool_ops)
277 {
278         return zs_create_pool(name, gfp);
279 }
280 
281 static void zs_zpool_destroy(void *pool)
282 {
283         zs_destroy_pool(pool);
284 }
285 
286 static int zs_zpool_malloc(void *pool, size_t size, gfp_t gfp,
287                         unsigned long *handle)
288 {
289         *handle = zs_malloc(pool, size);
290         return *handle ? 0 : -1;
291 }
292 static void zs_zpool_free(void *pool, unsigned long handle)
293 {
294         zs_free(pool, handle);
295 }
296 
297 static int zs_zpool_shrink(void *pool, unsigned int pages,
298                         unsigned int *reclaimed)
299 {
300         return -EINVAL;
301 }
302 
303 static void *zs_zpool_map(void *pool, unsigned long handle,
304                         enum zpool_mapmode mm)
305 {
306         enum zs_mapmode zs_mm;
307 
308         switch (mm) {
309         case ZPOOL_MM_RO:
310                 zs_mm = ZS_MM_RO;
311                 break;
312         case ZPOOL_MM_WO:
313                 zs_mm = ZS_MM_WO;
314                 break;
315         case ZPOOL_MM_RW: /* fallthru */
316         default:
317                 zs_mm = ZS_MM_RW;
318                 break;
319         }
320 
321         return zs_map_object(pool, handle, zs_mm);
322 }
323 static void zs_zpool_unmap(void *pool, unsigned long handle)
324 {
325         zs_unmap_object(pool, handle);
326 }
327 
328 static u64 zs_zpool_total_size(void *pool)
329 {
330         return zs_get_total_pages(pool) << PAGE_SHIFT;
331 }
332 
333 static struct zpool_driver zs_zpool_driver = {
334         .type =         "zsmalloc",
335         .owner =        THIS_MODULE,
336         .create =       zs_zpool_create,
337         .destroy =      zs_zpool_destroy,
338         .malloc =       zs_zpool_malloc,
339         .free =         zs_zpool_free,
340         .shrink =       zs_zpool_shrink,
341         .map =          zs_zpool_map,
342         .unmap =        zs_zpool_unmap,
343         .total_size =   zs_zpool_total_size,
344 };
345 
346 MODULE_ALIAS("zpool-zsmalloc");
347 #endif /* CONFIG_ZPOOL */
348 
349 /* per-cpu VM mapping areas for zspage accesses that cross page boundaries */
350 static DEFINE_PER_CPU(struct mapping_area, zs_map_area);
351 
352 static int is_first_page(struct page *page)
353 {
354         return PagePrivate(page);
355 }
356 
357 static int is_last_page(struct page *page)
358 {
359         return PagePrivate2(page);
360 }
361 
362 static void get_zspage_mapping(struct page *page, unsigned int *class_idx,
363                                 enum fullness_group *fullness)
364 {
365         unsigned long m;
366         BUG_ON(!is_first_page(page));
367 
368         m = (unsigned long)page->mapping;
369         *fullness = m & FULLNESS_MASK;
370         *class_idx = (m >> FULLNESS_BITS) & CLASS_IDX_MASK;
371 }
372 
373 static void set_zspage_mapping(struct page *page, unsigned int class_idx,
374                                 enum fullness_group fullness)
375 {
376         unsigned long m;
377         BUG_ON(!is_first_page(page));
378 
379         m = ((class_idx & CLASS_IDX_MASK) << FULLNESS_BITS) |
380                         (fullness & FULLNESS_MASK);
381         page->mapping = (struct address_space *)m;
382 }
383 
384 /*
385  * zsmalloc divides the pool into various size classes where each
386  * class maintains a list of zspages where each zspage is divided
387  * into equal sized chunks. Each allocation falls into one of these
388  * classes depending on its size. This function returns index of the
389  * size class which has chunk size big enough to hold the give size.
390  */
391 static int get_size_class_index(int size)
392 {
393         int idx = 0;
394 
395         if (likely(size > ZS_MIN_ALLOC_SIZE))
396                 idx = DIV_ROUND_UP(size - ZS_MIN_ALLOC_SIZE,
397                                 ZS_SIZE_CLASS_DELTA);
398 
399         return idx;
400 }
401 
402 /*
403  * For each size class, zspages are divided into different groups
404  * depending on how "full" they are. This was done so that we could
405  * easily find empty or nearly empty zspages when we try to shrink
406  * the pool (not yet implemented). This function returns fullness
407  * status of the given page.
408  */
409 static enum fullness_group get_fullness_group(struct page *page)
410 {
411         int inuse, max_objects;
412         enum fullness_group fg;
413         BUG_ON(!is_first_page(page));
414 
415         inuse = page->inuse;
416         max_objects = page->objects;
417 
418         if (inuse == 0)
419                 fg = ZS_EMPTY;
420         else if (inuse == max_objects)
421                 fg = ZS_FULL;
422         else if (inuse <= max_objects / fullness_threshold_frac)
423                 fg = ZS_ALMOST_EMPTY;
424         else
425                 fg = ZS_ALMOST_FULL;
426 
427         return fg;
428 }
429 
430 /*
431  * Each size class maintains various freelists and zspages are assigned
432  * to one of these freelists based on the number of live objects they
433  * have. This functions inserts the given zspage into the freelist
434  * identified by <class, fullness_group>.
435  */
436 static void insert_zspage(struct page *page, struct size_class *class,
437                                 enum fullness_group fullness)
438 {
439         struct page **head;
440 
441         BUG_ON(!is_first_page(page));
442 
443         if (fullness >= _ZS_NR_FULLNESS_GROUPS)
444                 return;
445 
446         head = &class->fullness_list[fullness];
447         if (*head)
448                 list_add_tail(&page->lru, &(*head)->lru);
449 
450         *head = page;
451 }
452 
453 /*
454  * This function removes the given zspage from the freelist identified
455  * by <class, fullness_group>.
456  */
457 static void remove_zspage(struct page *page, struct size_class *class,
458                                 enum fullness_group fullness)
459 {
460         struct page **head;
461 
462         BUG_ON(!is_first_page(page));
463 
464         if (fullness >= _ZS_NR_FULLNESS_GROUPS)
465                 return;
466 
467         head = &class->fullness_list[fullness];
468         BUG_ON(!*head);
469         if (list_empty(&(*head)->lru))
470                 *head = NULL;
471         else if (*head == page)
472                 *head = (struct page *)list_entry((*head)->lru.next,
473                                         struct page, lru);
474 
475         list_del_init(&page->lru);
476 }
477 
478 /*
479  * Each size class maintains zspages in different fullness groups depending
480  * on the number of live objects they contain. When allocating or freeing
481  * objects, the fullness status of the page can change, say, from ALMOST_FULL
482  * to ALMOST_EMPTY when freeing an object. This function checks if such
483  * a status change has occurred for the given page and accordingly moves the
484  * page from the freelist of the old fullness group to that of the new
485  * fullness group.
486  */
487 static enum fullness_group fix_fullness_group(struct zs_pool *pool,
488                                                 struct page *page)
489 {
490         int class_idx;
491         struct size_class *class;
492         enum fullness_group currfg, newfg;
493 
494         BUG_ON(!is_first_page(page));
495 
496         get_zspage_mapping(page, &class_idx, &currfg);
497         newfg = get_fullness_group(page);
498         if (newfg == currfg)
499                 goto out;
500 
501         class = pool->size_class[class_idx];
502         remove_zspage(page, class, currfg);
503         insert_zspage(page, class, newfg);
504         set_zspage_mapping(page, class_idx, newfg);
505 
506 out:
507         return newfg;
508 }
509 
510 /*
511  * We have to decide on how many pages to link together
512  * to form a zspage for each size class. This is important
513  * to reduce wastage due to unusable space left at end of
514  * each zspage which is given as:
515  *      wastage = Zp - Zp % size_class
516  * where Zp = zspage size = k * PAGE_SIZE where k = 1, 2, ...
517  *
518  * For example, for size class of 3/8 * PAGE_SIZE, we should
519  * link together 3 PAGE_SIZE sized pages to form a zspage
520  * since then we can perfectly fit in 8 such objects.
521  */
522 static int get_pages_per_zspage(int class_size)
523 {
524         int i, max_usedpc = 0;
525         /* zspage order which gives maximum used size per KB */
526         int max_usedpc_order = 1;
527 
528         for (i = 1; i <= ZS_MAX_PAGES_PER_ZSPAGE; i++) {
529                 int zspage_size;
530                 int waste, usedpc;
531 
532                 zspage_size = i * PAGE_SIZE;
533                 waste = zspage_size % class_size;
534                 usedpc = (zspage_size - waste) * 100 / zspage_size;
535 
536                 if (usedpc > max_usedpc) {
537                         max_usedpc = usedpc;
538                         max_usedpc_order = i;
539                 }
540         }
541 
542         return max_usedpc_order;
543 }
544 
545 /*
546  * A single 'zspage' is composed of many system pages which are
547  * linked together using fields in struct page. This function finds
548  * the first/head page, given any component page of a zspage.
549  */
550 static struct page *get_first_page(struct page *page)
551 {
552         if (is_first_page(page))
553                 return page;
554         else
555                 return page->first_page;
556 }
557 
558 static struct page *get_next_page(struct page *page)
559 {
560         struct page *next;
561 
562         if (is_last_page(page))
563                 next = NULL;
564         else if (is_first_page(page))
565                 next = (struct page *)page_private(page);
566         else
567                 next = list_entry(page->lru.next, struct page, lru);
568 
569         return next;
570 }
571 
572 /*
573  * Encode <page, obj_idx> as a single handle value.
574  * On hardware platforms with physical memory starting at 0x0 the pfn
575  * could be 0 so we ensure that the handle will never be 0 by adjusting the
576  * encoded obj_idx value before encoding.
577  */
578 static void *obj_location_to_handle(struct page *page, unsigned long obj_idx)
579 {
580         unsigned long handle;
581 
582         if (!page) {
583                 BUG_ON(obj_idx);
584                 return NULL;
585         }
586 
587         handle = page_to_pfn(page) << OBJ_INDEX_BITS;
588         handle |= ((obj_idx + 1) & OBJ_INDEX_MASK);
589 
590         return (void *)handle;
591 }
592 
593 /*
594  * Decode <page, obj_idx> pair from the given object handle. We adjust the
595  * decoded obj_idx back to its original value since it was adjusted in
596  * obj_location_to_handle().
597  */
598 static void obj_handle_to_location(unsigned long handle, struct page **page,
599                                 unsigned long *obj_idx)
600 {
601         *page = pfn_to_page(handle >> OBJ_INDEX_BITS);
602         *obj_idx = (handle & OBJ_INDEX_MASK) - 1;
603 }
604 
605 static unsigned long obj_idx_to_offset(struct page *page,
606                                 unsigned long obj_idx, int class_size)
607 {
608         unsigned long off = 0;
609 
610         if (!is_first_page(page))
611                 off = page->index;
612 
613         return off + obj_idx * class_size;
614 }
615 
616 static void reset_page(struct page *page)
617 {
618         clear_bit(PG_private, &page->flags);
619         clear_bit(PG_private_2, &page->flags);
620         set_page_private(page, 0);
621         page->mapping = NULL;
622         page->freelist = NULL;
623         page_mapcount_reset(page);
624 }
625 
626 static void free_zspage(struct page *first_page)
627 {
628         struct page *nextp, *tmp, *head_extra;
629 
630         BUG_ON(!is_first_page(first_page));
631         BUG_ON(first_page->inuse);
632 
633         head_extra = (struct page *)page_private(first_page);
634 
635         reset_page(first_page);
636         __free_page(first_page);
637 
638         /* zspage with only 1 system page */
639         if (!head_extra)
640                 return;
641 
642         list_for_each_entry_safe(nextp, tmp, &head_extra->lru, lru) {
643                 list_del(&nextp->lru);
644                 reset_page(nextp);
645                 __free_page(nextp);
646         }
647         reset_page(head_extra);
648         __free_page(head_extra);
649 }
650 
651 /* Initialize a newly allocated zspage */
652 static void init_zspage(struct page *first_page, struct size_class *class)
653 {
654         unsigned long off = 0;
655         struct page *page = first_page;
656 
657         BUG_ON(!is_first_page(first_page));
658         while (page) {
659                 struct page *next_page;
660                 struct link_free *link;
661                 unsigned int i = 1;
662                 void *vaddr;
663 
664                 /*
665                  * page->index stores offset of first object starting
666                  * in the page. For the first page, this is always 0,
667                  * so we use first_page->index (aka ->freelist) to store
668                  * head of corresponding zspage's freelist.
669                  */
670                 if (page != first_page)
671                         page->index = off;
672 
673                 vaddr = kmap_atomic(page);
674                 link = (struct link_free *)vaddr + off / sizeof(*link);
675 
676                 while ((off += class->size) < PAGE_SIZE) {
677                         link->next = obj_location_to_handle(page, i++);
678                         link += class->size / sizeof(*link);
679                 }
680 
681                 /*
682                  * We now come to the last (full or partial) object on this
683                  * page, which must point to the first object on the next
684                  * page (if present)
685                  */
686                 next_page = get_next_page(page);
687                 link->next = obj_location_to_handle(next_page, 0);
688                 kunmap_atomic(vaddr);
689                 page = next_page;
690                 off %= PAGE_SIZE;
691         }
692 }
693 
694 /*
695  * Allocate a zspage for the given size class
696  */
697 static struct page *alloc_zspage(struct size_class *class, gfp_t flags)
698 {
699         int i, error;
700         struct page *first_page = NULL, *uninitialized_var(prev_page);
701 
702         /*
703          * Allocate individual pages and link them together as:
704          * 1. first page->private = first sub-page
705          * 2. all sub-pages are linked together using page->lru
706          * 3. each sub-page is linked to the first page using page->first_page
707          *
708          * For each size class, First/Head pages are linked together using
709          * page->lru. Also, we set PG_private to identify the first page
710          * (i.e. no other sub-page has this flag set) and PG_private_2 to
711          * identify the last page.
712          */
713         error = -ENOMEM;
714         for (i = 0; i < class->pages_per_zspage; i++) {
715                 struct page *page;
716 
717                 page = alloc_page(flags);
718                 if (!page)
719                         goto cleanup;
720 
721                 INIT_LIST_HEAD(&page->lru);
722                 if (i == 0) {   /* first page */
723                         SetPagePrivate(page);
724                         set_page_private(page, 0);
725                         first_page = page;
726                         first_page->inuse = 0;
727                 }
728                 if (i == 1)
729                         set_page_private(first_page, (unsigned long)page);
730                 if (i >= 1)
731                         page->first_page = first_page;
732                 if (i >= 2)
733                         list_add(&page->lru, &prev_page->lru);
734                 if (i == class->pages_per_zspage - 1)   /* last page */
735                         SetPagePrivate2(page);
736                 prev_page = page;
737         }
738 
739         init_zspage(first_page, class);
740 
741         first_page->freelist = obj_location_to_handle(first_page, 0);
742         /* Maximum number of objects we can store in this zspage */
743         first_page->objects = class->pages_per_zspage * PAGE_SIZE / class->size;
744 
745         error = 0; /* Success */
746 
747 cleanup:
748         if (unlikely(error) && first_page) {
749                 free_zspage(first_page);
750                 first_page = NULL;
751         }
752 
753         return first_page;
754 }
755 
756 static struct page *find_get_zspage(struct size_class *class)
757 {
758         int i;
759         struct page *page;
760 
761         for (i = 0; i < _ZS_NR_FULLNESS_GROUPS; i++) {
762                 page = class->fullness_list[i];
763                 if (page)
764                         break;
765         }
766 
767         return page;
768 }
769 
770 #ifdef CONFIG_PGTABLE_MAPPING
771 static inline int __zs_cpu_up(struct mapping_area *area)
772 {
773         /*
774          * Make sure we don't leak memory if a cpu UP notification
775          * and zs_init() race and both call zs_cpu_up() on the same cpu
776          */
777         if (area->vm)
778                 return 0;
779         area->vm = alloc_vm_area(PAGE_SIZE * 2, NULL);
780         if (!area->vm)
781                 return -ENOMEM;
782         return 0;
783 }
784 
785 static inline void __zs_cpu_down(struct mapping_area *area)
786 {
787         if (area->vm)
788                 free_vm_area(area->vm);
789         area->vm = NULL;
790 }
791 
792 static inline void *__zs_map_object(struct mapping_area *area,
793                                 struct page *pages[2], int off, int size)
794 {
795         BUG_ON(map_vm_area(area->vm, PAGE_KERNEL, pages));
796         area->vm_addr = area->vm->addr;
797         return area->vm_addr + off;
798 }
799 
800 static inline void __zs_unmap_object(struct mapping_area *area,
801                                 struct page *pages[2], int off, int size)
802 {
803         unsigned long addr = (unsigned long)area->vm_addr;
804 
805         unmap_kernel_range(addr, PAGE_SIZE * 2);
806 }
807 
808 #else /* CONFIG_PGTABLE_MAPPING */
809 
810 static inline int __zs_cpu_up(struct mapping_area *area)
811 {
812         /*
813          * Make sure we don't leak memory if a cpu UP notification
814          * and zs_init() race and both call zs_cpu_up() on the same cpu
815          */
816         if (area->vm_buf)
817                 return 0;
818         area->vm_buf = kmalloc(ZS_MAX_ALLOC_SIZE, GFP_KERNEL);
819         if (!area->vm_buf)
820                 return -ENOMEM;
821         return 0;
822 }
823 
824 static inline void __zs_cpu_down(struct mapping_area *area)
825 {
826         kfree(area->vm_buf);
827         area->vm_buf = NULL;
828 }
829 
830 static void *__zs_map_object(struct mapping_area *area,
831                         struct page *pages[2], int off, int size)
832 {
833         int sizes[2];
834         void *addr;
835         char *buf = area->vm_buf;
836 
837         /* disable page faults to match kmap_atomic() return conditions */
838         pagefault_disable();
839 
840         /* no read fastpath */
841         if (area->vm_mm == ZS_MM_WO)
842                 goto out;
843 
844         sizes[0] = PAGE_SIZE - off;
845         sizes[1] = size - sizes[0];
846 
847         /* copy object to per-cpu buffer */
848         addr = kmap_atomic(pages[0]);
849         memcpy(buf, addr + off, sizes[0]);
850         kunmap_atomic(addr);
851         addr = kmap_atomic(pages[1]);
852         memcpy(buf + sizes[0], addr, sizes[1]);
853         kunmap_atomic(addr);
854 out:
855         return area->vm_buf;
856 }
857 
858 static void __zs_unmap_object(struct mapping_area *area,
859                         struct page *pages[2], int off, int size)
860 {
861         int sizes[2];
862         void *addr;
863         char *buf = area->vm_buf;
864 
865         /* no write fastpath */
866         if (area->vm_mm == ZS_MM_RO)
867                 goto out;
868 
869         sizes[0] = PAGE_SIZE - off;
870         sizes[1] = size - sizes[0];
871 
872         /* copy per-cpu buffer to object */
873         addr = kmap_atomic(pages[0]);
874         memcpy(addr + off, buf, sizes[0]);
875         kunmap_atomic(addr);
876         addr = kmap_atomic(pages[1]);
877         memcpy(addr, buf + sizes[0], sizes[1]);
878         kunmap_atomic(addr);
879 
880 out:
881         /* enable page faults to match kunmap_atomic() return conditions */
882         pagefault_enable();
883 }
884 
885 #endif /* CONFIG_PGTABLE_MAPPING */
886 
887 static int zs_cpu_notifier(struct notifier_block *nb, unsigned long action,
888                                 void *pcpu)
889 {
890         int ret, cpu = (long)pcpu;
891         struct mapping_area *area;
892 
893         switch (action) {
894         case CPU_UP_PREPARE:
895                 area = &per_cpu(zs_map_area, cpu);
896                 ret = __zs_cpu_up(area);
897                 if (ret)
898                         return notifier_from_errno(ret);
899                 break;
900         case CPU_DEAD:
901         case CPU_UP_CANCELED:
902                 area = &per_cpu(zs_map_area, cpu);
903                 __zs_cpu_down(area);
904                 break;
905         }
906 
907         return NOTIFY_OK;
908 }
909 
910 static struct notifier_block zs_cpu_nb = {
911         .notifier_call = zs_cpu_notifier
912 };
913 
914 static int zs_register_cpu_notifier(void)
915 {
916         int cpu, uninitialized_var(ret);
917 
918         cpu_notifier_register_begin();
919 
920         __register_cpu_notifier(&zs_cpu_nb);
921         for_each_online_cpu(cpu) {
922                 ret = zs_cpu_notifier(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
923                 if (notifier_to_errno(ret))
924                         break;
925         }
926 
927         cpu_notifier_register_done();
928         return notifier_to_errno(ret);
929 }
930 
931 static void zs_unregister_cpu_notifier(void)
932 {
933         int cpu;
934 
935         cpu_notifier_register_begin();
936 
937         for_each_online_cpu(cpu)
938                 zs_cpu_notifier(NULL, CPU_DEAD, (void *)(long)cpu);
939         __unregister_cpu_notifier(&zs_cpu_nb);
940 
941         cpu_notifier_register_done();
942 }
943 
944 static void init_zs_size_classes(void)
945 {
946         int nr;
947 
948         nr = (ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) / ZS_SIZE_CLASS_DELTA + 1;
949         if ((ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) % ZS_SIZE_CLASS_DELTA)
950                 nr += 1;
951 
952         zs_size_classes = nr;
953 }
954 
955 static unsigned int get_maxobj_per_zspage(int size, int pages_per_zspage)
956 {
957         return pages_per_zspage * PAGE_SIZE / size;
958 }
959 
960 static bool can_merge(struct size_class *prev, int size, int pages_per_zspage)
961 {
962         if (prev->pages_per_zspage != pages_per_zspage)
963                 return false;
964 
965         if (get_maxobj_per_zspage(prev->size, prev->pages_per_zspage)
966                 != get_maxobj_per_zspage(size, pages_per_zspage))
967                 return false;
968 
969         return true;
970 }
971 
972 #ifdef CONFIG_ZSMALLOC_STAT
973 
974 static inline void zs_stat_inc(struct size_class *class,
975                                 enum zs_stat_type type, unsigned long cnt)
976 {
977         class->stats.objs[type] += cnt;
978 }
979 
980 static inline void zs_stat_dec(struct size_class *class,
981                                 enum zs_stat_type type, unsigned long cnt)
982 {
983         class->stats.objs[type] -= cnt;
984 }
985 
986 static inline unsigned long zs_stat_get(struct size_class *class,
987                                 enum zs_stat_type type)
988 {
989         return class->stats.objs[type];
990 }
991 
992 static int __init zs_stat_init(void)
993 {
994         if (!debugfs_initialized())
995                 return -ENODEV;
996 
997         zs_stat_root = debugfs_create_dir("zsmalloc", NULL);
998         if (!zs_stat_root)
999                 return -ENOMEM;
1000 
1001         return 0;
1002 }
1003 
1004 static void __exit zs_stat_exit(void)
1005 {
1006         debugfs_remove_recursive(zs_stat_root);
1007 }
1008 
1009 static int zs_stats_size_show(struct seq_file *s, void *v)
1010 {
1011         int i;
1012         struct zs_pool *pool = s->private;
1013         struct size_class *class;
1014         int objs_per_zspage;
1015         unsigned long obj_allocated, obj_used, pages_used;
1016         unsigned long total_objs = 0, total_used_objs = 0, total_pages = 0;
1017 
1018         seq_printf(s, " %5s %5s %13s %10s %10s\n", "class", "size",
1019                                 "obj_allocated", "obj_used", "pages_used");
1020 
1021         for (i = 0; i < zs_size_classes; i++) {
1022                 class = pool->size_class[i];
1023 
1024                 if (class->index != i)
1025                         continue;
1026 
1027                 spin_lock(&class->lock);
1028                 obj_allocated = zs_stat_get(class, OBJ_ALLOCATED);
1029                 obj_used = zs_stat_get(class, OBJ_USED);
1030                 spin_unlock(&class->lock);
1031 
1032                 objs_per_zspage = get_maxobj_per_zspage(class->size,
1033                                 class->pages_per_zspage);
1034                 pages_used = obj_allocated / objs_per_zspage *
1035                                 class->pages_per_zspage;
1036 
1037                 seq_printf(s, " %5u %5u    %10lu %10lu %10lu\n", i,
1038                         class->size, obj_allocated, obj_used, pages_used);
1039 
1040                 total_objs += obj_allocated;
1041                 total_used_objs += obj_used;
1042                 total_pages += pages_used;
1043         }
1044 
1045         seq_puts(s, "\n");
1046         seq_printf(s, " %5s %5s    %10lu %10lu %10lu\n", "Total", "",
1047                         total_objs, total_used_objs, total_pages);
1048 
1049         return 0;
1050 }
1051 
1052 static int zs_stats_size_open(struct inode *inode, struct file *file)
1053 {
1054         return single_open(file, zs_stats_size_show, inode->i_private);
1055 }
1056 
1057 static const struct file_operations zs_stat_size_ops = {
1058         .open           = zs_stats_size_open,
1059         .read           = seq_read,
1060         .llseek         = seq_lseek,
1061         .release        = single_release,
1062 };
1063 
1064 static int zs_pool_stat_create(char *name, struct zs_pool *pool)
1065 {
1066         struct dentry *entry;
1067 
1068         if (!zs_stat_root)
1069                 return -ENODEV;
1070 
1071         entry = debugfs_create_dir(name, zs_stat_root);
1072         if (!entry) {
1073                 pr_warn("debugfs dir <%s> creation failed\n", name);
1074                 return -ENOMEM;
1075         }
1076         pool->stat_dentry = entry;
1077 
1078         entry = debugfs_create_file("obj_in_classes", S_IFREG | S_IRUGO,
1079                         pool->stat_dentry, pool, &zs_stat_size_ops);
1080         if (!entry) {
1081                 pr_warn("%s: debugfs file entry <%s> creation failed\n",
1082                                 name, "obj_in_classes");
1083                 return -ENOMEM;
1084         }
1085 
1086         return 0;
1087 }
1088 
1089 static void zs_pool_stat_destroy(struct zs_pool *pool)
1090 {
1091         debugfs_remove_recursive(pool->stat_dentry);
1092 }
1093 
1094 #else /* CONFIG_ZSMALLOC_STAT */
1095 
1096 static inline void zs_stat_inc(struct size_class *class,
1097                                 enum zs_stat_type type, unsigned long cnt)
1098 {
1099 }
1100 
1101 static inline void zs_stat_dec(struct size_class *class,
1102                                 enum zs_stat_type type, unsigned long cnt)
1103 {
1104 }
1105 
1106 static inline unsigned long zs_stat_get(struct size_class *class,
1107                                 enum zs_stat_type type)
1108 {
1109         return 0;
1110 }
1111 
1112 static int __init zs_stat_init(void)
1113 {
1114         return 0;
1115 }
1116 
1117 static void __exit zs_stat_exit(void)
1118 {
1119 }
1120 
1121 static inline int zs_pool_stat_create(char *name, struct zs_pool *pool)
1122 {
1123         return 0;
1124 }
1125 
1126 static inline void zs_pool_stat_destroy(struct zs_pool *pool)
1127 {
1128 }
1129 
1130 #endif
1131 
1132 unsigned long zs_get_total_pages(struct zs_pool *pool)
1133 {
1134         return atomic_long_read(&pool->pages_allocated);
1135 }
1136 EXPORT_SYMBOL_GPL(zs_get_total_pages);
1137 
1138 /**
1139  * zs_map_object - get address of allocated object from handle.
1140  * @pool: pool from which the object was allocated
1141  * @handle: handle returned from zs_malloc
1142  *
1143  * Before using an object allocated from zs_malloc, it must be mapped using
1144  * this function. When done with the object, it must be unmapped using
1145  * zs_unmap_object.
1146  *
1147  * Only one object can be mapped per cpu at a time. There is no protection
1148  * against nested mappings.
1149  *
1150  * This function returns with preemption and page faults disabled.
1151  */
1152 void *zs_map_object(struct zs_pool *pool, unsigned long handle,
1153                         enum zs_mapmode mm)
1154 {
1155         struct page *page;
1156         unsigned long obj_idx, off;
1157 
1158         unsigned int class_idx;
1159         enum fullness_group fg;
1160         struct size_class *class;
1161         struct mapping_area *area;
1162         struct page *pages[2];
1163 
1164         BUG_ON(!handle);
1165 
1166         /*
1167          * Because we use per-cpu mapping areas shared among the
1168          * pools/users, we can't allow mapping in interrupt context
1169          * because it can corrupt another users mappings.
1170          */
1171         BUG_ON(in_interrupt());
1172 
1173         obj_handle_to_location(handle, &page, &obj_idx);
1174         get_zspage_mapping(get_first_page(page), &class_idx, &fg);
1175         class = pool->size_class[class_idx];
1176         off = obj_idx_to_offset(page, obj_idx, class->size);
1177 
1178         area = &get_cpu_var(zs_map_area);
1179         area->vm_mm = mm;
1180         if (off + class->size <= PAGE_SIZE) {
1181                 /* this object is contained entirely within a page */
1182                 area->vm_addr = kmap_atomic(page);
1183                 return area->vm_addr + off;
1184         }
1185 
1186         /* this object spans two pages */
1187         pages[0] = page;
1188         pages[1] = get_next_page(page);
1189         BUG_ON(!pages[1]);
1190 
1191         return __zs_map_object(area, pages, off, class->size);
1192 }
1193 EXPORT_SYMBOL_GPL(zs_map_object);
1194 
1195 void zs_unmap_object(struct zs_pool *pool, unsigned long handle)
1196 {
1197         struct page *page;
1198         unsigned long obj_idx, off;
1199 
1200         unsigned int class_idx;
1201         enum fullness_group fg;
1202         struct size_class *class;
1203         struct mapping_area *area;
1204 
1205         BUG_ON(!handle);
1206 
1207         obj_handle_to_location(handle, &page, &obj_idx);
1208         get_zspage_mapping(get_first_page(page), &class_idx, &fg);
1209         class = pool->size_class[class_idx];
1210         off = obj_idx_to_offset(page, obj_idx, class->size);
1211 
1212         area = this_cpu_ptr(&zs_map_area);
1213         if (off + class->size <= PAGE_SIZE)
1214                 kunmap_atomic(area->vm_addr);
1215         else {
1216                 struct page *pages[2];
1217 
1218                 pages[0] = page;
1219                 pages[1] = get_next_page(page);
1220                 BUG_ON(!pages[1]);
1221 
1222                 __zs_unmap_object(area, pages, off, class->size);
1223         }
1224         put_cpu_var(zs_map_area);
1225 }
1226 EXPORT_SYMBOL_GPL(zs_unmap_object);
1227 
1228 /**
1229  * zs_malloc - Allocate block of given size from pool.
1230  * @pool: pool to allocate from
1231  * @size: size of block to allocate
1232  *
1233  * On success, handle to the allocated object is returned,
1234  * otherwise 0.
1235  * Allocation requests with size > ZS_MAX_ALLOC_SIZE will fail.
1236  */
1237 unsigned long zs_malloc(struct zs_pool *pool, size_t size)
1238 {
1239         unsigned long obj;
1240         struct link_free *link;
1241         struct size_class *class;
1242         void *vaddr;
1243 
1244         struct page *first_page, *m_page;
1245         unsigned long m_objidx, m_offset;
1246 
1247         if (unlikely(!size || size > ZS_MAX_ALLOC_SIZE))
1248                 return 0;
1249 
1250         class = pool->size_class[get_size_class_index(size)];
1251 
1252         spin_lock(&class->lock);
1253         first_page = find_get_zspage(class);
1254 
1255         if (!first_page) {
1256                 spin_unlock(&class->lock);
1257                 first_page = alloc_zspage(class, pool->flags);
1258                 if (unlikely(!first_page))
1259                         return 0;
1260 
1261                 set_zspage_mapping(first_page, class->index, ZS_EMPTY);
1262                 atomic_long_add(class->pages_per_zspage,
1263                                         &pool->pages_allocated);
1264 
1265                 spin_lock(&class->lock);
1266                 zs_stat_inc(class, OBJ_ALLOCATED, get_maxobj_per_zspage(
1267                                 class->size, class->pages_per_zspage));
1268         }
1269 
1270         obj = (unsigned long)first_page->freelist;
1271         obj_handle_to_location(obj, &m_page, &m_objidx);
1272         m_offset = obj_idx_to_offset(m_page, m_objidx, class->size);
1273 
1274         vaddr = kmap_atomic(m_page);
1275         link = (struct link_free *)vaddr + m_offset / sizeof(*link);
1276         first_page->freelist = link->next;
1277         memset(link, POISON_INUSE, sizeof(*link));
1278         kunmap_atomic(vaddr);
1279 
1280         first_page->inuse++;
1281         zs_stat_inc(class, OBJ_USED, 1);
1282         /* Now move the zspage to another fullness group, if required */
1283         fix_fullness_group(pool, first_page);
1284         spin_unlock(&class->lock);
1285 
1286         return obj;
1287 }
1288 EXPORT_SYMBOL_GPL(zs_malloc);
1289 
1290 void zs_free(struct zs_pool *pool, unsigned long obj)
1291 {
1292         struct link_free *link;
1293         struct page *first_page, *f_page;
1294         unsigned long f_objidx, f_offset;
1295         void *vaddr;
1296 
1297         int class_idx;
1298         struct size_class *class;
1299         enum fullness_group fullness;
1300 
1301         if (unlikely(!obj))
1302                 return;
1303 
1304         obj_handle_to_location(obj, &f_page, &f_objidx);
1305         first_page = get_first_page(f_page);
1306 
1307         get_zspage_mapping(first_page, &class_idx, &fullness);
1308         class = pool->size_class[class_idx];
1309         f_offset = obj_idx_to_offset(f_page, f_objidx, class->size);
1310 
1311         spin_lock(&class->lock);
1312 
1313         /* Insert this object in containing zspage's freelist */
1314         vaddr = kmap_atomic(f_page);
1315         link = (struct link_free *)(vaddr + f_offset);
1316         link->next = first_page->freelist;
1317         kunmap_atomic(vaddr);
1318         first_page->freelist = (void *)obj;
1319 
1320         first_page->inuse--;
1321         fullness = fix_fullness_group(pool, first_page);
1322 
1323         zs_stat_dec(class, OBJ_USED, 1);
1324         if (fullness == ZS_EMPTY)
1325                 zs_stat_dec(class, OBJ_ALLOCATED, get_maxobj_per_zspage(
1326                                 class->size, class->pages_per_zspage));
1327 
1328         spin_unlock(&class->lock);
1329 
1330         if (fullness == ZS_EMPTY) {
1331                 atomic_long_sub(class->pages_per_zspage,
1332                                 &pool->pages_allocated);
1333                 free_zspage(first_page);
1334         }
1335 }
1336 EXPORT_SYMBOL_GPL(zs_free);
1337 
1338 /**
1339  * zs_create_pool - Creates an allocation pool to work from.
1340  * @flags: allocation flags used to allocate pool metadata
1341  *
1342  * This function must be called before anything when using
1343  * the zsmalloc allocator.
1344  *
1345  * On success, a pointer to the newly created pool is returned,
1346  * otherwise NULL.
1347  */
1348 struct zs_pool *zs_create_pool(char *name, gfp_t flags)
1349 {
1350         int i;
1351         struct zs_pool *pool;
1352         struct size_class *prev_class = NULL;
1353 
1354         pool = kzalloc(sizeof(*pool), GFP_KERNEL);
1355         if (!pool)
1356                 return NULL;
1357 
1358         pool->name = kstrdup(name, GFP_KERNEL);
1359         if (!pool->name) {
1360                 kfree(pool);
1361                 return NULL;
1362         }
1363 
1364         pool->size_class = kcalloc(zs_size_classes, sizeof(struct size_class *),
1365                         GFP_KERNEL);
1366         if (!pool->size_class) {
1367                 kfree(pool->name);
1368                 kfree(pool);
1369                 return NULL;
1370         }
1371 
1372         /*
1373          * Iterate reversly, because, size of size_class that we want to use
1374          * for merging should be larger or equal to current size.
1375          */
1376         for (i = zs_size_classes - 1; i >= 0; i--) {
1377                 int size;
1378                 int pages_per_zspage;
1379                 struct size_class *class;
1380 
1381                 size = ZS_MIN_ALLOC_SIZE + i * ZS_SIZE_CLASS_DELTA;
1382                 if (size > ZS_MAX_ALLOC_SIZE)
1383                         size = ZS_MAX_ALLOC_SIZE;
1384                 pages_per_zspage = get_pages_per_zspage(size);
1385 
1386                 /*
1387                  * size_class is used for normal zsmalloc operation such
1388                  * as alloc/free for that size. Although it is natural that we
1389                  * have one size_class for each size, there is a chance that we
1390                  * can get more memory utilization if we use one size_class for
1391                  * many different sizes whose size_class have same
1392                  * characteristics. So, we makes size_class point to
1393                  * previous size_class if possible.
1394                  */
1395                 if (prev_class) {
1396                         if (can_merge(prev_class, size, pages_per_zspage)) {
1397                                 pool->size_class[i] = prev_class;
1398                                 continue;
1399                         }
1400                 }
1401 
1402                 class = kzalloc(sizeof(struct size_class), GFP_KERNEL);
1403                 if (!class)
1404                         goto err;
1405 
1406                 class->size = size;
1407                 class->index = i;
1408                 class->pages_per_zspage = pages_per_zspage;
1409                 spin_lock_init(&class->lock);
1410                 pool->size_class[i] = class;
1411 
1412                 prev_class = class;
1413         }
1414 
1415         pool->flags = flags;
1416 
1417         if (zs_pool_stat_create(name, pool))
1418                 goto err;
1419 
1420         return pool;
1421 
1422 err:
1423         zs_destroy_pool(pool);
1424         return NULL;
1425 }
1426 EXPORT_SYMBOL_GPL(zs_create_pool);
1427 
1428 void zs_destroy_pool(struct zs_pool *pool)
1429 {
1430         int i;
1431 
1432         zs_pool_stat_destroy(pool);
1433 
1434         for (i = 0; i < zs_size_classes; i++) {
1435                 int fg;
1436                 struct size_class *class = pool->size_class[i];
1437 
1438                 if (!class)
1439                         continue;
1440 
1441                 if (class->index != i)
1442                         continue;
1443 
1444                 for (fg = 0; fg < _ZS_NR_FULLNESS_GROUPS; fg++) {
1445                         if (class->fullness_list[fg]) {
1446                                 pr_info("Freeing non-empty class with size %db, fullness group %d\n",
1447                                         class->size, fg);
1448                         }
1449                 }
1450                 kfree(class);
1451         }
1452 
1453         kfree(pool->size_class);
1454         kfree(pool->name);
1455         kfree(pool);
1456 }
1457 EXPORT_SYMBOL_GPL(zs_destroy_pool);
1458 
1459 static int __init zs_init(void)
1460 {
1461         int ret = zs_register_cpu_notifier();
1462 
1463         if (ret)
1464                 goto notifier_fail;
1465 
1466         init_zs_size_classes();
1467 
1468 #ifdef CONFIG_ZPOOL
1469         zpool_register_driver(&zs_zpool_driver);
1470 #endif
1471 
1472         ret = zs_stat_init();
1473         if (ret) {
1474                 pr_err("zs stat initialization failed\n");
1475                 goto stat_fail;
1476         }
1477         return 0;
1478 
1479 stat_fail:
1480 #ifdef CONFIG_ZPOOL
1481         zpool_unregister_driver(&zs_zpool_driver);
1482 #endif
1483 notifier_fail:
1484         zs_unregister_cpu_notifier();
1485 
1486         return ret;
1487 }
1488 
1489 static void __exit zs_exit(void)
1490 {
1491 #ifdef CONFIG_ZPOOL
1492         zpool_unregister_driver(&zs_zpool_driver);
1493 #endif
1494         zs_unregister_cpu_notifier();
1495 
1496         zs_stat_exit();
1497 }
1498 
1499 module_init(zs_init);
1500 module_exit(zs_exit);
1501 
1502 MODULE_LICENSE("Dual BSD/GPL");
1503 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1504 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp