2 * zsmalloc memory allocator
4 * Copyright (C) 2011 Nitin Gupta
6 * This code is released using a dual license strategy: BSD/GPL
7 * You can choose the license that better fits your requirements.
9 * Released under the terms of 3-clause BSD License
10 * Released under the terms of GNU General Public License Version 2.0
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
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
30 * Following is how we use various fields and flags of underlying
31 * struct page(s) to form a zspage.
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)
42 * For _first_ page only:
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
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
55 * Usage of struct page flags:
56 * PG_private: identifies the first component page
57 * PG_private2: identifies the last component page
61 #ifdef CONFIG_ZSMALLOC_DEBUG
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>
77 #include <linux/vmalloc.h>
78 #include <linux/hardirq.h>
79 #include <linux/spinlock.h>
80 #include <linux/types.h>
85 * This must be power of 2 and greater than of equal to sizeof(link_free).
86 * These two conditions ensure that any 'struct link_free' itself doesn't
87 * span more than 1 page which avoids complex case of mapping 2 pages simply
88 * to restore link_free pointer values.
93 * A single 'zspage' is composed of up to 2^N discontiguous 0-order (single)
94 * pages. ZS_MAX_ZSPAGE_ORDER defines upper limit on N.
96 #define ZS_MAX_ZSPAGE_ORDER 2
97 #define ZS_MAX_PAGES_PER_ZSPAGE (_AC(1, UL) << ZS_MAX_ZSPAGE_ORDER)
100 * Object location (<PFN>, <obj_idx>) is encoded as
101 * as single (void *) handle value.
103 * Note that object index <obj_idx> is relative to system
104 * page <PFN> it is stored in, so for each sub-page belonging
105 * to a zspage, obj_idx starts with 0.
107 * This is made more complicated by various memory models and PAE.
110 #ifndef MAX_PHYSMEM_BITS
111 #ifdef CONFIG_HIGHMEM64G
112 #define MAX_PHYSMEM_BITS 36
113 #else /* !CONFIG_HIGHMEM64G */
115 * If this definition of MAX_PHYSMEM_BITS is used, OBJ_INDEX_BITS will just
118 #define MAX_PHYSMEM_BITS BITS_PER_LONG
121 #define _PFN_BITS (MAX_PHYSMEM_BITS - PAGE_SHIFT)
122 #define OBJ_INDEX_BITS (BITS_PER_LONG - _PFN_BITS)
123 #define OBJ_INDEX_MASK ((_AC(1, UL) << OBJ_INDEX_BITS) - 1)
125 #define MAX(a, b) ((a) >= (b) ? (a) : (b))
126 /* ZS_MIN_ALLOC_SIZE must be multiple of ZS_ALIGN */
127 #define ZS_MIN_ALLOC_SIZE \
128 MAX(32, (ZS_MAX_PAGES_PER_ZSPAGE << PAGE_SHIFT >> OBJ_INDEX_BITS))
129 #define ZS_MAX_ALLOC_SIZE PAGE_SIZE
132 * On systems with 4K page size, this gives 254 size classes! There is a
134 * - Large number of size classes is potentially wasteful as free page are
135 * spread across these classes
136 * - Small number of size classes causes large internal fragmentation
137 * - Probably its better to use specific size classes (empirically
138 * determined). NOTE: all those class sizes must be set as multiple of
139 * ZS_ALIGN to make sure link_free itself never has to span 2 pages.
141 * ZS_MIN_ALLOC_SIZE and ZS_SIZE_CLASS_DELTA must be multiple of ZS_ALIGN
144 #define ZS_SIZE_CLASS_DELTA 16
145 #define ZS_SIZE_CLASSES ((ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) / \
146 ZS_SIZE_CLASS_DELTA + 1)
149 * We do not maintain any list for completely empty or full pages
151 enum fullness_group {
154 _ZS_NR_FULLNESS_GROUPS,
161 * We assign a page to ZS_ALMOST_EMPTY fullness group when:
163 * n = number of allocated objects
164 * N = total number of objects zspage can store
165 * f = 1/fullness_threshold_frac
167 * Similarly, we assign zspage to:
168 * ZS_ALMOST_FULL when n > N / f
169 * ZS_EMPTY when n == 0
170 * ZS_FULL when n == N
172 * (see: fix_fullness_group())
174 static const int fullness_threshold_frac = 4;
178 * Size of objects stored in this class. Must be multiple
184 /* Number of PAGE_SIZE sized pages to combine to form a 'zspage' */
185 int pages_per_zspage;
192 struct page *fullness_list[_ZS_NR_FULLNESS_GROUPS];
196 * Placed within free objects to form a singly linked list.
197 * For every zspage, first_page->freelist gives head of this list.
199 * This must be power of 2 and less than or equal to ZS_ALIGN
202 /* Handle of next free chunk (encodes <PFN, obj_idx>) */
207 struct size_class size_class[ZS_SIZE_CLASSES];
209 gfp_t flags; /* allocation flags used when growing pool */
214 * A zspage's class index and fullness group
215 * are encoded in its (first)page->mapping
217 #define CLASS_IDX_BITS 28
218 #define FULLNESS_BITS 4
219 #define CLASS_IDX_MASK ((1 << CLASS_IDX_BITS) - 1)
220 #define FULLNESS_MASK ((1 << FULLNESS_BITS) - 1)
223 * By default, zsmalloc uses a copy-based object mapping method to access
224 * allocations that span two pages. However, if a particular architecture
225 * 1) Implements local_flush_tlb_kernel_range() and 2) Performs VM mapping
226 * faster than copying, then it should be added here so that
227 * USE_PGTABLE_MAPPING is defined. This causes zsmalloc to use page table
228 * mapping rather than copying
229 * for object mapping.
231 #if defined(CONFIG_ARM)
232 #define USE_PGTABLE_MAPPING
235 struct mapping_area {
236 #ifdef USE_PGTABLE_MAPPING
237 struct vm_struct *vm; /* vm area for mapping object that span pages */
239 char *vm_buf; /* copy buffer for objects that span pages */
241 char *vm_addr; /* address of kmap_atomic()'ed pages */
242 enum zs_mapmode vm_mm; /* mapping mode */
246 /* per-cpu VM mapping areas for zspage accesses that cross page boundaries */
247 static DEFINE_PER_CPU(struct mapping_area, zs_map_area);
249 static int is_first_page(struct page *page)
251 return PagePrivate(page);
254 static int is_last_page(struct page *page)
256 return PagePrivate2(page);
259 static void get_zspage_mapping(struct page *page, unsigned int *class_idx,
260 enum fullness_group *fullness)
263 BUG_ON(!is_first_page(page));
265 m = (unsigned long)page->mapping;
266 *fullness = m & FULLNESS_MASK;
267 *class_idx = (m >> FULLNESS_BITS) & CLASS_IDX_MASK;
270 static void set_zspage_mapping(struct page *page, unsigned int class_idx,
271 enum fullness_group fullness)
274 BUG_ON(!is_first_page(page));
276 m = ((class_idx & CLASS_IDX_MASK) << FULLNESS_BITS) |
277 (fullness & FULLNESS_MASK);
278 page->mapping = (struct address_space *)m;
281 static int get_size_class_index(int size)
285 if (likely(size > ZS_MIN_ALLOC_SIZE))
286 idx = DIV_ROUND_UP(size - ZS_MIN_ALLOC_SIZE,
287 ZS_SIZE_CLASS_DELTA);
292 static enum fullness_group get_fullness_group(struct page *page)
294 int inuse, max_objects;
295 enum fullness_group fg;
296 BUG_ON(!is_first_page(page));
299 max_objects = page->objects;
303 else if (inuse == max_objects)
305 else if (inuse <= max_objects / fullness_threshold_frac)
306 fg = ZS_ALMOST_EMPTY;
313 static void insert_zspage(struct page *page, struct size_class *class,
314 enum fullness_group fullness)
318 BUG_ON(!is_first_page(page));
320 if (fullness >= _ZS_NR_FULLNESS_GROUPS)
323 head = &class->fullness_list[fullness];
325 list_add_tail(&page->lru, &(*head)->lru);
330 static void remove_zspage(struct page *page, struct size_class *class,
331 enum fullness_group fullness)
335 BUG_ON(!is_first_page(page));
337 if (fullness >= _ZS_NR_FULLNESS_GROUPS)
340 head = &class->fullness_list[fullness];
342 if (list_empty(&(*head)->lru))
344 else if (*head == page)
345 *head = (struct page *)list_entry((*head)->lru.next,
348 list_del_init(&page->lru);
351 static enum fullness_group fix_fullness_group(struct zs_pool *pool,
355 struct size_class *class;
356 enum fullness_group currfg, newfg;
358 BUG_ON(!is_first_page(page));
360 get_zspage_mapping(page, &class_idx, &currfg);
361 newfg = get_fullness_group(page);
365 class = &pool->size_class[class_idx];
366 remove_zspage(page, class, currfg);
367 insert_zspage(page, class, newfg);
368 set_zspage_mapping(page, class_idx, newfg);
375 * We have to decide on how many pages to link together
376 * to form a zspage for each size class. This is important
377 * to reduce wastage due to unusable space left at end of
378 * each zspage which is given as:
379 * wastage = Zp - Zp % size_class
380 * where Zp = zspage size = k * PAGE_SIZE where k = 1, 2, ...
382 * For example, for size class of 3/8 * PAGE_SIZE, we should
383 * link together 3 PAGE_SIZE sized pages to form a zspage
384 * since then we can perfectly fit in 8 such objects.
386 static int get_pages_per_zspage(int class_size)
388 int i, max_usedpc = 0;
389 /* zspage order which gives maximum used size per KB */
390 int max_usedpc_order = 1;
392 for (i = 1; i <= ZS_MAX_PAGES_PER_ZSPAGE; i++) {
396 zspage_size = i * PAGE_SIZE;
397 waste = zspage_size % class_size;
398 usedpc = (zspage_size - waste) * 100 / zspage_size;
400 if (usedpc > max_usedpc) {
402 max_usedpc_order = i;
406 return max_usedpc_order;
410 * A single 'zspage' is composed of many system pages which are
411 * linked together using fields in struct page. This function finds
412 * the first/head page, given any component page of a zspage.
414 static struct page *get_first_page(struct page *page)
416 if (is_first_page(page))
419 return page->first_page;
422 static struct page *get_next_page(struct page *page)
426 if (is_last_page(page))
428 else if (is_first_page(page))
429 next = (struct page *)page->private;
431 next = list_entry(page->lru.next, struct page, lru);
436 /* Encode <page, obj_idx> as a single handle value */
437 static void *obj_location_to_handle(struct page *page, unsigned long obj_idx)
439 unsigned long handle;
446 handle = page_to_pfn(page) << OBJ_INDEX_BITS;
447 handle |= (obj_idx & OBJ_INDEX_MASK);
449 return (void *)handle;
452 /* Decode <page, obj_idx> pair from the given object handle */
453 static void obj_handle_to_location(unsigned long handle, struct page **page,
454 unsigned long *obj_idx)
456 *page = pfn_to_page(handle >> OBJ_INDEX_BITS);
457 *obj_idx = handle & OBJ_INDEX_MASK;
460 static unsigned long obj_idx_to_offset(struct page *page,
461 unsigned long obj_idx, int class_size)
463 unsigned long off = 0;
465 if (!is_first_page(page))
468 return off + obj_idx * class_size;
471 static void reset_page(struct page *page)
473 clear_bit(PG_private, &page->flags);
474 clear_bit(PG_private_2, &page->flags);
475 set_page_private(page, 0);
476 page->mapping = NULL;
477 page->freelist = NULL;
478 reset_page_mapcount(page);
481 static void free_zspage(struct page *first_page)
483 struct page *nextp, *tmp, *head_extra;
485 BUG_ON(!is_first_page(first_page));
486 BUG_ON(first_page->inuse);
488 head_extra = (struct page *)page_private(first_page);
490 reset_page(first_page);
491 __free_page(first_page);
493 /* zspage with only 1 system page */
497 list_for_each_entry_safe(nextp, tmp, &head_extra->lru, lru) {
498 list_del(&nextp->lru);
502 reset_page(head_extra);
503 __free_page(head_extra);
506 /* Initialize a newly allocated zspage */
507 static void init_zspage(struct page *first_page, struct size_class *class)
509 unsigned long off = 0;
510 struct page *page = first_page;
512 BUG_ON(!is_first_page(first_page));
514 struct page *next_page;
515 struct link_free *link;
516 unsigned int i, objs_on_page;
519 * page->index stores offset of first object starting
520 * in the page. For the first page, this is always 0,
521 * so we use first_page->index (aka ->freelist) to store
522 * head of corresponding zspage's freelist.
524 if (page != first_page)
527 link = (struct link_free *)kmap_atomic(page) +
529 objs_on_page = (PAGE_SIZE - off) / class->size;
531 for (i = 1; i <= objs_on_page; i++) {
533 if (off < PAGE_SIZE) {
534 link->next = obj_location_to_handle(page, i);
535 link += class->size / sizeof(*link);
540 * We now come to the last (full or partial) object on this
541 * page, which must point to the first object on the next
544 next_page = get_next_page(page);
545 link->next = obj_location_to_handle(next_page, 0);
548 off = (off + class->size) % PAGE_SIZE;
553 * Allocate a zspage for the given size class
555 static struct page *alloc_zspage(struct size_class *class, gfp_t flags)
558 struct page *first_page = NULL, *uninitialized_var(prev_page);
561 * Allocate individual pages and link them together as:
562 * 1. first page->private = first sub-page
563 * 2. all sub-pages are linked together using page->lru
564 * 3. each sub-page is linked to the first page using page->first_page
566 * For each size class, First/Head pages are linked together using
567 * page->lru. Also, we set PG_private to identify the first page
568 * (i.e. no other sub-page has this flag set) and PG_private_2 to
569 * identify the last page.
572 for (i = 0; i < class->pages_per_zspage; i++) {
575 page = alloc_page(flags);
579 INIT_LIST_HEAD(&page->lru);
580 if (i == 0) { /* first page */
581 SetPagePrivate(page);
582 set_page_private(page, 0);
584 first_page->inuse = 0;
587 first_page->private = (unsigned long)page;
589 page->first_page = first_page;
591 list_add(&page->lru, &prev_page->lru);
592 if (i == class->pages_per_zspage - 1) /* last page */
593 SetPagePrivate2(page);
597 init_zspage(first_page, class);
599 first_page->freelist = obj_location_to_handle(first_page, 0);
600 /* Maximum number of objects we can store in this zspage */
601 first_page->objects = class->pages_per_zspage * PAGE_SIZE / class->size;
603 error = 0; /* Success */
606 if (unlikely(error) && first_page) {
607 free_zspage(first_page);
614 static struct page *find_get_zspage(struct size_class *class)
619 for (i = 0; i < _ZS_NR_FULLNESS_GROUPS; i++) {
620 page = class->fullness_list[i];
628 #ifdef USE_PGTABLE_MAPPING
629 static inline int __zs_cpu_up(struct mapping_area *area)
632 * Make sure we don't leak memory if a cpu UP notification
633 * and zs_init() race and both call zs_cpu_up() on the same cpu
637 area->vm = alloc_vm_area(PAGE_SIZE * 2, NULL);
643 static inline void __zs_cpu_down(struct mapping_area *area)
646 free_vm_area(area->vm);
650 static inline void *__zs_map_object(struct mapping_area *area,
651 struct page *pages[2], int off, int size)
653 BUG_ON(map_vm_area(area->vm, PAGE_KERNEL, &pages));
654 area->vm_addr = area->vm->addr;
655 return area->vm_addr + off;
658 static inline void __zs_unmap_object(struct mapping_area *area,
659 struct page *pages[2], int off, int size)
661 unsigned long addr = (unsigned long)area->vm_addr;
662 unsigned long end = addr + (PAGE_SIZE * 2);
664 flush_cache_vunmap(addr, end);
665 unmap_kernel_range_noflush(addr, PAGE_SIZE * 2);
666 local_flush_tlb_kernel_range(addr, end);
669 #else /* USE_PGTABLE_MAPPING */
671 static inline int __zs_cpu_up(struct mapping_area *area)
674 * Make sure we don't leak memory if a cpu UP notification
675 * and zs_init() race and both call zs_cpu_up() on the same cpu
679 area->vm_buf = (char *)__get_free_page(GFP_KERNEL);
685 static inline void __zs_cpu_down(struct mapping_area *area)
688 free_page((unsigned long)area->vm_buf);
692 static void *__zs_map_object(struct mapping_area *area,
693 struct page *pages[2], int off, int size)
697 char *buf = area->vm_buf;
699 /* disable page faults to match kmap_atomic() return conditions */
702 /* no read fastpath */
703 if (area->vm_mm == ZS_MM_WO)
706 sizes[0] = PAGE_SIZE - off;
707 sizes[1] = size - sizes[0];
709 /* copy object to per-cpu buffer */
710 addr = kmap_atomic(pages[0]);
711 memcpy(buf, addr + off, sizes[0]);
713 addr = kmap_atomic(pages[1]);
714 memcpy(buf + sizes[0], addr, sizes[1]);
720 static void __zs_unmap_object(struct mapping_area *area,
721 struct page *pages[2], int off, int size)
725 char *buf = area->vm_buf;
727 /* no write fastpath */
728 if (area->vm_mm == ZS_MM_RO)
731 sizes[0] = PAGE_SIZE - off;
732 sizes[1] = size - sizes[0];
734 /* copy per-cpu buffer to object */
735 addr = kmap_atomic(pages[0]);
736 memcpy(addr + off, buf, sizes[0]);
738 addr = kmap_atomic(pages[1]);
739 memcpy(addr, buf + sizes[0], sizes[1]);
743 /* enable page faults to match kunmap_atomic() return conditions */
747 #endif /* USE_PGTABLE_MAPPING */
749 static int zs_cpu_notifier(struct notifier_block *nb, unsigned long action,
752 int ret, cpu = (long)pcpu;
753 struct mapping_area *area;
757 area = &per_cpu(zs_map_area, cpu);
758 ret = __zs_cpu_up(area);
760 return notifier_from_errno(ret);
763 case CPU_UP_CANCELED:
764 area = &per_cpu(zs_map_area, cpu);
772 static struct notifier_block zs_cpu_nb = {
773 .notifier_call = zs_cpu_notifier
776 static void zs_exit(void)
780 for_each_online_cpu(cpu)
781 zs_cpu_notifier(NULL, CPU_DEAD, (void *)(long)cpu);
782 unregister_cpu_notifier(&zs_cpu_nb);
785 static int zs_init(void)
789 register_cpu_notifier(&zs_cpu_nb);
790 for_each_online_cpu(cpu) {
791 ret = zs_cpu_notifier(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
792 if (notifier_to_errno(ret))
798 return notifier_to_errno(ret);
801 struct zs_pool *zs_create_pool(const char *name, gfp_t flags)
804 struct zs_pool *pool;
809 ovhd_size = roundup(sizeof(*pool), PAGE_SIZE);
810 pool = kzalloc(ovhd_size, GFP_KERNEL);
814 for (i = 0; i < ZS_SIZE_CLASSES; i++) {
816 struct size_class *class;
818 size = ZS_MIN_ALLOC_SIZE + i * ZS_SIZE_CLASS_DELTA;
819 if (size > ZS_MAX_ALLOC_SIZE)
820 size = ZS_MAX_ALLOC_SIZE;
822 class = &pool->size_class[i];
825 spin_lock_init(&class->lock);
826 class->pages_per_zspage = get_pages_per_zspage(size);
835 EXPORT_SYMBOL_GPL(zs_create_pool);
837 void zs_destroy_pool(struct zs_pool *pool)
841 for (i = 0; i < ZS_SIZE_CLASSES; i++) {
843 struct size_class *class = &pool->size_class[i];
845 for (fg = 0; fg < _ZS_NR_FULLNESS_GROUPS; fg++) {
846 if (class->fullness_list[fg]) {
847 pr_info("Freeing non-empty class with size "
848 "%db, fullness group %d\n",
855 EXPORT_SYMBOL_GPL(zs_destroy_pool);
858 * zs_malloc - Allocate block of given size from pool.
859 * @pool: pool to allocate from
860 * @size: size of block to allocate
862 * On success, handle to the allocated object is returned,
864 * Allocation requests with size > ZS_MAX_ALLOC_SIZE will fail.
866 unsigned long zs_malloc(struct zs_pool *pool, size_t size)
869 struct link_free *link;
871 struct size_class *class;
873 struct page *first_page, *m_page;
874 unsigned long m_objidx, m_offset;
876 if (unlikely(!size || size > ZS_MAX_ALLOC_SIZE))
879 class_idx = get_size_class_index(size);
880 class = &pool->size_class[class_idx];
881 BUG_ON(class_idx != class->index);
883 spin_lock(&class->lock);
884 first_page = find_get_zspage(class);
887 spin_unlock(&class->lock);
888 first_page = alloc_zspage(class, pool->flags);
889 if (unlikely(!first_page))
892 set_zspage_mapping(first_page, class->index, ZS_EMPTY);
893 spin_lock(&class->lock);
894 class->pages_allocated += class->pages_per_zspage;
897 obj = (unsigned long)first_page->freelist;
898 obj_handle_to_location(obj, &m_page, &m_objidx);
899 m_offset = obj_idx_to_offset(m_page, m_objidx, class->size);
901 link = (struct link_free *)kmap_atomic(m_page) +
902 m_offset / sizeof(*link);
903 first_page->freelist = link->next;
904 memset(link, POISON_INUSE, sizeof(*link));
908 /* Now move the zspage to another fullness group, if required */
909 fix_fullness_group(pool, first_page);
910 spin_unlock(&class->lock);
914 EXPORT_SYMBOL_GPL(zs_malloc);
916 void zs_free(struct zs_pool *pool, unsigned long obj)
918 struct link_free *link;
919 struct page *first_page, *f_page;
920 unsigned long f_objidx, f_offset;
923 struct size_class *class;
924 enum fullness_group fullness;
929 obj_handle_to_location(obj, &f_page, &f_objidx);
930 first_page = get_first_page(f_page);
932 get_zspage_mapping(first_page, &class_idx, &fullness);
933 class = &pool->size_class[class_idx];
934 f_offset = obj_idx_to_offset(f_page, f_objidx, class->size);
936 spin_lock(&class->lock);
938 /* Insert this object in containing zspage's freelist */
939 link = (struct link_free *)((unsigned char *)kmap_atomic(f_page)
941 link->next = first_page->freelist;
943 first_page->freelist = (void *)obj;
946 fullness = fix_fullness_group(pool, first_page);
948 if (fullness == ZS_EMPTY)
949 class->pages_allocated -= class->pages_per_zspage;
951 spin_unlock(&class->lock);
953 if (fullness == ZS_EMPTY)
954 free_zspage(first_page);
956 EXPORT_SYMBOL_GPL(zs_free);
959 * zs_map_object - get address of allocated object from handle.
960 * @pool: pool from which the object was allocated
961 * @handle: handle returned from zs_malloc
963 * Before using an object allocated from zs_malloc, it must be mapped using
964 * this function. When done with the object, it must be unmapped using
967 * Only one object can be mapped per cpu at a time. There is no protection
968 * against nested mappings.
970 * This function returns with preemption and page faults disabled.
972 void *zs_map_object(struct zs_pool *pool, unsigned long handle,
976 unsigned long obj_idx, off;
978 unsigned int class_idx;
979 enum fullness_group fg;
980 struct size_class *class;
981 struct mapping_area *area;
982 struct page *pages[2];
987 * Because we use per-cpu mapping areas shared among the
988 * pools/users, we can't allow mapping in interrupt context
989 * because it can corrupt another users mappings.
991 BUG_ON(in_interrupt());
993 obj_handle_to_location(handle, &page, &obj_idx);
994 get_zspage_mapping(get_first_page(page), &class_idx, &fg);
995 class = &pool->size_class[class_idx];
996 off = obj_idx_to_offset(page, obj_idx, class->size);
998 area = &get_cpu_var(zs_map_area);
1000 if (off + class->size <= PAGE_SIZE) {
1001 /* this object is contained entirely within a page */
1002 area->vm_addr = kmap_atomic(page);
1003 return area->vm_addr + off;
1006 /* this object spans two pages */
1008 pages[1] = get_next_page(page);
1011 return __zs_map_object(area, pages, off, class->size);
1013 EXPORT_SYMBOL_GPL(zs_map_object);
1015 void zs_unmap_object(struct zs_pool *pool, unsigned long handle)
1018 unsigned long obj_idx, off;
1020 unsigned int class_idx;
1021 enum fullness_group fg;
1022 struct size_class *class;
1023 struct mapping_area *area;
1027 obj_handle_to_location(handle, &page, &obj_idx);
1028 get_zspage_mapping(get_first_page(page), &class_idx, &fg);
1029 class = &pool->size_class[class_idx];
1030 off = obj_idx_to_offset(page, obj_idx, class->size);
1032 area = &__get_cpu_var(zs_map_area);
1033 if (off + class->size <= PAGE_SIZE)
1034 kunmap_atomic(area->vm_addr);
1036 struct page *pages[2];
1039 pages[1] = get_next_page(page);
1042 __zs_unmap_object(area, pages, off, class->size);
1044 put_cpu_var(zs_map_area);
1046 EXPORT_SYMBOL_GPL(zs_unmap_object);
1048 u64 zs_get_total_size_bytes(struct zs_pool *pool)
1053 for (i = 0; i < ZS_SIZE_CLASSES; i++)
1054 npages += pool->size_class[i].pages_allocated;
1056 return npages << PAGE_SHIFT;
1058 EXPORT_SYMBOL_GPL(zs_get_total_size_bytes);
1060 module_init(zs_init);
1061 module_exit(zs_exit);
1063 MODULE_LICENSE("Dual BSD/GPL");
1064 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");